From 6e9b7d5f00ae822dc89c1573a6df0214cf164442 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 4 May 2026 09:54:25 -0600 Subject: [PATCH 01/35] refactor(db): composite PK on M2M association tables (sc-105349) Replace synthetic id INTEGER PRIMARY KEY with composite PRIMARY KEY (fk1, fk2) on the eight pure-junction tables: dashboard_roles, dashboard_slices, dashboard_user, report_schedule_user, rls_filter_roles, rls_filter_tables, slice_user, sqlatable_user. The redundant UNIQUE(fk1, fk2) on dashboard_slices and report_schedule_user is dropped (subsumed by the new PK). Migration handles dialect quirks: copy_from for tables with pre-existing UNIQUE (so SQLite's anonymous-constraint reflection doesn't matter), wrapped- subquery dedupe for MySQL (ERROR 1093), sa.Identity(always=False) on downgrade to backfill the restored id column without NOT NULL violations, and distinct PK names per direction (pk_ on upgrade,
_pkey on downgrade) to avoid round-trip index-name collisions on Postgres. ORM Table() definitions updated to match. UPDATING.md entry added with operator runbook (BI-tool impact, pre-flight inventory queries, dedupe-row- loss notice, pg_dump workaround, FK-NOT-NULL downgrade asymmetry note). Tests: 8 schema-shape assertions (post-upgrade), 8 duplicate-rejection unit tests, 8 distinct-pair sanity tests, 1 round-trip + idempotency test (in-memory SQLite via Alembic MigrationContext). Continuum-restore verification against the new shape is out of scope for this PR; it is the responsibility of the versioning epic (sc-103156). Co-Authored-By: Claude Opus 4.7 (1M context) --- UPDATING.md | 47 +++ superset/connectors/sqla/models.py | 35 ++- ...3611e32_composite_pk_association_tables.py | 289 ++++++++++++++++++ superset/models/dashboard.py | 37 ++- superset/models/slice.py | 15 +- superset/reports/models.py | 6 +- .../composite_pk_association_tables__tests.py | 131 ++++++++ .../composite_pk_round_trip__tests.py | 168 ++++++++++ .../composite_pk_association_tables_test.py | 132 ++++++++ 9 files changed, 833 insertions(+), 27 deletions(-) create mode 100644 superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py create mode 100644 tests/integration_tests/migrations/composite_pk_association_tables__tests.py create mode 100644 tests/integration_tests/migrations/composite_pk_round_trip__tests.py create mode 100644 tests/unit_tests/migrations/composite_pk_association_tables_test.py diff --git a/UPDATING.md b/UPDATING.md index 8825c5d6573c..895e0d86c7d0 100644 --- a/UPDATING.md +++ b/UPDATING.md @@ -487,6 +487,53 @@ See `superset/mcp_service/PRODUCTION.md` for deployment guides. } ``` +### Composite primary keys on many-to-many association tables + +The eight M:N association tables listed below have been changed from a synthetic surrogate `id INTEGER PRIMARY KEY` to a composite `PRIMARY KEY (fk1, fk2)` on the two foreign-key columns. The `id` column is dropped, and the two tables that previously carried a redundant `UNIQUE (fk1, fk2)` constraint have that constraint removed (it is now subsumed by the composite primary key). + +**Affected tables and their composite-PK column pairs:** + +| Table | Composite PK | +|---|---| +| `dashboard_roles` | `(dashboard_id, role_id)` | +| `dashboard_slices` | `(dashboard_id, slice_id)` | +| `dashboard_user` | `(user_id, dashboard_id)` | +| `report_schedule_user` | `(user_id, report_schedule_id)` | +| `rls_filter_roles` | `(role_id, rls_filter_id)` | +| `rls_filter_tables` | `(table_id, rls_filter_id)` | +| `slice_user` | `(user_id, slice_id)` | +| `sqlatable_user` | `(user_id, table_id)` | + +**Impact on external readers:** Any BI tool, custom report, backup script, or external integration that references these tables by their old surrogate `id` column (e.g., `SELECT id FROM dashboard_slices WHERE …`, `WHERE dashboard_slices.id IN (…)`) will break. Update such queries to project or filter on the FK pair (`dashboard_id, slice_id`) instead. The FK columns themselves are unchanged. + +**Pre-flight inventory queries.** Before applying the upgrade, operators are encouraged to run the queries below against their database to assess what the migration will change. Two classes of pre-existing data are not preserved by the migration: duplicate `(fk1, fk2)` rows (the migration keeps `MIN(id)` and deletes the rest) and rows with `NULL` in either FK column (the migration deletes them, since FK columns are promoted to `NOT NULL` for the composite PK). Compliance- or audit-sensitive operators should also `\copy` (Postgres) or `SELECT … INTO OUTFILE` (MySQL) the affected rows for their own records before upgrading. + +```sql +-- Duplicate (fk1, fk2) pairs (the migration will keep MIN(id) per group, delete the rest) +SELECT dashboard_id, role_id, COUNT(*) FROM dashboard_roles GROUP BY dashboard_id, role_id HAVING COUNT(*) > 1; +SELECT dashboard_id, slice_id, COUNT(*) FROM dashboard_slices GROUP BY dashboard_id, slice_id HAVING COUNT(*) > 1; +SELECT user_id, dashboard_id, COUNT(*) FROM dashboard_user GROUP BY user_id, dashboard_id HAVING COUNT(*) > 1; +SELECT user_id, report_schedule_id, COUNT(*) FROM report_schedule_user GROUP BY user_id, report_schedule_id HAVING COUNT(*) > 1; +SELECT role_id, rls_filter_id, COUNT(*) FROM rls_filter_roles GROUP BY role_id, rls_filter_id HAVING COUNT(*) > 1; +SELECT table_id, rls_filter_id, COUNT(*) FROM rls_filter_tables GROUP BY table_id, rls_filter_id HAVING COUNT(*) > 1; +SELECT user_id, slice_id, COUNT(*) FROM slice_user GROUP BY user_id, slice_id HAVING COUNT(*) > 1; +SELECT user_id, table_id, COUNT(*) FROM sqlatable_user GROUP BY user_id, table_id HAVING COUNT(*) > 1; + +-- Rows with a NULL in either FK (the migration will delete these) +SELECT COUNT(*) FROM dashboard_roles WHERE dashboard_id IS NULL OR role_id IS NULL; +SELECT COUNT(*) FROM dashboard_slices WHERE dashboard_id IS NULL OR slice_id IS NULL; +SELECT COUNT(*) FROM dashboard_user WHERE user_id IS NULL OR dashboard_id IS NULL; +SELECT COUNT(*) FROM report_schedule_user WHERE user_id IS NULL OR report_schedule_id IS NULL; +SELECT COUNT(*) FROM rls_filter_roles WHERE role_id IS NULL OR rls_filter_id IS NULL; +SELECT COUNT(*) FROM rls_filter_tables WHERE table_id IS NULL OR rls_filter_id IS NULL; +SELECT COUNT(*) FROM slice_user WHERE user_id IS NULL OR slice_id IS NULL; +SELECT COUNT(*) FROM sqlatable_user WHERE user_id IS NULL OR table_id IS NULL; +``` + +**Restoring an old `pg_dump` (or equivalent) against the new schema.** A dump taken before the migration includes `INSERT` statements that populate the now-removed `id` column. Restoring such a dump against the post-migration schema will fail. The supported workaround is to dump only the schema and reference data, then re-create the M:N associations from application data after restore — for example with `pg_dump --exclude-table-data` (or per-table `--exclude-table-data=dashboard_slices` etc.) for the eight junction tables, restore the rest, then run a one-shot script that re-INSERTs `(fk1, fk2)` pairs derived from your application export. Operators who need to restore an old dump verbatim should restore against a pre-migration Superset and then re-run the upgrade. + +**Intentional downgrade asymmetry.** The migration's `downgrade()` restores the surrogate `id` column and (for `dashboard_slices` and `report_schedule_user`) the original `UNIQUE (fk1, fk2)` constraint, but it does **not** restore the original `NULL`-allowed state on the FK columns — they remain `NOT NULL`. This is intentional: under SQLAlchemy's `secondary=` semantics, a `NULL` in either FK column of a junction table is meaningless (it cannot participate in either side of the relationship). Operators downgrading are not expected to need this restored. The asymmetry is documented for completeness so that round-trip schema diffs are not mistaken for migration bugs. + ## 6.0.0 - [33055](https://github.com/apache/superset/pull/33055): Upgrades Flask-AppBuilder to 5.0.0. The AUTH_OID authentication type has been deprecated and is no longer available as an option in Flask-AppBuilder. OpenID (OID) is considered a deprecated authentication protocol - if you are using AUTH_OID, you will need to migrate to an alternative authentication method such as OAuth, LDAP, or database authentication before upgrading. - [34871](https://github.com/apache/superset/pull/34871): Fixed Jest test hanging issue from Ant Design v5 upgrade. MessageChannel is now mocked in test environment to prevent rc-overflow from causing Jest to hang. Test environment only - no production impact. diff --git a/superset/connectors/sqla/models.py b/superset/connectors/sqla/models.py index 81b5c2fc8d9d..ca8bada53804 100644 --- a/superset/connectors/sqla/models.py +++ b/superset/connectors/sqla/models.py @@ -1285,9 +1285,18 @@ def data(self) -> dict[str, Any]: sqlatable_user = DBTable( "sqlatable_user", metadata, - Column("id", Integer, primary_key=True), - Column("user_id", Integer, ForeignKey("ab_user.id", ondelete="CASCADE")), - Column("table_id", Integer, ForeignKey("tables.id", ondelete="CASCADE")), + Column( + "user_id", + Integer, + ForeignKey("ab_user.id", ondelete="CASCADE"), + primary_key=True, + ), + Column( + "table_id", + Integer, + ForeignKey("tables.id", ondelete="CASCADE"), + primary_key=True, + ), ) @@ -2218,17 +2227,25 @@ def text(self, clause: str) -> TextClause: RLSFilterRoles = DBTable( "rls_filter_roles", metadata, - Column("id", Integer, primary_key=True), - Column("role_id", Integer, ForeignKey("ab_role.id"), nullable=False), - Column("rls_filter_id", Integer, ForeignKey("row_level_security_filters.id")), + Column("role_id", Integer, ForeignKey("ab_role.id"), primary_key=True), + Column( + "rls_filter_id", + Integer, + ForeignKey("row_level_security_filters.id"), + primary_key=True, + ), ) RLSFilterTables = DBTable( "rls_filter_tables", metadata, - Column("id", Integer, primary_key=True), - Column("table_id", Integer, ForeignKey("tables.id")), - Column("rls_filter_id", Integer, ForeignKey("row_level_security_filters.id")), + Column("table_id", Integer, ForeignKey("tables.id"), primary_key=True), + Column( + "rls_filter_id", + Integer, + ForeignKey("row_level_security_filters.id"), + primary_key=True, + ), ) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py new file mode 100644 index 000000000000..2c841bc6171a --- /dev/null +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -0,0 +1,289 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""composite_pk_association_tables + +Replace the unused synthetic ``id INTEGER PRIMARY KEY`` on eight many-to-many +association tables with a composite primary key on the two FK columns. Drops +the now-redundant ``UniqueConstraint(fk1, fk2)`` on the two tables that +already carry one. Pre-flight: deletes rows with NULL FK values (six tables +allow them today) and any duplicate ``(fk1, fk2)`` rows. + +Motivated by SQLAlchemy-Continuum issue #129 (M2M restore against junction +tables with surrogate PKs); also closes the data-integrity hole where six +of the eight tables lacked DB-level uniqueness. + +Revision ID: 2bee73611e32 +Revises: ce6bd21901ab +Create Date: 2026-05-01 23:36:34.050058 + +""" + +import logging +from typing import NamedTuple + +import sqlalchemy as sa +from alembic import op +from sqlalchemy import inspect +from sqlalchemy.engine import Connection + +# revision identifiers, used by Alembic. +revision = "2bee73611e32" +down_revision = "ce6bd21901ab" + +logger = logging.getLogger("alembic.env") + + +class AssociationTable(NamedTuple): + """A junction table being converted from surrogate-id PK to composite-FK PK.""" + + name: str + fk1: str + fk2: str + + +# Order is alphabetical by table name; deterministic for review and bisection. +AFFECTED_TABLES: list[AssociationTable] = [ + AssociationTable("dashboard_roles", "dashboard_id", "role_id"), + AssociationTable("dashboard_slices", "dashboard_id", "slice_id"), + AssociationTable("dashboard_user", "user_id", "dashboard_id"), + AssociationTable("report_schedule_user", "user_id", "report_schedule_id"), + AssociationTable("rls_filter_roles", "role_id", "rls_filter_id"), + AssociationTable("rls_filter_tables", "table_id", "rls_filter_id"), + AssociationTable("slice_user", "user_id", "slice_id"), + AssociationTable("sqlatable_user", "user_id", "table_id"), +] + +# These two tables already declare ``UniqueConstraint(fk1, fk2)`` in the model; +# the composite PK subsumes it, so the migration drops the redundant constraint. +TABLES_WITH_PRE_EXISTING_UNIQUE: set[str] = { + "dashboard_slices", + "report_schedule_user", +} + +# Six tables whose FK columns are nullable today. Promoting an FK to a primary +# key column makes it NOT NULL, so any existing NULL-FK rows would block the +# PK-add. We delete them in pre-flight (a junction-table row with a NULL FK +# is meaningless under SQLAlchemy ``secondary=`` semantics anyway). +TABLES_WITH_NULLABLE_FKS: set[str] = { + "dashboard_slices", + "dashboard_user", + "rls_filter_roles", + "rls_filter_tables", + "slice_user", + "sqlatable_user", +} + + +def _check_no_external_fks_to_id(conn: Connection) -> None: + """Raise ``RuntimeError`` if any foreign key in the database references one + of the eight junction-table ``id`` columns. Uses SQLAlchemy's ``Inspector`` + for dialect-agnostic introspection across PostgreSQL, MySQL, and SQLite.""" + affected = {t.name for t in AFFECTED_TABLES} + insp = inspect(conn) + for table_name in insp.get_table_names(): + if table_name in affected: + continue + for fk in insp.get_foreign_keys(table_name): + if fk["referred_table"] in affected and "id" in fk["referred_columns"]: + raise RuntimeError( + f"Cannot drop synthetic id from {fk['referred_table']}: " + f"external FK {fk.get('name', '')} on {table_name} " + f"references {fk['referred_table']}({fk['referred_columns']}). " + f"Drop or migrate the referencing FK before applying this " + f"migration." + ) + + +def _delete_null_fk_rows(conn: Connection, t: AssociationTable) -> int: + """Delete rows where ``t.fk1`` or ``t.fk2`` is NULL on ``t.name``. + + Returns the deletion count. Called only on tables in + ``TABLES_WITH_NULLABLE_FKS``. Required because primary-key columns must be + NOT NULL; the PK-add downstream would fail with a cryptic constraint + violation if any NULL-FK rows survived. + """ + # Identifiers come from the AFFECTED_TABLES whitelist, not user input. + sql = sa.text( + f"DELETE FROM {t.name} WHERE {t.fk1} IS NULL OR {t.fk2} IS NULL" # noqa: S608 + ) + result = conn.execute(sql) + n = result.rowcount or 0 + if n: + logger.warning( + "Deleted %d row(s) with NULL FK from %s before composite-PK promotion", + n, + t.name, + ) + return n + + +def _dedupe_by_min_id(conn: Connection, t: AssociationTable) -> int: + """Delete duplicate ``(t.fk1, t.fk2)`` rows from ``t.name`` keeping ``MIN(id)``. + + Returns the deletion count. Uses the wrapped-subquery form for MySQL + portability — MySQL rejects ``DELETE FROM t WHERE id NOT IN (SELECT MIN(id) + FROM t GROUP BY ...)`` with ERROR 1093 unless the inner SELECT is wrapped + to force materialization. + """ + # Identifiers come from the AFFECTED_TABLES whitelist, not user input. + sql = sa.text( + f"DELETE FROM {t.name} WHERE id NOT IN (" # noqa: S608 + f" SELECT keep_id FROM (" + f" SELECT MIN(id) AS keep_id FROM {t.name} " + f"GROUP BY {t.fk1}, {t.fk2}" + f" ) AS s" + f")" + ) + result = conn.execute(sql) + n = result.rowcount or 0 + if n: + logger.warning("Deduped %d duplicate row(s) from %s", n, t.name) + return n + + +def _assert_no_duplicates(conn: Connection, t: AssociationTable) -> None: + """Raise ``RuntimeError`` if any ``(t.fk1, t.fk2)`` duplicate group remains. + + Called after ``_dedupe_by_min_id`` to surface silent dialect-dependent + dedupe failures (e.g., a MySQL syntax issue) as an actionable error + before the PK-add fires with a less-helpful constraint-violation message. + """ + # Identifiers come from the AFFECTED_TABLES whitelist, not user input. + sql = sa.text( + f"SELECT COUNT(*) FROM (" # noqa: S608 + f" SELECT 1 FROM {t.name} GROUP BY {t.fk1}, {t.fk2} HAVING COUNT(*) > 1" + f") AS s" + ) + if remaining := conn.scalar(sql) or 0: + raise RuntimeError( + f"Dedupe failed for {t.name}: {remaining} duplicate " + f"({t.fk1}, {t.fk2}) groups remain after _dedupe_by_min_id. " + f"Check the dedupe SQL for dialect {conn.dialect.name}." + ) + + +def _build_pre_upgrade_table( + insp: sa.engine.reflection.Inspector, t: AssociationTable +) -> sa.Table: + """Build a ``Table`` object representing the pre-upgrade schema of ``t``, + explicitly *without* any redundant ``UniqueConstraint(t.fk1, t.fk2)``. + Used as ``copy_from`` to ``batch_alter_table`` so the rebuilt table + omits the unnamed UNIQUE constraint deterministically across dialects + (SQLite reflects unnamed UNIQUEs with ``name=None``, defeating the + standard ``batch_op.drop_constraint(name)`` path). + + Reflects column types and FK targets (with original FK constraint names + preserved) from the live database; only the redundant UNIQUE is omitted. + """ + md = sa.MetaData() + fks_for_col: dict[str, list[dict]] = {} + for fk in insp.get_foreign_keys(t.name): + for col_name in fk["constrained_columns"]: + fks_for_col.setdefault(col_name, []).append(fk) + + cols: list[sa.Column] = [] + for c in insp.get_columns(t.name): + col_kwargs = {"nullable": c.get("nullable", True)} + if c["name"] == "id": + col_kwargs["primary_key"] = True + col_kwargs["autoincrement"] = True + fk_args = [] + for fk in fks_for_col.get(c["name"], []): + idx = fk["constrained_columns"].index(c["name"]) + target = f"{fk['referred_table']}.{fk['referred_columns'][idx]}" + options = {} + if fk.get("options", {}).get("ondelete"): + options["ondelete"] = fk["options"]["ondelete"] + if fk.get("name"): + options["name"] = fk["name"] + fk_args.append(sa.ForeignKey(target, **options)) + cols.append(sa.Column(c["name"], c["type"], *fk_args, **col_kwargs)) + return sa.Table(t.name, md, *cols) + + +def upgrade() -> None: + conn = op.get_bind() + _check_no_external_fks_to_id(conn) + insp = inspect(conn) + + for t in AFFECTED_TABLES: + if t.name in TABLES_WITH_NULLABLE_FKS: + _delete_null_fk_rows(conn, t) + _dedupe_by_min_id(conn, t) + _assert_no_duplicates(conn, t) + + # For the two tables with a pre-existing redundant UNIQUE + # (``dashboard_slices``, ``report_schedule_user``) build an explicit + # ``copy_from`` Table that omits the UNIQUE; this deterministically + # drops it across all dialects, including SQLite where unnamed + # constraints reflect with ``name=None`` and can't be dropped by + # name. For the other six tables, reflection-based default + # ``batch_alter_table`` (auto-detect) is fine since there's no + # UNIQUE to drop. On PostgreSQL/MySQL, direct ALTER avoids the + # temp-table index-name collision; on SQLite, the auto-detect picks + # ``recreate=True`` because PK changes need it. + if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: + with op.batch_alter_table( + t.name, + recreate="always", + copy_from=_build_pre_upgrade_table(insp, t), + ) as batch_op: + batch_op.drop_column("id") + batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) + else: + with op.batch_alter_table(t.name) as batch_op: + batch_op.drop_column("id") + batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) + + +def downgrade() -> None: + # Inverse order: undo upgrade transformations from last-applied to + # first-applied. Within each table, drop the composite PK, restore the + # surrogate ``id`` column, and re-add the original ``UNIQUE`` constraint + # on the two tables that previously carried one. + # + # Note: FK columns remain NOT NULL after downgrade (intentional asymmetry + # — see UPDATING.md). Restoring the original nullable state would require + # an explicit ``alter_column`` per FK per table for no operator value; + # junction-table NULL FKs were always meaningless under ``secondary=`` + # semantics. + # The downgrade names the restored PK ``
_pkey`` (matching Postgres' + # default constraint-naming convention, which was the original constraint + # name before this migration ran) so a downgrade-then-upgrade round-trip + # doesn't collide on the upgrade's ``pk_
`` name. + # + # Adding a NOT NULL ``id`` column to a table with existing rows requires + # a default that fires on the existing rows. ``sa.Identity()`` (Postgres + # 10+ / MySQL 8+) and ``sa.Sequence`` (with explicit nextval) both + # backfill existing rows during ALTER TABLE; bare ``autoincrement=True`` + # does not. ``Identity`` is the modern portable choice. + for t in reversed(AFFECTED_TABLES): + with op.batch_alter_table(t.name) as batch_op: + batch_op.drop_constraint(f"pk_{t.name}", type_="primary") + batch_op.add_column( + sa.Column( + "id", + sa.Integer, + sa.Identity(always=False), + nullable=False, + ) + ) + batch_op.create_primary_key(f"{t.name}_pkey", ["id"]) + if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: + batch_op.create_unique_constraint( + f"uq_{t.name}_{t.fk1}_{t.fk2}", [t.fk1, t.fk2] + ) diff --git a/superset/models/dashboard.py b/superset/models/dashboard.py index 4653272fcbf3..559ff273194d 100644 --- a/superset/models/dashboard.py +++ b/superset/models/dashboard.py @@ -35,7 +35,6 @@ String, Table, Text, - UniqueConstraint, ) from sqlalchemy.engine.base import Connection from sqlalchemy.orm import relationship, subqueryload @@ -93,37 +92,53 @@ def copy_dashboard(_mapper: Mapper, _connection: Connection, target: Dashboard) dashboard_slices = Table( "dashboard_slices", metadata, - Column("id", Integer, primary_key=True), - Column("dashboard_id", Integer, ForeignKey("dashboards.id", ondelete="CASCADE")), - Column("slice_id", Integer, ForeignKey("slices.id", ondelete="CASCADE")), - UniqueConstraint("dashboard_id", "slice_id"), + Column( + "dashboard_id", + Integer, + ForeignKey("dashboards.id", ondelete="CASCADE"), + primary_key=True, + ), + Column( + "slice_id", + Integer, + ForeignKey("slices.id", ondelete="CASCADE"), + primary_key=True, + ), ) dashboard_user = Table( "dashboard_user", metadata, - Column("id", Integer, primary_key=True), - Column("user_id", Integer, ForeignKey("ab_user.id", ondelete="CASCADE")), - Column("dashboard_id", Integer, ForeignKey("dashboards.id", ondelete="CASCADE")), + Column( + "user_id", + Integer, + ForeignKey("ab_user.id", ondelete="CASCADE"), + primary_key=True, + ), + Column( + "dashboard_id", + Integer, + ForeignKey("dashboards.id", ondelete="CASCADE"), + primary_key=True, + ), ) DashboardRoles = Table( "dashboard_roles", metadata, - Column("id", Integer, primary_key=True), Column( "dashboard_id", Integer, ForeignKey("dashboards.id", ondelete="CASCADE"), - nullable=False, + primary_key=True, ), Column( "role_id", Integer, ForeignKey("ab_role.id", ondelete="CASCADE"), - nullable=False, + primary_key=True, ), ) diff --git a/superset/models/slice.py b/superset/models/slice.py index 40de049df2ab..d1a954791f1d 100644 --- a/superset/models/slice.py +++ b/superset/models/slice.py @@ -59,9 +59,18 @@ slice_user = Table( "slice_user", metadata, - Column("id", Integer, primary_key=True), - Column("user_id", Integer, ForeignKey("ab_user.id", ondelete="CASCADE")), - Column("slice_id", Integer, ForeignKey("slices.id", ondelete="CASCADE")), + Column( + "user_id", + Integer, + ForeignKey("ab_user.id", ondelete="CASCADE"), + primary_key=True, + ), + Column( + "slice_id", + Integer, + ForeignKey("slices.id", ondelete="CASCADE"), + primary_key=True, + ), ) logger = logging.getLogger(__name__) diff --git a/superset/reports/models.py b/superset/reports/models.py index f0abda8a9216..7564336ae11d 100644 --- a/superset/reports/models.py +++ b/superset/reports/models.py @@ -101,20 +101,18 @@ class ReportSourceFormat(StrEnum): report_schedule_user = Table( "report_schedule_user", metadata, - Column("id", Integer, primary_key=True), Column( "user_id", Integer, ForeignKey("ab_user.id", ondelete="CASCADE"), - nullable=False, + primary_key=True, ), Column( "report_schedule_id", Integer, ForeignKey("report_schedule.id", ondelete="CASCADE"), - nullable=False, + primary_key=True, ), - UniqueConstraint("user_id", "report_schedule_id"), ) diff --git a/tests/integration_tests/migrations/composite_pk_association_tables__tests.py b/tests/integration_tests/migrations/composite_pk_association_tables__tests.py new file mode 100644 index 000000000000..52b1942bdb24 --- /dev/null +++ b/tests/integration_tests/migrations/composite_pk_association_tables__tests.py @@ -0,0 +1,131 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Schema-shape assertion tests for the composite-PK association-tables +migration (revision 2bee73611e32). + +Builds the pre-migration shape against an isolated in-memory SQLite engine, +runs the migration's ``upgrade()``, and asserts the resulting shape matches +the data-model.md "After" specification: no ``id`` column, composite PK on +the two FK columns, and no redundant ``UNIQUE(fk1, fk2)`` on the two tables +that previously carried one. + +Continuum-restore verification is OUT OF SCOPE; that work lives in the +versioning epic (sc-103156). Cross-backend verification (PostgreSQL, MySQL) +is handled by the CI matrix (T034a). +""" + +from importlib import import_module + +import pytest +import sqlalchemy as sa +from alembic.migration import MigrationContext +from alembic.operations import Operations +from sqlalchemy import inspect + +# Import the migration module under test. +_migration = import_module( + "superset.migrations.versions." + "2026-05-01_23-36_2bee73611e32_composite_pk_association_tables" +) +AFFECTED_TABLES = _migration.AFFECTED_TABLES +TABLES_WITH_PRE_EXISTING_UNIQUE = _migration.TABLES_WITH_PRE_EXISTING_UNIQUE + + +@pytest.fixture(scope="module") +def post_upgrade_engine() -> sa.engine.Engine: + """An isolated in-memory SQLite engine with the migration applied to a + pre-migration-shaped seed schema. Used by the post-upgrade assertions + below. Module-scoped so the upgrade only runs once per test session.""" + engine = sa.create_engine("sqlite:///:memory:") + md = sa.MetaData() + for t in AFFECTED_TABLES: + cols: list[sa.SchemaItem] = [ + sa.Column("id", sa.Integer, primary_key=True), + sa.Column(t.fk1, sa.Integer, nullable=False), + sa.Column(t.fk2, sa.Integer, nullable=False), + ] + constraints: list[sa.SchemaItem] = [] + if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: + constraints.append(sa.UniqueConstraint(t.fk1, t.fk2)) + sa.Table(t.name, md, *cols, *constraints) + md.create_all(engine) + + # Apply the migration's upgrade() against this engine via Alembic's + # MigrationContext, patching the migration module's ``op`` reference. + with engine.connect() as conn: + ctx = MigrationContext.configure(conn) + ops = Operations(ctx) + original_op = _migration.op + _migration.op = ops # type: ignore[attr-defined] + try: + _migration.upgrade() + finally: + _migration.op = original_op # type: ignore[attr-defined] + return engine + + +@pytest.mark.parametrize("t", AFFECTED_TABLES, ids=lambda t: t.name) +def test_no_id_column(post_upgrade_engine: sa.engine.Engine, t) -> None: + """The synthetic ``id`` column is gone from each affected table.""" + insp = inspect(post_upgrade_engine) + column_names = {c["name"] for c in insp.get_columns(t.name)} + assert "id" not in column_names, ( + f"{t.name} still has an 'id' column after migration; " + f"composite-PK conversion incomplete" + ) + + +@pytest.mark.parametrize("t", AFFECTED_TABLES, ids=lambda t: t.name) +def test_primary_key_is_composite_fks(post_upgrade_engine: sa.engine.Engine, t) -> None: + """The primary key of each affected table is exactly ``(fk1, fk2)``.""" + insp = inspect(post_upgrade_engine) + pk_cols = set(insp.get_pk_constraint(t.name).get("constrained_columns", [])) + assert pk_cols == {t.fk1, t.fk2}, ( + f"{t.name} primary key is {pk_cols}, expected {{{t.fk1}, {t.fk2}}}" + ) + + +@pytest.mark.parametrize( + "t", + [t for t in AFFECTED_TABLES if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE], + ids=lambda t: t.name, +) +def test_redundant_unique_dropped(post_upgrade_engine: sa.engine.Engine, t) -> None: + """For the two tables that previously carried a UNIQUE(fk1, fk2), that + constraint is now subsumed by the composite PK and must not appear + separately in the unique-constraint list.""" + insp = inspect(post_upgrade_engine) + redundant_pair = {t.fk1, t.fk2} + for uc in insp.get_unique_constraints(t.name): + cols = set(uc.get("column_names", [])) + assert cols != redundant_pair, ( + f"{t.name} still carries a redundant UniqueConstraint over " + f"{redundant_pair} (name={uc.get('name')!r}); " + f"composite-PK conversion incomplete" + ) + + +@pytest.mark.parametrize("t", AFFECTED_TABLES, ids=lambda t: t.name) +def test_fk_columns_not_null(post_upgrade_engine: sa.engine.Engine, t) -> None: + """PK promotion implicitly tightens the FK columns to NOT NULL.""" + insp = inspect(post_upgrade_engine) + cols_by_name = {c["name"]: c for c in insp.get_columns(t.name)} + for col in (t.fk1, t.fk2): + assert col in cols_by_name, f"{t.name} missing column {col}" + assert cols_by_name[col].get("nullable") is False, ( + f"{t.name}.{col} is nullable; expected NOT NULL after PK promotion" + ) diff --git a/tests/integration_tests/migrations/composite_pk_round_trip__tests.py b/tests/integration_tests/migrations/composite_pk_round_trip__tests.py new file mode 100644 index 000000000000..d83c9d113c3f --- /dev/null +++ b/tests/integration_tests/migrations/composite_pk_round_trip__tests.py @@ -0,0 +1,168 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Schema round-trip tests for the composite-PK association-tables migration +(revision 2bee73611e32). Builds the pre-migration shape against an in-memory +SQLite engine, runs the migration's ``upgrade()``, asserts the post-upgrade +shape, runs ``downgrade()``, asserts the prior shape is restored (modulo the +documented FK NOT NULL asymmetry), and re-runs ``upgrade()`` to verify +idempotency. + +This is run against an isolated in-memory engine via Alembic's +``MigrationContext`` so the test does not perturb the project's test DB. + +Cross-backend verification of the same migration against PostgreSQL and +MySQL is delegated to the CI matrix (see T034a in tasks.md) and to the +quickstart.md verification (T033). This file covers the SQLite slice. +""" + +from importlib import import_module +from typing import Any + +import pytest +import sqlalchemy as sa +from alembic.migration import MigrationContext +from alembic.operations import Operations +from sqlalchemy import inspect + +# Import the migration module under test. +_migration = import_module( + "superset.migrations.versions." + "2026-05-01_23-36_2bee73611e32_composite_pk_association_tables" +) +AFFECTED_TABLES = _migration.AFFECTED_TABLES +TABLES_WITH_PRE_EXISTING_UNIQUE = _migration.TABLES_WITH_PRE_EXISTING_UNIQUE + + +def _build_pre_migration_schema(engine: sa.engine.Engine) -> None: + """Recreate the eight tables in their pre-migration shape (surrogate + ``id INTEGER PRIMARY KEY`` plus an optional ``UNIQUE(fk1, fk2)`` on the + two tables that previously carried one). FKs to parent tables are + omitted to keep the test self-contained — we're testing schema + transformations, not FK enforcement.""" + md = sa.MetaData() + for t in AFFECTED_TABLES: + cols: list[sa.Column] = [ + sa.Column("id", sa.Integer, primary_key=True), + sa.Column(t.fk1, sa.Integer, nullable=False), + sa.Column(t.fk2, sa.Integer, nullable=False), + ] + constraints: list[sa.SchemaItem] = [] + if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: + constraints.append(sa.UniqueConstraint(t.fk1, t.fk2)) + sa.Table(t.name, md, *cols, *constraints) + md.create_all(engine) + + +def _shape(engine: sa.engine.Engine, table: str) -> dict[str, Any]: + """Return a structural summary for asserting equality across runs.""" + insp = inspect(engine) + pk = insp.get_pk_constraint(table).get("constrained_columns", []) + columns = sorted(c["name"] for c in insp.get_columns(table)) + uniques = sorted( + tuple(sorted(uc.get("column_names", []))) + for uc in insp.get_unique_constraints(table) + ) + return {"columns": columns, "pk": sorted(pk), "uniques": uniques} + + +def _run_with_alembic_context(engine: sa.engine.Engine, fn) -> None: + """Run ``fn()`` (the migration's upgrade/downgrade body) inside a fresh + Alembic ``MigrationContext`` bound to ``engine``. Patches the + migration module's ``op`` to point at this context so its + ``op.get_bind()`` and ``op.batch_alter_table`` calls execute against + the in-memory engine.""" + with engine.connect() as conn: + ctx = MigrationContext.configure(conn) + ops = Operations(ctx) + original_op = _migration.op + _migration.op = ops # type: ignore[attr-defined] + try: + fn() + finally: + _migration.op = original_op # type: ignore[attr-defined] + + +def test_round_trip_against_in_memory_sqlite() -> None: + """Round-trip: pre-migration → upgrade → downgrade → upgrade again. + + Asserts: + - Post-upgrade shape: no ``id``, composite PK on (fk1, fk2), no + UNIQUE(fk1, fk2) on the two tables that previously carried one. + - Post-downgrade shape: ``id`` restored, PK back on (id), UNIQUE + re-added on the two tables. (FK columns remain NOT NULL — the + documented intentional asymmetry.) + - Post-re-upgrade idempotency: shape matches the first post-upgrade. + """ + engine = sa.create_engine("sqlite:///:memory:") + _build_pre_migration_schema(engine) + + pre_shape = {t.name: _shape(engine, t.name) for t in AFFECTED_TABLES} + + _run_with_alembic_context(engine, _migration.upgrade) + + for t in AFFECTED_TABLES: + s = _shape(engine, t.name) + assert "id" not in s["columns"], f"{t.name}: id still present post-upgrade: {s}" + assert s["pk"] == sorted([t.fk1, t.fk2]), ( + f"{t.name}: PK is {s['pk']}, expected {sorted([t.fk1, t.fk2])}" + ) + assert tuple(sorted([t.fk1, t.fk2])) not in s["uniques"], ( + f"{t.name}: redundant UNIQUE not dropped post-upgrade: {s['uniques']}" + ) + + post_upgrade_shape = {t.name: _shape(engine, t.name) for t in AFFECTED_TABLES} + + _run_with_alembic_context(engine, _migration.downgrade) + + for t in AFFECTED_TABLES: + s = _shape(engine, t.name) + assert "id" in s["columns"], f"{t.name}: id not restored post-downgrade: {s}" + assert s["pk"] == ["id"], f"{t.name}: PK is {s['pk']}, expected ['id']" + if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: + assert tuple(sorted([t.fk1, t.fk2])) in s["uniques"], ( + f"{t.name}: UNIQUE not restored post-downgrade: {s['uniques']}" + ) + + _run_with_alembic_context(engine, _migration.upgrade) + + re_upgrade_shape = {t.name: _shape(engine, t.name) for t in AFFECTED_TABLES} + assert re_upgrade_shape == post_upgrade_shape, ( + "Re-upgrade shape differs from initial upgrade shape — " + "migration is not idempotent. " + f"diff: {set(re_upgrade_shape.items()) ^ set(post_upgrade_shape.items())}" + ) + + # Use pre_shape only to demonstrate it was captured (not asserted against + # because the round-trip downgrade intentionally diverges on FK NOT NULL). + _ = pre_shape + + +def test_migration_module_constants_are_consistent() -> None: + """Sanity-check the migration module's exported constants. Catches + accidental edits that misalign AFFECTED_TABLES with the auxiliary sets.""" + affected_names = {t.name for t in AFFECTED_TABLES} + assert _migration.TABLES_WITH_PRE_EXISTING_UNIQUE.issubset(affected_names) + assert _migration.TABLES_WITH_NULLABLE_FKS.issubset(affected_names) + # Order is alphabetical (deterministic for review/bisection). + assert [t.name for t in AFFECTED_TABLES] == sorted(affected_names) + + +@pytest.mark.skipif(True, reason="placeholder — see test_round_trip above") +def test_placeholder_for_future_postgres_round_trip() -> None: + """Reserved slot for a future Postgres-specific round-trip if local + SQLite divergence ever needs to be cross-checked against the real + backend. Today's CI matrix (T034a) handles this implicitly.""" diff --git a/tests/unit_tests/migrations/composite_pk_association_tables_test.py b/tests/unit_tests/migrations/composite_pk_association_tables_test.py new file mode 100644 index 000000000000..6c3115edaf65 --- /dev/null +++ b/tests/unit_tests/migrations/composite_pk_association_tables_test.py @@ -0,0 +1,132 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Unit tests for the composite-PK association-tables migration (revision +2bee73611e32). Verifies the post-migration constraint enforcement: duplicate +``(fk1, fk2)`` insertions fail with IntegrityError, distinct pairs succeed. + +Schema is built from the live ORM ``Table`` definitions via +``metadata.create_all(engine)`` against in-memory SQLite. This reflects the +post-T015–T018 ORM model state (composite-PK), independent of whether the +Alembic migration has run against the test DB. The two should agree. +""" + +import pytest +import sqlalchemy as sa +from sqlalchemy.exc import IntegrityError + +# (table_name, fk1_col, fk2_col, fk1_parent_table, fk2_parent_table) +# Parent-table names are needed to build the FK targets in the in-memory schema. +AFFECTED_TABLES = [ + ("dashboard_roles", "dashboard_id", "role_id", "dashboards", "ab_role"), + ("dashboard_slices", "dashboard_id", "slice_id", "dashboards", "slices"), + ("dashboard_user", "user_id", "dashboard_id", "ab_user", "dashboards"), + ( + "report_schedule_user", + "user_id", + "report_schedule_id", + "ab_user", + "report_schedule", + ), + ("rls_filter_roles", "role_id", "rls_filter_id", "ab_role", "rls_filter"), + ("rls_filter_tables", "table_id", "rls_filter_id", "tables", "rls_filter"), + ("slice_user", "user_id", "slice_id", "ab_user", "slices"), + ("sqlatable_user", "user_id", "table_id", "ab_user", "tables"), +] + + +def _build_in_memory_schema( + table_name: str, fk1: str, fk2: str, fk1_parent: str, fk2_parent: str +) -> tuple[sa.engine.Engine, sa.Table]: + """Build an in-memory SQLite schema with two minimal parent tables and + the junction table under test (composite-PK shape). Returns the engine + and the junction-table object for inserts.""" + metadata = sa.MetaData() + sa.Table( + fk1_parent, + metadata, + sa.Column("id", sa.Integer, primary_key=True), + ) + if fk2_parent != fk1_parent: + sa.Table( + fk2_parent, + metadata, + sa.Column("id", sa.Integer, primary_key=True), + ) + junction = sa.Table( + table_name, + metadata, + sa.Column( + fk1, + sa.Integer, + sa.ForeignKey(f"{fk1_parent}.id"), + primary_key=True, + ), + sa.Column( + fk2, + sa.Integer, + sa.ForeignKey(f"{fk2_parent}.id"), + primary_key=True, + ), + ) + engine = sa.create_engine("sqlite:///:memory:") + metadata.create_all(engine) + # Seed parent rows so the FK constraints can be satisfied. + # Identifiers come from the AFFECTED_TABLES test parameter list, not user input. + with engine.begin() as conn: + conn.execute( + sa.text(f"INSERT INTO {fk1_parent} (id) VALUES (1), (2)") # noqa: S608 + ) + if fk2_parent != fk1_parent: + conn.execute( + sa.text(f"INSERT INTO {fk2_parent} (id) VALUES (1), (2)") # noqa: S608 + ) + return engine, junction + + +@pytest.mark.parametrize("table,fk1,fk2,fk1_parent,fk2_parent", AFFECTED_TABLES) +def test_duplicate_insert_rejected( + table: str, fk1: str, fk2: str, fk1_parent: str, fk2_parent: str +) -> None: + """Inserting the same ``(fk1, fk2)`` pair twice raises ``IntegrityError``. + + Verifies SC-004 / FR-007 — the composite primary key enforces uniqueness + at the database level on every affected table. + """ + engine, junction = _build_in_memory_schema(table, fk1, fk2, fk1_parent, fk2_parent) + with engine.begin() as conn: + conn.execute(junction.insert().values({fk1: 1, fk2: 1})) + with pytest.raises(IntegrityError): + conn.execute(junction.insert().values({fk1: 1, fk2: 1})) + + +@pytest.mark.parametrize("table,fk1,fk2,fk1_parent,fk2_parent", AFFECTED_TABLES) +def test_distinct_pairs_accepted( + table: str, fk1: str, fk2: str, fk1_parent: str, fk2_parent: str +) -> None: + """Two distinct ``(fk1, fk2)`` pairs both succeed. + + Sanity check that the PK isn't accidentally a single-column constraint + (which would reject ``(1, 1)`` and ``(1, 2)`` as a duplicate on column 1). + """ + engine, junction = _build_in_memory_schema(table, fk1, fk2, fk1_parent, fk2_parent) + with engine.begin() as conn: + conn.execute(junction.insert().values({fk1: 1, fk2: 1})) + conn.execute(junction.insert().values({fk1: 1, fk2: 2})) + result = conn.execute( + sa.text(f"SELECT COUNT(*) FROM {table}") # noqa: S608 + ).scalar_one() + assert result == 2 From d75a30596a7391c2c8938e866c7d3bffad847298 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 4 May 2026 10:14:59 -0600 Subject: [PATCH 02/35] fix(migration): always run NULL-FK cleanup; correct RLS test parent name MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two cleanups from PR review: 1. ``dashboard_roles.dashboard_id`` was created nullable in revision e11ccdd12658 but was missing from ``TABLES_WITH_NULLABLE_FKS``. A production database with a stray NULL ``dashboard_id`` row would have failed the PK-add with a cryptic constraint violation. Fix by running the NULL-FK cleanup on every affected table — it is a no-op DELETE on tables whose FK columns are already NOT NULL, and it eliminates the risk of further drift in the hardcoded set. ``dashboard_roles`` is added to the documentation set; the runtime now does not consult it. 2. The unit-test parent-table name for ``rls_filter_roles`` and ``rls_filter_tables`` was ``rls_filter`` (does not exist) instead of the real parent ``row_level_security_filters``. Test passes either way (the in-memory FK is self-consistent), but the parameter is now accurate. Co-Authored-By: Claude Opus 4.7 (1M context) --- ...3611e32_composite_pk_association_tables.py | 20 +++++++++++++------ .../composite_pk_association_tables_test.py | 16 +++++++++++++-- 2 files changed, 28 insertions(+), 8 deletions(-) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index 2c841bc6171a..ec637de0118d 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -74,11 +74,15 @@ class AssociationTable(NamedTuple): "report_schedule_user", } -# Six tables whose FK columns are nullable today. Promoting an FK to a primary -# key column makes it NOT NULL, so any existing NULL-FK rows would block the -# PK-add. We delete them in pre-flight (a junction-table row with a NULL FK -# is meaningless under SQLAlchemy ``secondary=`` semantics anyway). +# Tables whose FK columns are nullable in their original create_table +# migrations. ``dashboard_roles.dashboard_id`` (created in revision +# e11ccdd12658) is nullable; ``report_schedule_user`` is the only association +# table that was created with both FK columns ``NOT NULL``. The pre-flight +# NULL-FK cleanup is a cheap no-op DELETE when run against tables whose FKs +# are already NOT NULL, so we run it on every affected table to avoid drift +# bugs from this set going stale. TABLES_WITH_NULLABLE_FKS: set[str] = { + "dashboard_roles", "dashboard_slices", "dashboard_user", "rls_filter_roles", @@ -221,8 +225,12 @@ def upgrade() -> None: insp = inspect(conn) for t in AFFECTED_TABLES: - if t.name in TABLES_WITH_NULLABLE_FKS: - _delete_null_fk_rows(conn, t) + # Run NULL-FK cleanup unconditionally: it is a no-op DELETE on tables + # whose FK columns are already NOT NULL (cheap), and skipping it on a + # table whose FK was nullable would leave the PK-add to fail with a + # cryptic constraint violation. Cf. ``TABLES_WITH_NULLABLE_FKS`` above + # for documentation of which tables are known to have nullable FKs. + _delete_null_fk_rows(conn, t) _dedupe_by_min_id(conn, t) _assert_no_duplicates(conn, t) diff --git a/tests/unit_tests/migrations/composite_pk_association_tables_test.py b/tests/unit_tests/migrations/composite_pk_association_tables_test.py index 6c3115edaf65..05a69293a23b 100644 --- a/tests/unit_tests/migrations/composite_pk_association_tables_test.py +++ b/tests/unit_tests/migrations/composite_pk_association_tables_test.py @@ -41,8 +41,20 @@ "ab_user", "report_schedule", ), - ("rls_filter_roles", "role_id", "rls_filter_id", "ab_role", "rls_filter"), - ("rls_filter_tables", "table_id", "rls_filter_id", "tables", "rls_filter"), + ( + "rls_filter_roles", + "role_id", + "rls_filter_id", + "ab_role", + "row_level_security_filters", + ), + ( + "rls_filter_tables", + "table_id", + "rls_filter_id", + "tables", + "row_level_security_filters", + ), ("slice_user", "user_id", "slice_id", "ab_user", "slices"), ("sqlatable_user", "user_id", "table_id", "ab_user", "tables"), ] From 621eb5492d788257c6c074904a123b95521eb5ab Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 4 May 2026 10:38:40 -0600 Subject: [PATCH 03/35] docs(migration): address SQLAlchemy review follow-ups Four operator-experience improvements from the second review pass: 1. ``TABLES_WITH_NULLABLE_FKS`` is now explicitly documented as an informational set that is not consulted at runtime; the comment explains the previous ``dashboard_roles`` omission was the bug that motivated the always-run cleanup. 2. ``_delete_null_fk_rows`` docstring updated to match the "always run" semantics (was still claiming "called only on tables in TABLES_WITH_NULLABLE_FKS"). 3. ``_check_no_external_fks_to_id`` now documents its scope limitation: ``Inspector.get_table_names()`` returns the default schema only, so cross-schema FKs in non-standard multi-schema PostgreSQL deployments would not be caught. The single-schema case (Superset's documented deployment) is fully covered. 4. ``_dedupe_by_min_id`` now logs a sample of up to 10 discarded ``(fk1, fk2, id)`` tuples at WARN before deletion, so operators can audit which rows the ``MIN(id)`` policy drops. The keep- original policy is correct in practice but discards later re-grants on ownership tables; the sample makes that visible. 5. ``UPDATING.md`` documents the upgrade/downgrade primary-key name divergence (``pk_
`` vs ``
_pkey``) so operators using schema-comparison tools don't mistake it for migration drift. No schema or runtime-behaviour changes. All 44 migration tests pass. Co-Authored-By: Claude Opus 4.7 (1M context) --- UPDATING.md | 2 + ...3611e32_composite_pk_association_tables.py | 59 +++++++++++++++---- 2 files changed, 49 insertions(+), 12 deletions(-) diff --git a/UPDATING.md b/UPDATING.md index 895e0d86c7d0..d08b878474de 100644 --- a/UPDATING.md +++ b/UPDATING.md @@ -534,6 +534,8 @@ SELECT COUNT(*) FROM sqlatable_user WHERE user_id IS NULL OR table_id IS NULL; **Intentional downgrade asymmetry.** The migration's `downgrade()` restores the surrogate `id` column and (for `dashboard_slices` and `report_schedule_user`) the original `UNIQUE (fk1, fk2)` constraint, but it does **not** restore the original `NULL`-allowed state on the FK columns — they remain `NOT NULL`. This is intentional: under SQLAlchemy's `secondary=` semantics, a `NULL` in either FK column of a junction table is meaningless (it cannot participate in either side of the relationship). Operators downgrading are not expected to need this restored. The asymmetry is documented for completeness so that round-trip schema diffs are not mistaken for migration bugs. +**Constraint-name divergence between upgrade and downgrade.** The composite primary key created on upgrade is named `pk_
` (Alembic's default for `op.create_primary_key("pk_
", ...)`), while the surrogate `id` primary key restored on downgrade is named `
_pkey` (PostgreSQL's default convention for `PrimaryKeyConstraint("id")`). The two names alternate so that a round-trip (upgrade → downgrade → upgrade) does not collide on a pre-existing constraint name. Operators using schema-comparison tools (e.g. `pg_diff`, `migra`) against a downgraded database may see this as drift versus a fresh-install schema. It is cosmetic — no application code references either constraint name. + ## 6.0.0 - [33055](https://github.com/apache/superset/pull/33055): Upgrades Flask-AppBuilder to 5.0.0. The AUTH_OID authentication type has been deprecated and is no longer available as an option in Flask-AppBuilder. OpenID (OID) is considered a deprecated authentication protocol - if you are using AUTH_OID, you will need to migrate to an alternative authentication method such as OAuth, LDAP, or database authentication before upgrading. - [34871](https://github.com/apache/superset/pull/34871): Fixed Jest test hanging issue from Ant Design v5 upgrade. MessageChannel is now mocked in test environment to prevent rc-overflow from causing Jest to hang. Test environment only - no production impact. diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index ec637de0118d..398e96cb755f 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -74,13 +74,15 @@ class AssociationTable(NamedTuple): "report_schedule_user", } -# Tables whose FK columns are nullable in their original create_table -# migrations. ``dashboard_roles.dashboard_id`` (created in revision -# e11ccdd12658) is nullable; ``report_schedule_user`` is the only association -# table that was created with both FK columns ``NOT NULL``. The pre-flight -# NULL-FK cleanup is a cheap no-op DELETE when run against tables whose FKs -# are already NOT NULL, so we run it on every affected table to avoid drift -# bugs from this set going stale. +# Documentation set: tables whose FK columns are nullable in their original +# create_table migrations (``dashboard_roles.dashboard_id`` from revision +# e11ccdd12658 is the most recent addition). ``report_schedule_user`` is the +# only affected table created with both FK columns ``NOT NULL`` and is +# intentionally absent here. This set is no longer consulted at runtime — the +# upgrade now runs the NULL-FK cleanup on every affected table because the +# DELETE is a cheap no-op when the columns are already NOT NULL, and that +# eliminates the risk of bugs from this set going stale (the +# ``dashboard_roles`` omission caught in PR review was exactly that bug). TABLES_WITH_NULLABLE_FKS: set[str] = { "dashboard_roles", "dashboard_slices", @@ -95,7 +97,18 @@ class AssociationTable(NamedTuple): def _check_no_external_fks_to_id(conn: Connection) -> None: """Raise ``RuntimeError`` if any foreign key in the database references one of the eight junction-table ``id`` columns. Uses SQLAlchemy's ``Inspector`` - for dialect-agnostic introspection across PostgreSQL, MySQL, and SQLite.""" + for dialect-agnostic introspection across PostgreSQL, MySQL, and SQLite. + + Scope limitation: ``Inspector.get_table_names()`` returns tables in the + connection's default schema only. On PostgreSQL deployments where Superset + metadata lives in a non-default schema, or on multi-schema deployments + that allow cross-schema FKs, an external FK in another schema would not + be detected. This is acceptable for the standard single-schema + deployment that Superset documents; operators with multi-schema + metadata should run the equivalent inventory query against + ``information_schema.referential_constraints`` themselves before + applying. + """ affected = {t.name for t in AFFECTED_TABLES} insp = inspect(conn) for table_name in insp.get_table_names(): @@ -115,10 +128,10 @@ def _check_no_external_fks_to_id(conn: Connection) -> None: def _delete_null_fk_rows(conn: Connection, t: AssociationTable) -> int: """Delete rows where ``t.fk1`` or ``t.fk2`` is NULL on ``t.name``. - Returns the deletion count. Called only on tables in - ``TABLES_WITH_NULLABLE_FKS``. Required because primary-key columns must be + Returns the deletion count. Required because primary-key columns must be NOT NULL; the PK-add downstream would fail with a cryptic constraint - violation if any NULL-FK rows survived. + violation if any NULL-FK rows survived. Run unconditionally on every + affected table — see ``TABLES_WITH_NULLABLE_FKS`` above for the rationale. """ # Identifiers come from the AFFECTED_TABLES whitelist, not user input. sql = sa.text( @@ -142,8 +155,22 @@ def _dedupe_by_min_id(conn: Connection, t: AssociationTable) -> int: portability — MySQL rejects ``DELETE FROM t WHERE id NOT IN (SELECT MIN(id) FROM t GROUP BY ...)`` with ERROR 1093 unless the inner SELECT is wrapped to force materialization. + + Logs a sample (up to 10) of the discarded ``(fk1, fk2, id)`` tuples at + WARN before deletion, so operators can audit which rows are dropped — the + "keep ``MIN(id)``" policy preserves the original row, which is correct + in practice but discards any later, semantically-identical re-grants. """ # Identifiers come from the AFFECTED_TABLES whitelist, not user input. + sample_sql = sa.text( + f"SELECT {t.fk1}, {t.fk2}, id FROM {t.name} WHERE id NOT IN (" # noqa: S608 + f" SELECT keep_id FROM (" + f" SELECT MIN(id) AS keep_id FROM {t.name} " + f"GROUP BY {t.fk1}, {t.fk2}" + f" ) AS s" + f") LIMIT 10" + ) + sample = list(conn.execute(sample_sql)) sql = sa.text( f"DELETE FROM {t.name} WHERE id NOT IN (" # noqa: S608 f" SELECT keep_id FROM (" @@ -155,7 +182,15 @@ def _dedupe_by_min_id(conn: Connection, t: AssociationTable) -> int: result = conn.execute(sql) n = result.rowcount or 0 if n: - logger.warning("Deduped %d duplicate row(s) from %s", n, t.name) + logger.warning( + "Deduped %d duplicate row(s) from %s; sample of discarded " + "(%s, %s, id) tuples (up to 10): %s", + n, + t.name, + t.fk1, + t.fk2, + sample, + ) return n From be6d3d183ccddb69f09157dbc144e38332a69c17 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 4 May 2026 15:35:14 -0600 Subject: [PATCH 04/35] refactor(migration): build pre-flight SQL via SQLAlchemy core (review) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Address Beto's review comments on apache/superset#39859: replace ``sa.text(f"...")`` SQL construction in the three pre-flight helpers (``_delete_null_fk_rows``, ``_dedupe_by_min_id``, ``_assert_no_duplicates``) with SQLAlchemy core constructs (``sa.delete``, ``sa.select``, ``sa.func``, ``.subquery()``, ``.notin_()``). A small ``_table_clause()`` helper builds a lightweight ``TableClause`` exposing the columns the queries reference; the three helpers consume it. Removes all ``# noqa: S608`` comments — they are no longer needed because there is no string-interpolated SQL. Verified the compiled SQL is identical on Postgres, MySQL, and SQLite, including the MySQL ERROR 1093 workaround (the inner aggregation is wrapped in a derived table via ``.subquery()``, producing ``... NOT IN (SELECT keep_id FROM (SELECT min(id) ...) AS keep_min)``). Also drops the redundant ``f`` prefix on the two non-interpolating lines of the ``_check_no_external_fks_to_id`` error message. 44 migration tests still pass. Co-Authored-By: Claude Opus 4.7 (1M context) --- ...3611e32_composite_pk_association_tables.py | 83 ++++++++++--------- 1 file changed, 46 insertions(+), 37 deletions(-) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index 398e96cb755f..8a128bfd7461 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -120,11 +120,19 @@ def _check_no_external_fks_to_id(conn: Connection) -> None: f"Cannot drop synthetic id from {fk['referred_table']}: " f"external FK {fk.get('name', '')} on {table_name} " f"references {fk['referred_table']}({fk['referred_columns']}). " - f"Drop or migrate the referencing FK before applying this " - f"migration." + "Drop or migrate the referencing FK before applying this " + "migration." ) +def _table_clause(t: AssociationTable) -> sa.sql.expression.TableClause: + """Build a lightweight SQLAlchemy ``TableClause`` for ``t`` exposing the + columns the helper queries reference (``id``, ``fk1``, ``fk2``). Used so + that the dedupe / cleanup / assert SQL can be expressed via SQLAlchemy + core constructs rather than via string interpolation.""" + return sa.table(t.name, sa.column("id"), sa.column(t.fk1), sa.column(t.fk2)) + + def _delete_null_fk_rows(conn: Connection, t: AssociationTable) -> int: """Delete rows where ``t.fk1`` or ``t.fk2`` is NULL on ``t.name``. @@ -133,11 +141,9 @@ def _delete_null_fk_rows(conn: Connection, t: AssociationTable) -> int: violation if any NULL-FK rows survived. Run unconditionally on every affected table — see ``TABLES_WITH_NULLABLE_FKS`` above for the rationale. """ - # Identifiers come from the AFFECTED_TABLES whitelist, not user input. - sql = sa.text( - f"DELETE FROM {t.name} WHERE {t.fk1} IS NULL OR {t.fk2} IS NULL" # noqa: S608 - ) - result = conn.execute(sql) + tbl = _table_clause(t) + stmt = sa.delete(tbl).where(sa.or_(tbl.c[t.fk1].is_(None), tbl.c[t.fk2].is_(None))) + result = conn.execute(stmt) n = result.rowcount or 0 if n: logger.warning( @@ -151,35 +157,35 @@ def _delete_null_fk_rows(conn: Connection, t: AssociationTable) -> int: def _dedupe_by_min_id(conn: Connection, t: AssociationTable) -> int: """Delete duplicate ``(t.fk1, t.fk2)`` rows from ``t.name`` keeping ``MIN(id)``. - Returns the deletion count. Uses the wrapped-subquery form for MySQL - portability — MySQL rejects ``DELETE FROM t WHERE id NOT IN (SELECT MIN(id) - FROM t GROUP BY ...)`` with ERROR 1093 unless the inner SELECT is wrapped - to force materialization. + Returns the deletion count. The ``NOT IN`` argument is wrapped in an + extra ``SELECT keep_id FROM (...) AS s`` derived table because MySQL + rejects ``DELETE FROM t WHERE id NOT IN (SELECT MIN(id) FROM t GROUP BY + ...)`` with ERROR 1093 unless the inner SELECT is materialized through + a derived table. SQLAlchemy's ``.subquery()`` produces that wrap. Logs a sample (up to 10) of the discarded ``(fk1, fk2, id)`` tuples at - WARN before deletion, so operators can audit which rows are dropped — the - "keep ``MIN(id)``" policy preserves the original row, which is correct - in practice but discards any later, semantically-identical re-grants. + WARN before deletion, so operators can audit which rows are dropped — + the "keep ``MIN(id)``" policy preserves the original row, which is + correct in practice but discards any later, semantically-identical + re-grants. """ - # Identifiers come from the AFFECTED_TABLES whitelist, not user input. - sample_sql = sa.text( - f"SELECT {t.fk1}, {t.fk2}, id FROM {t.name} WHERE id NOT IN (" # noqa: S608 - f" SELECT keep_id FROM (" - f" SELECT MIN(id) AS keep_id FROM {t.name} " - f"GROUP BY {t.fk1}, {t.fk2}" - f" ) AS s" - f") LIMIT 10" + tbl = _table_clause(t) + + keep_min = ( + sa.select(sa.func.min(tbl.c.id).label("keep_id")) + .group_by(tbl.c[t.fk1], tbl.c[t.fk2]) + .subquery("keep_min") ) - sample = list(conn.execute(sample_sql)) - sql = sa.text( - f"DELETE FROM {t.name} WHERE id NOT IN (" # noqa: S608 - f" SELECT keep_id FROM (" - f" SELECT MIN(id) AS keep_id FROM {t.name} " - f"GROUP BY {t.fk1}, {t.fk2}" - f" ) AS s" - f")" + keep_ids = sa.select(keep_min.c.keep_id) + discarded = tbl.c.id.notin_(keep_ids) + + sample_stmt = ( + sa.select(tbl.c[t.fk1], tbl.c[t.fk2], tbl.c.id).where(discarded).limit(10) ) - result = conn.execute(sql) + sample = list(conn.execute(sample_stmt)) + + delete_stmt = sa.delete(tbl).where(discarded) + result = conn.execute(delete_stmt) n = result.rowcount or 0 if n: logger.warning( @@ -201,13 +207,16 @@ def _assert_no_duplicates(conn: Connection, t: AssociationTable) -> None: dedupe failures (e.g., a MySQL syntax issue) as an actionable error before the PK-add fires with a less-helpful constraint-violation message. """ - # Identifiers come from the AFFECTED_TABLES whitelist, not user input. - sql = sa.text( - f"SELECT COUNT(*) FROM (" # noqa: S608 - f" SELECT 1 FROM {t.name} GROUP BY {t.fk1}, {t.fk2} HAVING COUNT(*) > 1" - f") AS s" + tbl = _table_clause(t) + duplicate_groups = ( + sa.select(sa.literal(1)) + .select_from(tbl) + .group_by(tbl.c[t.fk1], tbl.c[t.fk2]) + .having(sa.func.count() > 1) + .subquery("duplicate_groups") ) - if remaining := conn.scalar(sql) or 0: + count_stmt = sa.select(sa.func.count()).select_from(duplicate_groups) + if remaining := conn.scalar(count_stmt) or 0: raise RuntimeError( f"Dedupe failed for {t.name}: {remaining} duplicate " f"({t.fk1}, {t.fk2}) groups remain after _dedupe_by_min_id. " From 0da7e81487758ab7224606162c7dd8ea09eb4d69 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 4 May 2026 16:01:58 -0600 Subject: [PATCH 05/35] fix(migration): drop FKs before recreate on MySQL (sc-105349) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit CI test-mysql failed with: MySQLdb.OperationalError: (1826, "Duplicate foreign key constraint name 'fk_dashboard_slices_slice_id_slices'") Root cause: MySQL scopes foreign-key constraint names per-database, not per-table (PostgreSQL and SQLite scope per-table). The ``batch_alter_table(... recreate="always", copy_from=...)`` path used for ``dashboard_slices`` and ``report_schedule_user`` builds ``_alembic_tmp_
`` carrying the original FK names from ``copy_from`` while the original table still holds those names — MySQL rejects the temp-table creation with ERROR 1826. Fix: on MySQL only, drop the original FK constraints by name before the ``batch_alter_table`` runs. The ``copy_from`` re-creates them on the rebuilt table with their original names, so the post-migration shape is unchanged. On PostgreSQL and SQLite the original code path still runs unchanged. Local SQLite tests (44 passed, 1 skipped) still pass; CI will validate on MySQL. Co-Authored-By: Claude Opus 4.7 (1M context) --- ...2bee73611e32_composite_pk_association_tables.py | 14 ++++++++++++++ 1 file changed, 14 insertions(+) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index 8a128bfd7461..8d7b2846d342 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -289,6 +289,20 @@ def upgrade() -> None: # temp-table index-name collision; on SQLite, the auto-detect picks # ``recreate=True`` because PK changes need it. if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: + # MySQL ERROR 1826: foreign-key constraint names are unique + # per-database, not per-table. ``recreate="always"`` builds + # ``_alembic_tmp_
`` with the original FK names from + # ``copy_from``, but the original table still holds those + # names until it's dropped, which fails on MySQL with + # ``Duplicate foreign key constraint name``. PostgreSQL and + # SQLite scope FK names per-table, so the recreate path + # works there as-is. Drop the original FKs by name first + # on MySQL; ``copy_from`` re-creates them on the rebuilt + # table with their original names. + if conn.dialect.name == "mysql": + for fk in insp.get_foreign_keys(t.name): + if fk_name := fk.get("name"): + op.drop_constraint(fk_name, t.name, type_="foreignkey") with op.batch_alter_table( t.name, recreate="always", From f436efd1e941efc80afb8307389fe763806a909c Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Tue, 5 May 2026 10:41:03 -0600 Subject: [PATCH 06/35] fix(migration): MySQL downgrade FK + AUTO_INCREMENT (sc-105349) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two MySQL-only failures in the downgrade path, found by running the full migration history against a fresh MySQL 8 container: 1. ``MySQLdb.OperationalError: (1553, "Cannot drop index 'PRIMARY': needed in a foreign key constraint")``. InnoDB uses the composite PK index to back the FK on the leftmost column. The downgrade tried to drop the composite PK before dropping the FKs, orphaning the FK's backing index. PostgreSQL and SQLite create separate indexes for FK columns and don't trip on this. 2. ``Field 'id' doesn't have a default value`` on subsequent INSERT. ``sa.Identity(always=False)`` only emits ``AUTO_INCREMENT`` on MySQL when the column is created with ``primary_key=True`` — our portable path adds the column first then creates the PK separately, so MySQL leaves the column without auto-generation. Existing rows would all collide on id=0; future inserts fail because no default. Postgres' ``GENERATED BY DEFAULT AS IDENTITY`` and SQLite's ``INTEGER PRIMARY KEY`` rowid alias don't have this gap. Fix: extract ``_downgrade_mysql_table()`` that emits the canonical MySQL idiom — drop FKs, then a single ALTER combining ``DROP PRIMARY KEY, ADD COLUMN id INT NOT NULL AUTO_INCREMENT, ADD PRIMARY KEY (id)`` (which backfills existing rows with sequential ids and preserves AUTO_INCREMENT), restore the redundant UNIQUE on the 2 tables that originally had it, and re-add the FKs with their original names. Postgres and SQLite keep the existing portable ``batch_alter_table`` path. Raw SQL is unavoidable for the combined-ALTER form; per the constitution it's allowed for dialect-specific DDL with no SQLA equivalent, with triple-quoted strings for legibility. Verified end-to-end: upgrade → downgrade → upgrade against a fresh MySQL 8 container with INSERT-without-id sanity check showing the restored ``id`` column auto-increments correctly. Co-Authored-By: Claude Opus 4.7 (1M context) --- ...3611e32_composite_pk_association_tables.py | 108 +++++++++++++++--- 1 file changed, 94 insertions(+), 14 deletions(-) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index 8d7b2846d342..e8a77614561c 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -337,19 +337,99 @@ def downgrade() -> None: # 10+ / MySQL 8+) and ``sa.Sequence`` (with explicit nextval) both # backfill existing rows during ALTER TABLE; bare ``autoincrement=True`` # does not. ``Identity`` is the modern portable choice. + conn = op.get_bind() + insp = inspect(conn) + is_mysql = conn.dialect.name == "mysql" for t in reversed(AFFECTED_TABLES): - with op.batch_alter_table(t.name) as batch_op: - batch_op.drop_constraint(f"pk_{t.name}", type_="primary") - batch_op.add_column( - sa.Column( - "id", - sa.Integer, - sa.Identity(always=False), - nullable=False, - ) - ) - batch_op.create_primary_key(f"{t.name}_pkey", ["id"]) - if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: - batch_op.create_unique_constraint( - f"uq_{t.name}_{t.fk1}_{t.fk2}", [t.fk1, t.fk2] + if is_mysql: + _downgrade_mysql_table(insp, t) + else: + with op.batch_alter_table(t.name) as batch_op: + batch_op.drop_constraint(f"pk_{t.name}", type_="primary") + batch_op.add_column( + sa.Column( + "id", + sa.Integer, + sa.Identity(always=False), + nullable=False, + ) ) + batch_op.create_primary_key(f"{t.name}_pkey", ["id"]) + if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: + batch_op.create_unique_constraint( + f"uq_{t.name}_{t.fk1}_{t.fk2}", [t.fk1, t.fk2] + ) + + +def _downgrade_mysql_table( + insp: sa.engine.reflection.Inspector, t: AssociationTable +) -> None: + """MySQL-specific downgrade for one table. + + Two MySQL quirks force a dialect-specific path here: + + 1. **ERROR 1553 — ``Cannot drop index 'PRIMARY': needed in a foreign + key constraint``**. InnoDB uses the composite PK index to back the + FK on the leftmost column. Dropping the PK before the FKs orphans + that backing index. PostgreSQL and SQLite create separate indexes + for FK columns and don't need this dance. We drop the FKs first + and re-add them after the structural change. + + 2. **``Identity(always=False)`` on a non-PK column add does not emit + ``AUTO_INCREMENT`` on MySQL.** SQLAlchemy 1.4 only emits + ``AUTO_INCREMENT`` when the column has both ``Identity()`` and + ``primary_key=True`` at create time. Our portable path adds the + column first, then creates the PK separately — which works on + Postgres (the column gets ``GENERATED BY DEFAULT AS IDENTITY``) + and SQLite (``INTEGER PRIMARY KEY`` becomes a rowid alias) but + leaves MySQL without auto-generation, so existing rows can't be + backfilled and future ``INSERT`` statements fail with + ``Field 'id' doesn't have a default value``. The combined + ``DROP PRIMARY KEY, ADD COLUMN AUTO_INCREMENT, ADD PRIMARY KEY`` + in a single ALTER statement is the canonical MySQL idiom: MySQL + backfills existing rows with sequential values and the column + remains auto-incrementing for future inserts. + + Raw SQL is unavoidable here — there is no SQLAlchemy core equivalent + for the combined-ALTER form, and the constitution allows raw SQL for + dialect-specific DDL with no programmatic equivalent (preferring + triple-quoted strings for legibility). + """ + fks = insp.get_foreign_keys(t.name) + + for fk in fks: + if fk_name := fk.get("name"): + op.execute(f"ALTER TABLE `{t.name}` DROP FOREIGN KEY `{fk_name}`") + + op.execute( + f""" + ALTER TABLE `{t.name}` + DROP PRIMARY KEY, + ADD COLUMN id INT NOT NULL AUTO_INCREMENT, + ADD PRIMARY KEY (id) + """ + ) + + if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: + op.execute( + f""" + ALTER TABLE `{t.name}` + ADD UNIQUE INDEX `uq_{t.name}_{t.fk1}_{t.fk2}` + (`{t.fk1}`, `{t.fk2}`) + """ + ) + + for fk in fks: + ondelete = fk.get("options", {}).get("ondelete") + ondelete_clause = f" ON DELETE {ondelete}" if ondelete else "" + local_cols = ", ".join(f"`{c}`" for c in fk["constrained_columns"]) + ref_cols = ", ".join(f"`{c}`" for c in fk["referred_columns"]) + op.execute( + f""" + ALTER TABLE `{t.name}` + ADD CONSTRAINT `{fk["name"]}` + FOREIGN KEY ({local_cols}) + REFERENCES `{fk["referred_table"]}` ({ref_cols}) + {ondelete_clause} + """ + ) From d994ebf60f1b481d760c98136eedb53178c5c3e3 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Tue, 5 May 2026 10:46:01 -0600 Subject: [PATCH 07/35] fix(migration): explicit NOT NULL on FK columns for SQLite (sc-105349) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Found by running fresh-install + round-trip against a real SQLite DB: 6 of the 8 affected tables had FK columns that were originally declared nullable. PostgreSQL and MySQL implicitly promote the constituent columns of an ``ALTER TABLE ... ADD PRIMARY KEY`` to ``NOT NULL``; SQLite does not (it's a long-standing SQLite quirk — only ``INTEGER PRIMARY KEY`` enforces NOT NULL on a composite-PK column). Result: a fresh SQLite install would accept ``INSERT INTO dashboard_slices (NULL, 5)`` despite both columns being part of the composite PK. Our integration tests previously masked this: the test fixture seeds columns with ``nullable=False``, so the post-upgrade NOT NULL assertion passed regardless of whether the migration enforced it. Fix: add explicit ``batch_op.alter_column(fk, nullable=False)`` for both FK columns inside the per-table batch_alter_table block. On PostgreSQL and MySQL this is a no-op (PK already implies NOT NULL); on SQLite it adds the missing NOT NULL declaration so a fresh install matches the data-model.md "After" contract. Verified end-to-end: - Postgres + MySQL: column shape unchanged (still NOT NULL) - SQLite fresh install + round-trip: all 8 tables now have NOT NULL on FK columns, ``INSERT (NULL, 5)`` correctly rejected with IntegrityError on dashboard_slices, dashboard_user, sqlatable_user Co-Authored-By: Claude Opus 4.7 (1M context) --- ..._2bee73611e32_composite_pk_association_tables.py | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index e8a77614561c..210a419d0eea 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -310,10 +310,23 @@ def upgrade() -> None: ) as batch_op: batch_op.drop_column("id") batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) + # SQLite quirk: composite PRIMARY KEY does not promote the + # constituent columns to NOT NULL (only ``INTEGER PRIMARY + # KEY`` does). PostgreSQL and MySQL implicitly promote the + # PK columns to NOT NULL when the constraint is added, + # so the explicit ``alter_column`` is a no-op on those + # backends but enforces the post-upgrade contract on + # SQLite. Without it, ``INSERT (NULL, 5)`` would succeed + # on SQLite despite the columns being part of the PK. + batch_op.alter_column(t.fk1, existing_type=sa.Integer, nullable=False) + batch_op.alter_column(t.fk2, existing_type=sa.Integer, nullable=False) else: with op.batch_alter_table(t.name) as batch_op: batch_op.drop_column("id") batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) + # See comment above re: SQLite composite-PK NOT NULL quirk. + batch_op.alter_column(t.fk1, existing_type=sa.Integer, nullable=False) + batch_op.alter_column(t.fk2, existing_type=sa.Integer, nullable=False) def downgrade() -> None: From d6b695973dd17e37fbd02dde7d4a67f4a25c460c Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Tue, 5 May 2026 11:07:10 -0600 Subject: [PATCH 08/35] fix(migration): rebase down_revision onto 33d7e0e21daa (sc-105349) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit CI cypress + playwright shards were red with: ERROR [flask_migrate] Error: Multiple head revisions are present for given argument 'head' The recent rebase onto master pulled in ``33d7e0e21daa_add_semantic_layers_and_views.py`` (from PR #37815, "semantic layer extension"), which had been authored against ``ce6bd21901ab`` as its parent — the same parent our migration referenced. After the rebase both migrations point at ``ce6bd21901ab``, producing two heads and breaking ``flask db upgrade head`` for any downstream consumer (CI's Cypress / Playwright shards spin up a real Superset instance via ``superset db upgrade``, which is why those shards failed first; the integration shards run against a precomputed schema and didn't surface this). Fix: chain our migration after the semantic-layer migration by pointing ``down_revision`` at ``33d7e0e21daa``. The chain is now linear: ... → ce6bd21901ab → 33d7e0e21daa (semantic layers) → 2bee73611e32 (composite PK, this PR) Verified with ``superset db heads`` (returns single head ``2bee73611e32``) and the local migration test suite (44 passed, 1 skipped). Co-Authored-By: Claude Opus 4.7 (1M context) --- ...5-01_23-36_2bee73611e32_composite_pk_association_tables.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index 210a419d0eea..055ecd3c9700 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -27,7 +27,7 @@ of the eight tables lacked DB-level uniqueness. Revision ID: 2bee73611e32 -Revises: ce6bd21901ab +Revises: 33d7e0e21daa Create Date: 2026-05-01 23:36:34.050058 """ @@ -42,7 +42,7 @@ # revision identifiers, used by Alembic. revision = "2bee73611e32" -down_revision = "ce6bd21901ab" +down_revision = "33d7e0e21daa" logger = logging.getLogger("alembic.env") From 02662c5f46f82c49b5e6f00a5d7035c423016dd7 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 7 May 2026 10:23:21 -0600 Subject: [PATCH 09/35] docs(UPDATING): add Postgres-targeted maintenance-window queries (sc-105349) Add a "Sizing the maintenance window on PostgreSQL" sub-section to the operator runbook. The simple per-table COUNT/duplicate/NULL queries that were already there are dialect-portable but only count rows; operators on PostgreSQL with large deployments need to characterize the migration's runtime cost before scheduling it. Adds four diagnostic queries: - Per-table size, row count (from pg_class.reltuples), and which migration path each table will take (recreate-rewrite vs direct ALTER). Sizes the work concretely. - Aggregated duplicate-row roll-up: dup_groups + total rows_dropped per table. Replaces eight separate per-table queries with one consolidated result for audit/dump-before-apply decisions. - External-FK pre-flight check (the same one the migration runs at upgrade time and aborts on). Lets operators surface any blocking external reference ahead of the maintenance window. Should be empty on a stock install. - Lock-window estimate for the two full-rewrite tables, using pg_relation_size and a conservative 100 MB/s rewrite throughput assumption. The other six use direct ALTER and are dominated by composite-index build time (seconds for low-millions-of-rows tables). Prompted by reviewer feedback on apache/superset#39859 from a large deployment asking how to size the maintenance window. The original pre-flight queries are kept for cross-dialect operators (MySQL, SQLite) since the new queries use PostgreSQL-specific catalog views. Co-Authored-By: Claude Opus 4.7 (1M context) --- UPDATING.md | 102 ++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 102 insertions(+) diff --git a/UPDATING.md b/UPDATING.md index d08b878474de..b5c09de1ff9c 100644 --- a/UPDATING.md +++ b/UPDATING.md @@ -530,6 +530,108 @@ SELECT COUNT(*) FROM slice_user WHERE user_id IS NULL OR slice_id IS NULL; SELECT COUNT(*) FROM sqlatable_user WHERE user_id IS NULL OR table_id IS NULL; ``` +**Sizing the maintenance window on PostgreSQL.** The queries above are dialect-portable but only count rows. Operators on PostgreSQL can run the diagnostic queries below to characterize the migration's runtime cost ahead of time: per-table row count and on-disk size, an aggregated duplicate roll-up, the external-FK pre-flight check (the migration runs the same check and aborts if it returns rows), and a rewrite-time estimate for the two tables that go through the slower full-table-rebuild path. + +```sql +-- Per-table size, row count, and which migration path each will take. +-- Two tables ("dashboard_slices", "report_schedule_user") have a +-- redundant UNIQUE constraint that the migration drops via a full +-- table rewrite (op.batch_alter_table(recreate="always")). The other +-- six use direct ALTER TABLE, which is much cheaper. +WITH affected(name, has_unique) AS ( + VALUES + ('dashboard_roles', false), + ('dashboard_slices', true), + ('dashboard_user', false), + ('report_schedule_user', true), + ('rls_filter_roles', false), + ('rls_filter_tables', false), + ('slice_user', false), + ('sqlatable_user', false) +) +SELECT + a.name AS table_name, + CASE WHEN a.has_unique THEN 'recreate (full rewrite)' + ELSE 'direct ALTER' END AS migration_path, + c.reltuples::bigint AS estimated_rows, + pg_size_pretty(pg_total_relation_size(c.oid)) AS total_size, + pg_size_pretty(pg_relation_size(c.oid)) AS heap_size, + pg_size_pretty(pg_indexes_size(c.oid)) AS index_size +FROM affected a +JOIN pg_class c ON c.relname = a.name AND c.relkind = 'r' +ORDER BY pg_total_relation_size(c.oid) DESC; +``` + +```sql +-- Aggregated duplicate-row roll-up. +-- "dup_groups" is the number of (fk1, fk2) pairs that appear more +-- than once; "rows_dropped" is the total number of rows the +-- migration will delete during the dedupe pass (it keeps MIN(id) per +-- group and discards the rest). +SELECT 'dashboard_roles' AS t, COUNT(*) AS dup_groups, SUM(c) - COUNT(*) AS rows_dropped + FROM (SELECT COUNT(*) c FROM dashboard_roles GROUP BY dashboard_id, role_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'dashboard_slices', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM dashboard_slices GROUP BY dashboard_id, slice_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'dashboard_user', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM dashboard_user GROUP BY user_id, dashboard_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'report_schedule_user',COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM report_schedule_user GROUP BY user_id, report_schedule_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'rls_filter_roles', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM rls_filter_roles GROUP BY role_id, rls_filter_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'rls_filter_tables', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM rls_filter_tables GROUP BY table_id, rls_filter_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'slice_user', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM slice_user GROUP BY user_id, slice_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'sqlatable_user', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM sqlatable_user GROUP BY user_id, table_id HAVING COUNT(*) > 1) g +ORDER BY rows_dropped DESC NULLS LAST; +``` + +```sql +-- External-FK pre-flight check. +-- The migration runs the equivalent check at upgrade time and aborts +-- if any external FK references one of the soon-to-be-removed `id` +-- columns. Running it ahead of time lets you discover (and migrate) +-- any such reference before the maintenance window. On a stock +-- Superset install this should return zero rows. (Default schema +-- only; multi-schema deployments need to broaden the lookup.) +SELECT + rc.constraint_name, + kcu.table_schema || '.' || kcu.table_name AS referencing_table, + kcu.column_name AS referencing_column, + ccu.table_name AS referenced_table, + ccu.column_name AS referenced_column +FROM information_schema.referential_constraints rc +JOIN information_schema.key_column_usage kcu + ON kcu.constraint_name = rc.constraint_name + AND kcu.constraint_schema = rc.constraint_schema +JOIN information_schema.constraint_column_usage ccu + ON ccu.constraint_name = rc.constraint_name + AND ccu.constraint_schema = rc.constraint_schema +WHERE ccu.table_name IN ( + 'dashboard_roles','dashboard_slices','dashboard_user', + 'report_schedule_user','rls_filter_roles','rls_filter_tables', + 'slice_user','sqlatable_user') + AND ccu.column_name = 'id'; +``` + +```sql +-- Lock-window estimate for the two full-rewrite tables. +-- recreate="always" takes ACCESS EXCLUSIVE on the table for the full +-- rewrite. Use heap size combined with your hardware's effective +-- write throughput (~100-200 MB/s on commodity SSD; faster on NVMe) +-- to size the maintenance window. The other six tables use direct +-- ALTER and are dominated by composite-index build time, typically +-- seconds for tables in the low millions of rows. +SELECT + c.relname AS table_name, + pg_size_pretty(pg_relation_size(c.oid)) AS heap_size, + pg_relation_size(c.oid) / 1024 / 1024 AS heap_size_mb, + ROUND(pg_relation_size(c.oid) / 1024 / 1024 / 100.0, 1) AS est_rewrite_seconds_at_100mbs +FROM pg_class c +WHERE c.relname IN ('dashboard_slices', 'report_schedule_user'); +``` + **Restoring an old `pg_dump` (or equivalent) against the new schema.** A dump taken before the migration includes `INSERT` statements that populate the now-removed `id` column. Restoring such a dump against the post-migration schema will fail. The supported workaround is to dump only the schema and reference data, then re-create the M:N associations from application data after restore — for example with `pg_dump --exclude-table-data` (or per-table `--exclude-table-data=dashboard_slices` etc.) for the eight junction tables, restore the rest, then run a one-shot script that re-INSERTs `(fk1, fk2)` pairs derived from your application export. Operators who need to restore an old dump verbatim should restore against a pre-migration Superset and then re-run the upgrade. **Intentional downgrade asymmetry.** The migration's `downgrade()` restores the surrogate `id` column and (for `dashboard_slices` and `report_schedule_user`) the original `UNIQUE (fk1, fk2)` constraint, but it does **not** restore the original `NULL`-allowed state on the FK columns — they remain `NOT NULL`. This is intentional: under SQLAlchemy's `secondary=` semantics, a `NULL` in either FK column of a junction table is meaningless (it cannot participate in either side of the relationship). Operators downgrading are not expected to need this restored. The asymmetry is documented for completeness so that round-trip schema diffs are not mistaken for migration bugs. From 4c4c891f1d76634274b2634323b32098df45a49c Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 7 May 2026 10:58:21 -0600 Subject: [PATCH 10/35] docs(UPDATING): add MySQL-targeted maintenance-window queries (sc-105349) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Mirror of the PostgreSQL diagnostic queries added in 11148779ed, adapted for MySQL/InnoDB. One important difference: InnoDB rebuilds the clustered index on every PK change, so all eight tables undergo a full table rebuild on MySQL — not just the two that go through the explicit ``recreate="always"`` path. The lock-window estimate query is updated to cover all eight rather than just two, and the "migration_path" column makes the rebuild expectation explicit ("direct ALTER (still rebuilds InnoDB clustered index)"). Other notes: - ``information_schema.TABLES.TABLE_ROWS`` is an InnoDB estimate, analogous to PostgreSQL's ``reltuples``; documented inline. - ``KEY_COLUMN_USAGE`` carries both sides of the FK in a single row on MySQL, so the external-FK pre-flight check is simpler than the PostgreSQL version (no joins between three views). - The aggregated dedupe query is portable standard SQL; included verbatim for copy-paste convenience. Co-Authored-By: Claude Opus 4.7 (1M context) --- UPDATING.md | 89 +++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 89 insertions(+) diff --git a/UPDATING.md b/UPDATING.md index b5c09de1ff9c..7f0f97078a1b 100644 --- a/UPDATING.md +++ b/UPDATING.md @@ -632,6 +632,95 @@ FROM pg_class c WHERE c.relname IN ('dashboard_slices', 'report_schedule_user'); ``` +**Sizing the maintenance window on MySQL.** Equivalent diagnostic queries for MySQL/InnoDB. One important difference from PostgreSQL: InnoDB rebuilds the clustered index on every PK change, so *all eight* tables undergo a full table rebuild on MySQL — not just the two that go through the explicit `recreate="always"` path. The lock-window estimate query below therefore covers all eight tables. + +```sql +-- Per-table size, row count, and which migration path each will take. +-- TABLE_ROWS is an InnoDB estimate (analogous to PostgreSQL's reltuples); +-- run SELECT COUNT(*) per table for an exact count if needed. +SELECT + TABLE_NAME AS table_name, + CASE WHEN TABLE_NAME IN ('dashboard_slices', 'report_schedule_user') + THEN 'recreate (explicit, drops UNIQUE)' + ELSE 'direct ALTER (still rebuilds InnoDB clustered index)' + END AS migration_path, + TABLE_ROWS AS estimated_rows, + CONCAT(ROUND((DATA_LENGTH + INDEX_LENGTH) / 1024 / 1024, 1), ' MB') AS total_size, + CONCAT(ROUND(DATA_LENGTH / 1024 / 1024, 1), ' MB') AS heap_size, + CONCAT(ROUND(INDEX_LENGTH / 1024 / 1024, 1), ' MB') AS index_size +FROM information_schema.TABLES +WHERE TABLE_SCHEMA = DATABASE() + AND TABLE_NAME IN ( + 'dashboard_roles', 'dashboard_slices', 'dashboard_user', + 'report_schedule_user', 'rls_filter_roles', 'rls_filter_tables', + 'slice_user', 'sqlatable_user' + ) +ORDER BY (DATA_LENGTH + INDEX_LENGTH) DESC; +``` + +```sql +-- Aggregated duplicate-row roll-up. Same SQL as the PostgreSQL version +-- (standard SQL); included here for copy-paste convenience. +SELECT 'dashboard_roles' AS t, COUNT(*) AS dup_groups, SUM(c) - COUNT(*) AS rows_dropped + FROM (SELECT COUNT(*) c FROM dashboard_roles GROUP BY dashboard_id, role_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'dashboard_slices', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM dashboard_slices GROUP BY dashboard_id, slice_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'dashboard_user', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM dashboard_user GROUP BY user_id, dashboard_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'report_schedule_user',COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM report_schedule_user GROUP BY user_id, report_schedule_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'rls_filter_roles', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM rls_filter_roles GROUP BY role_id, rls_filter_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'rls_filter_tables', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM rls_filter_tables GROUP BY table_id, rls_filter_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'slice_user', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM slice_user GROUP BY user_id, slice_id HAVING COUNT(*) > 1) g +UNION ALL SELECT 'sqlatable_user', COUNT(*), SUM(c) - COUNT(*) + FROM (SELECT COUNT(*) c FROM sqlatable_user GROUP BY user_id, table_id HAVING COUNT(*) > 1) g +ORDER BY rows_dropped DESC; +``` + +```sql +-- External-FK pre-flight check. KEY_COLUMN_USAGE on MySQL carries +-- both sides of the FK in a single row, so this is simpler than the +-- PostgreSQL version. Should return zero rows on a stock install. +SELECT + CONSTRAINT_NAME, + CONCAT(TABLE_SCHEMA, '.', TABLE_NAME) AS referencing_table, + COLUMN_NAME AS referencing_column, + REFERENCED_TABLE_NAME AS referenced_table, + REFERENCED_COLUMN_NAME AS referenced_column +FROM information_schema.KEY_COLUMN_USAGE +WHERE TABLE_SCHEMA = DATABASE() + AND REFERENCED_TABLE_NAME IN ( + 'dashboard_roles', 'dashboard_slices', 'dashboard_user', + 'report_schedule_user', 'rls_filter_roles', 'rls_filter_tables', + 'slice_user', 'sqlatable_user' + ) + AND REFERENCED_COLUMN_NAME = 'id'; +``` + +```sql +-- Lock-window estimate for ALL EIGHT tables (InnoDB rebuilds the +-- clustered index on PK change, so even "direct ALTER" is a rewrite). +-- ADD PRIMARY KEY is INPLACE but not LOCK=NONE — it allows concurrent +-- reads but blocks writes. Use heap size combined with your effective +-- rebuild throughput (~100-200 MB/s on commodity SSD; higher on NVMe). +SELECT + TABLE_NAME AS table_name, + CONCAT(ROUND(DATA_LENGTH / 1024 / 1024, 1), ' MB') AS heap_size, + ROUND(DATA_LENGTH / 1024 / 1024, 1) AS heap_size_mb, + ROUND(DATA_LENGTH / 1024 / 1024 / 100.0, 1) AS est_rewrite_seconds_at_100mbs +FROM information_schema.TABLES +WHERE TABLE_SCHEMA = DATABASE() + AND TABLE_NAME IN ( + 'dashboard_roles', 'dashboard_slices', 'dashboard_user', + 'report_schedule_user', 'rls_filter_roles', 'rls_filter_tables', + 'slice_user', 'sqlatable_user' + ) +ORDER BY DATA_LENGTH DESC; +``` + **Restoring an old `pg_dump` (or equivalent) against the new schema.** A dump taken before the migration includes `INSERT` statements that populate the now-removed `id` column. Restoring such a dump against the post-migration schema will fail. The supported workaround is to dump only the schema and reference data, then re-create the M:N associations from application data after restore — for example with `pg_dump --exclude-table-data` (or per-table `--exclude-table-data=dashboard_slices` etc.) for the eight junction tables, restore the rest, then run a one-shot script that re-INSERTs `(fk1, fk2)` pairs derived from your application export. Operators who need to restore an old dump verbatim should restore against a pre-migration Superset and then re-run the upgrade. **Intentional downgrade asymmetry.** The migration's `downgrade()` restores the surrogate `id` column and (for `dashboard_slices` and `report_schedule_user`) the original `UNIQUE (fk1, fk2)` constraint, but it does **not** restore the original `NULL`-allowed state on the FK columns — they remain `NOT NULL`. This is intentional: under SQLAlchemy's `secondary=` semantics, a `NULL` in either FK column of a junction table is meaningless (it cannot participate in either side of the relationship). Operators downgrading are not expected to need this restored. The asymmetry is documented for completeness so that round-trip schema diffs are not mistaken for migration bugs. From 7258debe0bb579a138a5d4e59fdf70d85b3e6447 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 7 May 2026 11:41:53 -0600 Subject: [PATCH 11/35] build(docker): add MySQL compose override for dialect-swap evaluation MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Adds ``docker-compose-mysql.yml``, a compose-override file that swaps the default Postgres metadata DB for MySQL 8 with one extra ``-f`` flag: docker compose -f docker-compose.yml -f docker-compose-mysql.yml up Useful for evaluating dialect-specific behaviour (e.g., the runtime cost of DDL migrations on a deployment whose production metadata DB is MySQL — the question raised by review feedback on this PR). Mirrors the connection settings used by CI's ``test-mysql`` shard: ``mysql+mysqldb`` dialect, charset ``utf8mb4`` with binary_prefix. Host port defaults to 13306 (configurable via ``DATABASE_PORT_MYSQL``) to avoid colliding with a native MySQL install on 3306. A separate volume (``db_home_mysql``) keeps MySQL data isolated from the Postgres ``db_home`` volume, so switching between the two with ``-f`` flag toggles doesn't corrupt either side. The Postgres-specific init scripts under ``docker/docker-entrypoint-initdb.d/`` are not mounted on the MySQL service (they are postgres-only). Examples / cypress fixtures still load via ``superset-init``'s post-startup steps, which run ``superset load-examples`` against whichever metadata DB is in use. Co-Authored-By: Claude Opus 4.7 (1M context) --- docker-compose-mysql.yml | 93 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 93 insertions(+) create mode 100644 docker-compose-mysql.yml diff --git a/docker-compose-mysql.yml b/docker-compose-mysql.yml new file mode 100644 index 000000000000..4617eaaf0e2e --- /dev/null +++ b/docker-compose-mysql.yml @@ -0,0 +1,93 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +# +# Compose override that swaps the default Postgres metadata DB for MySQL 8. +# Useful for evaluating dialect-specific behaviour (e.g., DDL-migration +# cost on a deployment whose production metadata DB is MySQL). +# +# Usage: +# docker compose -f docker-compose.yml -f docker-compose-mysql.yml up +# docker compose -f docker-compose.yml -f docker-compose-mysql.yml down +# +# To switch back to Postgres, just drop the second `-f` flag — the MySQL +# data lives in a separate volume (`db_home_mysql`) so neither side is +# corrupted by switching dialects. +# +# Notes: +# - Mirrors the connection settings used by CI's `test-mysql` shard: +# dialect ``mysql+mysqldb``, charset utf8mb4 with binary_prefix. +# - Host port 13306 (configurable via DATABASE_PORT_MYSQL) to avoid +# colliding with a native MySQL install on 3306. +# - The Postgres-specific init scripts under +# docker/docker-entrypoint-initdb.d/ are not mounted (they are +# postgres-only); examples / cypress fixtures still load via +# `superset-init`'s post-startup steps. + +# Shared environment override applied to every Superset-side service that +# connects to the metadata DB. ``environment:`` takes precedence over the +# values inherited from the env_file in docker-compose.yml. +x-mysql-env: &mysql-env + DATABASE_DIALECT: mysql+mysqldb + DATABASE_HOST: db + DATABASE_PORT: "3306" + DATABASE_DB: superset + DATABASE_USER: superset + DATABASE_PASSWORD: superset + SQLALCHEMY_DATABASE_URI: "mysql+mysqldb://superset:superset@db:3306/superset?charset=utf8mb4&binary_prefix=true" + +services: + db: + image: mysql:8.0 + environment: + MYSQL_DATABASE: superset + MYSQL_USER: superset + MYSQL_PASSWORD: superset + MYSQL_ROOT_PASSWORD: root + ports: + - "127.0.0.1:${DATABASE_PORT_MYSQL:-13306}:3306" + volumes: + - db_home_mysql:/var/lib/mysql + command: + - --default-authentication-plugin=caching_sha2_password + - --character-set-server=utf8mb4 + - --collation-server=utf8mb4_0900_ai_ci + healthcheck: + test: ["CMD-SHELL", "mysqladmin ping -h localhost -uroot -proot --silent"] + interval: 5s + timeout: 5s + retries: 20 + + superset: + environment: *mysql-env + + superset-init: + environment: *mysql-env + + superset-worker: + environment: *mysql-env + + superset-worker-beat: + environment: *mysql-env + + superset-node: + environment: *mysql-env + + superset-tests-worker: + environment: *mysql-env + +volumes: + db_home_mysql: From 6b59619f369eddf5cb631196d76bd04843a36c37 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 7 May 2026 11:53:57 -0600 Subject: [PATCH 12/35] fix(docker): MySQL examples DB + EXAMPLES_PORT override (sc-105349) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Fix two follow-on issues reported when starting the dev stack with docker-compose-mysql.yml: 1. ``superset-init`` step 4 (load-examples) fails with ``MySQLdb.OperationalError: (2002, "Can't connect to server on 'db'")`` because the analytics-examples DB connection inherits ``EXAMPLES_PORT=5432`` (Postgres port) from ``docker/.env``. The override flipped ``DATABASE_DIALECT`` to ``mysql+mysqldb`` but left the EXAMPLES_* group on Postgres defaults, so the URI became ``mysql+mysqldb://examples:examples@db:5432/examples`` — MySQL container has no listener on 5432. Fix: add ``EXAMPLES_HOST/PORT/DB/USER/PASSWORD`` and a complete ``SUPERSET__SQLALCHEMY_EXAMPLES_URI`` to the ``mysql-env`` anchor. 2. The Postgres init scripts under ``docker/docker-entrypoint-initdb.d/`` (``cypress-init.sh``, ``examples-init.sh``) get mounted on the MySQL container too — compose merges volume lists. They invoke ``psql`` which doesn't exist in the MySQL image, abort with ``psql: command not found``, and prevent the ``examples`` DB from being created. Fix: add a MySQL-specific init script ``docker/mysql-init/examples-init.sql`` that creates the ``examples`` database and user, and mount it at ``/docker-entrypoint-initdb.d`` in the override. Compose's later-takes-precedence rule on duplicate volume targets displaces the Postgres init dir, so the MySQL container only sees the MySQL-compatible script. (Used a plain duplicate-target mount rather than the ``!override`` tag because pre-commit's ``check-yaml`` doesn't recognize Compose's custom YAML tags.) Recovery for an existing failed MySQL stack: ``docker compose -f docker-compose.yml -f docker-compose-mysql.yml down``, then ``docker volume rm superset_db_home_mysql`` (so the new init script runs on the next fresh boot), then ``up`` again. Co-Authored-By: Claude Opus 4.7 (1M context) --- docker-compose-mysql.yml | 24 ++++++++++++++++++++++ docker/mysql-init/examples-init.sql | 32 +++++++++++++++++++++++++++++ 2 files changed, 56 insertions(+) create mode 100644 docker/mysql-init/examples-init.sql diff --git a/docker-compose-mysql.yml b/docker-compose-mysql.yml index 4617eaaf0e2e..13f4c99236cb 100644 --- a/docker-compose-mysql.yml +++ b/docker-compose-mysql.yml @@ -48,6 +48,17 @@ x-mysql-env: &mysql-env DATABASE_USER: superset DATABASE_PASSWORD: superset SQLALCHEMY_DATABASE_URI: "mysql+mysqldb://superset:superset@db:3306/superset?charset=utf8mb4&binary_prefix=true" + # Override the analytics-examples DB connection too. ``EXAMPLES_PORT`` + # in docker/.env is hardcoded to 5432 (the Postgres port); without + # this override the examples connection would try MySQL on 5432 and + # fail. The examples user/DB are created by docker/mysql-init/ + # examples-init.sql on first MySQL boot. + EXAMPLES_HOST: db + EXAMPLES_PORT: "3306" + EXAMPLES_DB: examples + EXAMPLES_USER: examples + EXAMPLES_PASSWORD: examples + SUPERSET__SQLALCHEMY_EXAMPLES_URI: "mysql+mysqldb://examples:examples@db:3306/examples?charset=utf8mb4&binary_prefix=true" services: db: @@ -57,10 +68,23 @@ services: MYSQL_USER: superset MYSQL_PASSWORD: superset MYSQL_ROOT_PASSWORD: root + # The original 5432 port mapping is harmless on a MySQL container + # (nothing listens on 5432 inside it) but we add 13306->3306 so the + # MySQL port is reachable from the host without colliding with a + # native MySQL on 3306. Compose merges port lists. ports: - "127.0.0.1:${DATABASE_PORT_MYSQL:-13306}:3306" + # Override the init-scripts mount by re-binding the same target path + # to a MySQL-compatible directory. Compose merges volume lists by + # target path; later definitions win on conflict, so this displaces + # the Postgres-specific ``./docker/docker-entrypoint-initdb.d`` mount + # from docker-compose.yml. Without this, MySQL would try to run + # ``cypress-init.sh`` (which invokes ``psql``, not in the MySQL + # image), abort the init phase, and never create the ``examples`` + # database. Add the MySQL data volume separately. volumes: - db_home_mysql:/var/lib/mysql + - ./docker/mysql-init:/docker-entrypoint-initdb.d command: - --default-authentication-plugin=caching_sha2_password - --character-set-server=utf8mb4 diff --git a/docker/mysql-init/examples-init.sql b/docker/mysql-init/examples-init.sql new file mode 100644 index 000000000000..68dabe38671d --- /dev/null +++ b/docker/mysql-init/examples-init.sql @@ -0,0 +1,32 @@ +-- Licensed to the Apache Software Foundation (ASF) under one +-- or more contributor license agreements. See the NOTICE file +-- distributed with this work for additional information +-- regarding copyright ownership. The ASF licenses this file +-- to you under the Apache License, Version 2.0 (the +-- "License"); you may not use this file except in compliance +-- with the License. You may obtain a copy of the License at +-- +-- http://www.apache.org/licenses/LICENSE-2.0 +-- +-- Unless required by applicable law or agreed to in writing, +-- software distributed under the License is distributed on an +-- "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +-- KIND, either express or implied. See the License for the +-- specific language governing permissions and limitations +-- under the License. + +-- MySQL counterpart to docker/docker-entrypoint-initdb.d/examples-init.sh. +-- Creates the analytics-examples database and user that Superset's +-- ``load-examples`` command writes to. Mounted by docker-compose-mysql.yml +-- at /docker-entrypoint-initdb.d/ so the MySQL image's first-boot +-- entrypoint runs it automatically. (The Postgres init scripts under +-- docker/docker-entrypoint-initdb.d/ are NOT mounted on the MySQL +-- service — they invoke psql, which doesn't exist in the MySQL image.) + +CREATE DATABASE IF NOT EXISTS examples + CHARACTER SET utf8mb4 + COLLATE utf8mb4_0900_ai_ci; + +CREATE USER IF NOT EXISTS 'examples'@'%' IDENTIFIED BY 'examples'; +GRANT ALL PRIVILEGES ON examples.* TO 'examples'@'%'; +FLUSH PRIVILEGES; From 0107d7df618d345dcbf78c4c22375c6302b4b7ac Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 7 May 2026 13:35:59 -0600 Subject: [PATCH 13/35] build(scripts): add stress-test data generator for migration timing MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Add ``scripts/seed_junction_load.py``, a backend-agnostic script that bulk-inserts synthetic parent rows (dashboards, slices, users, roles, tables, dbs) and many-to-many junction rows for the four largest association tables targeted by the composite-PK migration: ``dashboard_slices``, ``slice_user``, ``dashboard_user``, ``dashboard_roles``. Designed for measuring migration runtime at varying scales — run with a series of size flags (100K / 1M / 5M / 10M for the target table) and time the migration at each scale to verify the predicted ``O(N log N)`` extrapolation against real numbers. Properties: - **Reproducible**: deterministic cross-product walk through parent IDs produces a stable pair sequence; re-running is replayable. - **Idempotent**: re-running with the same target is a no-op; with a higher target, only new rows are added. - **Backend-agnostic**: connects via Superset's standard ``DATABASE_*`` env vars (or ``SUPERSET__SQLALCHEMY_DATABASE_URI``). Branches on dialect for ``BINARY(16)`` vs ``UUID`` vs TEXT/BLOB UUID columns. - **Batched**: bulk INSERT 10K rows per statement. - **Per-phase timing**: logs elapsed wall time for the parents phase, the junctions phase as a whole, and per junction-table. - **Avoidance set**: loads existing junction pairs into a Python set so re-runs on top of pre-existing data don't collide on the uniqueness constraint. Usage (inside the Superset container): docker exec superset-superset-1 \\ /app/.venv/bin/python /app/scripts/seed_junction_load.py \\ --dashboard-slices 1000000 Defaults target a "large multi-team install" shape: 1M ``dashboard_slices``, 100K each ``slice_user`` / ``dashboard_user``, 10K ``dashboard_roles``. Override per-table via flags. Tested locally on MySQL (the user's current eval stack): - 200/100/100/50 row mini-run produced expected counts. - Re-running at the same target is a no-op (idempotent). - ``--dry-run`` plans without writing. Junction tables not yet covered (``sqlatable_user``, ``rls_filter_*``, ``report_schedule_user``) are typically small in production and require additional parent seeding (RLS filters, report schedules) that wasn't worth the scope here. Adding them is straightforward by extending ``JUNCTIONS`` and writing the corresponding parent seeder. Co-Authored-By: Claude Opus 4.7 (1M context) --- scripts/seed_junction_load.py | 567 ++++++++++++++++++++++++++++++++++ 1 file changed, 567 insertions(+) create mode 100644 scripts/seed_junction_load.py diff --git a/scripts/seed_junction_load.py b/scripts/seed_junction_load.py new file mode 100644 index 000000000000..74a891c5035d --- /dev/null +++ b/scripts/seed_junction_load.py @@ -0,0 +1,567 @@ +#!/usr/bin/env python3 +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +# +# ---------------------------------------------------------------------- +# Stress-test data generator for the composite-PK migration (sc-105349). +# +# Bulk-inserts synthetic parent rows and many-to-many junction rows for +# the eight association tables that the composite-PK migration touches. +# Useful for measuring migration runtime at varying scales — run this at +# 100K / 1M / 5M / 10M rows and time the migration at each scale to +# verify the O(N log N) extrapolation. +# +# Idempotent: rerunning with the same target is a no-op; rerunning with +# a higher target adds rows up to the new total. Batched bulk INSERTs +# (10K rows per statement) make it fast on Postgres, MySQL, and SQLite. +# +# Usage (inside the Superset container): +# +# docker exec superset-superset-1 \\ +# /app/.venv/bin/python /app/scripts/seed_junction_load.py \\ +# --dashboard-slices 1000000 \\ +# --slice-user 100000 \\ +# --dashboard-user 100000 +# +# Run with no flags for the defaults shown below. Use ``--dry-run`` to +# print the planned inserts without writing anything. +# +# The script connects via Superset's standard ``DATABASE_*`` env vars +# (or ``SUPERSET__SQLALCHEMY_DATABASE_URI`` if set), so it works +# automatically inside the Superset container regardless of which +# metadata DB backend is in use. + +from __future__ import annotations + +import argparse +import logging +import os +import sys +import time +from contextlib import contextmanager +from typing import Iterator +from uuid import uuid4 + +import sqlalchemy as sa +from sqlalchemy.engine import Connection, Engine + +logger = logging.getLogger("seed_junction_load") + +# Bulk INSERT batch size. Larger values = fewer statements but more memory. +BATCH = 10_000 + +# Default per-junction-table target row counts. Tuned to mimic the shape +# of a large multi-team Superset install. Override via CLI flags. +DEFAULTS: dict[str, int] = { + "dashboard_slices": 1_000_000, + "slice_user": 100_000, + "dashboard_user": 100_000, + "dashboard_roles": 10_000, +} + +# (junction_table, fk1_col, fk2_col, parent1_table, parent2_table) +# parents reference id columns; we generate (fk1, fk2) pairs by sampling +# from the parents' existing IDs. +JUNCTIONS: list[tuple[str, str, str, str, str]] = [ + ("dashboard_slices", "dashboard_id", "slice_id", "dashboards", "slices"), + ("slice_user", "user_id", "slice_id", "ab_user", "slices"), + ("dashboard_user", "user_id", "dashboard_id", "ab_user", "dashboards"), + ("dashboard_roles", "dashboard_id", "role_id", "dashboards", "ab_role"), +] + + +# ---------------------------------------------------------------------- +# Connection setup +# ---------------------------------------------------------------------- + + +def build_engine() -> Engine: + """Build a SQLAlchemy engine from Superset env vars.""" + if uri := os.environ.get("SUPERSET__SQLALCHEMY_DATABASE_URI"): + logger.info("Using SUPERSET__SQLALCHEMY_DATABASE_URI from env") + return sa.create_engine(uri) + + try: + dialect = os.environ["DATABASE_DIALECT"] + user = os.environ["DATABASE_USER"] + password = os.environ["DATABASE_PASSWORD"] + host = os.environ["DATABASE_HOST"] + port = os.environ["DATABASE_PORT"] + db = os.environ["DATABASE_DB"] + except KeyError as exc: + sys.exit( + f"Missing env var {exc}; either set DATABASE_DIALECT/USER/PASSWORD/" + f"HOST/PORT/DB or SUPERSET__SQLALCHEMY_DATABASE_URI before running." + ) + + uri = f"{dialect}://{user}:{password}@{host}:{port}/{db}" + logger.info( + "Built URI from DATABASE_* env vars (dialect=%s, host=%s)", dialect, host + ) + return sa.create_engine(uri) + + +# ---------------------------------------------------------------------- +# Helpers +# ---------------------------------------------------------------------- + + +def uuid_value(dialect_name: str) -> bytes | str: + """Return a UUID in the form the active dialect expects. + + MySQL stores UUIDs as ``BINARY(16)`` (16 raw bytes); Postgres has a + native ``UUID`` type that accepts strings; SQLite stores them as + BLOB/TEXT and accepts either. Branching here keeps the seed script + backend-agnostic without depending on Superset's custom column types. + """ + if dialect_name.startswith("mysql"): + return uuid4().bytes + return str(uuid4()) + + +@contextmanager +def time_phase(name: str) -> Iterator[None]: + """Log elapsed wall time for a named phase.""" + start = time.monotonic() + logger.info("[%s] starting", name) + try: + yield + finally: + elapsed = time.monotonic() - start + logger.info("[%s] done in %.2fs", name, elapsed) + + +def count_rows(conn: Connection, table: str) -> int: + return conn.scalar(sa.text(f"SELECT COUNT(*) FROM {table}")) or 0 # noqa: S608 + + +def existing_ids(conn: Connection, table: str, limit: int | None = None) -> list[int]: + sql = f"SELECT id FROM {table} ORDER BY id" # noqa: S608 + if limit is not None: + sql += f" LIMIT {limit}" + return [row[0] for row in conn.execute(sa.text(sql))] + + +# ---------------------------------------------------------------------- +# Parent seeders +# +# Each function ensures the named parent table has at least ``target`` +# rows by inserting synthetic ones with minimal-but-valid columns. +# Returns nothing; subsequent code reads back IDs via ``existing_ids``. +# ---------------------------------------------------------------------- + + +def seed_dashboards(conn: Connection, target: int, dry_run: bool) -> None: + current = count_rows(conn, "dashboards") + if current >= target: + logger.info( + "dashboards: %d rows (target %d) — no insert needed", current, target + ) + return + needed = target - current + logger.info("dashboards: %d → %d (+%d)", current, target, needed) + if dry_run: + return + + dialect = conn.engine.dialect.name + sql = sa.text( + "INSERT INTO dashboards (uuid, dashboard_title, slug, published) " + "VALUES (:uuid, :title, :slug, :published)" + ) + for batch_start in range(0, needed, BATCH): + rows = [ + { + "uuid": uuid_value(dialect), + "title": f"seed_dashboard_{current + i}", + "slug": f"seed-dashboard-{current + i}-{uuid4().hex[:8]}", + "published": False, + } + for i in range(batch_start, min(batch_start + BATCH, needed)) + ] + conn.execute(sql, rows) + logger.info(" dashboards: inserted %d / %d", batch_start + len(rows), needed) + + +def seed_dbs(conn: Connection, dry_run: bool) -> int: + """Ensure at least one row exists in ``dbs`` (parent of ``tables``). + Returns the id to use as ``database_id`` when seeding ``tables``.""" + ids = existing_ids(conn, "dbs", limit=1) + if ids: + return ids[0] + if dry_run: + return -1 # placeholder + dialect = conn.engine.dialect.name + logger.info("dbs: inserting one synthetic database (no rows present)") + conn.execute( + sa.text( + "INSERT INTO dbs (uuid, database_name, sqlalchemy_uri, expose_in_sqllab) " + "VALUES (:uuid, :name, :uri, :expose)" + ), + { + "uuid": uuid_value(dialect), + "name": f"seed_db_{uuid4().hex[:8]}", + "uri": "sqlite:///seed.db", + "expose": False, + }, + ) + return existing_ids(conn, "dbs", limit=1)[0] + + +def seed_tables(conn: Connection, target: int, dry_run: bool) -> None: + current = count_rows(conn, "tables") + if current >= target: + logger.info("tables: %d rows (target %d) — no insert needed", current, target) + return + needed = target - current + logger.info("tables: %d → %d (+%d)", current, target, needed) + if dry_run: + return + + database_id = seed_dbs(conn, dry_run=False) + dialect = conn.engine.dialect.name + sql = sa.text( + "INSERT INTO tables (uuid, table_name, database_id) " + "VALUES (:uuid, :name, :db_id)" + ) + for batch_start in range(0, needed, BATCH): + rows = [ + { + "uuid": uuid_value(dialect), + "name": f"seed_table_{current + i}", + "db_id": database_id, + } + for i in range(batch_start, min(batch_start + BATCH, needed)) + ] + conn.execute(sql, rows) + logger.info(" tables: inserted %d / %d", batch_start + len(rows), needed) + + +def seed_slices(conn: Connection, target: int, dry_run: bool) -> None: + current = count_rows(conn, "slices") + if current >= target: + logger.info("slices: %d rows (target %d) — no insert needed", current, target) + return + needed = target - current + logger.info("slices: %d → %d (+%d)", current, target, needed) + if dry_run: + return + + # Slices reference tables.id; ensure at least one ``tables`` row exists + # so the FK is satisfiable (datasource_id is nullable but we set it for + # realism). The migration test doesn't care, but a real Superset that + # re-renders these slices does. + seed_tables(conn, target=1, dry_run=False) + table_id = existing_ids(conn, "tables", limit=1)[0] + dialect = conn.engine.dialect.name + sql = sa.text( + "INSERT INTO slices " + "(uuid, slice_name, datasource_id, datasource_type, viz_type) " + "VALUES (:uuid, :name, :ds_id, :ds_type, :viz)" + ) + for batch_start in range(0, needed, BATCH): + rows = [ + { + "uuid": uuid_value(dialect), + "name": f"seed_slice_{current + i}", + "ds_id": table_id, + "ds_type": "table", + "viz": "table", + } + for i in range(batch_start, min(batch_start + BATCH, needed)) + ] + conn.execute(sql, rows) + logger.info(" slices: inserted %d / %d", batch_start + len(rows), needed) + + +def seed_users(conn: Connection, target: int, dry_run: bool) -> None: + current = count_rows(conn, "ab_user") + if current >= target: + logger.info("ab_user: %d rows (target %d) — no insert needed", current, target) + return + needed = target - current + logger.info("ab_user: %d → %d (+%d)", current, target, needed) + if dry_run: + return + + sql = sa.text( + "INSERT INTO ab_user (first_name, last_name, username, email, active) " + "VALUES (:first, :last, :username, :email, :active)" + ) + for batch_start in range(0, needed, BATCH): + rows = [ + { + "first": "seed", + "last": f"user_{current + i}", + "username": f"seed_user_{current + i}_{uuid4().hex[:8]}", + "email": f"seed_user_{current + i}_{uuid4().hex[:8]}@example.invalid", + "active": True, + } + for i in range(batch_start, min(batch_start + BATCH, needed)) + ] + conn.execute(sql, rows) + logger.info(" ab_user: inserted %d / %d", batch_start + len(rows), needed) + + +def seed_roles(conn: Connection, target: int, dry_run: bool) -> None: + current = count_rows(conn, "ab_role") + if current >= target: + logger.info("ab_role: %d rows (target %d) — no insert needed", current, target) + return + needed = target - current + logger.info("ab_role: %d → %d (+%d)", current, target, needed) + if dry_run: + return + + sql = sa.text("INSERT INTO ab_role (name) VALUES (:name)") + for batch_start in range(0, needed, BATCH): + rows = [ + {"name": f"seed_role_{current + i}_{uuid4().hex[:8]}"} + for i in range(batch_start, min(batch_start + BATCH, needed)) + ] + conn.execute(sql, rows) + logger.info(" ab_role: inserted %d / %d", batch_start + len(rows), needed) + + +# ---------------------------------------------------------------------- +# Junction seeder +# ---------------------------------------------------------------------- + + +def _load_existing_pairs( + conn: Connection, junction: str, fk1_col: str, fk2_col: str +) -> set[tuple[int, int]]: + """Load existing ``(fk1, fk2)`` pairs from a junction table into a set. + + Used so the seeder can skip them when generating new pairs (junction + tables enforce uniqueness on the FK pair). Memory is ~32 bytes/tuple + on CPython, so 10M existing pairs is ~320MB — acceptable for a dev + machine. The junction / column names come from ``JUNCTIONS``, not + user input, so the f-string interpolation is safe. + """ + sql_text = f"SELECT {fk1_col}, {fk2_col} FROM {junction}" # noqa: S608 + return {(row[0], row[1]) for row in conn.execute(sa.text(sql_text))} + + +def _generate_new_pairs( + p1_ids: list[int], + p2_ids: list[int], + existing_pairs: set[tuple[int, int]], +) -> Iterator[tuple[int, int]]: + """Yield ``(fk1, fk2)`` pairs from the parent1 × parent2 cross-product + that are not already in ``existing_pairs``.""" + for fk1 in p1_ids: + for fk2 in p2_ids: + if (fk1, fk2) not in existing_pairs: + yield (fk1, fk2) + + +def seed_junction( + conn: Connection, + junction: str, + fk1_col: str, + fk2_col: str, + parent1: str, + parent2: str, + target: int, + dry_run: bool, +) -> None: + """Bulk-insert junction rows up to ``target`` rows total. + + Generates ``(fk1, fk2)`` pairs by walking the cross-product of + parent1 IDs × parent2 IDs in row-major order, skipping pairs that + already exist. Walking the cross-product deterministically keeps + the script replayable: re-running with the same target is a no-op, + and re-running with a higher target appends new pairs in a stable + order regardless of how many runs preceded. + """ + current = count_rows(conn, junction) + if current >= target: + logger.info( + "%s: %d rows (target %d) — no insert needed", junction, current, target + ) + return + needed = target - current + logger.info("%s: %d → %d (+%d)", junction, current, target, needed) + if dry_run: + return + + p1_ids = existing_ids(conn, parent1) + p2_ids = existing_ids(conn, parent2) + max_pairs = len(p1_ids) * len(p2_ids) + if max_pairs < target: + sys.exit( + f"Cannot reach {target} rows in {junction}: " + f"only {max_pairs} unique pairs available " + f"({len(p1_ids)} × {len(p2_ids)}). " + f"Increase parent targets and rerun." + ) + + existing_pairs: set[tuple[int, int]] = ( + _load_existing_pairs(conn, junction, fk1_col, fk2_col) if current > 0 else set() + ) + if existing_pairs: + logger.info( + " %s: loaded %d existing pairs into avoidance set", + junction, + len(existing_pairs), + ) + + insert_sql = sa.text( + f"INSERT INTO {junction} ({fk1_col}, {fk2_col}) " # noqa: S608 + f"VALUES (:fk1, :fk2)" + ) + + inserted = 0 + batch: list[dict[str, int]] = [] + for fk1, fk2 in _generate_new_pairs(p1_ids, p2_ids, existing_pairs): + batch.append({"fk1": fk1, "fk2": fk2}) + inserted += 1 + if len(batch) == BATCH or inserted == needed: + conn.execute(insert_sql, batch) + logger.info(" %s: inserted %d / %d", junction, inserted, needed) + batch = [] + if inserted == needed: + return + if inserted < needed: + sys.exit( + f"Ran out of unique pairs at {inserted}/{needed} for {junction} " + f"(parents have {len(p1_ids)} × {len(p2_ids)} = {max_pairs} pairs, " + f"{len(existing_pairs)} already present)" + ) + + +# ---------------------------------------------------------------------- +# Orchestration +# ---------------------------------------------------------------------- + + +def required_parent_count(target_pairs: int, other_parent: int) -> int: + """How many rows we need in this parent so that + (this_parent × other_parent) ≥ target_pairs.""" + if other_parent == 0: + # Bootstrapping: assume we'll create at least 1 + other_parent = 1 + return -(-target_pairs // other_parent) # ceil(target_pairs / other_parent) + + +def _compute_parent_requirements(targets: dict[str, int]) -> dict[str, int]: + """For each parent table, return the minimum row count needed so that + parent1 × parent2 ≥ target for every junction it participates in. + + Allocates ceil(sqrt(target)) rows per parent, balanced across the two + parents of each junction. The actual junction seeder will then walk + the cross-product to produce the target number of unique pairs. + """ + parent_req: dict[str, int] = {} + for junction, _, _, p1, p2 in JUNCTIONS: + target = targets.get(junction, 0) + if target == 0: + continue + sqrt_n = int(target**0.5) + 1 + parent_req[p1] = max(parent_req.get(p1, 0), sqrt_n) + parent_req[p2] = max(parent_req.get(p2, 0), sqrt_n) + return parent_req + + +def _seed_parents(conn: Connection, parent_req: dict[str, int], dry_run: bool) -> None: + """Seed parent tables in dependency order: + independent parents (ab_user, ab_role) first, then dashboards / slices / + tables (which transitively depend on dbs, seeded inside seed_tables).""" + if "ab_user" in parent_req: + seed_users(conn, parent_req["ab_user"], dry_run) + if "ab_role" in parent_req: + seed_roles(conn, parent_req["ab_role"], dry_run) + if "dashboards" in parent_req: + seed_dashboards(conn, parent_req["dashboards"], dry_run) + if "slices" in parent_req: + seed_slices(conn, parent_req["slices"], dry_run) + if "tables" in parent_req: + seed_tables(conn, parent_req["tables"], dry_run) + + +def _seed_all_junctions( + conn: Connection, targets: dict[str, int], dry_run: bool +) -> None: + for junction, fk1, fk2, p1, p2 in JUNCTIONS: + target = targets.get(junction, 0) + if target == 0: + continue + with time_phase(f"junction:{junction}"): + seed_junction(conn, junction, fk1, fk2, p1, p2, target, dry_run) + + +def run(targets: dict[str, int], dry_run: bool) -> None: + engine = build_engine() + with engine.begin() as conn: + parent_req = _compute_parent_requirements(targets) + logger.info("Required parent row counts: %s", parent_req) + + with time_phase("parents"): + _seed_parents(conn, parent_req, dry_run) + + with time_phase("junctions"): + _seed_all_junctions(conn, targets, dry_run) + + +# ---------------------------------------------------------------------- +# CLI +# ---------------------------------------------------------------------- + + +def main() -> None: + parser = argparse.ArgumentParser( + description=__doc__, + formatter_class=argparse.RawDescriptionHelpFormatter, + ) + for table, default in DEFAULTS.items(): + parser.add_argument( + f"--{table.replace('_', '-')}", + type=int, + default=default, + help=f"target row count for {table} (default: {default:,})", + ) + parser.add_argument( + "--dry-run", + "-n", + action="store_true", + help="print planned inserts without writing to the DB", + ) + parser.add_argument( + "--verbose", + "-v", + action="store_true", + help="increase log verbosity", + ) + args = parser.parse_args() + + logging.basicConfig( + level=logging.DEBUG if args.verbose else logging.INFO, + format="%(asctime)s [%(levelname)s] %(message)s", + datefmt="%H:%M:%S", + ) + + targets = {table: getattr(args, table) for table in DEFAULTS} + + logger.info("Targets: %s", targets) + logger.info("Dry run: %s", args.dry_run) + + with time_phase("total"): + run(targets, dry_run=args.dry_run) + + +if __name__ == "__main__": + main() From dd9487d944e82c318b1513539167ecd1f62f2cbe Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 7 May 2026 14:17:03 -0600 Subject: [PATCH 14/35] feat(scripts): add --dirty-duplicates-pct to seed_junction_load.py MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Extends the stress-test seed script with an optional duplicate-row injection step, used to measure the empirical cost of the migration's ``_dedupe_by_min_id`` phase. Usage: after running the normal seed at a given scale, add ``--dirty-duplicates-pct 5`` (or any non-zero value) to inject that percentage of duplicate ``(fk1, fk2)`` rows into each non-UNIQUE junction (slice_user, dashboard_user, dashboard_roles — dashboard_slices is skipped because its UNIQUE constraint, present both pre- and post-migration, rejects duplicates). Pre-condition: requires the DB to be at the pre-migration revision (33d7e0e21daa). The post-migration composite PK rejects duplicates, so attempting to inject on the upgraded schema errors out. Empirical result on MySQL @ 10M dashboard_slices + ~2.1M other junction rows + 105K injected duplicates (5% on the 3 non-UNIQUE tables): Upgrade time: 1m 36s vs clean baseline 1m 37s → dedupe cost is within measurement noise; the table-scan that the migration already performs dominates whether or not duplicates exist. This empirically confirms what the cost-model predicted: the ``_dedupe_by_min_id`` GROUP BY scan is the dominant cost of that phase, and the actual per-duplicate DELETE is negligible. NULL-FK injection deliberately skipped — would require altering the six non-UNIQUE FK columns from NOT NULL back to nullable (the migration's downgrade keeps them NOT NULL by design), which adds per-backend ALTER complexity for a code path that's structurally identical in cost shape (DELETE WHERE col IS NULL is the same scan shape as the dedupe scan). Co-Authored-By: Claude Opus 4.7 (1M context) --- scripts/seed_junction_load.py | 119 +++++++++++++++++++++++++++++++++- 1 file changed, 117 insertions(+), 2 deletions(-) diff --git a/scripts/seed_junction_load.py b/scripts/seed_junction_load.py index 74a891c5035d..cc42a6bfce9c 100644 --- a/scripts/seed_junction_load.py +++ b/scripts/seed_junction_load.py @@ -83,6 +83,11 @@ ("dashboard_roles", "dashboard_id", "role_id", "dashboards", "ab_role"), ] +# Junction tables that originally carried ``UNIQUE(fk1, fk2)`` and therefore +# cannot accept duplicate ``(fk1, fk2)`` pairs even on the pre-migration +# (downgrade) schema. The other JUNCTIONS allow duplicates pre-migration. +JUNCTIONS_WITH_UNIQUE: set[str] = {"dashboard_slices", "report_schedule_user"} + # ---------------------------------------------------------------------- # Connection setup @@ -504,7 +509,95 @@ def _seed_all_junctions( seed_junction(conn, junction, fk1, fk2, p1, p2, target, dry_run) -def run(targets: dict[str, int], dry_run: bool) -> None: +def inject_duplicates( + conn: Connection, + junction: str, + fk1_col: str, + fk2_col: str, + pct: float, + dry_run: bool, +) -> None: + """Insert duplicate ``(fk1, fk2)`` rows on a non-UNIQUE junction table. + + Used to stress-test the migration's ``_dedupe_by_min_id`` phase, which + is otherwise a no-op on cleanly-seeded data. Computes ``count = + current_rows * pct / 100`` and inserts that many rows by re-sampling + existing ``(fk1, fk2)`` pairs in row-major order. The synthetic + duplicates land on top of distinct existing pairs (one duplicate per + distinct pair, then wraps), so the migration's dedupe finds and + deletes them. + + **Pre-condition: the table must NOT have UNIQUE on (fk1, fk2)**, i.e., + the schema must be the pre-migration shape (after running + ``superset db downgrade``). On the post-migration schema the composite + PK rejects duplicates and this function will error. + """ + if pct == 0: + return + current = count_rows(conn, junction) + count = int(current * pct / 100) + if count == 0: + logger.info( + "%s: 0 duplicates to inject (current=%d, pct=%g)", + junction, + current, + pct, + ) + return + logger.info( + "%s: injecting %d duplicate rows (%g%% of %d existing)", + junction, + count, + pct, + current, + ) + if dry_run: + return + + select_sql = sa.text( + f"SELECT {fk1_col}, {fk2_col} FROM {junction} ORDER BY id LIMIT :n" # noqa: S608 + ) + sample = conn.execute(select_sql, {"n": count}).fetchall() + if not sample: + logger.warning("%s: no rows to duplicate (table is empty)", junction) + return + + insert_sql = sa.text( + f"INSERT INTO {junction} ({fk1_col}, {fk2_col}) " # noqa: S608 + f"VALUES (:fk1, :fk2)" + ) + inserted = 0 + while inserted < count: + batch: list[dict[str, int]] = [] + while len(batch) < BATCH and inserted < count: + row = sample[inserted % len(sample)] + batch.append({"fk1": row[0], "fk2": row[1]}) + inserted += 1 + conn.execute(insert_sql, batch) + logger.info(" %s: injected %d / %d duplicates", junction, inserted, count) + + +def _inject_dirty_data(conn: Connection, dirty_pct: float, dry_run: bool) -> None: + """Inject duplicate rows on every non-UNIQUE seeded junction. + + The two tables that originally carried ``UNIQUE(fk1, fk2)`` are + skipped because their composite-PK successor (and their pre-migration + UNIQUE constraint) both reject duplicate inserts. + """ + if dirty_pct == 0: + return + for junction, fk1, fk2, _, _ in JUNCTIONS: + if junction in JUNCTIONS_WITH_UNIQUE: + logger.info( + "%s: skipping duplicate injection (table has UNIQUE on FK pair)", + junction, + ) + continue + with time_phase(f"dirty:{junction}"): + inject_duplicates(conn, junction, fk1, fk2, dirty_pct, dry_run) + + +def run(targets: dict[str, int], dry_run: bool, dirty_duplicates_pct: float) -> None: engine = build_engine() with engine.begin() as conn: parent_req = _compute_parent_requirements(targets) @@ -516,6 +609,10 @@ def run(targets: dict[str, int], dry_run: bool) -> None: with time_phase("junctions"): _seed_all_junctions(conn, targets, dry_run) + if dirty_duplicates_pct > 0: + with time_phase("dirty-duplicates"): + _inject_dirty_data(conn, dirty_duplicates_pct, dry_run) + # ---------------------------------------------------------------------- # CLI @@ -540,6 +637,19 @@ def main() -> None: action="store_true", help="print planned inserts without writing to the DB", ) + parser.add_argument( + "--dirty-duplicates-pct", + type=float, + default=0, + help=( + "after seeding distinct pairs, inject this percentage of duplicate " + "rows on each non-UNIQUE junction (slice_user, dashboard_user, " + "dashboard_roles). Stress-tests the migration's _dedupe_by_min_id " + "phase. Requires the DB to be at the pre-migration revision " + "(33d7e0e21daa) — the post-migration composite PK rejects " + "duplicates and this will error. Default: 0 (no duplicates)." + ), + ) parser.add_argument( "--verbose", "-v", @@ -558,9 +668,14 @@ def main() -> None: logger.info("Targets: %s", targets) logger.info("Dry run: %s", args.dry_run) + logger.info("Dirty duplicates pct: %g", args.dirty_duplicates_pct) with time_phase("total"): - run(targets, dry_run=args.dry_run) + run( + targets, + dry_run=args.dry_run, + dirty_duplicates_pct=args.dirty_duplicates_pct, + ) if __name__ == "__main__": From ba7a4502a4954044be4ed572b045f688936ad57e Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Wed, 20 May 2026 09:35:52 -0600 Subject: [PATCH 15/35] fix(migration): skip alter_column nullable=False on non-SQLite (sc-105349) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Justin Park (@justinpark) reported on apache/superset#39859: MySQLdb.OperationalError: (1832, "Cannot change column 'dashboard_id': used in a foreign key constraint 'fk_dashboard_roles_dashboard_id_dashboards'") Root cause: ``batch_op.alter_column(fk1, nullable=False)`` for the six non-UNIQUE association tables emits ``ALTER COLUMN`` on a column that participates in an FK constraint. MySQL 8 rejects this with ERROR 1832 when the table has data — even when the change is just ``NULL`` → ``NOT NULL`` and the column is already part of a freshly-added composite primary key (which InnoDB has just made implicitly NOT NULL anyway). The error fires on populated tables only; CI's ``test-mysql`` shard runs against empty tables and so didn't catch this, while a real production-shaped install does. The ``alter_column`` was only ever needed for SQLite, where composite ``PRIMARY KEY`` does not promote constituent columns to ``NOT NULL`` (a long-standing SQLite quirk — only ``INTEGER PRIMARY KEY`` does). PostgreSQL and MySQL implicitly promote PK columns to ``NOT NULL`` as part of ``ADD PRIMARY KEY``, so the explicit step is unnecessary on both — and on MySQL it's actively broken on populated tables. Fix: extract the ``alter_column`` pair into a helper ``_enforce_not_null_for_sqlite()`` that no-ops on Postgres and MySQL. Both branches of the per-table upgrade (the ``recreate="always"`` path for the two UNIQUE-bearing tables, and the direct-ALTER path for the other six) now call the helper instead of inlining the ``alter_column``. Verified end-to-end: downgrade-then-upgrade against MySQL with ~12M total junction rows (10M dashboard_slices + 1M each slice_user/dashboard_user + 100K dashboard_roles) completes in 1m 39s with no ERROR 1832. The 44 in-memory SQLite tests still pass. Considered Justin's alternative (drop FKs on MySQL across all eight tables, unifying the two branches) but rejected as more invasive — it would require capturing FK metadata and explicitly re-creating the FKs for the six non-recreate tables, since they don't go through the ``copy_from`` path that re-creates FKs automatically. The SQLite-only approach is more targeted: it removes the operation that MySQL rejects rather than working around the rejection. Co-Authored-By: Claude Opus 4.7 (1M context) --- ...3611e32_composite_pk_association_tables.py | 46 +++++++++++++------ 1 file changed, 33 insertions(+), 13 deletions(-) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index 055ecd3c9700..c9fceaed0eaf 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -37,6 +37,7 @@ import sqlalchemy as sa from alembic import op +from alembic.operations.base import BatchOperations from sqlalchemy import inspect from sqlalchemy.engine import Connection @@ -263,6 +264,36 @@ def _build_pre_upgrade_table( return sa.Table(t.name, md, *cols) +def _enforce_not_null_for_sqlite( + batch_op: BatchOperations, t: AssociationTable, conn: Connection +) -> None: + """Force ``NOT NULL`` on the FK columns post-PK-promotion on SQLite only. + + SQLite has a long-standing quirk: composite ``PRIMARY KEY`` does not + promote constituent columns to ``NOT NULL`` (only ``INTEGER PRIMARY KEY`` + does). PostgreSQL and MySQL implicitly promote the PK columns to + ``NOT NULL`` when the constraint is added, making the explicit + ``alter_column`` redundant there. + + Skipping the ``alter_column`` on MySQL is also functionally required: + MySQL 8 rejects ``ALTER COLUMN`` on a column that participates in a + foreign key constraint with ``ERROR 1832 (HY000): Cannot change column + 'X': used in a foreign key constraint 'Y'`` whenever the table has + data — even when the only change is ``NULL`` → ``NOT NULL`` and the + column is already part of a freshly-added composite primary key (which + InnoDB has just made implicitly ``NOT NULL`` anyway). The error fires + on populated tables but not on empty ones, which is why CI's + ``test-mysql`` shard (fresh schema) didn't catch this and a real + production-shaped install does. + + Only SQLite still needs the explicit step, and SQLite has no FK + enforcement objection. + """ + if conn.dialect.name == "sqlite": + batch_op.alter_column(t.fk1, existing_type=sa.Integer, nullable=False) + batch_op.alter_column(t.fk2, existing_type=sa.Integer, nullable=False) + + def upgrade() -> None: conn = op.get_bind() _check_no_external_fks_to_id(conn) @@ -310,23 +341,12 @@ def upgrade() -> None: ) as batch_op: batch_op.drop_column("id") batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) - # SQLite quirk: composite PRIMARY KEY does not promote the - # constituent columns to NOT NULL (only ``INTEGER PRIMARY - # KEY`` does). PostgreSQL and MySQL implicitly promote the - # PK columns to NOT NULL when the constraint is added, - # so the explicit ``alter_column`` is a no-op on those - # backends but enforces the post-upgrade contract on - # SQLite. Without it, ``INSERT (NULL, 5)`` would succeed - # on SQLite despite the columns being part of the PK. - batch_op.alter_column(t.fk1, existing_type=sa.Integer, nullable=False) - batch_op.alter_column(t.fk2, existing_type=sa.Integer, nullable=False) + _enforce_not_null_for_sqlite(batch_op, t, conn) else: with op.batch_alter_table(t.name) as batch_op: batch_op.drop_column("id") batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) - # See comment above re: SQLite composite-PK NOT NULL quirk. - batch_op.alter_column(t.fk1, existing_type=sa.Integer, nullable=False) - batch_op.alter_column(t.fk2, existing_type=sa.Integer, nullable=False) + _enforce_not_null_for_sqlite(batch_op, t, conn) def downgrade() -> None: From 550c560df799bf80af4a9f06b5f811b4d4925d49 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Wed, 20 May 2026 14:03:56 -0600 Subject: [PATCH 16/35] fix(migration): address aminghadersohi review feedback (sc-105349) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Three improvements from @aminghadersohi's review on apache/superset#39859: 1. **`fk["name"]` unguarded in ``_downgrade_mysql_table`` re-add loop** The drop loop gates on ``if fk_name := fk.get("name"):`` but the re-add loop accessed ``fk["name"]`` unconditionally in an f-string. MySQL/InnoDB always assigns FK names, so this branch was defensive, but the asymmetry was confusing. Symmetrized via ``continue`` at the top of the re-add loop. 2. **``ondelete`` whitelist before raw-SQL interpolation** The value comes from MySQL's ``information_schema`` (not user input), but interpolating a reflected string into raw SQL without a guard left a "what if an unexpected value appears" footgun. Added ``_VALID_ONDELETE_ACTIONS`` (the four SQL-standard actions) and a ``RuntimeError`` when an unexpected value is reflected. 3. **Direct ALTER on PostgreSQL for tables with pre-existing UNIQUE** ``recreate="always"`` is dialect-agnostic — on PostgreSQL it triggers ``CREATE TABLE AS SELECT → DROP → RENAME`` holding ``ACCESS EXCLUSIVE`` for the full table-copy duration. For a multi-million-row ``dashboard_slices``, that lock window can be noticeable. The reflected UNIQUE constraint has a stable name on PostgreSQL (default ``
__key`` convention), so dropping it directly and then running structural change as direct ALTER avoids the copy entirely. The reflected UNIQUE name is wrapped in a new ``_drop_redundant_unique_by_name()`` helper. Postgres takes the direct path; MySQL keeps ``recreate="always"`` because InnoDB binds FKs to the UNIQUE's underlying index for back-reference (``DROP CONSTRAINT`` on the UNIQUE there raises ``ERROR 1553``); SQLite keeps ``recreate="always"`` because unnamed UNIQUEs reflect with ``name=None`` and can't be dropped by name. Verified end-to-end: downgrade-then-upgrade against MySQL with ~12M total junction rows seeded completes in ~1m 41s (within the range of the prior measurements). Co-Authored-By: Claude Opus 4.7 (1M context) --- ...3611e32_composite_pk_association_tables.py | 131 +++++++++++++----- 1 file changed, 98 insertions(+), 33 deletions(-) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index c9fceaed0eaf..5ecbb54b4b9f 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -264,6 +264,40 @@ def _build_pre_upgrade_table( return sa.Table(t.name, md, *cols) +def _drop_redundant_unique_by_name( + conn: Connection, insp: sa.engine.reflection.Inspector, t: AssociationTable +) -> None: + """Drop the redundant ``UNIQUE(fk1, fk2)`` constraint by its reflected + name on PostgreSQL / MySQL. + + The two tables in ``TABLES_WITH_PRE_EXISTING_UNIQUE`` carry a UNIQUE + constraint that the composite primary key subsumes. PostgreSQL and + MySQL both auto-name UNIQUE constraints (``
__key`` on + Postgres, ``
__`` or the explicit ``uq_*`` we may have + given it on MySQL), so they're reflectable by name. SQLite is + handled separately via ``recreate="always"`` + ``copy_from`` because + it reflects unnamed UNIQUEs with ``name=None``. + + No-op if no matching UNIQUE is found (defensive — re-runs after a + partial application should not error). + """ + for uc in insp.get_unique_constraints(t.name): + if set(uc.get("column_names", [])) == {t.fk1, t.fk2} and uc.get("name"): + op.drop_constraint(uc["name"], t.name, type_="unique") + return + + +# MySQL ON DELETE actions that the downgrade re-create loop is allowed +# to interpolate into raw SQL. The reflected value comes from MySQL's +# information_schema (so not user input), but a whitelist eliminates +# the "what if an unexpected value appears" question entirely. The +# four entries are the SQL-standard set; SET DEFAULT is intentionally +# excluded because InnoDB silently downgrades it to NO ACTION. +_VALID_ONDELETE_ACTIONS: frozenset[str] = frozenset( + {"CASCADE", "SET NULL", "RESTRICT", "NO ACTION"} +) + + def _enforce_not_null_for_sqlite( batch_op: BatchOperations, t: AssociationTable, conn: Connection ) -> None: @@ -309,39 +343,53 @@ def upgrade() -> None: _dedupe_by_min_id(conn, t) _assert_no_duplicates(conn, t) - # For the two tables with a pre-existing redundant UNIQUE - # (``dashboard_slices``, ``report_schedule_user``) build an explicit - # ``copy_from`` Table that omits the UNIQUE; this deterministically - # drops it across all dialects, including SQLite where unnamed - # constraints reflect with ``name=None`` and can't be dropped by - # name. For the other six tables, reflection-based default - # ``batch_alter_table`` (auto-detect) is fine since there's no - # UNIQUE to drop. On PostgreSQL/MySQL, direct ALTER avoids the - # temp-table index-name collision; on SQLite, the auto-detect picks - # ``recreate=True`` because PK changes need it. + # Two tables (``dashboard_slices``, ``report_schedule_user``) + # carry a redundant ``UNIQUE(fk1, fk2)`` that the composite PK + # subsumes. Three dialect-specific paths: + # + # * **PostgreSQL** — the UNIQUE constraint has a stable + # reflected name (Postgres default convention), so we + # ``DROP CONSTRAINT`` by name and then run the structural + # change as direct ALTER. This avoids the full-table copy + # that ``recreate="always"`` would trigger + # (``CREATE TABLE AS SELECT → DROP → RENAME``), holding + # ``ACCESS EXCLUSIVE`` only for the (much shorter) PK + # index build instead of the full copy duration. + # + # * **MySQL** — InnoDB binds the FK constraints to the + # redundant UNIQUE's underlying index for back-reference, + # so a direct ``DROP CONSTRAINT`` of the UNIQUE raises + # ``ERROR 1553``. Use ``recreate="always"`` to rebuild the + # table without the UNIQUE; drop the FKs first to dodge + # the ``ERROR 1826`` (duplicate FK constraint name) that + # the temp-table phase would otherwise provoke. The FKs + # are re-created automatically as part of ``copy_from``. + # + # * **SQLite** — unnamed UNIQUE constraints reflect with + # ``name=None`` and can't be dropped by name. Use + # ``recreate="always"`` + ``copy_from`` (omits UNIQUE). + # SQLite always rebuilds for PK changes anyway, so the + # recreate isn't extra cost there. if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: - # MySQL ERROR 1826: foreign-key constraint names are unique - # per-database, not per-table. ``recreate="always"`` builds - # ``_alembic_tmp_
`` with the original FK names from - # ``copy_from``, but the original table still holds those - # names until it's dropped, which fails on MySQL with - # ``Duplicate foreign key constraint name``. PostgreSQL and - # SQLite scope FK names per-table, so the recreate path - # works there as-is. Drop the original FKs by name first - # on MySQL; ``copy_from`` re-creates them on the rebuilt - # table with their original names. - if conn.dialect.name == "mysql": - for fk in insp.get_foreign_keys(t.name): - if fk_name := fk.get("name"): - op.drop_constraint(fk_name, t.name, type_="foreignkey") - with op.batch_alter_table( - t.name, - recreate="always", - copy_from=_build_pre_upgrade_table(insp, t), - ) as batch_op: - batch_op.drop_column("id") - batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) - _enforce_not_null_for_sqlite(batch_op, t, conn) + if conn.dialect.name == "postgresql": + _drop_redundant_unique_by_name(conn, insp, t) + with op.batch_alter_table(t.name) as batch_op: + batch_op.drop_column("id") + batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) + _enforce_not_null_for_sqlite(batch_op, t, conn) + else: + if conn.dialect.name == "mysql": + for fk in insp.get_foreign_keys(t.name): + if fk_name := fk.get("name"): + op.drop_constraint(fk_name, t.name, type_="foreignkey") + with op.batch_alter_table( + t.name, + recreate="always", + copy_from=_build_pre_upgrade_table(insp, t), + ) as batch_op: + batch_op.drop_column("id") + batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) + _enforce_not_null_for_sqlite(batch_op, t, conn) else: with op.batch_alter_table(t.name) as batch_op: batch_op.drop_column("id") @@ -453,14 +501,31 @@ def _downgrade_mysql_table( ) for fk in fks: + # Guard the FK name for symmetry with the drop loop above. + # MySQL/InnoDB always reflects a name for FK constraints + # (auto-assigning ``
_ibfk_`` if none was specified), + # so this branch is defensive rather than reachable in practice. + fk_name = fk.get("name") + if not fk_name: + continue ondelete = fk.get("options", {}).get("ondelete") + # Defensive whitelist: ``ondelete`` is reflected from MySQL's + # information_schema (not user input), but interpolating it + # into raw SQL without a check leaves a "what if an + # unexpected value appears" footgun. The SQL standard defines + # exactly four actions; reject anything else loudly. + if ondelete and ondelete.upper() not in _VALID_ONDELETE_ACTIONS: + raise RuntimeError( + f"Unexpected ON DELETE action {ondelete!r} reflected from " + f"{t.name}.{fk_name}; refusing to interpolate into raw SQL." + ) ondelete_clause = f" ON DELETE {ondelete}" if ondelete else "" local_cols = ", ".join(f"`{c}`" for c in fk["constrained_columns"]) ref_cols = ", ".join(f"`{c}`" for c in fk["referred_columns"]) op.execute( f""" ALTER TABLE `{t.name}` - ADD CONSTRAINT `{fk["name"]}` + ADD CONSTRAINT `{fk_name}` FOREIGN KEY ({local_cols}) REFERENCES `{fk["referred_table"]}` ({ref_cols}) {ondelete_clause} From 65e5efe84db3cf15f5cf75b4c6475ea9269f2a78 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Tue, 2 Jun 2026 14:43:12 -0600 Subject: [PATCH 17/35] fix(migration): allowlist guard on _downgrade_mysql_table MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Belt-and-braces invariant: ``t.name`` is interpolated as a backtick-quoted identifier into the ALTER statements emitted by ``_downgrade_mysql_table``. The values originate from ``AFFECTED_TABLES`` (a module-level literal), so SQL injection is already structurally precluded at the call site. Adding an explicit ``allowed = {a.name for a in AFFECTED_TABLES}`` membership check makes that invariant load-bearing rather than implicit — a future refactor that loosens the call-site can't slip past review. Surfaced during a downstream SQLAlchemy review on the entity-versioning branch that stacks on top of this one; lifted onto sc-105349 because the patch is properly scoped to this branch's composite-PK migration. --- ...2bee73611e32_composite_pk_association_tables.py | 14 ++++++++++++++ 1 file changed, 14 insertions(+) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index 5ecbb54b4b9f..384f180a84d3 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -475,7 +475,21 @@ def _downgrade_mysql_table( for the combined-ALTER form, and the constitution allows raw SQL for dialect-specific DDL with no programmatic equivalent (preferring triple-quoted strings for legibility). + + Belt-and-braces guard: ``t.name`` is interpolated as a backtick-quoted + identifier in the ALTER statements below. The value comes from + ``AFFECTED_TABLES`` (a module-level literal), so SQL injection is + structurally precluded. The explicit ``allowed`` check here makes + that invariant load-bearing rather than implicit, so a future + refactor that loosens the call-site can't slip past review. """ + allowed = {a.name for a in AFFECTED_TABLES} + if t.name not in allowed: + raise RuntimeError( + f"Refusing to ALTER unknown table {t.name!r}: " + f"only AFFECTED_TABLES entries may flow through this path." + ) + fks = insp.get_foreign_keys(t.name) for fk in fks: From 519c6283ef384e09b89fd35b6eeb48b663bb23a7 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 11 Jun 2026 16:04:17 -0600 Subject: [PATCH 18/35] fix(versioning): re-point composite-PK migration at master's head After rebasing onto master, 2bee73611e32 and master's 31dae2559c05 both revised 33d7e0e21daa, forking the alembic chain into two heads ('superset db upgrade' refuses to run). Re-point down_revision at 31dae2559c05 so the versioning chain extends the real head. Co-Authored-By: Claude Opus 4.8 (1M context) --- ...-05-01_23-36_2bee73611e32_composite_pk_association_tables.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index 384f180a84d3..a0c8ce867bf8 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -43,7 +43,7 @@ # revision identifiers, used by Alembic. revision = "2bee73611e32" -down_revision = "33d7e0e21daa" +down_revision = "31dae2559c05" logger = logging.getLogger("alembic.env") From e7c7899464753c1ec465c71f5158152759d51196 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 11 Jun 2026 16:15:17 -0600 Subject: [PATCH 19/35] fix(versioning): capture FK list before dropping in composite-PK upgrade MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The MySQL branch dropped the live FK constraints and then re-reflected them for the copy_from table — which only returned the pre-drop list via the Inspector's per-instance info_cache, an implementation detail. Capture the list before dropping and pass it through explicitly (the downgrade path already did this). Co-Authored-By: Claude Opus 4.8 (1M context) --- ...3611e32_composite_pk_association_tables.py | 24 +++++++++++++++---- 1 file changed, 20 insertions(+), 4 deletions(-) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index a0c8ce867bf8..b8f85a768a8d 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -226,7 +226,9 @@ def _assert_no_duplicates(conn: Connection, t: AssociationTable) -> None: def _build_pre_upgrade_table( - insp: sa.engine.reflection.Inspector, t: AssociationTable + insp: sa.engine.reflection.Inspector, + t: AssociationTable, + fks: list[dict] | None = None, ) -> sa.Table: """Build a ``Table`` object representing the pre-upgrade schema of ``t``, explicitly *without* any redundant ``UniqueConstraint(t.fk1, t.fk2)``. @@ -237,10 +239,19 @@ def _build_pre_upgrade_table( Reflects column types and FK targets (with original FK constraint names preserved) from the live database; only the redundant UNIQUE is omitted. + + *fks* lets a caller pass a pre-captured ``get_foreign_keys`` result. + The MySQL upgrade path drops the live FK constraints before building + this table, so re-reflecting here would only see them via the + Inspector's per-instance ``info_cache`` — an implementation detail, + not a contract. Passing the pre-drop list makes the dependency + explicit instead of relying on reflection caching. """ md = sa.MetaData() + if fks is None: + fks = insp.get_foreign_keys(t.name) fks_for_col: dict[str, list[dict]] = {} - for fk in insp.get_foreign_keys(t.name): + for fk in fks: for col_name in fk["constrained_columns"]: fks_for_col.setdefault(col_name, []).append(fk) @@ -378,14 +389,19 @@ def upgrade() -> None: batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) _enforce_not_null_for_sqlite(batch_op, t, conn) else: + # Capture the FK list BEFORE dropping: the copy_from table + # below must embed these constraints, and re-reflecting + # after the drop only works via the Inspector's + # per-instance info_cache (see _build_pre_upgrade_table). + pre_drop_fks = insp.get_foreign_keys(t.name) if conn.dialect.name == "mysql": - for fk in insp.get_foreign_keys(t.name): + for fk in pre_drop_fks: if fk_name := fk.get("name"): op.drop_constraint(fk_name, t.name, type_="foreignkey") with op.batch_alter_table( t.name, recreate="always", - copy_from=_build_pre_upgrade_table(insp, t), + copy_from=_build_pre_upgrade_table(insp, t, fks=pre_drop_fks), ) as batch_op: batch_op.drop_column("id") batch_op.create_primary_key(f"pk_{t.name}", [t.fk1, t.fk2]) From 536ebfd0d8e8e32af5fffe12efa193629e678a97 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 11 Jun 2026 17:28:50 -0600 Subject: [PATCH 20/35] fix(migration): resumable MySQL upgrade + stale revision references MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two fixes from a 4-lens review pass: - Resumability guard: on MySQL every DDL statement auto-commits, so a failure at table N of 8 left tables 1..N-1 converted with alembic_version un-stamped — re-running failed at table 1 (drop_column('id') on a converted table) and downgrade couldn't run either. Skip tables whose id column is already gone, making re-runs safe on every dialect. - The down_revision re-point left two stale 33d7e0e21daa references: the migration docstring header, and — operationally worse — the seed script's --dirty-duplicates-pct help text, which instructed a downgrade that would unwind every migration since 2025-11. The help text now points at the migration's down_revision instead of hardcoding a hash. Also: drop the never-called required_parent_count helper and trim report_schedule_user from JUNCTIONS_WITH_UNIQUE (the script never seeds that table; the entry implied coverage that doesn't exist). Co-Authored-By: Claude Opus 4.8 (1M context) --- scripts/seed_junction_load.py | 23 ++++++++----------- ...3611e32_composite_pk_association_tables.py | 19 ++++++++++++++- 2 files changed, 28 insertions(+), 14 deletions(-) diff --git a/scripts/seed_junction_load.py b/scripts/seed_junction_load.py index cc42a6bfce9c..4e4e5d6b42c1 100644 --- a/scripts/seed_junction_load.py +++ b/scripts/seed_junction_load.py @@ -86,7 +86,11 @@ # Junction tables that originally carried ``UNIQUE(fk1, fk2)`` and therefore # cannot accept duplicate ``(fk1, fk2)`` pairs even on the pre-migration # (downgrade) schema. The other JUNCTIONS allow duplicates pre-migration. -JUNCTIONS_WITH_UNIQUE: set[str] = {"dashboard_slices", "report_schedule_user"} +# Only ``dashboard_slices`` is listed: the migration's other UNIQUE table +# (``report_schedule_user``) is not in JUNCTIONS — this script doesn't seed +# it — so listing it here would imply coverage that doesn't exist. Add it +# alongside a JUNCTIONS entry if that table ever gets seeded. +JUNCTIONS_WITH_UNIQUE: set[str] = {"dashboard_slices"} # ---------------------------------------------------------------------- @@ -454,15 +458,6 @@ def seed_junction( # ---------------------------------------------------------------------- -def required_parent_count(target_pairs: int, other_parent: int) -> int: - """How many rows we need in this parent so that - (this_parent × other_parent) ≥ target_pairs.""" - if other_parent == 0: - # Bootstrapping: assume we'll create at least 1 - other_parent = 1 - return -(-target_pairs // other_parent) # ceil(target_pairs / other_parent) - - def _compute_parent_requirements(targets: dict[str, int]) -> dict[str, int]: """For each parent table, return the minimum row count needed so that parent1 × parent2 ≥ target for every junction it participates in. @@ -645,9 +640,11 @@ def main() -> None: "after seeding distinct pairs, inject this percentage of duplicate " "rows on each non-UNIQUE junction (slice_user, dashboard_user, " "dashboard_roles). Stress-tests the migration's _dedupe_by_min_id " - "phase. Requires the DB to be at the pre-migration revision " - "(33d7e0e21daa) — the post-migration composite PK rejects " - "duplicates and this will error. Default: 0 (no duplicates)." + "phase. Requires 2bee73611e32 to NOT be applied: un-apply it by " + "downgrading to its parent (`superset db downgrade `, " + "where is read from the 2bee73611e32 migration " + "file) — the post-migration composite PK rejects duplicates and " + "this will error. Default: 0 (no duplicates)." ), ) parser.add_argument( diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index b8f85a768a8d..9d7035a06717 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -27,7 +27,7 @@ of the eight tables lacked DB-level uniqueness. Revision ID: 2bee73611e32 -Revises: 33d7e0e21daa +Revises: 31dae2559c05 Create Date: 2026-05-01 23:36:34.050058 """ @@ -345,6 +345,23 @@ def upgrade() -> None: insp = inspect(conn) for t in AFFECTED_TABLES: + # Resumability guard: on MySQL every DDL statement auto-commits, so + # a failure at table N of 8 leaves tables 1..N-1 already converted + # while ``alembic_version`` is still un-stamped. Without this guard + # a re-run would fail at table 1 (``drop_column("id")`` on a table + # that no longer has ``id``), and ``downgrade`` can't run either + # (the revision was never stamped) — recovery would need manual + # surgery. A converted table is identified by the absent ``id`` + # column; skipping it makes re-running the upgrade safe on every + # dialect (Postgres/SQLite wrap the migration in a transaction, so + # the guard is simply never hit there). + if "id" not in {c["name"] for c in insp.get_columns(t.name)}: + logger.info( + "%s: already converted (no surrogate id column); skipping", + t.name, + ) + continue + # Run NULL-FK cleanup unconditionally: it is a no-op DELETE on tables # whose FK columns are already NOT NULL (cheap), and skipping it on a # table whose FK was nullable would leave the PK-add to fail with a From 061b31604e519f19c5ce878bed84bef95f7fd76b Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 11 Jun 2026 17:28:50 -0600 Subject: [PATCH 21/35] test(migration): cover the composite-PK data-cleanup paths MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The migration's riskiest half — _delete_null_fk_rows / _dedupe_by_min_id / _assert_no_duplicates — had zero coverage: the fixtures seeded no rows, and both test schema builders created FK columns NOT NULL, diverging from the real pre-migration shape (six of eight tables allowed NULLs), so test_fk_columns_not_null passed trivially. - Build the pre-migration schema with historically-accurate nullable FKs (keyed on the migration's TABLES_WITH_NULLABLE_FKS, giving that documentation set a load-bearing consumer). - Add test_upgrade_scrubs_null_fks_and_duplicates: seeds NULL-FK rows and duplicate pairs, runs upgrade, asserts exactly the distinct non-NULL pairs survive. Verified deletable-detectable: commenting out _dedupe_by_min_id makes it fail. - Delete the permanently-skipped placeholder test and the captured-but-never-asserted pre_shape; replace spec-kit references (T034a/tasks.md/quickstart.md) with self-contained prose. Co-Authored-By: Claude Opus 4.8 (1M context) --- .../composite_pk_association_tables__tests.py | 24 +++--- .../composite_pk_round_trip__tests.py | 74 +++++++++++++------ 2 files changed, 68 insertions(+), 30 deletions(-) diff --git a/tests/integration_tests/migrations/composite_pk_association_tables__tests.py b/tests/integration_tests/migrations/composite_pk_association_tables__tests.py index 52b1942bdb24..05097ef308d6 100644 --- a/tests/integration_tests/migrations/composite_pk_association_tables__tests.py +++ b/tests/integration_tests/migrations/composite_pk_association_tables__tests.py @@ -18,14 +18,13 @@ migration (revision 2bee73611e32). Builds the pre-migration shape against an isolated in-memory SQLite engine, -runs the migration's ``upgrade()``, and asserts the resulting shape matches -the data-model.md "After" specification: no ``id`` column, composite PK on -the two FK columns, and no redundant ``UNIQUE(fk1, fk2)`` on the two tables -that previously carried one. +runs the migration's ``upgrade()``, and asserts the resulting shape: no +``id`` column, composite PK on the two FK columns, and no redundant +``UNIQUE(fk1, fk2)`` on the two tables that previously carried one. Continuum-restore verification is OUT OF SCOPE; that work lives in the -versioning epic (sc-103156). Cross-backend verification (PostgreSQL, MySQL) -is handled by the CI matrix (T034a). +entity-versioning follow-up. Cross-backend verification (PostgreSQL, +MySQL) is handled by CI's test-postgres / test-mysql shards. """ from importlib import import_module @@ -43,20 +42,27 @@ ) AFFECTED_TABLES = _migration.AFFECTED_TABLES TABLES_WITH_PRE_EXISTING_UNIQUE = _migration.TABLES_WITH_PRE_EXISTING_UNIQUE +TABLES_WITH_NULLABLE_FKS = _migration.TABLES_WITH_NULLABLE_FKS @pytest.fixture(scope="module") def post_upgrade_engine() -> sa.engine.Engine: """An isolated in-memory SQLite engine with the migration applied to a pre-migration-shaped seed schema. Used by the post-upgrade assertions - below. Module-scoped so the upgrade only runs once per test session.""" + below. Module-scoped so the upgrade only runs once per module. + + FK columns are NULLABLE on the six tables that historically allowed + NULLs — with ``nullable=False`` here, ``test_fk_columns_not_null`` + would pass trivially rather than because the migration promoted + anything.""" engine = sa.create_engine("sqlite:///:memory:") md = sa.MetaData() for t in AFFECTED_TABLES: + nullable = t.name in TABLES_WITH_NULLABLE_FKS cols: list[sa.SchemaItem] = [ sa.Column("id", sa.Integer, primary_key=True), - sa.Column(t.fk1, sa.Integer, nullable=False), - sa.Column(t.fk2, sa.Integer, nullable=False), + sa.Column(t.fk1, sa.Integer, nullable=nullable), + sa.Column(t.fk2, sa.Integer, nullable=nullable), ] constraints: list[sa.SchemaItem] = [] if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: diff --git a/tests/integration_tests/migrations/composite_pk_round_trip__tests.py b/tests/integration_tests/migrations/composite_pk_round_trip__tests.py index d83c9d113c3f..3a010889799f 100644 --- a/tests/integration_tests/migrations/composite_pk_round_trip__tests.py +++ b/tests/integration_tests/migrations/composite_pk_round_trip__tests.py @@ -24,15 +24,14 @@ This is run against an isolated in-memory engine via Alembic's ``MigrationContext`` so the test does not perturb the project's test DB. -Cross-backend verification of the same migration against PostgreSQL and -MySQL is delegated to the CI matrix (see T034a in tasks.md) and to the -quickstart.md verification (T033). This file covers the SQLite slice. +Cross-backend (Postgres/MySQL) verification is handled by CI's +test-postgres / test-mysql shards running ``superset db upgrade``. This +file covers the SQLite slice. """ from importlib import import_module from typing import Any -import pytest import sqlalchemy as sa from alembic.migration import MigrationContext from alembic.operations import Operations @@ -45,20 +44,26 @@ ) AFFECTED_TABLES = _migration.AFFECTED_TABLES TABLES_WITH_PRE_EXISTING_UNIQUE = _migration.TABLES_WITH_PRE_EXISTING_UNIQUE +TABLES_WITH_NULLABLE_FKS = _migration.TABLES_WITH_NULLABLE_FKS def _build_pre_migration_schema(engine: sa.engine.Engine) -> None: """Recreate the eight tables in their pre-migration shape (surrogate ``id INTEGER PRIMARY KEY`` plus an optional ``UNIQUE(fk1, fk2)`` on the - two tables that previously carried one). FKs to parent tables are - omitted to keep the test self-contained — we're testing schema - transformations, not FK enforcement.""" + two tables that previously carried one). FK columns are NULLABLE on + the six tables that historically allowed NULLs — fidelity matters: + with ``nullable=False`` here, the post-upgrade NOT NULL assertions + pass trivially rather than because the migration promoted anything, + and the NULL-row cleanup path can't be exercised. FKs to parent + tables are omitted to keep the test self-contained — we're testing + schema transformations, not FK enforcement.""" md = sa.MetaData() for t in AFFECTED_TABLES: + nullable = t.name in TABLES_WITH_NULLABLE_FKS cols: list[sa.Column] = [ sa.Column("id", sa.Integer, primary_key=True), - sa.Column(t.fk1, sa.Integer, nullable=False), - sa.Column(t.fk2, sa.Integer, nullable=False), + sa.Column(t.fk1, sa.Integer, nullable=nullable), + sa.Column(t.fk2, sa.Integer, nullable=nullable), ] constraints: list[sa.SchemaItem] = [] if t.name in TABLES_WITH_PRE_EXISTING_UNIQUE: @@ -110,8 +115,6 @@ def test_round_trip_against_in_memory_sqlite() -> None: engine = sa.create_engine("sqlite:///:memory:") _build_pre_migration_schema(engine) - pre_shape = {t.name: _shape(engine, t.name) for t in AFFECTED_TABLES} - _run_with_alembic_context(engine, _migration.upgrade) for t in AFFECTED_TABLES: @@ -146,9 +149,45 @@ def test_round_trip_against_in_memory_sqlite() -> None: f"diff: {set(re_upgrade_shape.items()) ^ set(post_upgrade_shape.items())}" ) - # Use pre_shape only to demonstrate it was captured (not asserted against - # because the round-trip downgrade intentionally diverges on FK NOT NULL). - _ = pre_shape + +def test_upgrade_scrubs_null_fks_and_duplicates() -> None: + """The pre-flight data surgery is the migration's riskiest half — and + it must be deletable-detectable: this test fails if + ``_delete_null_fk_rows`` or ``_dedupe_by_min_id`` is removed from + ``upgrade()``. + + Seeds a nullable-FK junction (``slice_user``) with NULL-FK rows and + duplicate ``(fk1, fk2)`` pairs in the true pre-migration shape, runs + the upgrade, and asserts exactly the distinct non-NULL pairs survive + (the composite PK could not even be created otherwise). + """ + engine = sa.create_engine("sqlite:///:memory:") + _build_pre_migration_schema(engine) + + md = sa.MetaData() + slice_user = sa.Table("slice_user", md, autoload_with=engine) + with engine.begin() as conn: + conn.execute( + slice_user.insert(), + [ + {"id": 1, "user_id": 1, "slice_id": 1}, # keeper (MIN id) + {"id": 2, "user_id": 1, "slice_id": 1}, # duplicate pair + {"id": 3, "user_id": 1, "slice_id": 1}, # duplicate pair + {"id": 4, "user_id": 2, "slice_id": 2}, # distinct keeper + {"id": 5, "user_id": None, "slice_id": 3}, # NULL fk1 + {"id": 6, "user_id": 3, "slice_id": None}, # NULL fk2 + ], + ) + + _run_with_alembic_context(engine, _migration.upgrade) + + with engine.connect() as conn: + survivors = sorted( + conn.execute(sa.text("SELECT user_id, slice_id FROM slice_user")).fetchall() + ) + assert survivors == [(1, 1), (2, 2)], ( + f"expected the two distinct non-NULL pairs to survive, got {survivors}" + ) def test_migration_module_constants_are_consistent() -> None: @@ -159,10 +198,3 @@ def test_migration_module_constants_are_consistent() -> None: assert _migration.TABLES_WITH_NULLABLE_FKS.issubset(affected_names) # Order is alphabetical (deterministic for review/bisection). assert [t.name for t in AFFECTED_TABLES] == sorted(affected_names) - - -@pytest.mark.skipif(True, reason="placeholder — see test_round_trip above") -def test_placeholder_for_future_postgres_round_trip() -> None: - """Reserved slot for a future Postgres-specific round-trip if local - SQLite divergence ever needs to be cross-checked against the real - backend. Today's CI matrix (T034a) handles this implicitly.""" From 494f5eebb8fc0eb450c75551ea5e8e40cef143ce Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Thu, 11 Jun 2026 17:28:50 -0600 Subject: [PATCH 22/35] docs(migration): correct composite-PK operator sizing guidance MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Four corrections to the maintenance-window guidance: - PostgreSQL takes the direct-ALTER path for ALL eight tables (the redundant UNIQUEs are dropped by name); the doc described a recreate='always' full rewrite the code deliberately avoids, and sized only those two tables. The lock-window query now covers all eight. - State the cumulative-lock property: Alembic runs the upgrade in one transaction on Postgres, so ACCESS EXCLUSIVE locks are held until commit — total unavailability is the sum of per-table windows; quiesce the app. - MySQL: DROP COLUMN and ADD PRIMARY KEY are separate ALTERs, so most tables pay the InnoDB clustered-index rebuild twice — budget ~2x the single-rebuild estimate. - Downgrade is a comparable maintenance window in its own right, not a quick undo. Co-Authored-By: Claude Opus 4.8 (1M context) --- UPDATING.md | 33 +++++++++++++++++---------------- 1 file changed, 17 insertions(+), 16 deletions(-) diff --git a/UPDATING.md b/UPDATING.md index 7f0f97078a1b..5daecd4e05f9 100644 --- a/UPDATING.md +++ b/UPDATING.md @@ -530,14 +530,14 @@ SELECT COUNT(*) FROM slice_user WHERE user_id IS NULL OR slice_id IS NULL; SELECT COUNT(*) FROM sqlatable_user WHERE user_id IS NULL OR table_id IS NULL; ``` -**Sizing the maintenance window on PostgreSQL.** The queries above are dialect-portable but only count rows. Operators on PostgreSQL can run the diagnostic queries below to characterize the migration's runtime cost ahead of time: per-table row count and on-disk size, an aggregated duplicate roll-up, the external-FK pre-flight check (the migration runs the same check and aborts if it returns rows), and a rewrite-time estimate for the two tables that go through the slower full-table-rebuild path. +**Sizing the maintenance window on PostgreSQL.** The queries above are dialect-portable but only count rows. Operators on PostgreSQL can run the diagnostic queries below to characterize the migration's runtime cost ahead of time: per-table row count and on-disk size, an aggregated duplicate roll-up, the external-FK pre-flight check (the migration runs the same check and aborts if it returns rows), and a lock-window estimate. On PostgreSQL **all eight tables take the direct-ALTER path** — the two redundant `UNIQUE` constraints are dropped by name (`DROP CONSTRAINT`), avoiding any full-table rewrite; the `recreate="always"` rewrite path applies only on MySQL/SQLite. Note also that Alembic runs the whole upgrade in **one transaction on PostgreSQL: the `ACCESS EXCLUSIVE` locks acquired per table are held cumulatively until commit**, so total unavailability of these RBAC/RLS junction tables is the *sum* of the per-table windows — and a waiting `ACCESS EXCLUSIVE` queues all later reads behind it. Run the migration with the application quiesced. ```sql --- Per-table size, row count, and which migration path each will take. --- Two tables ("dashboard_slices", "report_schedule_user") have a --- redundant UNIQUE constraint that the migration drops via a full --- table rewrite (op.batch_alter_table(recreate="always")). The other --- six use direct ALTER TABLE, which is much cheaper. +-- Per-table size and row count. Two tables ("dashboard_slices", +-- "report_schedule_user") carry a redundant UNIQUE constraint; on +-- PostgreSQL it is dropped by name (DROP CONSTRAINT) and every table +-- then takes the same direct-ALTER path — no full-table rewrite on +-- this dialect. has_unique only signals the extra DROP CONSTRAINT. WITH affected(name, has_unique) AS ( VALUES ('dashboard_roles', false), @@ -616,23 +616,24 @@ WHERE ccu.table_name IN ( ``` ```sql --- Lock-window estimate for the two full-rewrite tables. --- recreate="always" takes ACCESS EXCLUSIVE on the table for the full --- rewrite. Use heap size combined with your hardware's effective --- write throughput (~100-200 MB/s on commodity SSD; faster on NVMe) --- to size the maintenance window. The other six tables use direct --- ALTER and are dominated by composite-index build time, typically --- seconds for tables in the low millions of rows. +-- Lock-window estimate, all eight tables. Each direct ALTER takes +-- ACCESS EXCLUSIVE for the duration of the composite-PK index build +-- (plus the implicit NOT NULL validation scan) — typically seconds +-- for tables in the low millions of rows, but the locks are held +-- cumulatively until the migration's single transaction commits. SELECT c.relname AS table_name, pg_size_pretty(pg_relation_size(c.oid)) AS heap_size, pg_relation_size(c.oid) / 1024 / 1024 AS heap_size_mb, - ROUND(pg_relation_size(c.oid) / 1024 / 1024 / 100.0, 1) AS est_rewrite_seconds_at_100mbs + ROUND(pg_relation_size(c.oid) / 1024 / 1024 / 100.0, 1) AS est_seconds_at_100mbs FROM pg_class c -WHERE c.relname IN ('dashboard_slices', 'report_schedule_user'); +WHERE c.relname IN ( + 'dashboard_roles', 'dashboard_slices', 'dashboard_user', + 'report_schedule_user', 'rls_filter_roles', 'rls_filter_tables', + 'slice_user', 'sqlatable_user'); ``` -**Sizing the maintenance window on MySQL.** Equivalent diagnostic queries for MySQL/InnoDB. One important difference from PostgreSQL: InnoDB rebuilds the clustered index on every PK change, so *all eight* tables undergo a full table rebuild on MySQL — not just the two that go through the explicit `recreate="always"` path. The lock-window estimate query below therefore covers all eight tables. +**Sizing the maintenance window on MySQL.** Equivalent diagnostic queries for MySQL/InnoDB. One important difference from PostgreSQL: InnoDB rebuilds the clustered index on every PK change, so *all eight* tables undergo a full table rebuild on MySQL — not just the two that go through the explicit `recreate="always"` path. Additionally, the upgrade emits `DROP COLUMN id` and `ADD PRIMARY KEY (fk1, fk2)` as **separate ALTER statements, so most tables pay the clustered-index rebuild twice** — budget roughly 2× the single-rebuild estimate from the query below. The **downgrade is a comparable maintenance window in its own right** (it re-adds the `id` column and rebuilds every table on both dialects); plan rollback windows with the same sizing, not as a quick undo. ```sql -- Per-table size, row count, and which migration path each will take. From 43b29bb44a44f4ae960ab761f92782eada903302 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Wed, 17 Jun 2026 14:55:24 -0600 Subject: [PATCH 23/35] fix(db): re-point composite-PK migration onto current master head After rebasing onto current master, the migration root pointed at a stale master revision, forking alembic into multiple heads in the PR-merge CI. Re-point down_revision onto master's current head so the chain is linear. --- ...5-01_23-36_2bee73611e32_composite_pk_association_tables.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py index 9d7035a06717..e31296d241f4 100644 --- a/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py +++ b/superset/migrations/versions/2026-05-01_23-36_2bee73611e32_composite_pk_association_tables.py @@ -27,7 +27,7 @@ of the eight tables lacked DB-level uniqueness. Revision ID: 2bee73611e32 -Revises: 31dae2559c05 +Revises: 78a40c08b4be Create Date: 2026-05-01 23:36:34.050058 """ @@ -43,7 +43,7 @@ # revision identifiers, used by Alembic. revision = "2bee73611e32" -down_revision = "31dae2559c05" +down_revision = "78a40c08b4be" logger = logging.getLogger("alembic.env") From f6d5ed7091666d87aad6e91a0bd041f71ddd0352 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Wed, 17 Jun 2026 15:50:46 -0600 Subject: [PATCH 24/35] feat(versioning): entity-version base infrastructure (gated off) Land the version-history schema + SQLAlchemy-Continuum wiring inert: the ENABLE_VERSIONING_CAPTURE flag defaults OFF, so init_versioning detaches Continuum's write listeners and a save writes zero version_transaction / shadow rows (proven by capture_disabled_tests). The read-only /versions/ list + get endpoints are wired (return empty until capture is enabled). Restore and the version-history UI ship in follow-ups. - migration: version_transaction + *_version shadow tables (additive, inert) - Continuum wiring: make_versioned, VersionTransactionFactory, VersioningFlaskPlugin, the superset/versioning/ module (minus restore) - gate: ENABLE_VERSIONING_CAPTURE (default off; permanent kill-switch) - read endpoints: GET /api/v1/{chart,dashboard,dataset}//versions[/...] - behavioral test: capture-off writes nothing; capture-on control writes one Co-Authored-By: Claude Opus 4.8 (1M context) --- pyproject.toml | 1 + requirements/base.txt | 3 + requirements/development.txt | 5 + superset/charts/api.py | 186 ++- superset/commands/dashboard/copy.py | 13 + .../dashboard/importers/v1/__init__.py | 62 +- superset/commands/dashboard/update.py | 42 +- superset/commands/dataset/duplicate.py | 10 + superset/commands/importers/v1/__init__.py | 13 + superset/commands/importers/v1/assets.py | 37 +- superset/config.py | 22 +- superset/connectors/sqla/models.py | 42 +- superset/constants.py | 2 + superset/daos/dataset.py | 199 ++-- superset/daos/version.py | 70 ++ superset/dashboards/api.py | 182 ++- superset/datasets/api.py | 233 +++- superset/extensions/__init__.py | 25 + superset/initialization/__init__.py | 139 +++ ...9-50_56cd24c07170_add_versioning_tables.py | 567 +++++++++ ...00_8f3a1b2c4d5e_shadow_live_row_indexes.py | 173 +++ superset/models/dashboard.py | 21 + superset/models/helpers.py | 68 +- superset/models/slice.py | 38 +- superset/versioning/__init__.py | 16 + superset/versioning/api_helpers.py | 143 +++ superset/versioning/baseline/__init__.py | 69 ++ superset/versioning/baseline/children.py | 212 ++++ superset/versioning/baseline/collection.py | 153 +++ superset/versioning/baseline/dirty.py | 243 ++++ superset/versioning/baseline/insertion.py | 149 +++ superset/versioning/baseline/listener.py | 95 ++ superset/versioning/baseline/shadow.py | 72 ++ superset/versioning/changes/__init__.py | 76 ++ superset/versioning/changes/listener.py | 507 ++++++++ superset/versioning/changes/shadow_queries.py | 323 +++++ superset/versioning/changes/state.py | 237 ++++ superset/versioning/changes/table.py | 87 ++ superset/versioning/diff.py | 1036 +++++++++++++++++ superset/versioning/etag.py | 75 ++ superset/versioning/factory.py | 312 +++++ superset/versioning/queries.py | 516 ++++++++ superset/versioning/schemas.py | 140 +++ superset/versioning/utils.py | 81 ++ .../integration_tests/versioning/__init__.py | 16 + .../versioning/capture_disabled_tests.py | 143 +++ 46 files changed, 6707 insertions(+), 147 deletions(-) create mode 100644 superset/daos/version.py create mode 100644 superset/migrations/versions/2026-05-28_19-50_56cd24c07170_add_versioning_tables.py create mode 100644 superset/migrations/versions/2026-06-03_12-00_8f3a1b2c4d5e_shadow_live_row_indexes.py create mode 100644 superset/versioning/__init__.py create mode 100644 superset/versioning/api_helpers.py create mode 100644 superset/versioning/baseline/__init__.py create mode 100644 superset/versioning/baseline/children.py create mode 100644 superset/versioning/baseline/collection.py create mode 100644 superset/versioning/baseline/dirty.py create mode 100644 superset/versioning/baseline/insertion.py create mode 100644 superset/versioning/baseline/listener.py create mode 100644 superset/versioning/baseline/shadow.py create mode 100644 superset/versioning/changes/__init__.py create mode 100644 superset/versioning/changes/listener.py create mode 100644 superset/versioning/changes/shadow_queries.py create mode 100644 superset/versioning/changes/state.py create mode 100644 superset/versioning/changes/table.py create mode 100644 superset/versioning/diff.py create mode 100644 superset/versioning/etag.py create mode 100644 superset/versioning/factory.py create mode 100644 superset/versioning/queries.py create mode 100644 superset/versioning/schemas.py create mode 100644 superset/versioning/utils.py create mode 100644 tests/integration_tests/versioning/__init__.py create mode 100644 tests/integration_tests/versioning/capture_disabled_tests.py diff --git a/pyproject.toml b/pyproject.toml index 202f7a44fbaa..2b5ef0b4328b 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -108,6 +108,7 @@ dependencies = [ "simplejson>=4.1.1", "slack_sdk>=3.19.0, <4", "sqlalchemy>=1.4, <2", + "sqlalchemy-continuum>=1.6.0, <2.0.0", "sqlalchemy-utils>=0.38.0, <0.43", # expanding lowerbound to work with pydoris "sqlglot>=30.8.0, <31", # newer pandas needs 0.9+ diff --git a/requirements/base.txt b/requirements/base.txt index 230afa5023ab..522e2001167d 100644 --- a/requirements/base.txt +++ b/requirements/base.txt @@ -404,7 +404,10 @@ sqlalchemy==1.4.54 # flask-sqlalchemy # marshmallow-sqlalchemy # shillelagh + # sqlalchemy-continuum # sqlalchemy-utils +sqlalchemy-continuum==1.6.0 + # via apache-superset (pyproject.toml) sqlalchemy-utils==0.42.0 # via # apache-superset (pyproject.toml) diff --git a/requirements/development.txt b/requirements/development.txt index 81c47a31ef85..1d55c5d3614b 100644 --- a/requirements/development.txt +++ b/requirements/development.txt @@ -973,9 +973,14 @@ sqlalchemy==1.4.54 # marshmallow-sqlalchemy # shillelagh # sqlalchemy-bigquery + # sqlalchemy-continuum # sqlalchemy-utils sqlalchemy-bigquery==1.17.0 # via apache-superset +sqlalchemy-continuum==1.6.0 + # via + # -c requirements/base-constraint.txt + # apache-superset sqlalchemy-utils==0.42.0 # via # -c requirements/base-constraint.txt diff --git a/superset/charts/api.py b/superset/charts/api.py index 0098996a5022..85f509516ac3 100644 --- a/superset/charts/api.py +++ b/superset/charts/api.py @@ -81,7 +81,10 @@ from superset.commands.importers.v1.utils import get_contents_from_bundle from superset.constants import MODEL_API_RW_METHOD_PERMISSION_MAP, RouteMethod from superset.daos.chart import ChartDAO -from superset.exceptions import ScreenshotImageNotAvailableException +from superset.daos.version import VersionDAO +from superset.exceptions import ( + ScreenshotImageNotAvailableException, +) from superset.extensions import event_logger, security_manager from superset.models.slice import Slice from superset.tasks.thumbnails import cache_chart_thumbnail @@ -95,6 +98,12 @@ StatusValues, ) from superset.utils.urls import get_url_path +from superset.versioning.api_helpers import ( + get_version_endpoint, + list_versions_endpoint, +) +from superset.versioning.etag import set_version_etag +from superset.versioning.schemas import VersionListItemSchema from superset.views.base_api import ( BaseSupersetModelRestApi, RelatedFieldFilter, @@ -132,6 +141,8 @@ def ensure_thumbnails_enabled(self) -> Optional[Response]: "screenshot", "cache_screenshot", "warm_up_cache", + "list_versions", + "get_version", } class_permission_name = "Chart" method_permission_name = MODEL_API_RW_METHOD_PERMISSION_MAP @@ -238,7 +249,7 @@ def ensure_thumbnails_enabled(self) -> Optional[Response]: openapi_spec_tag = "Charts" """ Override the name set for this collection of endpoints """ - openapi_spec_component_schemas = CHART_SCHEMAS + openapi_spec_component_schemas = CHART_SCHEMAS + (VersionListItemSchema,) apispec_parameter_schemas = { "screenshot_query_schema": screenshot_query_schema, @@ -312,7 +323,11 @@ def get(self, id_or_uuid: str) -> Response: result = self.chart_get_response_schema.dump(dash) if resolver := current_app.config.get("EXTRA_OWNERS_RESOLVER"): result["extra_owners"] = resolver(dash) - return self.response(200, result=result) + + return set_version_etag( + self.response(200, result=result), + VersionDAO.current_live_version_uuid(Slice, dash.id, dash.uuid), + ) except ChartNotFoundError: return self.response_404() @@ -419,6 +434,34 @@ def put(self, pk: int) -> Response: type: number result: $ref: '#/components/schemas/{{self.__class__.__name__}}.put' + old_version: + type: integer + nullable: true + description: >- + 0-based version_number of the live row before this + update. Unstable under retention pruning — see + old_transaction_id for a stable identifier. + new_version: + type: integer + nullable: true + description: >- + 0-based version_number of the newly-live row after + this update. Can equal old_version when no + versioned column changed, or when retention + pruning dropped an older closed row in the same + commit. + old_transaction_id: + type: integer + nullable: true + description: Continuum transaction_id of the live + row before this update. Stable across pruning. + new_transaction_id: + type: integer + nullable: true + description: Continuum transaction_id of the live + row after this update. Differs from + old_transaction_id when the update produced a new + version row. 400: $ref: '#/components/responses/400' 401: @@ -437,9 +480,40 @@ def put(self, pk: int) -> Response: # This validates custom Schema with custom validations except ValidationError as error: return self.response_400(message=error.messages) + + # pylint: disable=import-outside-toplevel + from superset.extensions import db as _db + + pre_chart = _db.session.query(Slice).filter(Slice.id == pk).one_or_none() + old_version = VersionDAO.current_version_number(Slice, pk) + old_transaction_id = VersionDAO.current_live_transaction_id(Slice, pk) + old_version_uuid = ( + VersionDAO.current_live_version_uuid(Slice, pk, pre_chart.uuid) + if pre_chart is not None + else None + ) + try: changed_model = UpdateChartCommand(pk, item).run() - response = self.response(200, id=changed_model.id, result=item) + new_version = VersionDAO.current_version_number(Slice, changed_model.id) + new_transaction_id = VersionDAO.current_live_transaction_id( + Slice, changed_model.id + ) + new_version_uuid = VersionDAO.current_live_version_uuid( + Slice, changed_model.id, changed_model.uuid + ) + response = self.response( + 200, + id=changed_model.id, + result=item, + old_version=old_version, + new_version=new_version, + old_transaction_id=old_transaction_id, + new_transaction_id=new_transaction_id, + old_version_uuid=str(old_version_uuid) if old_version_uuid else None, + new_version_uuid=str(new_version_uuid) if new_version_uuid else None, + ) + set_version_etag(response, new_version_uuid) except ChartNotFoundError: response = self.response_404() except ChartForbiddenError: @@ -1214,3 +1288,107 @@ def import_(self) -> Response: ) command.run() return self.response(200, message="OK") + + @expose("//versions/", methods=("GET",)) + @protect() + @safe + @statsd_metrics + @event_logger.log_this_with_context( + action=lambda self, *args, **kwargs: f"{self.__class__.__name__}.list_versions", + log_to_statsd=False, + ) + def list_versions(self, uuid_str: str) -> Response: + """List version history for a chart. + --- + get: + summary: Return the version history for a chart + parameters: + - in: path + schema: + type: string + format: uuid + name: uuid_str + description: Chart UUID + responses: + 200: + description: Version history ordered by oldest first + content: + application/json: + schema: + type: object + properties: + result: + type: array + items: + $ref: '#/components/schemas/VersionListItemSchema' + count: + type: integer + 400: + $ref: '#/components/responses/400' + 401: + $ref: '#/components/responses/401' + 403: + $ref: '#/components/responses/403' + 404: + $ref: '#/components/responses/404' + """ + return list_versions_endpoint(self, Slice, uuid_str, access_kwarg="chart") + + @expose( + "//versions//", + methods=("GET",), + ) + @protect() + @safe + @statsd_metrics + @event_logger.log_this_with_context( + action=lambda self, *args, **kwargs: f"{self.__class__.__name__}.get_version", # noqa: E501 + log_to_statsd=False, + ) + def get_version(self, uuid_str: str, version_uuid_str: str) -> Response: + """Return the chart's state at a specific version. + --- + get: + summary: Read-only snapshot of the chart at a given version + parameters: + - in: path + schema: + type: string + format: uuid + name: uuid_str + description: Chart UUID + - in: path + schema: + type: string + format: uuid + name: version_uuid_str + description: Version UUID as returned by the list endpoint + responses: + 200: + description: Snapshot of the chart at the target version + content: + application/json: + schema: + type: object + properties: + result: + type: object + description: >- + The chart's scalar fields at the target version + (entity-specific keys), plus a `_version` block + with the version-level metadata. + properties: + _version: + $ref: '#/components/schemas/VersionListItemSchema' + 400: + $ref: '#/components/responses/400' + 401: + $ref: '#/components/responses/401' + 403: + $ref: '#/components/responses/403' + 404: + $ref: '#/components/responses/404' + """ + return get_version_endpoint( + self, Slice, uuid_str, version_uuid_str, access_kwarg="chart" + ) diff --git a/superset/commands/dashboard/copy.py b/superset/commands/dashboard/copy.py index b694d3686788..f8239751cce7 100644 --- a/superset/commands/dashboard/copy.py +++ b/superset/commands/dashboard/copy.py @@ -40,6 +40,19 @@ def __init__(self, original_dash: Dashboard, data: dict[str, Any]) -> None: @transaction(on_error=partial(on_error, reraise=DashboardCopyError)) def run(self) -> Dashboard: self.validate() + # Declare the high-level avenue before the copy touches the + # session. The change-record listener stamps + # ``version_transaction.action_kind = 'clone'`` so the new + # dashboard's baseline records read as "Cloned from " + # in the timeline instead of "Dashboard created". + # Method-scoped imports — defer the versioning bootstrap path + # (``Model.metadata`` and Continuum-adjacent setup) out of this + # command's module-load graph; see ``changes.py`` module + # docstring for the broader init-order rationale. + from superset import db + from superset.versioning.changes import ACTION_KIND_CLONE, ACTION_KIND_KEY + + db.session.info[ACTION_KIND_KEY] = ACTION_KIND_CLONE return DashboardDAO.copy_dashboard(self._original_dash, self._properties) def validate(self) -> None: diff --git a/superset/commands/dashboard/importers/v1/__init__.py b/superset/commands/dashboard/importers/v1/__init__.py index c3063ed5437f..044f4d63ac60 100644 --- a/superset/commands/dashboard/importers/v1/__init__.py +++ b/superset/commands/dashboard/importers/v1/__init__.py @@ -22,7 +22,7 @@ from marshmallow import Schema from sqlalchemy.orm import Session # noqa: F401 -from sqlalchemy.sql import delete, select +from sqlalchemy.sql import select from superset import db from superset.charts.schemas import ImportV1ChartSchema @@ -47,6 +47,7 @@ from superset.extensions import feature_flag_manager from superset.migrations.shared.native_filters import migrate_dashboard from superset.models.dashboard import Dashboard, dashboard_slices +from superset.models.slice import Slice from superset.themes.schemas import ImportV1ThemeSchema logger = logging.getLogger(__name__) @@ -167,8 +168,18 @@ def _import( ) # import dashboards + # + # Dashboard → charts associations go through the ORM relationship + # (``dashboard.slices = [...]``) rather than Core + # ``delete()``/``insert()`` on the ``dashboard_slices`` table. + # Bulk DML via Core would emit a malformed INSERT into + # ``dashboard_slices_version`` (missing the composite-PK columns) + # because SQLAlchemy-Continuum's M2M tracker can't see per-row + # column values when the DELETE/INSERT goes through the Core + # layer. The same pattern is applied in + # ``superset/commands/importers/v1/assets.py`` and the spike's + # ``DatasetDAO.update_columns`` rewrite. dashboards: list[Dashboard] = [] - dashboard_chart_ids: list[tuple[int, int]] = [] for file_name, config in configs.items(): if file_name.startswith("dashboards/"): config = update_id_refs(config, chart_ids, dataset_info) @@ -183,16 +194,9 @@ def _import( dashboard = import_dashboard(config, overwrite=overwrite) dashboards.append(dashboard) - # When overwriting, first delete all existing chart relationships - # so the dashboard is replaced rather than merged - if overwrite: - db.session.execute( - delete(dashboard_slices).where( - dashboard_slices.c.dashboard_id == dashboard.id - ) - ) - - # Collect chart IDs to associate with this dashboard + # Resolve the dashboard's chart membership from the imported + # position_json and apply it to the ORM relationship. + target_chart_ids: list[int] = [] for uuid in find_chart_uuids(config["position"]): if uuid not in chart_ids: continue @@ -201,7 +205,31 @@ def _import( overwrite or (dashboard.id, chart_id) not in existing_relationships ): - dashboard_chart_ids.append((dashboard.id, chart_id)) + target_chart_ids.append(chart_id) + + if overwrite: + # Replace the dashboard's chart membership entirely. + dashboard.slices = ( + db.session.query(Slice) + .filter(Slice.id.in_(target_chart_ids)) + .all() + if target_chart_ids + else [] + ) + # Flush eagerly so the M2M rows land in + # ``dashboard_slices`` before any subsequent + # autoflush fires an inner-flush event handler + # that would reset the relationship change. + db.session.flush() + elif target_chart_ids: + # Append only the new associations to existing ones. + new_slices = ( + db.session.query(Slice) + .filter(Slice.id.in_(target_chart_ids)) + .all() + ) + dashboard.slices = list(dashboard.slices) + new_slices + db.session.flush() # Handle tags using import_tag function if feature_flag_manager.is_feature_enabled("TAGGING_SYSTEM"): @@ -215,14 +243,6 @@ def _import( db.session, ) - # set ref in the dashboard_slices table - if dashboard_chart_ids: - values = [ - {"dashboard_id": dashboard_id, "slice_id": chart_id} - for (dashboard_id, chart_id) in dashboard_chart_ids - ] - db.session.execute(dashboard_slices.insert(), values) - # Migrate any filter-box charts to native dashboard filters. for dashboard in dashboards: migrate_dashboard(dashboard) diff --git a/superset/commands/dashboard/update.py b/superset/commands/dashboard/update.py index dd81d96deeb1..b39ef4af9463 100644 --- a/superset/commands/dashboard/update.py +++ b/superset/commands/dashboard/update.py @@ -59,23 +59,31 @@ def __init__(self, model_id: int, data: dict[str, Any]): def run(self) -> Model: self.validate() assert self._model is not None - self.process_tab_diff() - self.process_native_filter_diff() - - # Update tags - if (tags := self._properties.pop("tags", None)) is not None: - update_tags(ObjectType.dashboard, self._model.id, self._model.tags, tags) - - # Re-serialize position_json to escape 4-byte Unicode characters - if position_json := self._properties.get("position_json"): - self._properties["position_json"] = json.dumps(json.loads(position_json)) - - dashboard = DashboardDAO.update(self._model, self._properties) - if self._properties.get("json_metadata"): - DashboardDAO.set_dash_metadata( - dashboard, - data=json.loads(self._properties.get("json_metadata", "{}")), - ) + # Suppress autoflush during the update body so that Continuum's + # before_flush baseline listener does not fire mid-operation while + # the session is only partially populated. + with db.session.no_autoflush: + self.process_tab_diff() + self.process_native_filter_diff() + + # Update tags + if (tags := self._properties.pop("tags", None)) is not None: + update_tags( + ObjectType.dashboard, self._model.id, self._model.tags, tags + ) + + # Re-serialize position_json to escape 4-byte Unicode characters + if position_json := self._properties.get("position_json"): + self._properties["position_json"] = json.dumps( + json.loads(position_json) + ) + + dashboard = DashboardDAO.update(self._model, self._properties) + if self._properties.get("json_metadata"): + DashboardDAO.set_dash_metadata( + dashboard, + data=json.loads(self._properties.get("json_metadata", "{}")), + ) return dashboard def validate(self) -> None: diff --git a/superset/commands/dataset/duplicate.py b/superset/commands/dataset/duplicate.py index 2be7be5690b9..8371610fe55b 100644 --- a/superset/commands/dataset/duplicate.py +++ b/superset/commands/dataset/duplicate.py @@ -52,6 +52,16 @@ def __init__(self, data: dict[str, Any]) -> None: @transaction(on_error=partial(on_error, reraise=DatasetDuplicateFailedError)) def run(self) -> Model: self.validate() + # Declare the high-level avenue before the duplicate touches + # the session. The change-record listener stamps + # ``version_transaction.action_kind = 'clone'`` so the new + # dataset's baseline records read as a clone in the timeline. + # Method-scoped import — defers the versioning bootstrap path + # out of this command's module-load graph; see ``changes.py`` + # module docstring for the broader init-order rationale. + from superset.versioning.changes import ACTION_KIND_CLONE, ACTION_KIND_KEY + + db.session.info[ACTION_KIND_KEY] = ACTION_KIND_CLONE database_id = self._base_model.database_id table_name = self._properties["table_name"] owners = self._properties["owners"] diff --git a/superset/commands/importers/v1/__init__.py b/superset/commands/importers/v1/__init__.py index d8d010408761..5cf0f8ade990 100644 --- a/superset/commands/importers/v1/__init__.py +++ b/superset/commands/importers/v1/__init__.py @@ -86,6 +86,19 @@ def _get_uuids(cls) -> set[str]: def run(self) -> None: self.validate() + # Declare the high-level avenue before any session writes. The + # change-record listener reads this on its first after_flush + # for the resulting ``version_transaction`` row and stamps + # ``version_transaction.action_kind = 'import'``. Lets operators + # explain otherwise-confusing diffs ("Cleared default_filters") + # as "this was an import". See data-model.md §"Three dimensions". + # Method-scoped import — defers the versioning bootstrap path + # out of this command's module-load graph; see ``changes.py`` + # module docstring for the broader init-order rationale. + from superset.versioning.changes import ACTION_KIND_IMPORT, ACTION_KIND_KEY + + db.session.info[ACTION_KIND_KEY] = ACTION_KIND_IMPORT + try: self._import(self._configs, self.overwrite, self.contents) except CommandException: diff --git a/superset/commands/importers/v1/assets.py b/superset/commands/importers/v1/assets.py index 99e28b38f964..1b7b4b20b573 100644 --- a/superset/commands/importers/v1/assets.py +++ b/superset/commands/importers/v1/assets.py @@ -19,7 +19,6 @@ from marshmallow import Schema from marshmallow.exceptions import ValidationError -from sqlalchemy.sql import delete, insert from superset import db from superset.charts.schemas import ImportV1ChartSchema @@ -49,7 +48,7 @@ from superset.extensions import feature_flag_manager from superset.migrations.shared.native_filters import migrate_dashboard from superset.models.core import Database -from superset.models.dashboard import Dashboard, dashboard_slices +from superset.models.dashboard import Dashboard from superset.models.slice import Slice from superset.models.sql_lab import SavedQuery from superset.queries.saved_queries.schemas import ImportV1SavedQuerySchema @@ -165,23 +164,33 @@ def _import( # noqa: C901 dashboard = import_dashboard(config, overwrite=overwrite) # set ref in the dashboard_slices table - dashboard_chart_ids: list[dict[str, int]] = [] + # Use ORM-level reassignment instead of Core + # delete()/insert() so SQLAlchemy-Continuum's M2M tracker + # sees per-row changes through the ORM. Bulk DML via Core + # would emit a malformed INSERT into + # ``dashboard_slices_version`` (missing the composite-PK + # columns) — see the parallel rewrite in + # ``DatasetDAO.update_columns`` and the test-factory's + # ``delete_dashboard_slices_associations`` for the same + # reason. + slice_ids: list[int] = [] for uuid in find_chart_uuids(config["position"]): if uuid not in chart_ids: break - chart_id = chart_ids[uuid] - dashboard_chart_id = { - "dashboard_id": dashboard.id, - "slice_id": chart_id, - } - dashboard_chart_ids.append(dashboard_chart_id) + slice_ids.append(chart_ids[uuid]) - db.session.execute( - delete(dashboard_slices).where( - dashboard_slices.c.dashboard_id == dashboard.id - ) + dashboard.slices = ( + db.session.query(Slice).filter(Slice.id.in_(slice_ids)).all() + if slice_ids + else [] ) - db.session.execute(insert(dashboard_slices).values(dashboard_chart_ids)) + # Flush eagerly so the M2M rows land in + # ``dashboard_slices`` before any subsequent autoflush + # fires an inner-flush event handler that would reset + # the relationship change (cf. the SAWarning at + # ``superset/models/helpers.py`` re. "attribute history + # events accumulated ... have been reset"). + db.session.flush() # Handle tags using import_tag function if feature_flag_manager.is_feature_enabled("TAGGING_SYSTEM"): diff --git a/superset/config.py b/superset/config.py index 2a44424afc6a..356fb8c41a2e 100644 --- a/superset/config.py +++ b/superset/config.py @@ -1247,7 +1247,11 @@ class D3TimeFormat(TypedDict, total=False): "origins": [ "https://tile.openstreetmap.org", "https://tile.osm.ch", - ] + ], + # Make the entity-version-history `ETag` header readable by cross-origin + # browser clients. Without this, `fetch()` callers cannot read the header + # even when CORS is otherwise permissive. + "expose_headers": ["ETag"], } # Sanitizes the HTML content used in markdowns to allow its rendering in a safe manner. @@ -1427,6 +1431,22 @@ class D3TimeFormat(TypedDict, total=False): # The limit for the Superset Meta DB when the feature flag ENABLE_SUPERSET_META_DB is on SUPERSET_META_DB_LIMIT: int | None = 1000 +# Master switch for entity-version-history capture. Default ``True`` — +# every save of a chart, dashboard, or dataset writes shadow rows + a +# ``version_changes`` record. Set to ``False`` in ``superset_config.py`` +# (or via the env var of the same name) to disable the two before-flush +# listeners that drive capture; existing shadow tables stay intact and +# the /versions/ + /activity/ endpoints continue to work read-only. +# This is an operational escape hatch — for use when a versioning-induced +# regression needs a 30-second recovery instead of revert-and-redeploy — +# not a feature flag. It ships defaulted OFF so the versioning infrastructure +# (schema + Continuum wiring) lands inert; capture is activated by flipping +# the default to on once validated in production. It then remains as the +# permanent kill-switch. +ENABLE_VERSIONING_CAPTURE: bool = utils.parse_boolean_string( + os.environ.get("ENABLE_VERSIONING_CAPTURE", "false") +) + # Adds a warning message on sqllab save query and schedule query modals. SQLLAB_SAVE_WARNING_MESSAGE = None SQLLAB_SCHEDULE_WARNING_MESSAGE = None diff --git a/superset/connectors/sqla/models.py b/superset/connectors/sqla/models.py index ca8bada53804..cdf903473de2 100644 --- a/superset/connectors/sqla/models.py +++ b/superset/connectors/sqla/models.py @@ -945,6 +945,15 @@ class TableColumn(AuditMixinNullable, ImportExportMixin, CertificationMixin, Mod __tablename__ = "table_columns" __table_args__ = (UniqueConstraint("table_id", "column_name"),) + # SPIKE (full-Continuum): Continuum-versioned + # again, with audit-field exclusions to suppress the per-column-per-save + # noise rows that ADR-004 flagged as Failure 3. ``changed_on`` refreshes + # on every parent dataset save even when the column itself wasn't user- + # edited; capturing it produced one shadow row per column per save with + # no user signal. + __versioned__: dict[str, Any] = { + "exclude": ["changed_on", "created_on", "changed_by_fk", "created_by_fk"] + } id = Column(Integer, primary_key=True) column_name = Column(String(255), nullable=False) @@ -1190,6 +1199,10 @@ class SqlMetric(AuditMixinNullable, ImportExportMixin, CertificationMixin, Model __tablename__ = "sql_metrics" __table_args__ = (UniqueConstraint("table_id", "metric_name"),) + # SPIKE: same audit-field exclusions as TableColumn (see above). + __versioned__: dict[str, Any] = { + "exclude": ["changed_on", "created_on", "changed_by_fk", "created_by_fk"] + } id = Column(Integer, primary_key=True) metric_name = Column(String(255), nullable=False) @@ -1327,6 +1340,33 @@ class SqlaTable( owner_class = security_manager.user_model __tablename__ = "tables" + # Exclude M2M association relationships: Continuum only captures FK columns on + # association INSERTs (not the auto-increment id), which breaks the NOT NULL PK. + # deleted_at exclusion will be added when soft delete is merged. + # Audit columns are auto-bumped on every save. Excluding them lets + # Continuum's is_modified() return False on no-op saves (e.g. owners-only + # edits) so we don't create empty version rows. version_transaction.user_id + # / issued_at preserve "who/when". + # The perm-string class (perm / schema_perm / catalog_perm) is derived + # security state, not user-authored content: permission maintenance + # rewrites it in bulk, and versioning it produced phantom transactions + # flooding the activity stream (one "updated" row per touched entity + # with no user edit — surfaced by the version-history UI, PR #40988). + # Excluding it also means a restore can't resurrect stale permission + # strings; the live, derived values stay authoritative. + __versioned__: dict[str, Any] = { + "exclude": [ + "owners", + "row_level_security_filters", + "changed_on", + "created_on", + "changed_by_fk", + "created_by_fk", + "perm", + "schema_perm", + "catalog_perm", + ] + } # Note this uniqueness constraint is not part of the physical schema, i.e., it does # not exist in the migrations, but is required by `import_from_dict` to ensure the @@ -1455,7 +1495,7 @@ def link(self) -> Markup: name = escape(self.name) url = escape(self.explore_url) anchor = f'{name}' - return Markup(anchor) + return Markup(anchor) # noqa: S704 def get_catalog_perm(self) -> str | None: """Returns catalog permission if present, database one otherwise.""" diff --git a/superset/constants.py b/superset/constants.py index 863d0cf82ba1..3525abc882c1 100644 --- a/superset/constants.py +++ b/superset/constants.py @@ -179,6 +179,8 @@ class RouteMethod: # pylint: disable=too-few-public-methods "put_colors": "write", "sync_permissions": "write", "restore": "write", + "list_versions": "read", + "get_version": "read", } EXTRA_FORM_DATA_APPEND_KEYS = { diff --git a/superset/daos/dataset.py b/superset/daos/dataset.py index 4d819deb5bd4..e7394099e3b7 100644 --- a/superset/daos/dataset.py +++ b/superset/daos/dataset.py @@ -275,6 +275,103 @@ def update( return super().update(item, attributes) + @classmethod + def _validate_column_date_formats( + cls, property_columns: list[dict[str, Any]] + ) -> None: + for column in property_columns: + if column.get("python_date_format") is None: + continue + if not DatasetDAO.validate_python_date_format(column["python_date_format"]): + raise ValueError( + "python_date_format is an invalid date/timestamp format." + ) + + @classmethod + def _override_columns( + cls, model: SqlaTable, property_columns: list[dict[str, Any]] + ) -> None: + """Replace columns by natural key (``column_name``) — update in place + rather than delete-and-reinsert. + + SPIKE (full-Continuum): the previous + delete-and-reinsert pattern produced overlapping shadow rows in + ``table_columns_version`` (the same ``column_name`` had a DELETE + shadow at tx N alongside an INSERT shadow at tx N for a fresh PK). + Continuum's ``Reverter`` couldn't unwind this on restore: its flush + ordering inserts the historical row before deleting the live one, + hitting the ``UNIQUE (table_id, column_name)`` constraint mid-flush + (ADR-004 Failure 1). + + The natural-key upsert keeps PKs stable across metadata refresh. + Continuum captures only real field changes; new columns get plain + INSERT shadows; removed columns get plain DELETE shadows. No + natural-key collisions, so Reverter can restore cleanly. + + Behaviour change vs. the previous implementation: PKs of unchanged + columns are preserved. Charts that reference columns by their + ``id`` continue to work across a metadata refresh — previously + such references would be invalidated. + """ + existing_by_name = {c.column_name: c for c in model.columns} + incoming_by_name = {p["column_name"]: p for p in property_columns} + + # Identity is the natural key here, never the payload's ``id``: + # setattr-ing an incoming ``id`` onto a name-matched row would + # rewrite a live primary key, and a renamed column whose payload + # still carries its old ``id`` would INSERT with a live PK while + # the old-named row is deleted in the same flush — INSERTs flush + # before DELETEs, so that collides on the PK / UNIQUE(table_id, + # column_name) constraints. ``table_id`` is pinned to *model*. + protected_keys = ("id", "table_id") + + # Update columns present in both: in-place setattr. + for name, col in existing_by_name.items(): + if name in incoming_by_name: + for key, value in incoming_by_name[name].items(): + if key not in protected_keys: + setattr(col, key, value) + + # Insert columns present only in incoming. + for name, properties in incoming_by_name.items(): + if name not in existing_by_name: + cleaned = { + key: value + for key, value in properties.items() + if key not in protected_keys + } + db.session.add(TableColumn(**{**cleaned, "table_id": model.id})) + + # Delete columns present only in existing. + for name, col in existing_by_name.items(): + if name not in incoming_by_name: + db.session.delete(col) + + @classmethod + def _upsert_columns( + cls, model: SqlaTable, property_columns: list[dict[str, Any]] + ) -> None: + columns_by_id = {column.id: column for column in model.columns} + property_columns_by_id = { + properties["id"]: properties + for properties in property_columns + if "id" in properties + } + + for properties in property_columns: + if "id" not in properties: + db.session.add(TableColumn(**{**properties, "table_id": model.id})) + + for properties in property_columns_by_id.values(): + col = columns_by_id[properties["id"]] + for key, value in properties.items(): + setattr(col, key, value) + + ids_to_keep = property_columns_by_id.keys() + for col in model.columns: + if col.id not in ids_to_keep: + db.session.delete(col) + @classmethod def update_columns( cls, @@ -290,64 +387,15 @@ def update_columns( - If a column Dict does not have an `id` then we create a new metric. - If there are extra columns on the metadata db that are not defined on the List then we delete. - """ - - for column in property_columns: - if ( - "python_date_format" in column - and column["python_date_format"] is not None - ): - if not DatasetDAO.validate_python_date_format( - column["python_date_format"] - ): - raise ValueError( - "python_date_format is an invalid date/timestamp format." - ) + Uses individual ORM operations (not bulk) so that SQLAlchemy-Continuum + can capture each row change in the version history. + """ + cls._validate_column_date_formats(property_columns) if override_columns: - db.session.query(TableColumn).filter( - TableColumn.table_id == model.id - ).delete(synchronize_session="fetch") - - db.session.bulk_insert_mappings( - TableColumn, - [ - {**properties, "table_id": model.id} - for properties in property_columns - ], - ) + cls._override_columns(model, property_columns) else: - columns_by_id = {column.id: column for column in model.columns} - - property_columns_by_id = { - properties["id"]: properties - for properties in property_columns - if "id" in properties - } - - db.session.bulk_insert_mappings( - TableColumn, - [ - {**properties, "table_id": model.id} - for properties in property_columns - if "id" not in properties - ], - ) - - db.session.bulk_update_mappings( - TableColumn, - [ - {**columns_by_id[properties["id"]].__dict__, **properties} - for properties in property_columns_by_id.values() - ], - ) - - db.session.query(TableColumn).filter( - TableColumn.id.in_( - {column.id for column in model.columns} - - property_columns_by_id.keys() - ) - ).delete(synchronize_session="fetch") + cls._upsert_columns(model, property_columns) @classmethod def update_metrics( @@ -363,6 +411,9 @@ def update_metrics( - If a metric Dict does not have an `id` then we create a new metric. - If there are extra metrics on the metadata db that are not defined on the List then we delete. + + Uses individual ORM operations (not bulk) so that SQLAlchemy-Continuum + can capture each row change in the version history. """ metrics_by_id = {metric.id: metric for metric in model.metrics} @@ -373,28 +424,22 @@ def update_metrics( if "id" in properties } - db.session.bulk_insert_mappings( - SqlMetric, - [ - {**properties, "table_id": model.id} - for properties in property_metrics - if "id" not in properties - ], - ) - - db.session.bulk_update_mappings( - SqlMetric, - [ - {**metrics_by_id[properties["id"]].__dict__, **properties} - for properties in property_metrics_by_id.values() - ], - ) - - db.session.query(SqlMetric).filter( - SqlMetric.id.in_( - {metric.id for metric in model.metrics} - property_metrics_by_id.keys() - ) - ).delete(synchronize_session="fetch") + # Insert new metrics + for properties in property_metrics: + if "id" not in properties: + db.session.add(SqlMetric(**{**properties, "table_id": model.id})) + + # Update existing metrics + for properties in property_metrics_by_id.values(): + metric = metrics_by_id[properties["id"]] + for key, value in properties.items(): + setattr(metric, key, value) + + # Delete removed metrics + ids_to_keep = property_metrics_by_id.keys() + for metric in model.metrics: + if metric.id not in ids_to_keep: + db.session.delete(metric) @classmethod def find_dataset_column(cls, dataset_id: int, column_id: int) -> TableColumn | None: diff --git a/superset/daos/version.py b/superset/daos/version.py new file mode 100644 index 000000000000..91a174439e72 --- /dev/null +++ b/superset/daos/version.py @@ -0,0 +1,70 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Backward-compat façade for the entity-versioning DAO surface. + +The actual implementation lives in :mod:`superset.versioning.queries` +(read side: list/get/resolve/find/UUID derivation) and +:mod:`superset.versioning.restore` (write side: restore + audit +stamping). This module re-exports both under a single ``VersionDAO`` +class plus the module-level UUID helpers so existing callers keep +working without changes. + +New code should import from the versioning sub-modules directly. +""" + +from __future__ import annotations + +from superset.versioning.queries import ( + current_live_transaction_id, + current_live_version_uuid, + current_version_number, + derive_version_uuid, + derive_version_uuid as _derive_version_uuid, # noqa: F401 + find_active_by_uuid, + get_version, + list_change_records_batch, + list_versions, + resolve_version_uuid, + VERSION_UUID_NAMESPACE, +) + +# Re-exports for ``from superset.daos.version import …`` consumers. +__all__ = [ + "VERSION_UUID_NAMESPACE", + "VersionDAO", + "derive_version_uuid", +] + + +class VersionDAO: + """Thin façade over :mod:`superset.versioning.queries` and + :mod:`superset.versioning.restore`. + + Preserved as a single namespace for ergonomic access from API + handlers and command classes; the underlying functions are + importable directly from their respective sub-modules. + """ + + # --- read side (queries.py) ------------------------------------------- + find_active_by_uuid = staticmethod(find_active_by_uuid) + current_version_number = staticmethod(current_version_number) + current_live_transaction_id = staticmethod(current_live_transaction_id) + current_live_version_uuid = staticmethod(current_live_version_uuid) + list_change_records_batch = staticmethod(list_change_records_batch) + list_versions = staticmethod(list_versions) + resolve_version_uuid = staticmethod(resolve_version_uuid) + get_version = staticmethod(get_version) diff --git a/superset/dashboards/api.py b/superset/dashboards/api.py index b8c437b74167..0284f69c09ab 100644 --- a/superset/dashboards/api.py +++ b/superset/dashboards/api.py @@ -84,6 +84,7 @@ from superset.commands.importers.v1.utils import get_contents_from_bundle from superset.constants import MODEL_API_RW_METHOD_PERMISSION_MAP, RouteMethod from superset.daos.dashboard import DashboardDAO, EmbeddedDashboardDAO +from superset.daos.version import VersionDAO from superset.dashboards.filters import ( DashboardAccessFilter, DashboardCertifiedFilter, @@ -119,7 +120,9 @@ TabsPayloadSchema, thumbnail_query_schema, ) -from superset.exceptions import ScreenshotImageNotAvailableException +from superset.exceptions import ( + ScreenshotImageNotAvailableException, +) from superset.extensions import event_logger, security_manager from superset.models.dashboard import Dashboard from superset.models.embedded_dashboard import EmbeddedDashboard @@ -139,6 +142,12 @@ ScreenshotCachePayload, ) from superset.utils.urls import get_url_path +from superset.versioning.api_helpers import ( + get_version_endpoint, + list_versions_endpoint, +) +from superset.versioning.etag import set_version_etag +from superset.versioning.schemas import VersionListItemSchema from superset.views.base_api import ( BaseSupersetModelRestApi, RelatedFieldFilter, @@ -252,6 +261,8 @@ class DashboardRestApi(CustomTagsOptimizationMixin, BaseSupersetModelRestApi): "put_chart_customizations", "put_colors", "export_as_example", + "list_versions", + "get_version", } resource_name = "dashboard" allow_browser_login = True @@ -436,6 +447,7 @@ def get_list(self, **kwargs: Any) -> Response: GetFavStarIdsSchema, EmbeddedDashboardResponseSchema, DashboardScreenshotPostSchema, + VersionListItemSchema, ) apispec_parameter_schemas = { "get_delete_ids_schema": get_delete_ids_schema, @@ -526,7 +538,10 @@ def get( add_extra_log_payload( dashboard_id=dash.id, action=f"{self.__class__.__name__}.get" ) - return self.response(200, result=result) + return set_version_etag( + self.response(200, result=result), + VersionDAO.current_live_version_uuid(Dashboard, dash.id, dash.uuid), + ) @expose("//datasets", methods=("GET",)) @protect() @@ -810,6 +825,34 @@ def put(self, pk: int) -> Response: $ref: '#/components/schemas/{{self.__class__.__name__}}.put' last_modified_time: type: number + old_version: + type: integer + nullable: true + description: >- + 0-based version_number of the live row before this + update. Unstable under retention pruning — see + old_transaction_id for a stable identifier. + new_version: + type: integer + nullable: true + description: >- + 0-based version_number of the newly-live row after + this update. Can equal old_version when no + versioned column changed, or when retention + pruning dropped an older closed row in the same + commit. + old_transaction_id: + type: integer + nullable: true + description: Continuum transaction_id of the live + row before this update. Stable across pruning. + new_transaction_id: + type: integer + nullable: true + description: Continuum transaction_id of the live + row after this update. Differs from + old_transaction_id when the update produced a new + version row. 400: $ref: '#/components/responses/400' 401: @@ -828,17 +871,46 @@ def put(self, pk: int) -> Response: # This validates custom Schema with custom validations except ValidationError as error: return self.response_400(message=error.messages) + + # pylint: disable=import-outside-toplevel + from superset.extensions import db as _db + + pre_dashboard = ( + _db.session.query(Dashboard).filter(Dashboard.id == pk).one_or_none() + ) + old_version = VersionDAO.current_version_number(Dashboard, pk) + old_transaction_id = VersionDAO.current_live_transaction_id(Dashboard, pk) + old_version_uuid = ( + VersionDAO.current_live_version_uuid(Dashboard, pk, pre_dashboard.uuid) + if pre_dashboard is not None + else None + ) + try: changed_model = UpdateDashboardCommand(pk, item).run() last_modified_time = changed_model.changed_on.replace( microsecond=0 ).timestamp() + new_version = VersionDAO.current_version_number(Dashboard, changed_model.id) + new_transaction_id = VersionDAO.current_live_transaction_id( + Dashboard, changed_model.id + ) + new_version_uuid = VersionDAO.current_live_version_uuid( + Dashboard, changed_model.id, changed_model.uuid + ) response = self.response( 200, id=changed_model.id, result=item, last_modified_time=last_modified_time, + old_version=old_version, + new_version=new_version, + old_transaction_id=old_transaction_id, + new_transaction_id=new_transaction_id, + old_version_uuid=str(old_version_uuid) if old_version_uuid else None, + new_version_uuid=str(new_version_uuid) if new_version_uuid else None, ) + set_version_etag(response, new_version_uuid) except DashboardNotFoundError: response = self.response_404() except DashboardForbiddenError: @@ -2231,3 +2303,109 @@ def copy_dash(self, original_dash: Dashboard) -> Response: ).timestamp(), }, ) + + @expose("//versions/", methods=("GET",)) + @protect() + @safe + @statsd_metrics + @event_logger.log_this_with_context( + action=lambda self, *args, **kwargs: f"{self.__class__.__name__}.list_versions", + log_to_statsd=False, + ) + def list_versions(self, uuid_str: str) -> Response: + """List version history for a dashboard. + --- + get: + summary: Return the version history for a dashboard + parameters: + - in: path + schema: + type: string + format: uuid + name: uuid_str + description: Dashboard UUID + responses: + 200: + description: Version history ordered by oldest first + content: + application/json: + schema: + type: object + properties: + result: + type: array + items: + $ref: '#/components/schemas/VersionListItemSchema' + count: + type: integer + 400: + $ref: '#/components/responses/400' + 401: + $ref: '#/components/responses/401' + 403: + $ref: '#/components/responses/403' + 404: + $ref: '#/components/responses/404' + """ + return list_versions_endpoint( + self, Dashboard, uuid_str, access_kwarg="dashboard" + ) + + @expose( + "//versions//", + methods=("GET",), + ) + @protect() + @safe + @statsd_metrics + @event_logger.log_this_with_context( + action=lambda self, *args, **kwargs: f"{self.__class__.__name__}.get_version", # noqa: E501 + log_to_statsd=False, + ) + def get_version(self, uuid_str: str, version_uuid_str: str) -> Response: + """Return the dashboard's state at a specific version. + --- + get: + summary: Read-only snapshot of the dashboard at a given version + parameters: + - in: path + schema: + type: string + format: uuid + name: uuid_str + description: Dashboard UUID + - in: path + schema: + type: string + format: uuid + name: version_uuid_str + description: Version UUID as returned by the list endpoint + responses: + 200: + description: Snapshot of the dashboard at the target version + content: + application/json: + schema: + type: object + properties: + result: + type: object + description: >- + The dashboard's scalar fields at the target version + (entity-specific keys), plus a `_version` block + with the version-level metadata. + properties: + _version: + $ref: '#/components/schemas/VersionListItemSchema' + 400: + $ref: '#/components/responses/400' + 401: + $ref: '#/components/responses/401' + 403: + $ref: '#/components/responses/403' + 404: + $ref: '#/components/responses/404' + """ + return get_version_endpoint( + self, Dashboard, uuid_str, version_uuid_str, access_kwarg="dashboard" + ) diff --git a/superset/datasets/api.py b/superset/datasets/api.py index 4801e3afcc5c..ac4aaae7d6cd 100644 --- a/superset/datasets/api.py +++ b/superset/datasets/api.py @@ -58,6 +58,7 @@ from superset.constants import MODEL_API_RW_METHOD_PERMISSION_MAP, RouteMethod from superset.daos.dashboard import DashboardDAO from superset.daos.dataset import DatasetDAO +from superset.daos.version import VersionDAO from superset.databases.filters import DatabaseFilter from superset.datasets.filters import DatasetCertifiedFilter, DatasetIsNullOrEmptyFilter from superset.datasets.schemas import ( @@ -74,10 +75,19 @@ GetOrCreateDatasetSchema, openapi_spec_methods_override, ) -from superset.exceptions import SupersetSyntaxErrorException, SupersetTemplateException +from superset.exceptions import ( + SupersetSyntaxErrorException, + SupersetTemplateException, +) from superset.jinja_context import BaseTemplateProcessor, get_template_processor from superset.utils import json from superset.utils.core import parse_boolean_string, sanitize_cookie_token +from superset.versioning.api_helpers import ( + get_version_endpoint, + list_versions_endpoint, +) +from superset.versioning.etag import set_version_etag +from superset.versioning.schemas import VersionListItemSchema from superset.views.base import DatasourceFilter from superset.views.base_api import ( BaseSupersetModelRestApi, @@ -112,6 +122,8 @@ class DatasetRestApi(BaseSupersetModelRestApi): "get_or_create_dataset", "warm_up_cache", "get_drill_info", + "list_versions", + "get_version", } list_columns = [ "id", @@ -300,6 +312,7 @@ class DatasetRestApi(BaseSupersetModelRestApi): DatasetRelatedObjectsResponse, DatasetDuplicateSchema, GetOrCreateDatasetSchema, + VersionListItemSchema, ) openapi_spec_methods = openapi_spec_methods_override @@ -417,6 +430,40 @@ def put(self, pk: int) -> Response: type: number result: $ref: '#/components/schemas/{{self.__class__.__name__}}.put' + old_version: + type: integer + nullable: true + description: >- + 0-based version_number of the live row before this + update (null if the dataset had no prior history). + Matches the ``version_number`` field of the list + versions endpoint. Unstable under retention + pruning — see ``old_transaction_id`` for a stable + identifier. + new_version: + type: integer + nullable: true + description: >- + 0-based version_number of the newly-live row after + this update. Can equal ``old_version`` when no + versioned column changed, or when retention + pruning dropped an older closed row in the same + commit. + old_transaction_id: + type: integer + nullable: true + description: >- + Continuum transaction_id of the live row before + this update. Stable across retention pruning. + new_transaction_id: + type: integer + nullable: true + description: >- + Continuum transaction_id of the live row after + this update. When this differs from + ``old_transaction_id`` the update produced a new + version row (regardless of whether ``new_version`` + changed). 400: $ref: '#/components/responses/400' 401: @@ -440,17 +487,83 @@ def put(self, pk: int) -> Response: # This validates custom Schema with custom validations except ValidationError as error: return self.response_400(message=error.messages) + + # pylint: disable=import-outside-toplevel + from superset.extensions import db as _db + + pre_dataset = ( + _db.session.query(SqlaTable).filter(SqlaTable.id == pk).one_or_none() + ) + old_version = VersionDAO.current_version_number(SqlaTable, pk) + old_transaction_id = VersionDAO.current_live_transaction_id(SqlaTable, pk) + old_version_uuid = ( + VersionDAO.current_live_version_uuid(SqlaTable, pk, pre_dataset.uuid) + if pre_dataset is not None + else None + ) + try: + # Two commands, two commits, two Continuum transactions for an + # ``override_columns`` save — deliberately NOT merged into one + # transaction. A single-transaction design was attempted and + # reverted: ``DBEventLogger`` writes request logs through the + # SHARED scoped session and calls ``commit()`` / + # ``rollback()`` on it mid-request (superset/utils/log.py), + # so any save held uncommitted across a logged sub-action can + # be committed half-done (Postgres/MySQL) or rolled back + # entirely on a transient logger failure (SQLite's + # "database is locked"). Until the event logger gets its own + # session, per-command commit boundaries are the only shape + # whose failure modes are honest. Consequence the + # version-history UI must tolerate: one logical save can + # surface as two version transactions stamped the same second. changed_model = UpdateDatasetCommand(pk, item, override_columns).run() + # Capture the post-update identifiers BEFORE the refresh: + # RefreshDatasetCommand commits its own transaction, so reading + # afterwards would attribute the refresh's version to the + # user's update (and old→new would span two transactions). + new_version = VersionDAO.current_version_number(SqlaTable, changed_model.id) + new_transaction_id = VersionDAO.current_live_transaction_id( + SqlaTable, changed_model.id + ) + new_version_uuid = VersionDAO.current_live_version_uuid( + SqlaTable, changed_model.id, changed_model.uuid + ) + etag_version_uuid = new_version_uuid if override_columns: RefreshDatasetCommand(pk).run() - response = self.response(200, id=changed_model.id, result=item) + # The ETag must reflect the entity's *current live* version, + # which after the refresh is the refresh's transaction — + # re-read it rather than reusing the pre-refresh uuid. + etag_version_uuid = VersionDAO.current_live_version_uuid( + SqlaTable, changed_model.id, changed_model.uuid + ) + response = self.response( + 200, + id=changed_model.id, + result=item, + old_version=old_version, + new_version=new_version, + old_transaction_id=old_transaction_id, + new_transaction_id=new_transaction_id, + old_version_uuid=str(old_version_uuid) if old_version_uuid else None, + new_version_uuid=str(new_version_uuid) if new_version_uuid else None, + ) + set_version_etag(response, etag_version_uuid) except DatasetNotFoundError: response = self.response_404() except DatasetForbiddenError: response = self.response_403() except DatasetInvalidError as ex: response = self.response_422(message=ex.normalized_messages()) + except DatasetRefreshFailedError as ex: + logger.error( + "Error refreshing dataset during update %s: %s", + self.__class__.__name__, + str(ex), + exc_info=True, + ) + response = self.response_422(message=str(ex)) except DatasetUpdateFailedError as ex: logger.error( "Error updating model %s: %s", @@ -1302,7 +1415,10 @@ def get(self, id_or_uuid: str, **kwargs: Any) -> Response: except SupersetTemplateException as ex: return self.response(ex.status, message=str(ex)) - return self.response(200, **response) + return set_version_etag( + self.response(200, **response), + VersionDAO.current_live_version_uuid(SqlaTable, table.id, table.uuid), + ) @expose("//drill_info/", methods=("GET",)) @protect() @@ -1447,3 +1563,114 @@ def render_item_list(item_list: list[dict[str, Any]]) -> list[dict[str, Any]]: raise template_exception from ex return data + + @expose("//versions/", methods=("GET",)) + @protect() + @safe + @statsd_metrics + @event_logger.log_this_with_context( + action=lambda self, *args, **kwargs: f"{self.__class__.__name__}.list_versions", + log_to_statsd=False, + ) + def list_versions(self, uuid_str: str) -> Response: + """List version history for a dataset. + --- + get: + summary: Return the version history for a dataset + parameters: + - in: path + schema: + type: string + format: uuid + name: uuid_str + description: Dataset UUID + responses: + 200: + description: Version history ordered by oldest first + content: + application/json: + schema: + type: object + properties: + result: + type: array + items: + $ref: '#/components/schemas/VersionListItemSchema' + count: + type: integer + 400: + $ref: '#/components/responses/400' + 401: + $ref: '#/components/responses/401' + 403: + $ref: '#/components/responses/403' + 404: + $ref: '#/components/responses/404' + """ + return list_versions_endpoint( + self, SqlaTable, uuid_str, access_kwarg="datasource" + ) + + @expose( + "//versions//", + methods=("GET",), + ) + @protect() + @safe + @statsd_metrics + @event_logger.log_this_with_context( + action=lambda self, *args, **kwargs: f"{self.__class__.__name__}.get_version", # noqa: E501 + log_to_statsd=False, + ) + def get_version(self, uuid_str: str, version_uuid_str: str) -> Response: + """Return the dataset's state at a specific version. + --- + get: + summary: Read-only snapshot of the dataset at a given version + description: >- + Returns the dataset's scalar fields plus reconstructed + ``columns`` and ``metrics`` lists as they were at the target + version. Does not modify live state. + parameters: + - in: path + schema: + type: string + format: uuid + name: uuid_str + description: Dataset UUID + - in: path + schema: + type: string + format: uuid + name: version_uuid_str + description: Version UUID as returned by the list endpoint + responses: + 200: + description: Snapshot of the dataset at the target version + content: + application/json: + schema: + type: object + properties: + result: + type: object + description: >- + The dataset's scalar fields at the target version + (entity-specific keys), plus `columns` / `metrics` + as they were at that version, plus a `_version` + block with the version-level metadata. + properties: + _version: + $ref: '#/components/schemas/VersionListItemSchema' + 400: + $ref: '#/components/responses/400' + 401: + $ref: '#/components/responses/401' + 403: + $ref: '#/components/responses/403' + 404: + $ref: '#/components/responses/404' + """ + return get_version_endpoint( + self, SqlaTable, uuid_str, version_uuid_str, access_kwarg="datasource" + ) diff --git a/superset/extensions/__init__.py b/superset/extensions/__init__.py index e704a2a4048f..32ca2bf2c6ed 100644 --- a/superset/extensions/__init__.py +++ b/superset/extensions/__init__.py @@ -146,6 +146,31 @@ def init_app(self, app: Flask) -> None: celery_app = celery.Celery() csrf = CSRFProtect() db = get_sqla_class()() + +# make_versioned() MUST be called immediately after db is constructed and before +# any versioned model class is defined. Continuum patches the SQLAlchemy +# metaclass at call time; models constructed before this call are silently skipped. +from sqlalchemy_continuum import ( # noqa: E402 + make_versioned, + versioning_manager as _continuum_manager, +) + +from superset.versioning.factory import ( # noqa: E402 + SkipUnmodifiedPlugin, + VersioningFlaskPlugin, + VersionTransactionFactory, +) + +# Rename the transaction table from "transaction" (SQL reserved word) to +# "version_transaction" via the custom factory before make_versioned() fires. +_continuum_manager.transaction_cls = VersionTransactionFactory() + +make_versioned( + user_cls=None, + plugins=[VersioningFlaskPlugin(), SkipUnmodifiedPlugin()], + options={"strategy": "validity"}, +) + _event_logger: dict[str, Any] = {} encrypted_field_factory = EncryptedFieldFactory() event_logger = LocalProxy(lambda: _event_logger.get("event_logger")) diff --git a/superset/initialization/__init__.py b/superset/initialization/__init__.py index a0c17d78e1f1..a9e8269bcd24 100644 --- a/superset/initialization/__init__.py +++ b/superset/initialization/__init__.py @@ -613,6 +613,142 @@ def init_extensions(self) -> None: # Surface exceptions during initialization of extensions print(ex) + @staticmethod + def _remove_continuum_write_listeners() -> None: + """Detach SQLAlchemy-Continuum's own write listeners. + + ``make_versioned()`` runs unconditionally at import of + ``superset.extensions`` and registers Continuum's mapper, session, + and engine listeners — the ones that write shadow rows and + ``version_transaction`` rows on every flush. Skipping only the + custom baseline/change-record listeners would leave those running, + so with the kill-switch off the shadow tables would silently keep + accumulating, contradicting the documented contract. + + This is deliberately a *targeted subset* of + ``sqlalchemy_continuum.remove_versioning()``: that helper also + calls ``manager.reset()``, which clears ``version_class_map`` — + and ``version_class()`` would then silently return the live model + class, breaking the read-only ``/versions/`` endpoints this flag + promises to keep working. + + Idempotent: guarded on a representative listener so repeated app + initializations in one process (test fixtures) don't raise on + double-removal. + """ + # pylint: disable=import-outside-toplevel + import sqlalchemy as sa + from sqlalchemy_continuum import versioning_manager + + if not sa.event.contains( + sa.orm.Mapper, "after_insert", versioning_manager.track_inserts + ): + return # already detached by a prior init + versioning_manager.remove_operations_tracking(sa.orm.Mapper) + versioning_manager.remove_session_tracking(sa.orm.session.Session) + sa.event.remove( + sa.engine.Engine, + "before_execute", + versioning_manager.track_association_operations, + ) + sa.event.remove( + sa.engine.Engine, "rollback", versioning_manager.clear_connection + ) + sa.event.remove( + sa.engine.Engine, + "set_connection_execution_options", + versioning_manager.track_cloned_connections, + ) + + def init_versioning(self) -> None: + """Register SQLAlchemy-Continuum baseline and retention listeners. + + Must be called after all versioned model classes have been imported so + that VERSIONED_MODELS can be populated and configure_mappers() has run. + + ``ENABLE_VERSIONING_CAPTURE`` (default ``True``) gates the two + before-flush listener registrations. The flag is operational, not + feature: every deployment captures version history by default. The + switch exists so an operator who observes a versioning-induced + regression (e.g. a save-path slowdown attributable to the + change-record listener) can disable capture in + ``superset_config.py`` and restart workers — a 30-second recovery + instead of revert-and-redeploy. Shadow tables already created by + the migration stay; they just stop accumulating new rows. + """ + if not self.config.get("ENABLE_VERSIONING_CAPTURE", True): + logger.warning( + "versioning: ENABLE_VERSIONING_CAPTURE is False; " + "skipping baseline + change-record listener registration " + "and detaching Continuum's write listeners. Save-path " + "capture is disabled; existing shadow tables and " + "/versions/ endpoints continue to work read-only." + ) + self._remove_continuum_write_listeners() + return + + from sqlalchemy.orm import Session # noqa: F401 + from sqlalchemy_continuum import version_class + + from superset.connectors.sqla.models import SqlaTable + from superset.models.dashboard import Dashboard + from superset.models.slice import Slice + from superset.versioning.baseline import ( + register_baseline_listener, + VERSIONED_MODELS, + ) + + # Note: previously this block called ``configure_mappers()`` before + # importing the snapshot modules, believing their Table declarations + # needed ``version_transaction`` to exist. That's not actually the + # case — the snapshot tables reference ``version_transaction.id`` + # only at the DB level (via the migration); the SQLAlchemy Table + # objects here intentionally declare ``transaction_id`` as a plain + # ``BigInteger`` without a FK to avoid the resolution dependency. + # Removing the global ``configure_mappers()`` avoids eagerly + # resolving relationships in other unrelated models (notably + # Flask-AppBuilder's AuditMixin on classes like Tag, whose + # ``created_by`` primaryjoin only resolves under specific class + # registry states in SQLAlchemy 1.4). + from superset.versioning.changes import ( # noqa: E402 + register_change_record_listener, + ) + + # All versioned models — Dashboard / Slice / SqlaTable plus their + # children (TableColumn / SqlMetric) and the dashboard_slices + # M2M — go through Continuum's shadow tables. The JSON-snapshot + # path that previously backed dataset / dashboard child diffs + # has been removed (full-Continuum spike). + for model_cls in (Dashboard, Slice, SqlaTable): + try: + version_class(model_cls) # ensure Continuum wired this model + # Dedup guard: VERSIONED_MODELS is module-level state, and + # test fixtures initialize multiple Superset apps per + # process — without the check each re-init appends + # duplicate entries. + if model_cls not in VERSIONED_MODELS: + VERSIONED_MODELS.append(model_cls) + except Exception: # pylint: disable=broad-except + # Continuum failed to wire versioning for this model. We + # boot in degraded mode rather than failing startup, but a + # silent skip would hide that change capture has stopped for + # the model — so surface it at WARNING with the traceback. + logger.warning( + "Versioning is not wired for %s; change capture will be " + "skipped for it. This usually means Continuum did not " + "register a version class for the model.", + model_cls.__name__, + exc_info=True, + ) + + register_baseline_listener() + register_change_record_listener() + + # Retention pruning runs out-of-band as a scheduled Celery beat + # task, shipped as a separate stacked PR. The previous + # synchronous after_commit listener was retired so retention work + # doesn't add latency to user saves. + def init_app_in_ctx(self) -> None: """ Runs init logic in the context of the app @@ -639,6 +775,9 @@ def init_app_in_ctx(self) -> None: self.init_all_dependencies_and_extensions() + # Must run after all versioned models are imported and mappers configured. + self.init_versioning() + @staticmethod def _log_config_warning(message: str) -> None: top_banner = 80 * "-" + "\n" + 36 * " " + "WARNING\n" + 80 * "-" diff --git a/superset/migrations/versions/2026-05-28_19-50_56cd24c07170_add_versioning_tables.py b/superset/migrations/versions/2026-05-28_19-50_56cd24c07170_add_versioning_tables.py new file mode 100644 index 000000000000..141edafcec4c --- /dev/null +++ b/superset/migrations/versions/2026-05-28_19-50_56cd24c07170_add_versioning_tables.py @@ -0,0 +1,567 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""add_versioning_tables + +Creates the full schema backing entity versioning in a single +migration: + +1. ``version_transaction`` — audit log keyed by Continuum's per-flush + transaction id (plus a Postgres-specific id sequence). +2. **Parent shadow tables** mirroring each versioned entity's columns: + ``dashboards_version`` / ``slices_version`` / ``tables_version``. +3. ``version_changes`` — field-level diff log keyed to a + ``(transaction, entity)`` pair; each row describes one atomic change + (one field or one child-collection element) that occurred during a + save. +4. **Child shadow tables** for the collections Continuum auto-registers + when ``__versioned__`` is applied to ``TableColumn`` / ``SqlMetric`` + and the ``slices`` exclude is removed from + ``Dashboard.__versioned__``: ``table_columns_version`` / + ``sql_metrics_version`` / ``dashboard_slices_version``. + +All shadow tables follow the validity-strategy shape (mirrored columns ++ ``transaction_id`` / ``end_transaction_id`` / ``operation_type`` +bookkeeping with FKs to ``version_transaction.id``). The current +version row has ``end_transaction_id = NULL``. + +This migration replaces three iterative migrations from the spike phase +(``56cd24c07170``, ``e1f3c5a7b9d0``, ``f7a2b3c4d5e6``) that captured the +same schema in three steps as the feature was developed. Compacting +gives downstream operators one migration to apply / reverse and one +review surface. The ``revision`` hash is reused from the original first +migration so anyone still tracking the chain by that hash lands on the +same logical change set. + +Generated by hand because the current Continuum + Alembic-autogenerate +interaction trips on the renamed ``transaction`` -> ``version_transaction`` +table key (``KeyError`` lookups in ``table_key_to_table``). Column +inventories were sourced from the live model ``__table__`` definitions +and ``version_class(...).__table__`` / Continuum association metadata. + +Primary key choice. Both ``version_transaction.id`` and +``version_changes.id`` are ``BigInteger`` autoincrement — a deliberate +carveout from the project's UUID-PK convention for new models (see +``CLAUDE.md`` §"UUID Migration"). ``version_transaction`` is keyed +externally by SQLAlchemy-Continuum via +``nextval('version_transaction_id_seq')`` on every INSERT; matching +that contract is required for ``versioning_manager`` to function. +``version_changes`` follows the same shape because the user-facing +identity is the ``(transaction_id, entity_kind, entity_id, sequence)`` +composite unique key, not the row id; the API surfaces a deterministic +UUIDv5 ``version_uuid`` derived from ``entity.uuid`` and +``transaction_id`` for stable external references. + +See spec FR-016..FR-021, data-model.md §``version_changes`` / +§"Storage architecture", and the spike notes in +``spike-continuum-restore.md``. + +Revision ID: 56cd24c07170 +Revises: 2bee73611e32 +Create Date: 2026-05-28 19:50:00.000000 + +""" + +from __future__ import annotations + +import sqlalchemy as sa +from alembic import op +from sqlalchemy_utils import UUIDType + +from superset.utils.core import MediumText + +revision = "56cd24c07170" +# Stacked on the composite-PK association-tables change (2bee73611e32) so the +# Continuum shadow tables this migration creates can mirror the +# composite-PK shape of the live association tables. If that change +# is removed from the stack, this should be reverted to "ce6bd21901ab". +down_revision = "2bee73611e32" + + +def upgrade() -> None: + bind = op.get_bind() + + # ------------------------------------------------------------------ + # version_transaction + # + # Audit log for each versioning event. Continuum emits + # ``nextval('version_transaction_id_seq')`` on every INSERT, so the + # sequence must exist before the table on Postgres. SQLite/MySQL + # ignore the explicit CREATE SEQUENCE (they auto-increment natively). + # ------------------------------------------------------------------ + if bind.dialect.name == "postgresql": + op.execute("CREATE SEQUENCE IF NOT EXISTS version_transaction_id_seq") + + op.create_table( + "version_transaction", + sa.Column( + "id", + sa.BigInteger(), + sa.Sequence("version_transaction_id_seq"), + primary_key=True, + autoincrement=True, + nullable=False, + ), + sa.Column("issued_at", sa.DateTime(), nullable=True), + sa.Column("remote_addr", sa.String(50), nullable=True), + sa.Column("user_id", sa.Integer(), nullable=True), + # ``action_kind`` carries the high-level avenue that produced + # this transaction (``restore`` / ``import`` / ``clone``). + # ``NULL`` is the default "ordinary save" — most rows leave + # this empty. Commands set + # ``session.info["_versioning_action_kind"]`` before commit; + # the change-record listener stamps the value here. Parallel + # to ``version_changes.entity_kind`` and ``version_changes.kind`` + # — the schema's third ``*_kind`` column, at transaction scope. + sa.Column("action_kind", sa.String(32), nullable=True), + ) + + if bind.dialect.name == "postgresql": + op.execute( + "ALTER SEQUENCE version_transaction_id_seq OWNED BY version_transaction.id" + ) + + # ------------------------------------------------------------------ + # dashboards_version + # ------------------------------------------------------------------ + op.create_table( + "dashboards_version", + sa.Column("uuid", UUIDType(binary=True), nullable=True), + sa.Column("id", sa.Integer(), nullable=False), + sa.Column("dashboard_title", sa.String(500), nullable=True), + # ``MediumText()`` mirrors the live column type — on MySQL plain + # ``TEXT`` caps at 64 KB, which large dashboards exceed; an + # oversized live write would then fail the shadow INSERT under + # ``STRICT_TRANS_TABLES`` (or silently truncate without it) and + # corrupt the history. Postgres ``TEXT`` is unbounded and SQLite + # ignores the length annotation so this is MySQL-driven. + sa.Column("position_json", MediumText(), nullable=True), + sa.Column("description", sa.Text(), nullable=True), + sa.Column("css", MediumText(), nullable=True), + sa.Column("theme_id", sa.Integer(), nullable=True), + sa.Column("certified_by", sa.Text(), nullable=True), + sa.Column("certification_details", sa.Text(), nullable=True), + sa.Column("json_metadata", MediumText(), nullable=True), + sa.Column("slug", sa.String(255), nullable=True), + sa.Column("published", sa.Boolean(), nullable=True), + sa.Column("is_managed_externally", sa.Boolean(), nullable=True), + sa.Column("external_url", sa.Text(), nullable=True), + sa.Column("transaction_id", sa.BigInteger(), nullable=False), + sa.Column("end_transaction_id", sa.BigInteger(), nullable=True), + sa.Column("operation_type", sa.SmallInteger(), nullable=False), + sa.PrimaryKeyConstraint("id", "transaction_id"), + sa.ForeignKeyConstraint( + ["transaction_id"], + ["version_transaction.id"], + name="fk_dashboards_version_transaction_id", + ), + sa.ForeignKeyConstraint( + ["end_transaction_id"], + ["version_transaction.id"], + name="fk_dashboards_version_end_transaction_id", + ), + ) + op.create_index( + "ix_dashboards_version_end_transaction_id", + "dashboards_version", + ["end_transaction_id"], + ) + op.create_index( + "ix_dashboards_version_operation_type", + "dashboards_version", + ["operation_type"], + ) + op.create_index( + "ix_dashboards_version_transaction_id", + "dashboards_version", + ["transaction_id"], + ) + + # ------------------------------------------------------------------ + # slices_version (Charts) + # ------------------------------------------------------------------ + op.create_table( + "slices_version", + sa.Column("uuid", UUIDType(binary=True), nullable=True), + sa.Column("id", sa.Integer(), nullable=False), + sa.Column("slice_name", sa.String(250), nullable=True), + sa.Column("datasource_id", sa.Integer(), nullable=True), + sa.Column("datasource_type", sa.String(200), nullable=True), + sa.Column("datasource_name", sa.String(2000), nullable=True), + sa.Column("viz_type", sa.String(250), nullable=True), + sa.Column("params", MediumText(), nullable=True), + sa.Column("description", sa.Text(), nullable=True), + sa.Column("cache_timeout", sa.Integer(), nullable=True), + sa.Column("certified_by", sa.Text(), nullable=True), + sa.Column("certification_details", sa.Text(), nullable=True), + sa.Column("is_managed_externally", sa.Boolean(), nullable=True), + sa.Column("external_url", sa.Text(), nullable=True), + sa.Column("transaction_id", sa.BigInteger(), nullable=False), + sa.Column("end_transaction_id", sa.BigInteger(), nullable=True), + sa.Column("operation_type", sa.SmallInteger(), nullable=False), + sa.PrimaryKeyConstraint("id", "transaction_id"), + sa.ForeignKeyConstraint( + ["transaction_id"], + ["version_transaction.id"], + name="fk_slices_version_transaction_id", + ), + sa.ForeignKeyConstraint( + ["end_transaction_id"], + ["version_transaction.id"], + name="fk_slices_version_end_transaction_id", + ), + ) + op.create_index( + "ix_slices_version_end_transaction_id", + "slices_version", + ["end_transaction_id"], + ) + op.create_index( + "ix_slices_version_operation_type", + "slices_version", + ["operation_type"], + ) + op.create_index( + "ix_slices_version_transaction_id", + "slices_version", + ["transaction_id"], + ) + + # ------------------------------------------------------------------ + # tables_version (SqlaTable / Datasets) + # ------------------------------------------------------------------ + op.create_table( + "tables_version", + sa.Column("uuid", UUIDType(binary=True), nullable=True), + sa.Column("id", sa.Integer(), nullable=False), + sa.Column("description", sa.Text(), nullable=True), + sa.Column("default_endpoint", sa.Text(), nullable=True), + sa.Column("is_featured", sa.Boolean(), nullable=True), + sa.Column("filter_select_enabled", sa.Boolean(), nullable=True), + sa.Column("offset", sa.Integer(), nullable=True), + sa.Column("cache_timeout", sa.Integer(), nullable=True), + sa.Column("params", sa.String(1000), nullable=True), + sa.Column("is_managed_externally", sa.Boolean(), nullable=True), + sa.Column("external_url", sa.Text(), nullable=True), + sa.Column("table_name", sa.String(250), nullable=True), + sa.Column("main_dttm_col", sa.String(250), nullable=True), + sa.Column("currency_code_column", sa.String(250), nullable=True), + sa.Column("database_id", sa.Integer(), nullable=True), + sa.Column("fetch_values_predicate", sa.Text(), nullable=True), + sa.Column("schema", sa.String(255), nullable=True), + sa.Column("catalog", sa.String(256), nullable=True), + sa.Column("sql", MediumText(), nullable=True), + sa.Column("is_sqllab_view", sa.Boolean(), nullable=True), + sa.Column("template_params", sa.Text(), nullable=True), + sa.Column("extra", sa.Text(), nullable=True), + sa.Column("normalize_columns", sa.Boolean(), nullable=True), + sa.Column("always_filter_main_dttm", sa.Boolean(), nullable=True), + sa.Column("folders", sa.JSON(), nullable=True), + sa.Column("transaction_id", sa.BigInteger(), nullable=False), + sa.Column("end_transaction_id", sa.BigInteger(), nullable=True), + sa.Column("operation_type", sa.SmallInteger(), nullable=False), + sa.PrimaryKeyConstraint("id", "transaction_id"), + sa.ForeignKeyConstraint( + ["transaction_id"], + ["version_transaction.id"], + name="fk_tables_version_transaction_id", + ), + sa.ForeignKeyConstraint( + ["end_transaction_id"], + ["version_transaction.id"], + name="fk_tables_version_end_transaction_id", + ), + ) + op.create_index( + "ix_tables_version_end_transaction_id", + "tables_version", + ["end_transaction_id"], + ) + op.create_index( + "ix_tables_version_operation_type", + "tables_version", + ["operation_type"], + ) + op.create_index( + "ix_tables_version_transaction_id", + "tables_version", + ["transaction_id"], + ) + + # ------------------------------------------------------------------ + # version_changes + # + # Field-level diff log keyed to a (transaction, entity) pair. Each + # row describes one atomic change (one field or one child-collection + # element) that occurred to one entity during a save. See spec + # FR-016..FR-021 and data-model.md §version_changes. + # + # ``(entity_kind, entity_id)`` is a polymorphic reference: depending + # on ``entity_kind`` (``"chart"`` / ``"dashboard"`` / ``"dataset"``) + # the ``entity_id`` is the integer PK on ``slices`` / ``dashboards`` / + # ``tables`` respectively. SQL has no native polymorphic FK, so the + # constraint is intentionally omitted — cleanup relies on the + # ``CASCADE`` from ``version_transaction.id`` plus command-layer + # ordering for entity deletes (the command that hard-deletes the + # entity runs inside the same transaction that prunes its history). + # A bare ``DELETE FROM WHERE id = X`` outside that + # transactional boundary leaves orphan ``version_changes`` rows + # whose ``entity_id`` references a vanished row — the read-side + # tombstone-state lookup handles this gracefully. + # ------------------------------------------------------------------ + op.create_table( + "version_changes", + sa.Column( + "id", + sa.BigInteger(), + primary_key=True, + autoincrement=True, + nullable=False, + ), + sa.Column( + "transaction_id", + sa.BigInteger(), + sa.ForeignKey("version_transaction.id", ondelete="CASCADE"), + nullable=False, + ), + sa.Column( + "entity_kind", + sa.String(length=32), + nullable=False, + ), + sa.Column( + "entity_id", + sa.Integer(), + nullable=False, + ), + sa.Column( + # Integer, not SmallInteger: per-entity sequence within one + # transaction is assigned by unbounded enumerate(); a + # pathological diff (e.g. a giant position_json rewrite) could + # overflow SmallInteger's 32767 on Postgres/MySQL. + "sequence", + sa.Integer(), + nullable=False, + ), + sa.Column( + "kind", + sa.String(length=32), + nullable=False, + ), + # ``operation`` is the per-record verb: ``add`` / ``remove`` / + # ``move`` / ``edit``. ``move`` only fires for layout records; + # the other three apply across every emit site. Made explicit + # so consumers don't have to infer the verb from ``from_value`` + # / ``to_value`` null-tests or from ``path[0]`` for layout records. + sa.Column( + "operation", + sa.String(length=16), + nullable=False, + ), + sa.Column("path", sa.JSON(), nullable=False), + sa.Column("from_value", sa.JSON(), nullable=True), + sa.Column("to_value", sa.JSON(), nullable=True), + sa.UniqueConstraint( + "transaction_id", + "entity_kind", + "entity_id", + "sequence", + name="uq_version_changes_tx_entity_sequence", + ), + ) + op.create_index( + "ix_version_changes_kind", + "version_changes", + ["kind"], + ) + op.create_index( + "ix_version_changes_entity", + "version_changes", + ["entity_kind", "entity_id"], + ) + + # ------------------------------------------------------------------ + # table_columns_version + # ------------------------------------------------------------------ + op.create_table( + "table_columns_version", + sa.Column("uuid", UUIDType(binary=True), nullable=True), + sa.Column("id", sa.Integer(), nullable=False), + sa.Column("column_name", sa.String(255), nullable=True), + sa.Column("verbose_name", sa.String(1024), nullable=True), + sa.Column("is_active", sa.Boolean(), nullable=True), + sa.Column("type", sa.Text(), nullable=True), + sa.Column("advanced_data_type", sa.String(255), nullable=True), + sa.Column("groupby", sa.Boolean(), nullable=True), + sa.Column("filterable", sa.Boolean(), nullable=True), + sa.Column("description", MediumText(), nullable=True), + sa.Column("table_id", sa.Integer(), nullable=True), + sa.Column("is_dttm", sa.Boolean(), nullable=True), + sa.Column("expression", MediumText(), nullable=True), + sa.Column("python_date_format", sa.String(255), nullable=True), + sa.Column("datetime_format", sa.String(100), nullable=True), + sa.Column("extra", sa.Text(), nullable=True), + sa.Column("transaction_id", sa.BigInteger(), nullable=False), + sa.Column("end_transaction_id", sa.BigInteger(), nullable=True), + sa.Column("operation_type", sa.SmallInteger(), nullable=False), + sa.PrimaryKeyConstraint("id", "transaction_id"), + sa.ForeignKeyConstraint( + ["transaction_id"], + ["version_transaction.id"], + name="fk_table_columns_version_transaction_id", + ), + sa.ForeignKeyConstraint( + ["end_transaction_id"], + ["version_transaction.id"], + name="fk_table_columns_version_end_transaction_id", + ), + ) + op.create_index( + "ix_table_columns_version_end_transaction_id", + "table_columns_version", + ["end_transaction_id"], + ) + op.create_index( + "ix_table_columns_version_operation_type", + "table_columns_version", + ["operation_type"], + ) + op.create_index( + "ix_table_columns_version_transaction_id", + "table_columns_version", + ["transaction_id"], + ) + + # ------------------------------------------------------------------ + # sql_metrics_version + # ------------------------------------------------------------------ + op.create_table( + "sql_metrics_version", + sa.Column("uuid", UUIDType(binary=True), nullable=True), + sa.Column("id", sa.Integer(), nullable=False), + sa.Column("metric_name", sa.String(255), nullable=True), + sa.Column("verbose_name", sa.String(1024), nullable=True), + sa.Column("metric_type", sa.String(32), nullable=True), + sa.Column("description", MediumText(), nullable=True), + sa.Column("d3format", sa.String(128), nullable=True), + sa.Column("currency", sa.JSON(), nullable=True), + sa.Column("warning_text", sa.Text(), nullable=True), + sa.Column("table_id", sa.Integer(), nullable=True), + sa.Column("expression", MediumText(), nullable=True), + sa.Column("extra", sa.Text(), nullable=True), + sa.Column("transaction_id", sa.BigInteger(), nullable=False), + sa.Column("end_transaction_id", sa.BigInteger(), nullable=True), + sa.Column("operation_type", sa.SmallInteger(), nullable=False), + sa.PrimaryKeyConstraint("id", "transaction_id"), + sa.ForeignKeyConstraint( + ["transaction_id"], + ["version_transaction.id"], + name="fk_sql_metrics_version_transaction_id", + ), + sa.ForeignKeyConstraint( + ["end_transaction_id"], + ["version_transaction.id"], + name="fk_sql_metrics_version_end_transaction_id", + ), + ) + op.create_index( + "ix_sql_metrics_version_end_transaction_id", + "sql_metrics_version", + ["end_transaction_id"], + ) + op.create_index( + "ix_sql_metrics_version_operation_type", + "sql_metrics_version", + ["operation_type"], + ) + op.create_index( + "ix_sql_metrics_version_transaction_id", + "sql_metrics_version", + ["transaction_id"], + ) + + # ------------------------------------------------------------------ + # dashboard_slices_version (M2M association) + # + # The live ``dashboard_slices`` table is reshaped to a + # composite PK on ``(dashboard_id, slice_id)`` — no surrogate ``id``. + # Continuum auto-mirrors the live columns into the shadow Table at + # ``make_versioned()`` time, so the shadow's SQLAlchemy metadata + # also has no ``id``. The DB shadow PK is the natural composite key + # plus Continuum's bookkeeping (``transaction_id``, ``operation_type``); + # ``operation_type`` is included because a single transaction can in + # principle produce both INSERT and DELETE shadows for the same + # ``(dashboard_id, slice_id)`` pair (slice removed and re-added in + # one save). + # + # If that reshape is removed from the stack, the live table reverts to + # carrying its surrogate ``id`` and this migration would need to + # match — see ``spike-continuum-restore.md`` "Branch maintenance". + # ------------------------------------------------------------------ + op.create_table( + "dashboard_slices_version", + sa.Column("dashboard_id", sa.Integer(), nullable=False), + sa.Column("slice_id", sa.Integer(), nullable=False), + sa.Column("transaction_id", sa.BigInteger(), nullable=False), + sa.Column("end_transaction_id", sa.BigInteger(), nullable=True), + sa.Column("operation_type", sa.SmallInteger(), nullable=False), + sa.PrimaryKeyConstraint( + "dashboard_id", "slice_id", "transaction_id", "operation_type" + ), + sa.ForeignKeyConstraint( + ["transaction_id"], + ["version_transaction.id"], + name="fk_dashboard_slices_version_transaction_id", + ), + sa.ForeignKeyConstraint( + ["end_transaction_id"], + ["version_transaction.id"], + name="fk_dashboard_slices_version_end_transaction_id", + ), + ) + op.create_index( + "ix_dashboard_slices_version_end_transaction_id", + "dashboard_slices_version", + ["end_transaction_id"], + ) + op.create_index( + "ix_dashboard_slices_version_operation_type", + "dashboard_slices_version", + ["operation_type"], + ) + op.create_index( + "ix_dashboard_slices_version_transaction_id", + "dashboard_slices_version", + ["transaction_id"], + ) + + +def downgrade() -> None: + # Drop in reverse dependency order: children with FKs to + # ``version_transaction`` drop first; ``version_transaction`` and its + # sequence drop last. + op.drop_table("dashboard_slices_version") + op.drop_table("sql_metrics_version") + op.drop_table("table_columns_version") + op.drop_table("version_changes") + op.drop_table("tables_version") + op.drop_table("slices_version") + op.drop_table("dashboards_version") + op.drop_table("version_transaction") + + bind = op.get_bind() + if bind.dialect.name == "postgresql": + op.execute("DROP SEQUENCE IF EXISTS version_transaction_id_seq") diff --git a/superset/migrations/versions/2026-06-03_12-00_8f3a1b2c4d5e_shadow_live_row_indexes.py b/superset/migrations/versions/2026-06-03_12-00_8f3a1b2c4d5e_shadow_live_row_indexes.py new file mode 100644 index 000000000000..68723578588c --- /dev/null +++ b/superset/migrations/versions/2026-06-03_12-00_8f3a1b2c4d5e_shadow_live_row_indexes.py @@ -0,0 +1,173 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""shadow_live_row_indexes + +Adds per-shadow-table indexes covering the canonical "current live row +of entity X" lookup that ``find_active_by_uuid`` / ``list_versions`` / +``get_version`` / restore validation / activity-view all funnel +through: + + SELECT ... FROM _version + WHERE id = ? AND end_transaction_id IS NULL + +The base migration (``56cd24c07170_add_versioning_tables``) created +single-column indexes on ``transaction_id``, ``end_transaction_id``, +and ``operation_type``, but nothing covering the predicate combination +that actually runs in hot paths. + +Index choice is dialect-specific: + +* **PostgreSQL / SQLite** — partial index over the entity ``id`` with + ``WHERE end_transaction_id IS NULL``. Cuts the index size to one row + per live entity (vs. one row per historical version) and turns the + hot lookup into a single index probe. +* **MySQL** — partial indexes aren't supported; use a plain composite + ``(id, end_transaction_id)``. MySQL's optimizer handles the + ``IS NULL`` predicate against the composite efficiently. + +It also adds a composite ``(table_id, transaction_id)`` index on the two +child shadow tables (``table_columns_version`` / ``sql_metrics_version``). +The dataset child-diff path queries these by parent ``table_id`` plus a +transaction-range bound, neither of which the base migration's +single-column indexes nor the ``id``-leading PK can serve: + + SELECT ... FROM table_columns_version + WHERE table_id = ? AND transaction_id <= ? AND ... (shadow_rows_valid_at) + + SELECT max(transaction_id) FROM table_columns_version + WHERE table_id = ? AND transaction_id < ? (prior-tx probe) + +A plain composite leading with ``table_id`` serves both on every dialect, +so no partial-index split is needed here. + +Surfaced by sqlalchemy-review pass W-NEW-4 (live-row lookup) and a +Codex sqlalchemy-review pass (child-diff ``table_id`` lookup). + +Revision ID: 8f3a1b2c4d5e +Revises: 56cd24c07170 +Create Date: 2026-06-03 12:00:00.000000 + +""" + +from __future__ import annotations + +import sqlalchemy as sa +from alembic import op + +revision = "8f3a1b2c4d5e" +down_revision = "56cd24c07170" + + +# The parent + child shadow tables, all of which carry an ``id`` +# column (mirroring the live entity's integer PK). ``dashboard_slices_version`` +# is intentionally excluded: it's the M2M association shadow with a +# composite PK ``(dashboard_id, slice_id, transaction_id, operation_type)`` +# and no ``id`` column. The canonical "live row" lookup doesn't apply to +# the M2M shadow — readers query it by ``transaction_id`` (already +# indexed by the base migration) when reconstructing per-tx changes. +SHADOW_TABLES: tuple[str, ...] = ( + "dashboards_version", + "slices_version", + "tables_version", + "table_columns_version", + "sql_metrics_version", +) + + +# Child shadow tables whose rows are looked up by parent ``table_id`` plus a +# transaction-range bound on the dataset child-diff path. Both carry a +# nullable ``table_id`` mirroring the live row's FK to ``tables.id``. +CHILD_SHADOW_TABLES: tuple[str, ...] = ( + "table_columns_version", + "sql_metrics_version", +) + + +def _index_name(table: str) -> str: + return f"ix_{table}_live_id" + + +def _child_index_name(table: str) -> str: + return f"ix_{table}_table_id_transaction_id" + + +def upgrade() -> None: + bind = op.get_bind() + dialect = bind.dialect.name + + where_clause = sa.text("end_transaction_id IS NULL") + + for table in SHADOW_TABLES: + index_name = _index_name(table) + if dialect == "postgresql": + op.create_index( + index_name, + table, + ["id"], + unique=False, + postgresql_where=where_clause, + ) + elif dialect == "sqlite": + op.create_index( + index_name, + table, + ["id"], + unique=False, + sqlite_where=where_clause, + ) + else: + # MySQL (and any unknown dialect): partial indexes aren't + # supported, so use a plain composite. MySQL's optimizer + # handles ``id = ? AND end_transaction_id IS NULL`` against + # the composite efficiently. + op.create_index( + index_name, + table, + ["id", "end_transaction_id"], + unique=False, + ) + + # Child-diff access pattern: filter by parent ``table_id`` plus a + # transaction-range bound. A plain composite serves this on every + # dialect, so no partial-index split is needed. + for table in CHILD_SHADOW_TABLES: + op.create_index( + _child_index_name(table), + table, + ["table_id", "transaction_id"], + unique=False, + ) + + +def downgrade() -> None: + # Probe the inspector instead of emitting ``DROP INDEX IF EXISTS``: + # stock MySQL (5.7/8.x) has no IF EXISTS grammar for DROP INDEX + # (it's a MariaDB extension), so the clause is not dialect-portable. + # The existence check keeps the downgrade robust against a + # partial-application failure on upgrade (e.g. the first + # ``op.create_index`` succeeded under Postgres' transactional DDL but + # a later one failed and rolled back the rest — repeated downgrade + # must not raise on the missing indexes). + inspector = sa.inspect(op.get_bind()) + for table in SHADOW_TABLES: + index_name = _index_name(table) + if any(ix["name"] == index_name for ix in inspector.get_indexes(table)): + op.drop_index(index_name, table_name=table) + for table in CHILD_SHADOW_TABLES: + index_name = _child_index_name(table) + if any(ix["name"] == index_name for ix in inspector.get_indexes(table)): + op.drop_index(index_name, table_name=table) diff --git a/superset/models/dashboard.py b/superset/models/dashboard.py index 559ff273194d..25395813d4a6 100644 --- a/superset/models/dashboard.py +++ b/superset/models/dashboard.py @@ -147,6 +147,27 @@ class Dashboard(CoreDashboard, AuditMixinNullable, ImportExportMixin): """The dashboard object!""" __tablename__ = "dashboards" + # deleted_at exclusion will be added when soft delete is merged. + # SPIKE (full-Continuum): ``slices`` removed from + # the exclude list so Continuum auto-creates an association version table + # for ``dashboard_slices`` and ``Reverter(relations=["slices"])`` can + # restore chart membership. Owners / roles stay excluded — access metadata, + # not user-authored content (ADR-005). + # Audit columns (changed_on/created_on/changed_by_fk/created_by_fk) are + # auto-bumped by AuditMixin on every save; excluding them lets Continuum's + # is_modified() return False on no-op saves (e.g. owners-only edits) so we + # don't create empty version rows. version_transaction.user_id / + # issued_at preserve "who/when" without per-row duplication. + __versioned__: dict[str, Any] = { + "exclude": [ + "owners", + "roles", + "changed_on", + "created_on", + "changed_by_fk", + "created_by_fk", + ] + } id = Column(Integer, primary_key=True) dashboard_title = Column(String(500)) position_json = Column(utils.MediumText()) diff --git a/superset/models/helpers.py b/superset/models/helpers.py index b603f0bb3cea..b8b0aa7100a1 100644 --- a/superset/models/helpers.py +++ b/superset/models/helpers.py @@ -265,6 +265,53 @@ class UUIDMixin: # pylint: disable=too-few-public-methods UUIDType(binary=True), primary_key=False, unique=True, default=uuid.uuid4 ) + @validates("uuid") + def _coerce_uuid(self, key: str, value: Any) -> Any: # noqa: ARG002 + """Coerce well-formed UUID strings to ``uuid.UUID`` on assignment; + pass everything else through untouched. + + **Why coerce.** ``UUIDType`` only converts at SQL bind / SQL + result time. Importers and ad-hoc construction + (``SqlMetric(uuid="…string…")``) leave the in-memory attribute + as a ``str`` until the next DB round-trip refreshes it. With + SQLAlchemy-Continuum versioning attached to a child mapper + (``TableColumn`` / ``SqlMetric``), the post-INSERT attribute- + expire behaviour changes enough that the refresh doesn't happen + before the caller reads the attribute — breaking equality + assertions like ``test_import_dataset``'s + ``metric.uuid == uuid.UUID(...)`` because str ≠ UUID. Coercing + defensively here makes the in-memory attribute always a UUID + regardless of provenance. + + **Why the non-UUID-string escape hatch.** Tightening this + validator to raise on non-UUID strings would break a small set + of existing unit tests that use human-readable placeholder + strings as fixture uuids (e.g. + ``test_dashboard_schemas.py``'s ``"dashboard-uuid-7"`` and + analogous placeholders in importer tests). The fixtures use + these placeholders for legibility — they're only ever compared + by string equality, never written to a real database. Letting + them through unchanged keeps the fixtures working at the cost + of deferring "real" UUID malformation to the SQL bind layer, + which raises a clearer "invalid input syntax for type uuid" + error keyed to the actual column. + + **Tightening path** (if amin M1 is ever revisited): replace + the ``return value`` in the ``except`` branch with + ``raise ValueError(f"Invalid UUID: {value!r}")``, then run the + unit test suite and migrate any remaining placeholder fixtures + to ``uuid.uuid4()`` (use + ``rg '''SqlMetric\\(uuid="[^"]*"|"dashboard-uuid|"slice-uuid'''`` + to find them). The full migration touches ~5–10 fixture files + and is non-breaking outside tests. + """ + if isinstance(value, str): + try: + return uuid.UUID(value) + except ValueError: + return value + return value + @property def short_uuid(self) -> str: return str(self.uuid)[:8] @@ -547,14 +594,23 @@ def remove_params(self, param_to_remove: str) -> None: def reset_ownership(self) -> None: """object will belong to the user the current user""" - # make sure the object doesn't have relations to a user - # it will be filled by appbuilder on save - self.created_by = None - self.changed_by = None - # flask global context might not exist (in cli or tests for example) + # Reset the audit pointers. When a Flask request context is + # available we explicitly stamp the current user, otherwise we + # leave the attributes unset so Flask-AppBuilder's column + # defaults fill them in on save. An explicit assignment is + # required because once the ``created_by`` / ``changed_by`` + # relationships are configured (which happens eagerly on models + # registered with SQLAlchemy-Continuum), setting them to + # ``None`` propagates to the FK column and suppresses the + # ``default=`` callable. self.owners = [] - if g and hasattr(g, "user"): + if g and hasattr(g, "user") and g.user: + self.created_by = g.user + self.changed_by = g.user self.owners = [g.user] + else: + self.created_by = None + self.changed_by = None @property def params_dict(self) -> dict[Any, Any]: diff --git a/superset/models/slice.py b/superset/models/slice.py index d1a954791f1d..1975ca43c5c6 100644 --- a/superset/models/slice.py +++ b/superset/models/slice.py @@ -83,6 +83,38 @@ class Slice( # pylint: disable=too-many-public-methods query_context_factory: QueryContextFactory | None = None __tablename__ = "slices" + # query_context is excluded: it is a cached/regenerated field, not user-authored. + # deleted_at exclusion will be added when soft delete is merged. + # Exclude M2M association relationships: Continuum only captures FK columns on + # association INSERTs (not the auto-increment id), which breaks the NOT NULL PK. + # Ownership changes are administrative metadata, not user-authored content. + # Audit / save-marker columns are auto-bumped on every save. Excluding + # them lets Continuum's is_modified() return False on no-op saves + # (e.g. owners-only edits) so we don't create empty version rows. + # version_transaction.user_id / issued_at preserve "who/when". + # The perm-string class (perm / schema_perm / catalog_perm) is derived + # security state, not user-authored content: permission maintenance + # rewrites it in bulk, and versioning it produced phantom transactions + # flooding the activity stream (10 "Chart updated" rows for one user + # save — surfaced by the version-history UI, PR #40988). Excluding it + # also means a restore can't resurrect stale permission strings; the + # live, derived values stay authoritative. + __versioned__: dict[str, Any] = { + "exclude": [ + "query_context", + "owners", + "dashboards", + "changed_on", + "created_on", + "changed_by_fk", + "created_by_fk", + "last_saved_at", + "last_saved_by_fk", + "perm", + "schema_perm", + "catalog_perm", + ] + } id = Column(Integer, primary_key=True) slice_name = Column(String(250)) datasource_id = Column(Integer) @@ -337,7 +369,11 @@ def chart(self) -> str: @property def slice_link(self) -> Markup: name = escape(self.chart) - return Markup(f'{name}') + # ``self.url`` is ``/explore/?slice_id=``; the only + # interpolation is the integer primary key, so the URL has no + # user-controlled segment to escape (unlike ``Dashboard.url`` + # which embeds the user-set slug). ``noqa: S704`` is safe. + return Markup(f'{name}') # noqa: S704 @property def icons(self) -> str: diff --git a/superset/versioning/__init__.py b/superset/versioning/__init__.py new file mode 100644 index 000000000000..13a83393a912 --- /dev/null +++ b/superset/versioning/__init__.py @@ -0,0 +1,16 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. diff --git a/superset/versioning/api_helpers.py b/superset/versioning/api_helpers.py new file mode 100644 index 000000000000..6c3debe1c2a7 --- /dev/null +++ b/superset/versioning/api_helpers.py @@ -0,0 +1,143 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Shared handlers for the ``/versions/`` REST endpoints. + +Each ``ChartRestApi`` / ``DashboardRestApi`` / ``DatasetRestApi`` carries +the same read endpoint methods — ``list_versions`` and ``get_version`` — +whose bodies are byte-for-byte identical apart from the model class and +the ``security_manager.raise_for_access`` kwarg. Extracting the bodies +here lets each per-resource method collapse to a single delegation call, +while the OpenAPI docstring + FAB decorators stay at the method site +where they belong. + +(The restore endpoint ships in a later PR; only the read endpoints are +wired here.) +""" + +from __future__ import annotations + +from typing import Any +from uuid import UUID + +from flask import Response +from flask_appbuilder import Model + +from superset.daos.version import VersionDAO +from superset.exceptions import SupersetSecurityException +from superset.extensions import security_manager +from superset.versioning.etag import set_version_etag_by_uuid +from superset.versioning.schemas import VersionListItemSchema + +#: Serializer for version rows (list items and the ``_version`` block of a +#: single-version snapshot — same shape). Dumping through marshmallow +#: instead of handing raw dicts to ``jsonify`` keeps ``issued_at`` +#: ISO-8601 (Flask's default JSON provider renders datetimes as RFC-1123 +#: http-dates) and ``version_uuid`` consistently a string (the list rows +#: carry UUID instances, the snapshot block pre-stringifies). +_version_item_schema = VersionListItemSchema() + + +def _resolve_entity( + api: Any, + model_cls: type[Model], + uuid_str: str, + access_kwarg: str, +) -> tuple[Any, UUID] | Response: + """Parse the path UUID, look up the live entity, run the read-access + gate. + + Returns ``(entity, entity_uuid)`` on success or a pre-built + ``Response`` (400 / 403 / 404) that the caller should return + directly. The split shape keeps the call site terse and lets the + three handler functions share the preflight without each repeating + the try / except dance. + """ + try: + entity_uuid = UUID(uuid_str) + except ValueError: + return api.response_400(message="Invalid UUID") + + entity = VersionDAO.find_active_by_uuid(model_cls, entity_uuid) + if entity is None: + return api.response_404() + + try: + security_manager.raise_for_access(**{access_kwarg: entity}) + except SupersetSecurityException: + return api.response_403() + + return entity, entity_uuid + + +def list_versions_endpoint( + api: Any, + model_cls: type[Model], + uuid_str: str, + access_kwarg: str, +) -> Response: + """Body of ``GET /api/v1/{resource}//versions/``.""" + resolved = _resolve_entity(api, model_cls, uuid_str, access_kwarg) + if isinstance(resolved, Response): + return resolved + entity, entity_uuid = resolved + + versions = VersionDAO.list_versions(model_cls, entity_uuid, entity=entity) + if versions is None: + return api.response_404() + result = _version_item_schema.dump(versions, many=True) + return set_version_etag_by_uuid( + api.response(200, result=result, count=len(result)), + model_cls, + entity_uuid, + entity_id=entity.id, + ) + + +def get_version_endpoint( + api: Any, + model_cls: type[Model], + uuid_str: str, + version_uuid_str: str, + access_kwarg: str, +) -> Response: + """Body of ``GET /api/v1/{resource}//versions//``.""" + resolved = _resolve_entity(api, model_cls, uuid_str, access_kwarg) + if isinstance(resolved, Response): + return resolved + entity, entity_uuid = resolved + + try: + version_uuid = UUID(version_uuid_str) + except ValueError: + return api.response_400(message="Invalid version UUID") + + snapshot = VersionDAO.get_version( + model_cls, entity_uuid, version_uuid, entity=entity + ) + if snapshot is None: + return api.response_404() + # Normalize the version-level block through the schema; the entity + # scalar fields stay as the DAO shaped them (their keys are + # entity-specific by design). + if "_version" in snapshot: + snapshot["_version"] = _version_item_schema.dump(snapshot["_version"]) + return set_version_etag_by_uuid( + api.response(200, result=snapshot), + model_cls, + entity_uuid, + entity_id=entity.id, + ) diff --git a/superset/versioning/baseline/__init__.py b/superset/versioning/baseline/__init__.py new file mode 100644 index 000000000000..664af88c4cb3 --- /dev/null +++ b/superset/versioning/baseline/__init__.py @@ -0,0 +1,69 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""``before_flush`` listener that captures a baseline version (version 0) +for entities being updated for the first time after the versioning +migration. + +Package layout (descends from public entry point to leaf builders): + +* :mod:`.listener` — public :func:`register_baseline_listener` that + wires the before-flush event on ``db.session``. +* :mod:`.dirty` — :func:`force_parent_dirty_on_child_change` and + :func:`pin_audit_columns`: promote a parent into ``session.dirty`` + when only its versioned children changed, and pin its audit columns + so the synthetic flush doesn't bump them. +* :mod:`.collection` — discovery: which parents need a baseline row? + Holds ``VERSIONED_MODELS`` (populated at app start), + :func:`collect_parents_to_baseline`, the + :func:`child_to_parent_registry` mapping, and the per-parent + Continuum-shadow-table lookups. +* :mod:`.insertion` — parent baseline insertion + child-handler + dispatch. +* :mod:`.children` — per-entity child baseline handlers + (``_baseline_dataset_children`` / ``_baseline_dashboard_children``) + plus the leaf helpers that synthesize child / slice shadow rows. +* :mod:`.shadow` — low-level :func:`insert_baseline_shadow_row` + helper used by every module that writes a shadow row, and the + :data:`CONTINUUM_BOOKKEEPING_COLUMNS` constant re-used outside this + package (the change-record listener and ``queries.py`` filter on it). + +The re-exports below preserve the prior ``from +superset.versioning.baseline import …`` call shape; no caller outside +this package needs to change. +""" + +from __future__ import annotations + +from superset.versioning.baseline.collection import ( + child_to_parent_registry, + VERSIONED_MODELS, +) +from superset.versioning.baseline.dirty import pin_audit_columns +from superset.versioning.baseline.listener import register_baseline_listener +from superset.versioning.baseline.shadow import ( + CONTINUUM_BOOKKEEPING_COLUMNS, + insert_baseline_shadow_row, +) + +__all__ = [ + "CONTINUUM_BOOKKEEPING_COLUMNS", + "VERSIONED_MODELS", + "child_to_parent_registry", + "insert_baseline_shadow_row", + "pin_audit_columns", + "register_baseline_listener", +] diff --git a/superset/versioning/baseline/children.py b/superset/versioning/baseline/children.py new file mode 100644 index 000000000000..e55a5cf4fb57 --- /dev/null +++ b/superset/versioning/baseline/children.py @@ -0,0 +1,212 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Per-entity child-baseline handlers. + +After a parent baseline row lands in :mod:`.insertion`, this module's +handlers write the parent's child baselines under the same transaction +id. The dispatch table :data:`CHILD_BASELINE_HANDLERS` is keyed on +the parent class name (avoids an import-cycle with the entity modules, +which can't be loaded at app-init time). + +The dataset handler baselines :class:`TableColumn` and +:class:`SqlMetric` children. The dashboard handler baselines the +``dashboard_slices`` M2M membership *and* synthesizes +``operation_type=0`` rows in ``slices_version`` for attached slices +that have no prior shadow — without those slice-side baselines, +Continuum's M2M revert query returns empty. + +Leaf-level helpers (:func:`_insert_child_baseline_rows`, +:func:`_baseline_attached_slices`, +:func:`_insert_synthetic_slice_baseline`) live here too — they're +shared between the two parent-specific handlers. +""" + +from __future__ import annotations + +from collections.abc import Callable +from typing import Any + +import sqlalchemy as sa +from sqlalchemy.orm import Session + +from superset.versioning.baseline.shadow import insert_baseline_shadow_row + + +def _baseline_dataset_children(session: Session, dataset: Any, tx_id: int) -> None: + """Baseline a dataset's ``TableColumn`` and ``SqlMetric`` children + under the dataset's baseline tx. + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.connectors.sqla.models import SqlMetric, TableColumn + + for child_cls in (TableColumn, SqlMetric): + _insert_child_baseline_rows( + session, + dataset, + child_cls.__table__, + version_class(child_cls).__table__, + "table_id", + tx_id, + ) + + +def _baseline_dashboard_children(session: Session, dashboard: Any, tx_id: int) -> None: + """Baseline a dashboard's ``dashboard_slices`` M2M plus synthesize + ``operation_type=0`` rows in ``slices_version`` for attached slices + with no prior shadow. + + Continuum's M2M version-side relationship for ``Dashboard.slices`` + joins through both ``dashboard_slices_version`` AND + ``slices_version``: the second exists clause filters slices by + "latest slices_version row with tx <= dashboard.tx". If a slice + has no slices_version rows at all, that join produces no match + and ``version_obj.slices`` returns empty — leaving the dashboard + restore with no slices to append. The synthetic slice baseline at + this dashboard's tx gives the M2M query a slice version it can match. + + Doesn't try to be clever about slices shared across dashboards: a + slice is baselined at this dashboard's tx_id only when it has no + shadow rows at all. If a later dashboard baseline references the + same slice, this baseline (now at lower tx) is still found by + that dashboard's restore. The reverse — a dashboard baselined + AFTER the slice was first baselined under another dashboard at + a higher tx — is a residual gap deferred to a future fix. + """ + metadata = type(dashboard).__table__.metadata + live_tbl = metadata.tables.get("dashboard_slices") + shadow_tbl = metadata.tables.get("dashboard_slices_version") + if live_tbl is None or shadow_tbl is None: + return + + _insert_child_baseline_rows( + session, dashboard, live_tbl, shadow_tbl, "dashboard_id", tx_id + ) + _baseline_attached_slices(session, dashboard, live_tbl, tx_id) + + +# Dispatch table keyed by parent CLASS NAME rather than class, to avoid +# the import-cycle between baseline.py (loaded at app init) and the +# entity modules. The class-name string is set once at app start by +# the model definitions — typo-prone if extended. Declared after the +# handlers it references because module-level dict literals evaluate +# at import time and need the names already bound. +_ChildBaselineHandler = Callable[[Session, Any, int], None] +CHILD_BASELINE_HANDLERS: dict[str, _ChildBaselineHandler] = { + "SqlaTable": _baseline_dataset_children, + "Dashboard": _baseline_dashboard_children, +} + + +def _insert_child_baseline_rows( + session: Session, + parent_obj: Any, + child_table: sa.Table, + child_version_table: sa.Table, + fk_column_name: str, + tx_id: int, +) -> None: + """Synthesize ``operation_type=0`` shadow rows for every live child of + *parent_obj* under transaction id *tx_id*. + + Parallels :func:`~superset.versioning.baseline.insertion._insert_baseline_row` + but iterates over child rows. Used to give Continuum's ``Reverter`` + baseline data for children of pre-existing parents (children that + predate this commit have no shadow rows otherwise, so Reverter + would treat them as "deleted at the target tx" and try to remove + them on revert — the ADR-004 Failure 1 reproduction scenario). + + :param child_table: the live child SQLAlchemy ``Table`` (e.g. + ``TableColumn.__table__`` or the bare ``dashboard_slices`` association) + :param child_version_table: the corresponding Continuum shadow ``Table`` + :param fk_column_name: column on *child_table* that points to the parent + (e.g. ``"table_id"`` for ``TableColumn``, ``"dashboard_id"`` for + ``dashboard_slices``) + """ + conn = session.connection() + fk_col = getattr(child_table.c, fk_column_name) + + rows = ( + conn.execute(sa.select(child_table).where(fk_col == parent_obj.id)) + .mappings() + .all() + ) + if not rows: + return + + for row in rows: + insert_baseline_shadow_row(conn, child_version_table, row, tx_id) + + +def _baseline_attached_slices( + session: Session, dashboard: Any, live_tbl: sa.Table, tx_id: int +) -> None: + """Insert ``operation_type=0`` rows in ``slices_version`` for each + slice attached to *dashboard* that has no shadow row yet. + + Batched: one membership SELECT, one existing-shadow SELECT, one live + SELECT for the missing slices. Per-slice work happens only on + ``_insert_synthetic_slice_baseline``. The previous per-slice + ``COUNT(*)`` + ``SELECT`` pattern was O(N) round-trips and surfaced + as a measurable first-save hotspot on dashboards with many charts. + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.models.slice import Slice + + slice_ver_table = version_class(Slice).__table__ + slice_table = Slice.__table__ + conn = session.connection() + + attached_slice_ids = [ + r.slice_id + for r in conn.execute( + sa.select(live_tbl.c.slice_id).where( + live_tbl.c.dashboard_id == dashboard.id + ) + ).all() + ] + if not attached_slice_ids: + return + + existing_shadow_ids = { + row[0] + for row in conn.execute( + sa.select(slice_ver_table.c.id.distinct()).where( + slice_ver_table.c.id.in_(attached_slice_ids) + ) + ).all() + } + missing_ids = [sid for sid in attached_slice_ids if sid not in existing_shadow_ids] + if not missing_ids: + return + + slice_rows = ( + conn.execute(sa.select(slice_table).where(slice_table.c.id.in_(missing_ids))) + .mappings() + .all() + ) + for slice_row in slice_rows: + _insert_synthetic_slice_baseline(conn, slice_ver_table, slice_row, tx_id) + + +def _insert_synthetic_slice_baseline( + conn: Any, slice_ver_table: sa.Table, slice_row: Any, tx_id: int +) -> None: + insert_baseline_shadow_row(conn, slice_ver_table, slice_row, tx_id) diff --git a/superset/versioning/baseline/collection.py b/superset/versioning/baseline/collection.py new file mode 100644 index 000000000000..7ab78a9d6087 --- /dev/null +++ b/superset/versioning/baseline/collection.py @@ -0,0 +1,153 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Discovery: figure out which parents need a baseline row. + +Three helpers cooperate on the listener's "should I baseline" decision: + +* :func:`collect_parents_to_baseline` — walks ``session.dirty`` / + ``new`` / ``deleted`` and returns the unique parent entities to + consider (directly-dirty versioned parents + parents reachable from + dirty children via :func:`child_to_parent_registry`). +* :func:`version_table_for` — resolves a Continuum shadow Table for + one parent object. +* :func:`shadow_row_count` — counts existing shadow rows for the + parent's id; ``0`` is the signal to insert a baseline. + +:func:`child_to_parent_registry` is also exposed because +:mod:`superset.versioning.factory` consumes it via inline import. + +**Inline imports.** ``versioning.baseline`` is imported during +``init_versioning()`` before all SQLAlchemy mappers are configured; +the lazy imports defer Continuum + model resolution until call time. +""" + +from __future__ import annotations + +import functools +import logging +from typing import Any + +import sqlalchemy as sa +from sqlalchemy.exc import OperationalError, ProgrammingError +from sqlalchemy.orm import Session + +# Populated at app startup (superset/initialization/__init__.py) before +# register_baseline_listener() is called. +VERSIONED_MODELS: list[type] = [] + +logger = logging.getLogger(__name__) + + +def collect_parents_to_baseline(session: Session) -> dict[int, Any]: + """Return parents-to-baseline as ``{id(obj): obj}`` keyed by Python + object identity to dedupe across ``session.dirty + new + deleted``. + + Includes both directly-dirty versioned parents and parents reachable + from dirty/new/deleted children via the child→parent registry. + """ + parents: dict[int, Any] = {} + child_map = child_to_parent_registry() + for obj in list(session.dirty) + list(session.new) + list(session.deleted): + if type(obj) in VERSIONED_MODELS: + parents[id(obj)] = obj + continue + entry = child_map.get(type(obj)) + if entry is None: + continue + parent_attr, parent_cls = entry + parent = getattr(obj, parent_attr, None) + if parent is not None and type(parent) is parent_cls: # noqa: E721 + parents[id(parent)] = parent + return parents + + +@functools.cache +def child_to_parent_registry() -> dict[type, tuple[str, type]]: + """Map child entity class → (parent-relationship-attr, parent class). + + When a dirty child of a known type appears in session.dirty/new/deleted, + we walk to its parent and baseline the parent (+ siblings) under the + SAME flush so pre-edit child values land in the baseline shadow rows. + Without this, edits that only touch child rows produce a "silent" flush + A (just ``TableColumn``) followed by flush B (``SqlaTable.changed_on``); + flush B reads children from DB AFTER flush A already pushed UPDATEs, + capturing post-edit state. + + Cached because this is called from ``force_parent_dirty_on_child_change`` + and ``collect_parents_to_baseline`` on every save flush. The returned + mapping depends only on the (fixed at import time) child model classes, + so an unbounded ``functools.cache`` is the right shape — no invalidation + needed. + """ + # Lazy import: ``baseline`` is imported during ``init_versioning``, which + # runs before all model mappers are configured. Importing model classes + # at module load would either cycle or hit unresolved mappers. + # pylint: disable=import-outside-toplevel + from superset.connectors.sqla.models import SqlaTable, SqlMetric, TableColumn + + return { + TableColumn: ("table", SqlaTable), + SqlMetric: ("table", SqlaTable), + } + + +def version_table_for(obj: Any) -> Any: + """Return Continuum's shadow ``Table`` for *obj*'s class, or ``None`` + when the class isn't registered (forks / plugins that subclass without + ``__versioned__``). + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + from sqlalchemy_continuum.exc import ClassNotVersioned + + try: + return version_class(type(obj)).__table__ + except ClassNotVersioned: + return None + + +def shadow_row_count(session: Session, obj: Any, version_table: Any) -> int | None: + """Return number of shadow rows for *obj.id* in *version_table*, or + ``None`` when the version table is missing (migration not yet applied) + or the count query raised unexpectedly. + """ + try: + # SAVEPOINT so a missing-table probe can't poison the enclosing + # transaction on PostgreSQL (a failed statement aborts the tx + # there; subsequent statements would raise InFailedSqlTransaction + # and fail the user's save despite the except below). + with session.no_autoflush, session.connection().begin_nested(): + return ( + session.connection() + .execute( + sa.select(sa.func.count()) + .select_from(version_table) + .where(version_table.c.id == obj.id) + ) + .scalar() + ) + except (OperationalError, ProgrammingError): + # Missing table: OperationalError on SQLite/MySQL, + # ProgrammingError (UndefinedTable) on PostgreSQL. + return None + except Exception: # pylint: disable=broad-except + logger.exception( + "baseline_listener: count query failed for %s id=%s", + type(obj).__name__, + getattr(obj, "id", None), + ) + return None diff --git a/superset/versioning/baseline/dirty.py b/superset/versioning/baseline/dirty.py new file mode 100644 index 000000000000..f351cbb38bbe --- /dev/null +++ b/superset/versioning/baseline/dirty.py @@ -0,0 +1,243 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Parent-dirty force machinery for child-only saves. + +When a versioned child (``TableColumn`` / ``SqlMetric``) is in +``session.dirty`` / ``new`` / ``deleted`` but its parent's scalar +columns haven't been touched, the parent is *missing* from the dirty +set — so Continuum's UnitOfWork never creates a parent UPDATE +operation, no parent shadow row is written, and the version-history +dropdown comes back empty for column/metric-only saves. + +:func:`force_parent_dirty_on_child_change` walks dirty/new/deleted +children, looks them up in the child→parent registry (in +:mod:`.collection`), and ``attributes.flag_modified``s a deterministic +non-excluded column on the parent. SQLAlchemy adds the parent to +``session.dirty``; Continuum then writes a parent shadow row whose +scalars mirror the previous version (only the children actually +changed). + +:func:`pin_audit_columns` is a companion: when the parent is force- +flagged, we pin ``changed_by_fk`` / ``changed_on`` to their current +in-memory values so the parent UPDATE doesn't invoke the audit +columns' ``onupdate=get_user_id`` / ``onupdate=datetime.now`` hooks +(which would attribute the synthetic flush to whoever ``g.user`` is +at the time, possibly a deleted test user under autoflush teardown). + +**Inline imports.** Same init-order rationale as +:mod:`superset.versioning.baseline.collection`. +""" + +from __future__ import annotations + +import logging +from collections.abc import Iterator +from typing import Any + +from sqlalchemy.exc import InvalidRequestError +from sqlalchemy.orm import attributes, Session + +from superset.versioning.baseline.collection import child_to_parent_registry + +logger = logging.getLogger(__name__) + + +def force_parent_dirty_on_child_change(session: Session) -> None: + """Mark a versioned parent as dirty whenever one of its versioned + children appears in ``session.dirty``/``new``/``deleted`` but the + parent's own scalars haven't been edited. + + Without this hook, edits that only touch ``TableColumn`` or + ``SqlMetric`` rows leave the parent ``SqlaTable`` out of + ``session.dirty`` — so Continuum's UnitOfWork never creates a + parent UPDATE operation and ``list_versions`` (which queries the + parent shadow ``tables_version``) returns just the baseline. The + user-visible symptom is "I edited a column description but the + dataset's version history dropdown is empty". + + For each child that represents a real edit, we resolve its parent + and ``attributes.flag_modified`` a deterministic non-excluded + column so SQLAlchemy adds the parent to ``session.dirty`` without + altering any column values. Continuum then writes a parent shadow + row at this transaction; its scalar columns mirror the previous + version (only the children changed). + ``SkipUnmodifiedPlugin._is_no_op_update`` is taught to recognize + the "scalars match but children dirty" case and keep the row. + """ + child_map = child_to_parent_registry() + new_set = session.new + for child in _real_dirty_versioned_children(session, child_map): + parent = _resolve_parent(child, child_map) + if parent is None: + continue + if parent in new_set: + # Already-new short-circuit. If the parent itself is in + # ``session.new`` (typical during an import that adds a + # ``SqlaTable`` plus 50 fresh ``TableColumn`` children), it + # will INSERT in this flush regardless — the + # ``flag_modified`` call is redundant (and the attribute- + # default-not-yet-fired case in ``_flag_parent`` would just + # swallow an ``InvalidRequestError``). Skip the work. + continue + if _flag_parent(parent): + pin_audit_columns(parent) + + +def _real_dirty_versioned_children( + session: Session, child_map: dict[type, Any] +) -> Iterator[Any]: + """Yield child instances that are (a) of a versioned-child class + registered in *child_map*, and (b) represent a real content edit — + not a phantom-dirty entry from lazy-load side effects or audit- + column auto-bumps. + + Phantom-dirty filter rationale: a child can appear in + ``session.dirty`` for reasons that don't represent real content + edits — lazy-load side effects, ``AuditMixin`` auto-bumps from + prior code paths, M2M relationship-cascade artifacts (e.g., + ``rls_entry.tables.extend([dataset])`` in setUp), Reverter side + passes. Force-touching the parent in those cases produces an + incidental ``UPDATE tables SET description=…, changed_on=…, + changed_by_fk=…`` that can violate FK integrity on some dialects + (observed in ``test_rls_filter_alters_no_role_user_birth_names_query``). + + The filter applies ONLY to persistent rows in ``session.dirty``: + ``session.new`` (creation) and ``session.deleted`` (removal) are + always real content changes — deletion in particular is a state + transition with no attribute history, so ``is_modified`` returns + False there even when the change is real (column-removed records + must still emit). + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import is_modified + + # ``session.dirty`` / ``session.new`` are IdentitySets — ``__contains__`` + # uses identity comparison, which is what we need for the phantom- + # dirty filter below. + dirty_set = session.dirty + for obj in list(session.dirty) + list(session.new) + list(session.deleted): + if type(obj) not in child_map: + continue + if obj in dirty_set and not is_modified(obj): + continue + yield obj + + +def _resolve_parent(child: Any, child_map: dict[type, Any]) -> Any | None: + """Resolve the versioned parent for *child* via the child→parent + registry; return ``None`` when the registered parent attribute + isn't loaded or has been swapped for an unexpected type.""" + parent_attr, parent_cls = child_map[type(child)] + parent = getattr(child, parent_attr, None) + if parent is None or type(parent) is not parent_cls: # noqa: E721 + return None + return parent + + +def _flag_parent(parent: Any) -> bool: + """``flag_modified`` a stable non-excluded column on *parent* so + SQLAlchemy adds it to ``session.dirty`` without altering values. + Returns ``True`` on success. + + Column choice: ``description`` is a plain ``Text`` column on all + three versioned parent classes (Dashboard, Slice, SqlaTable) and is + in none of their ``__versioned__`` excludes — pick it + deterministically so the flagged attribute is stable across + SQLAlchemy versions / mapper-configuration orders. We deliberately + avoid ``uuid``: when a versioned-parent UPDATE goes through with + ``uuid`` flagged, the column's ``UUIDType``/BLOB round-trip + produces a memoryview that fails an FK integrity check on some + dialects (observed in + ``test_rls_filter_alters_no_role_user_birth_names_query`` and + ``test_restore_applies_scalar_field``). ``description`` is a plain + text column with no marshaling layer, so flagging it safely + round-trips its current value. Falls back to ``uuid`` then + ``col_keys[0]`` for forks that excluded ``description``. + + Returns ``False`` for the freshly-constructed ``session.new`` + instance whose attribute defaults haven't fired yet — the + attribute is unloaded in instance state, so ``flag_modified`` + rejects it with ``InvalidRequestError``. The parent will INSERT in + this flush regardless, so the flag was redundant; safely skip. + Hit by ``test_create_dataset_item`` (POST /api/v1/dataset/). + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum.utils import versioned_column_properties + + col_keys = [prop.key for prop in versioned_column_properties(parent)] + if not col_keys: + return False + if "description" in col_keys: + flag_col = "description" + elif "uuid" in col_keys: + flag_col = "uuid" + else: + flag_col = col_keys[0] + try: + attributes.flag_modified(parent, flag_col) + except InvalidRequestError: + return False + return True + + +def pin_audit_columns(parent: Any) -> None: + """Pin ``changed_by_fk`` and ``changed_on`` to their current in-memory + values on a flag-flushed parent. + + ``changed_by_fk`` carries ``onupdate=get_user_id`` from ``AuditMixin``: + any UPDATE statement that doesn't explicitly set this column lets + SQLAlchemy invoke ``get_user_id()`` and write whoever ``g.user`` is + at flush time. When the flush is autoflush-triggered during an + earlier test's teardown (after the test user has been deleted from + ``ab_user``), the bumped value points at a non-existent row and the + parent UPDATE fails the FK to ``ab_user``. The same applies to + ``changed_on``'s ``onupdate=datetime.now`` (cosmetic only, but it's + cheap to pin together). + + ``flag_modified`` on both columns marks them as having dirty + attribute history, which tells SQLAlchemy to use the in-memory + (previously-committed) values instead of invoking ``onupdate`` — + the parent UPDATE then carries the existing audit values rather + than whatever ``g.user`` resolves to during the synthetic flag + flush. Hits ``test_rls_filter_alters_no_role_user_birth_names_query`` + and ``TestDatasetRestoreApi::test_restore_applies_scalar_field`` + in CI's full-suite ordering (autoflush during teardown). + """ + pinned_any = False + for audit_col in ("changed_by_fk", "changed_on"): + if hasattr(parent, audit_col): + try: + attributes.flag_modified(parent, audit_col) + pinned_any = True + except InvalidRequestError: + continue + if not pinned_any and hasattr(parent, "changed_by_fk"): + # Both audit columns are present on the parent but neither + # ``flag_modified`` succeeded — typically because the parent is + # a freshly-constructed ``session.new`` instance whose attribute + # defaults haven't fired yet. Without the pin, the synthetic + # parent UPDATE in this flush invokes ``onupdate=get_user_id`` + # and writes whoever ``g.user`` is at flush time, which under + # autoflush-during-teardown can point at a deleted test user + # and fail the FK to ``ab_user``. Surface this so the failure + # mode is debuggable from the log without inspection. + logger.info( + "baseline: skipped audit-column pin on %s id=%s " + "(attribute defaults not loaded)", + type(parent).__name__, + getattr(parent, "id", None), + ) diff --git a/superset/versioning/baseline/insertion.py b/superset/versioning/baseline/insertion.py new file mode 100644 index 000000000000..303a1e4f202c --- /dev/null +++ b/superset/versioning/baseline/insertion.py @@ -0,0 +1,149 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Parent baseline insertion + child-handler dispatch. + +Two complementary helpers: + +* :func:`insert_baseline_and_children` — top-level glue called by + the listener. Wraps the work in ``session.no_autoflush`` (so + ``session.connection()`` doesn't trigger a flush of Continuum's + pending Transaction object before our direct-SQL insert claims its + tx_id) and logs any failures as listener-boundary errors. +* :func:`_insert_baseline_row` — actually writes the + ``version_transaction`` row and the parent shadow row. Returns the + allocated ``transaction_id``. +* :func:`_baseline_children_for_parent` — dispatches to the per- + entity handler in :mod:`.children` under the same tx_id. +""" + +from __future__ import annotations + +import logging +from typing import Any + +import sqlalchemy as sa +from sqlalchemy.orm import Session + +from superset.versioning.baseline.children import CHILD_BASELINE_HANDLERS +from superset.versioning.baseline.shadow import insert_baseline_shadow_row +from superset.versioning.utils import read_row_outside_flush + +logger = logging.getLogger(__name__) + + +def insert_baseline_and_children( + session: Session, obj: Any, version_table: Any +) -> None: + """Insert the parent baseline row, then baseline the parent's child + collections under the same transaction id. + + Wrapped in ``no_autoflush`` so ``session.connection()`` inside + ``_insert_baseline_row`` does not trigger a flush of Continuum's + pending Transaction object before our direct-SQL insert claims its + tx_id. + """ + try: + with session.no_autoflush: + tx_id = _insert_baseline_row(session, obj, version_table) + if tx_id is None: + return + _baseline_children_for_parent(session, obj, tx_id) + logger.debug( + "baseline_listener: inserted baseline tx_id=%s for %s id=%s", + tx_id, + type(obj).__name__, + getattr(obj, "id", None), + ) + except Exception: # pylint: disable=broad-except + logger.exception( + "baseline_listener: failed to insert baseline for %s id=%s", + type(obj).__name__, + getattr(obj, "id", None), + ) + + +def _insert_baseline_row( + session: Session, obj: Any, version_table: sa.Table +) -> int | None: + """Insert a synthetic baseline row capturing the pre-edit DB state of *obj*. + + Creates a version_transaction entry and an operation_type=0 version row. + All writes use the session's existing connection so they share the same + database transaction as the triggering flush. + + Returns the allocated ``transaction_id`` so the caller can baseline child + collections under the same tx (see + :func:`~superset.versioning.baseline.children._insert_child_baseline_rows`), + or ``None`` when the entity has no live row. + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import versioning_manager + + main_table = type(obj).__table__ + row = read_row_outside_flush(session, main_table, obj.id) + if row is None: + return None + + conn = session.connection() + + # Insert a version_transaction row for the baseline. + # + # ``issued_at`` and ``user_id`` are sourced from the entity's audit fields + # (``changed_on`` / ``changed_by_fk``, falling back to ``created_on`` / + # ``created_by_fk`` if the row was never edited), so the baseline reads + # in the version-history UI as "this is the state at the time of the + # last pre-versioning edit, by that user." Using ``now()`` and the + # current user would have made the baseline look chronologically newer + # than subsequent edits and attributed historical content to the user + # who happened to trigger the first save under versioning. + baseline_issued_at = row.get("changed_on") or row.get("created_on") or sa.func.now() + baseline_user_id = row.get("changed_by_fk") or row.get("created_by_fk") + tx_table = versioning_manager.transaction_cls.__table__ + result = conn.execute( + tx_table.insert().values( + issued_at=baseline_issued_at, + user_id=baseline_user_id, + remote_addr=None, + ) + ) + tx_id = result.inserted_primary_key[0] + insert_baseline_shadow_row(conn, version_table, row, tx_id) + return tx_id + + +def _baseline_children_for_parent( + session: Session, parent_obj: Any, tx_id: int +) -> None: + """Baseline a parent's child collections under the parent's baseline tx. + + Dispatches via the + :data:`~superset.versioning.baseline.children.CHILD_BASELINE_HANDLERS` + table to per-entity handlers. A handler failure is logged but does + not block the parent baseline. + """ + parent_name = type(parent_obj).__name__ + handler = CHILD_BASELINE_HANDLERS.get(parent_name) + if handler is None: + return + try: + handler(session, parent_obj, tx_id) + except Exception: # pylint: disable=broad-except + logger.exception( + "baseline_listener: failed to baseline children of %s id=%s", + parent_name, + getattr(parent_obj, "id", None), + ) diff --git a/superset/versioning/baseline/listener.py b/superset/versioning/baseline/listener.py new file mode 100644 index 000000000000..0b16f7365c6a --- /dev/null +++ b/superset/versioning/baseline/listener.py @@ -0,0 +1,95 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Public entry point: attach the ``before_flush`` baseline listener. + +:func:`register_baseline_listener` is called from +:class:`superset.initialization.SupersetAppInitializer.init_versioning` +after ``make_versioned()`` has run and all versioned model classes +have been imported. It registers one ``before_flush`` listener on +``db.session`` that: + +1. force-dirties versioned parents whose only changes are + child-collection edits (:mod:`.dirty`); +2. collects the parents that need a baseline row + (:mod:`.collection`); +3. for each parent with no prior shadow row, inserts the synthetic + baseline row + its child baseline rows (:mod:`.insertion` + + :mod:`.children`). +""" + +from __future__ import annotations + +from typing import Any + +from sqlalchemy import event +from sqlalchemy.orm import Session + +from superset.versioning.baseline.collection import ( + collect_parents_to_baseline, + shadow_row_count, + version_table_for, + VERSIONED_MODELS, +) +from superset.versioning.baseline.dirty import force_parent_dirty_on_child_change +from superset.versioning.baseline.insertion import insert_baseline_and_children + +# Sentinel attribute set on the session target after first successful +# registration — same pattern as +# :mod:`superset.versioning.changes.listener`. Subsequent calls become +# no-ops so test fixtures that instantiate multiple Superset apps per +# process don't attach a second copy of the listener to the shared +# ``db.session`` (every flush would otherwise run the baseline pass +# twice). +_REGISTERED_SENTINEL = "_versioning_baseline_listener_registered" + + +def register_baseline_listener() -> None: + """Attach the before_flush listener that captures baseline versions. + + Call this after ``VERSIONED_MODELS`` has been populated and + ``make_versioned()`` has run. Idempotent — repeat calls are no-ops. + """ + # pylint: disable=import-outside-toplevel + from superset.extensions import db + + if getattr(db.session, _REGISTERED_SENTINEL, False): + return + + # insert=True prepends us in the listener chain so we run BEFORE + # Continuum's before_flush. Continuum's pending Transaction object + # (added in its own before_flush) would otherwise get a lower + # auto-increment tx_id than our direct-SQL baseline insert, placing the + # baseline row after the update in version_number order. Prepending + # ensures our baseline's tx_id comes first. + @event.listens_for(db.session, "before_flush", insert=True) + def capture_baseline(session: Session, flush_context: Any, instances: Any) -> None: + if not VERSIONED_MODELS: + return + # Make sure a child-only edit promotes the parent to ``session.dirty`` + # before Continuum's before_flush reads the dirty set. + force_parent_dirty_on_child_change(session) + for obj in collect_parents_to_baseline(session).values(): + if type(obj) not in VERSIONED_MODELS: + continue + version_table = version_table_for(obj) + if version_table is None: + continue + count = shadow_row_count(session, obj, version_table) + if count == 0: + insert_baseline_and_children(session, obj, version_table) + + setattr(db.session, _REGISTERED_SENTINEL, True) diff --git a/superset/versioning/baseline/shadow.py b/superset/versioning/baseline/shadow.py new file mode 100644 index 000000000000..c31a5835bcf7 --- /dev/null +++ b/superset/versioning/baseline/shadow.py @@ -0,0 +1,72 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Continuum-shaped shadow-row writer. + +Two pieces: + +* :data:`CONTINUUM_BOOKKEEPING_COLUMNS` — the set of column names + Continuum uses for per-row bookkeeping (``transaction_id`` / + ``end_transaction_id`` / ``operation_type``). Re-used outside this + package as a filter (the change-record listener strips these from + JSON record values). +* :func:`insert_baseline_shadow_row` — copies a live row into a + shadow ``Table`` as a synthetic ``operation_type=0`` baseline at + the given transaction id. The other modules in this package use it + for every parent and child baseline insert. +""" + +from __future__ import annotations + +from typing import Any + +import sqlalchemy as sa + +# Continuum's per-shadow-row bookkeeping columns. Skipped when copying +# content from a live row into a synthetic baseline shadow row; set +# explicitly by the baseline writer so the row reads as a freshly-created +# live row at the baseline transaction. +CONTINUUM_BOOKKEEPING_COLUMNS: frozenset[str] = frozenset( + {"transaction_id", "end_transaction_id", "operation_type"} +) + + +def insert_baseline_shadow_row( + conn: Any, + version_table: sa.Table, + source_row: Any, + tx_id: int, +) -> None: + """Copy *source_row* into *version_table* as a synthetic baseline + (``operation_type=0``) shadow row at *tx_id*. + + Content columns are copied through; the three Continuum bookkeeping + columns are set explicitly so the row reads as a freshly-created + live row at *tx_id*. Column objects (not names) are used as + ``values()`` keys to avoid the "Unconsumed column names" error that + a name-based dict hits when a Column's ``.key`` differs from its + ``.name`` — a thing Continuum-generated tables occasionally produce. + """ + col_values: dict[Any, Any] = {} + for col in version_table.columns: + if col.name in CONTINUUM_BOOKKEEPING_COLUMNS: + continue + if col.name in source_row: + col_values[col] = source_row[col.name] + col_values[version_table.c.transaction_id] = tx_id + col_values[version_table.c.end_transaction_id] = None + col_values[version_table.c.operation_type] = 0 + conn.execute(version_table.insert().values(col_values)) diff --git a/superset/versioning/changes/__init__.py b/superset/versioning/changes/__init__.py new file mode 100644 index 000000000000..29c1592b3e84 --- /dev/null +++ b/superset/versioning/changes/__init__.py @@ -0,0 +1,76 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Write-side change-record capture for ``version_changes``. + +The package is split into four submodules that descend from public +entry point to leaf helpers: + +* :mod:`.listener` — public ``register_change_record_listener`` plus + the session-event machinery (``before_flush`` / ``after_flush`` / + ``after_commit`` / ``after_rollback``) that drives the capture. + Holds ``ACTION_KIND_KEY``, the buffer-key constants, and the per-tx + ``action_kind`` stamper. +* :mod:`.state` — per-entity diff dispatch: pre-state read, + post-state serialisation, JSON-safety coercion (``jsonable``), + cached scalar-field discovery, and bulk-insert into the + ``version_changes`` table. +* :mod:`.shadow_queries` — shadow-table reads that drive child- + collection diffs (dataset columns/metrics, dashboard slice + membership). Includes the validity-strategy ``shadow_rows_valid_at`` + helper consumed externally by :mod:`superset.versioning.queries`. +* :mod:`.table` — the SQLAlchemy ``Table`` definition for + ``version_changes`` plus the ``ENTITY_KIND_BY_CLASS_NAME`` mapping + consumed by the API + activity-view modules. + +The re-exports below preserve the prior ``from +superset.versioning.changes import …`` call shape; no caller outside +this package needs to change. +""" + +from __future__ import annotations + +from superset.versioning.changes.listener import ( + ACTION_KIND_CLONE, + ACTION_KIND_IMPORT, + ACTION_KIND_KEY, + ACTION_KIND_RESTORE, + ACTION_KINDS, + ACTION_META_KEY, + build_action_headline, + OPERATION_ANNOUNCE, + register_change_record_listener, +) +from superset.versioning.changes.shadow_queries import shadow_rows_valid_at +from superset.versioning.changes.table import ( + ENTITY_KIND_BY_CLASS_NAME, + version_changes_table, +) + +__all__ = [ + "ACTION_KIND_CLONE", + "ACTION_KIND_IMPORT", + "ACTION_KIND_KEY", + "ACTION_KIND_RESTORE", + "ACTION_KINDS", + "ACTION_META_KEY", + "build_action_headline", + "OPERATION_ANNOUNCE", + "ENTITY_KIND_BY_CLASS_NAME", + "register_change_record_listener", + "shadow_rows_valid_at", + "version_changes_table", +] diff --git a/superset/versioning/changes/listener.py b/superset/versioning/changes/listener.py new file mode 100644 index 000000000000..fbd944eaaf70 --- /dev/null +++ b/superset/versioning/changes/listener.py @@ -0,0 +1,507 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Session-level listeners that drive ``version_changes`` writes. + +Two flush events cooperate, plus two post-commit / post-rollback +cleanups: + +- ``before_flush``: for each versioned entity in ``session.dirty``, + reads the pre-save scalar state from the DB via raw SQL inside + ``session.no_autoflush`` (same idiom as the baseline listener, not + Continuum's internal ``units_of_work`` which is a private API), reads + the post-save state from the in-memory ORM object, calls the diff + engine, and buffers the resulting :class:`ChangeRecord` list on + ``session.info``. This must run before the flush because after the + flush the DB already reflects the post-state; we can't recover the + pre-state from it. + +- ``after_flush``: drains the buffer, resolves the current Continuum + transaction id via ``versioning_manager.units_of_work``, and bulk- + inserts one ``version_changes`` row per record with a monotonic + ``sequence`` number. Records accumulated across multiple before_flush + calls within one transaction share the same ``transaction_id`` and + contiguous sequence numbers. + +- ``after_commit`` / ``after_rollback``: clean up session-scoped + state (processed-tx set, ``action_kind`` / ``action_meta`` keys, and + the pending-records buffer) so a long-lived session doesn't carry any + of it into the next transaction. + +Scope: + - Slice, Dashboard, SqlaTable **scalar fields** (via the cached + field set from :mod:`superset.versioning.changes.state` — new + columns are picked up automatically). + - ``Slice.params`` kind-classification (filter / metric / time_range + / color_palette / dimension, plus generic ``field`` fallback). + +Child-collection diffs (dataset ``TableColumn`` / ``SqlMetric``, +dashboard ``dashboard_slices``) read the pre- and post-state from +Continuum shadow tables via the helpers in +:mod:`superset.versioning.changes.shadow_queries`, executed in +``after_flush`` once Continuum has written its tx-N rows. + +``session.new`` entities are not processed in this listener: +operation_type=0 transactions (baseline capture and first-save INSERTs) +produce zero change records per spec §Clarifications 2026-04-24. +""" + +from __future__ import annotations + +import logging +from typing import Any + +import sqlalchemy as sa +from sqlalchemy import event +from sqlalchemy.exc import OperationalError, ProgrammingError +from sqlalchemy.orm import Session + +from superset.versioning.changes.shadow_queries import ( + _dashboard_child_records_for_tx_from_shadows, + _dataset_child_records_for_tx_from_shadows, +) +from superset.versioning.changes.state import ( + bulk_insert_records, + compute_records_for_entity, +) +from superset.versioning.changes.table import ENTITY_KIND_BY_CLASS_NAME +from superset.versioning.diff import ( + ChangeRecord, + fold_dashboard_layout_with_chart_changes, +) + +logger = logging.getLogger(__name__) + + +# Key under which the pending-records buffer is stored on ``session.info``. +# Using ``session.info`` (SQLAlchemy's user-data dict) avoids the need +# for a module-level WeakKeyDictionary and keeps buffers naturally scoped +# to the session's lifetime. +_BUFFER_KEY = "_version_changes_pending" + +# Key for the set of Continuum transaction ids whose change records +# have already been written in this session. ``after_flush`` can fire +# more than once for a single transaction (e.g. autoflush triggered by +# a mid-commit query), and our child-diff path reads snapshot tables +# that don't care about the buffer state — without this marker we'd +# re-insert the same child records on the second flush and hit the +# UNIQUE(transaction_id, entity_kind, entity_id, sequence) constraint. +_PROCESSED_TXS_KEY = "_version_changes_processed_txs" + +# Key on ``session.info`` that commands set to declare the high-level +# action that produced the current transaction. Read once per flush by +# the change-record listener and stamped onto the +# ``version_transaction.action_kind`` column via ``sa.update()``. +# ``None`` (the default) means "ordinary save". +# +# Commands set this immediately before ``db.session.commit()``: +# +# db.session.info[ACTION_KIND_KEY] = ACTION_KIND_RESTORE +# db.session.commit() +# +# The listener pops the key after stamping, and ``after_commit`` / +# ``after_rollback`` cleanup pop it again as a safety net, so a +# long-lived session can't accidentally carry the value into the next +# transaction. +ACTION_KIND_KEY = "_versioning_action_kind" + +# Recognised ``action_kind`` values — the single source of truth shared +# by the four command-side stampers (restore / import / clone) and the +# listener that writes them to ``version_transaction.action_kind``. +# Schemas / response decorators that need an allowlist read from +# ``ACTION_KINDS`` so a future addition (e.g. ``"thumbnail_warm"``) only +# has to update this one constant. ``None`` is *not* a member — it +# represents the default "ordinary save" path that never sets the key. +ACTION_KIND_RESTORE = "restore" +ACTION_KIND_IMPORT = "import" +ACTION_KIND_CLONE = "clone" +ACTION_KINDS: frozenset[str] = frozenset( + {ACTION_KIND_RESTORE, ACTION_KIND_IMPORT, ACTION_KIND_CLONE} +) + +# Key on ``session.info`` carrying a synthetic "headline" change record +# for the current transaction — the ``__meta__`` record convention. Set +# by commands alongside ``ACTION_KIND_KEY`` when the avenue has a payload +# the field-level diff can't express; the canonical case is restore, +# whose transaction otherwise carries no pointer to WHICH version was +# restored (surfaced by the version-history UI, PR #40988: "Restored to +# X from [date]" can't be rendered from API data alone). +# +# Build the value with :func:`build_action_headline` — the single owner +# of the record shape — rather than hand-rolling the dict; renderers +# dispatch on ``kind == "__meta__"`` plus the transaction's +# ``action_kind`` (the verb deliberately does NOT ride in ``path``, +# which stays pure navigation per the ChangeRecord contract). +# +# The listener pops the key on the first record-bearing firing for the +# transaction and PREPENDS the record to the entity's buffer (sequence 0 +# — headline first). Same lifecycle as ``ACTION_KIND_KEY``: popped on +# use, and the ``after_commit`` / ``after_rollback`` cleanups pop it as +# a safety net. +ACTION_META_KEY = "_versioning_action_meta" + +# ``operation`` value for synthetic headline records: a headline +# announces an action, it does not mutate a field, so the field-verb +# vocabulary (add / remove / move / edit) would be dishonest here. +OPERATION_ANNOUNCE = "announce" + + +def build_action_headline( + entity_kind: str, + entity_id: int, + to_value: dict[str, Any], +) -> dict[str, Any]: + """Build the ``ACTION_META_KEY`` payload — the single owner of the + ``__meta__`` headline record's shape. + + *entity_kind* is the table-kind (see ``ENTITY_KIND_BY_CLASS_NAME``); + *to_value* carries the action's payload (for restore: + ``{"version_uuid": ..., "version_number": ...}``). The action itself + is identified by the transaction's ``action_kind`` column, which the + same command stamps via ``ACTION_KIND_KEY`` — renderers join the + two rather than parsing the verb out of the record. + """ + return { + "entity_kind": entity_kind, + "entity_id": entity_id, + "record": ChangeRecord( + kind="__meta__", + operation=OPERATION_ANNOUNCE, + path=["__meta__"], + from_value=None, + to_value=to_value, + ), + } + + +# Sentinel attribute set on the session target after first successful +# registration. Subsequent calls become no-ops. Storing the flag on the +# target itself (rather than module-level state) keeps the guard +# naturally scoped — a fresh session proxy gets a fresh registration — +# and avoids the TOCTOU race between ``event.contains`` and +# ``event.listen`` that a module-level ref would have under concurrent +# init. In test fixtures that instantiate multiple Superset apps per +# process, the shared ``db.session`` carries the sentinel and re-entry +# is correctly deduped. +_REGISTERED_SENTINEL = "_versioning_change_listener_registered" + + +def _process_dirty_entity_into_buffer( + session: Session, + obj: Any, + buffer: dict[tuple[str, int], list[ChangeRecord]], +) -> None: + """Compute scalar change records for one dirty entity + append to buffer.""" + entity_kind = ENTITY_KIND_BY_CLASS_NAME.get(type(obj).__name__) + if entity_kind is None: + return + entity_id = getattr(obj, "id", None) + if entity_id is None: + return + try: + records = compute_records_for_entity(session, obj) + except Exception: # pylint: disable=broad-except + logger.exception( + "version_changes: diff failed for %s id=%s", + type(obj).__name__, + entity_id, + ) + return + if records: + buffer.setdefault((entity_kind, entity_id), []).extend(records) + + +def _append_child_records_to_buffer( + session: Session, + tx_id: int, + buffer: dict[tuple[str, int], list[ChangeRecord]], +) -> None: + """Compute dataset + dashboard child-collection records + append to buffer. + + Runs in ``after_flush`` so the shadow tables already have the + current-tx rows. Reads from Continuum shadow tables + (``table_columns_version`` / ``sql_metrics_version`` / + ``dashboard_slices_version`` / ``slices_version``). + """ + try: + for dataset_id, records in _dataset_child_records_for_tx_from_shadows( + session, tx_id + ).items(): + buffer.setdefault(("dataset", dataset_id), []).extend(records) + for dashboard_id, records in ( + _dashboard_child_records_for_tx_from_shadows(session, tx_id) + ).items(): + buffer.setdefault(("dashboard", dashboard_id), []).extend(records) + + # Post-merge fold: when a dashboard save adds/removes charts, + # drop the redundant ``position_json.*`` records that mirror + # the membership change. See + # ``diff.fold_dashboard_layout_with_chart_changes``. + for key in list(buffer.keys()): + if key[0] == "dashboard": + buffer[key] = fold_dashboard_layout_with_chart_changes(buffer[key]) + if not buffer[key]: + del buffer[key] + except Exception: # pylint: disable=broad-except + logger.exception("version_changes: child-diff failed for tx %s", tx_id) + + +def _current_transaction_id(session: Session) -> int | None: + """Return the Continuum transaction id for *session*'s current unit of + work, or ``None`` when Continuum has no active transaction (e.g. raw + SQL execution outside the ORM's flush flow). + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import versioning_manager + + uow = versioning_manager.units_of_work.get(session.connection()) + if uow is None or uow.current_transaction is None: + return None + return uow.current_transaction.id + + +def _inject_action_meta_record( + session: Session, + buffer: dict[tuple[str, int], list[ChangeRecord]], +) -> None: + """Pop ``ACTION_META_KEY`` and prepend its synthetic headline record + to the owning entity's buffer (the ``__meta__`` record convention). + + No-op when no command set the key — and, critically, no-op WITHOUT + popping when the buffer is empty: the buffer-empty short-circuit in + ``flush_change_records`` exists so a multi-flush transaction can + deliver its records on a later firing, and a headline-only buffer + would defeat it (the first firing would persist just the headline, + mark the tx processed, and the later flush's real records would be + silently dropped). Leaving the key in place parks the headline until + the record-bearing firing. Prepended (not appended) so the headline + gets ``sequence`` 0 and renders first. Malformed payloads are logged + and dropped — a headline is descriptive enrichment, never worth + failing the user's save over. + """ + if not buffer: + return + meta = session.info.pop(ACTION_META_KEY, None) + if meta is None: + return + try: + key = (meta["entity_kind"], meta["entity_id"]) + record = meta["record"] + buffer.setdefault(key, []).insert(0, record) + except (KeyError, TypeError): # pragma: no cover - defensive + logger.exception("version_changes: malformed ACTION_META_KEY payload") + + +def _stamp_action_kind_on_transaction(session: Session, tx_id: int) -> None: + """Pop the per-tx action_kind from ``session.info`` and stamp it + onto the ``version_transaction`` row identified by *tx_id*. + + No-op when no command set the action_kind (the default for + ordinary saves). Emits via ``sa.update()`` against Continuum's + transaction Table so the identifier is auto-quoted per dialect + (MySQL would otherwise reject the unquoted column name if it ever + collided with a reserved word) and the dialect-portable column + binding is reused instead of hand-written SQL. + + The action_kind is popped (not just read) so a long-lived session + can't accidentally carry the value into the next transaction. A + failed stamp is logged and swallowed — action_kind is a + descriptive enrichment, not a correctness invariant; refusing to + write change records because an UPDATE on a single column failed + would punish the user save for an audit-log nicety. + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import versioning_manager + + action_kind = session.info.pop(ACTION_KIND_KEY, None) + if action_kind is None: + return + tx_tbl = versioning_manager.transaction_cls.__table__ + try: + session.connection().execute( + sa.update(tx_tbl) + .where(tx_tbl.c.id == tx_id) + .values(action_kind=action_kind) + ) + except Exception: # pylint: disable=broad-except + logger.exception( + "version_changes: failed to stamp action_kind=%s on tx %s", + action_kind, + tx_id, + ) + + +def _persist_buffered_records( + session: Session, + tx_id: int, + buffer: dict[tuple[str, int], list[ChangeRecord]], +) -> None: + """Bulk-insert *buffer*'s records under *tx_id* and reset the buffer. + + Catches ``OperationalError`` / ``ProgrammingError`` to handle the + pre-migration startup race (version_changes table missing — the + former on SQLite/MySQL, the latter on PostgreSQL), and ``Exception`` + as the listener-boundary safety net so a malformed record can't + crash the user's save. + + The insert runs under a SAVEPOINT (``begin_nested`` on the + connection): on PostgreSQL a failed statement aborts the enclosing + transaction, so without it the swallowed exception would still + poison the user's save — the COMMIT that follows this listener + would raise ``InFailedSqlTransaction``, defeating the fail-open + guarantee exactly where it matters. + """ + try: + with session.connection().begin_nested(): + bulk_insert_records(session, tx_id, buffer) + except (OperationalError, ProgrammingError): + # version_changes table missing (migration not yet applied). + pass + except Exception: # pylint: disable=broad-except + logger.exception( + "version_changes: bulk insert failed for tx %s (%d entities)", + tx_id, + len(buffer), + ) + + +def register_change_record_listener() -> None: # noqa: C901 + """Attach the before_flush + after_flush listeners. + + Registered from :class:`superset.initialization.SupersetAppInitializer` + (``init_versioning``) alongside the baseline, dataset-snapshot, + and dashboard-snapshot listeners. Must run after Continuum's + ``make_versioned()`` so the ``versioning_manager`` is available + and has installed its own before_flush hook. + """ + # pylint: disable=import-outside-toplevel + from superset.connectors.sqla.models import SqlaTable + from superset.extensions import db + from superset.models.dashboard import Dashboard + from superset.models.slice import Slice + + if getattr(db.session, _REGISTERED_SENTINEL, False): + return + + versioned_classes: tuple[type, ...] = (Dashboard, Slice, SqlaTable) + + def compute_change_records( + session: Session, _flush_context: Any, _instances: Any + ) -> None: + # session.info persists across before_flush/after_flush within + # a single transaction. The buffer is keyed on + # ``(entity_kind, entity_id)`` so scalar records captured here + # and child records captured in after_flush merge + # under the same entity without duplication. + buffer: dict[tuple[str, int], list[ChangeRecord]] = session.info.setdefault( + _BUFFER_KEY, {} + ) + for obj in list(session.dirty): + if isinstance(obj, versioned_classes): + _process_dirty_entity_into_buffer(session, obj, buffer) + + def flush_change_records(session: Session, _flush_context: Any) -> None: + buffer: dict[tuple[str, int], list[ChangeRecord]] = session.info.setdefault( + _BUFFER_KEY, {} + ) + + tx_id = _current_transaction_id(session) + if tx_id is None: + session.info[_BUFFER_KEY] = {} + return + + # Skip if we've already written records for this tx (after_flush + # can fire more than once per commit — e.g. autoflush from a + # mid-commit query). Without this guard the child-diff path would + # re-read the same shadow rows and re-emit the same records, + # tripping the UNIQUE(transaction_id, entity_kind, entity_id, + # sequence) constraint on insert. + processed: set[int] = session.info.setdefault(_PROCESSED_TXS_KEY, set()) + if tx_id in processed: + # Drop anything buffered after the tx was persisted: records + # left here would otherwise survive on the long-lived scoped + # session and be inserted under the NEXT transaction's id. + session.info[_BUFFER_KEY] = {} + return + + # Stamp action_kind eagerly, before the buffer-empty short- + # circuit. Restores / imports / clones may flush across multiple + # cycles; the FIRST firing for this tx is the one with the + # value still on ``session.info``. The helper pops on success + # so subsequent firings see ``None`` and short-circuit cleanly. + _stamp_action_kind_on_transaction(session, tx_id) + + _append_child_records_to_buffer(session, tx_id, buffer) + + # After the child append and before the emptiness check: the + # headline joins whichever firing carries the transaction's real + # records (scalar or child), and its peek-don't-pop guard parks + # it across record-less firings instead of defeating the + # multi-flush short-circuit below. + _inject_action_meta_record(session, buffer) + + if not buffer: + # Don't mark tx as processed when nothing was inserted. A + # later after_flush firing for the same tx may carry the + # records — e.g. when an entity's edit lands across two + # flushes (a child-only flush followed by a parent-dirty + # flush): the parent shadow only lands in the parent-dirty + # flush, so the child-diff path can't find a prior tx to + # compare against until then. + session.info[_BUFFER_KEY] = {} + return + + try: + _persist_buffered_records(session, tx_id, buffer) + finally: + session.info[_BUFFER_KEY] = {} + processed.add(tx_id) + + def reset_processed_after_commit(session: Session) -> None: + # ``_PROCESSED_TXS_KEY`` accumulates Continuum tx ids whose change + # records have already been written, to dedup against multiple + # ``after_flush`` firings within one transaction. After commit + # the tx is closed and its id will never recur on this session + # — drop the set so a long-lived session (Celery worker, CLI) + # doesn't grow it without bound. + session.info.pop(_PROCESSED_TXS_KEY, None) + # If a command set the action_kind but no flush fired (e.g. a + # save that touched nothing versioned), the value would + # otherwise leak into the next transaction. Drop it here as a + # belt-and-suspenders cleanup; the + # ``_stamp_action_kind_on_transaction`` helper already pops on + # the normal path. + session.info.pop(ACTION_KIND_KEY, None) + session.info.pop(ACTION_META_KEY, None) + session.info.pop(_BUFFER_KEY, None) + + def reset_action_kind_after_rollback(session: Session) -> None: + # When a command sets ``ACTION_KIND_KEY`` and then an exception + # fires before flush (e.g. validation error after the key is + # set), the transaction rolls back without the listener ever + # popping the key. The next save on the same session would + # then inherit the stale value and label an unrelated commit + # as "restore" / "import" / "clone". Pop here so a rolled-back + # action's intent doesn't leak forward. + session.info.pop(ACTION_KIND_KEY, None) + session.info.pop(ACTION_META_KEY, None) + session.info.pop(_BUFFER_KEY, None) + + event.listen(db.session, "before_flush", compute_change_records) + event.listen(db.session, "after_flush", flush_change_records) + event.listen(db.session, "after_commit", reset_processed_after_commit) + event.listen(db.session, "after_rollback", reset_action_kind_after_rollback) + setattr(db.session, _REGISTERED_SENTINEL, True) diff --git a/superset/versioning/changes/shadow_queries.py b/superset/versioning/changes/shadow_queries.py new file mode 100644 index 000000000000..1f6a290007f1 --- /dev/null +++ b/superset/versioning/changes/shadow_queries.py @@ -0,0 +1,323 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Shadow-table queries that drive child-collection diffs. + +Reads Continuum shadow tables (``table_columns_version`` / +``sql_metrics_version`` / ``dashboard_slices_version`` / +``slices_version``) under the validity-strategy semantics to compute +the pre/post state of child collections at a given transaction. Used +by the change-record listener's ``after_flush`` path once Continuum +has written the current transaction's shadow rows. + +**Inline imports.** Continuum's ``version_class`` and the Superset +model classes are imported inside each helper because this package is +loaded from ``init_versioning()`` before all SQLAlchemy mappers are +configured. The deferred imports keep the module-load graph free of +mapper-resolution side effects. +""" + +from __future__ import annotations + +from typing import Any + +import sqlalchemy as sa +from sqlalchemy.orm import Session + +from superset.versioning.baseline import CONTINUUM_BOOKKEEPING_COLUMNS +from superset.versioning.changes.state import jsonable +from superset.versioning.diff import ( + ChangeRecord, + diff_dashboard_slices, + diff_dataset_columns, + diff_dataset_metrics, +) + + +def shadow_rows_valid_at( + session: Session, + shadow_table: sa.Table, + fk_col_name: str, + fk_value: int, + tx: int, +) -> list[dict[str, Any]]: + """Return the live state of *shadow_table* rows whose FK column + (``fk_col_name``) equals *fk_value*, as of transaction *tx*. + + Uses Continuum's validity-strategy semantics: a row is "valid at tx" + when ``transaction_id <= tx`` AND (``end_transaction_id`` IS NULL OR + ``end_transaction_id`` > tx) AND it isn't a DELETE shadow. + + The returned dicts mirror the live row's column set (no Continuum + bookkeeping columns), so they can be passed straight to the + natural-key diff helpers (``diff_dataset_columns`` etc.). + """ + fk_col = getattr(shadow_table.c, fk_col_name) + rows = ( + session.connection() + .execute( + sa.select(shadow_table).where( + fk_col == fk_value, + shadow_table.c.transaction_id <= tx, + sa.or_( + shadow_table.c.end_transaction_id.is_(None), + shadow_table.c.end_transaction_id > tx, + ), + shadow_table.c.operation_type != 2, + ) + ) + .mappings() + .all() + ) + # Coerce values to JSON-safe forms — raw shadow rows can carry + # ``UUID``, ``datetime``, ``bytes`` etc. that don't survive the + # ``version_changes.from_value/to_value`` JSON column write. + return [ + { + k: jsonable(v) + for k, v in dict(row).items() + if k not in CONTINUUM_BOOKKEEPING_COLUMNS + } + for row in rows + ] + + +def _affected_dataset_ids_at_tx(session: Session, tx: int) -> set[int]: + """Datasets touched at *tx* — directly (parent shadow at tx) or + indirectly (column / metric shadow at tx).""" + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.connectors.sqla.models import SqlaTable, SqlMetric, TableColumn + + dataset_ids: set[int] = set() + parent_tbl = version_class(SqlaTable).__table__ + for row in session.connection().execute( + sa.select(parent_tbl.c.id).where(parent_tbl.c.transaction_id == tx) + ): + dataset_ids.add(row[0]) + for child_cls in (TableColumn, SqlMetric): + child_tbl = version_class(child_cls).__table__ + for row in session.connection().execute( + sa.select(child_tbl.c.table_id).where(child_tbl.c.transaction_id == tx) + ): + if row[0] is not None: + dataset_ids.add(row[0]) + return dataset_ids + + +def _dataset_child_records_for_tx_from_shadows( + session: Session, transaction_id: int +) -> dict[int, list[ChangeRecord]]: + """Compute column + metric diff records for each dataset touched at + *transaction_id*, reading from Continuum shadow tables. + + For each dataset: + * Post-state = rows valid at ``transaction_id`` in + ``table_columns_version`` / ``sql_metrics_version``. + * Pre-state = rows valid at ``transaction_id - 1`` in the same + shadow tables. + + With Continuum's validity-strategy semantics, "valid at tx N - 1" + is the state immediately before this transaction's effects (the + row that gets superseded at tx=N has ``end_transaction_id=N``, so + it satisfies ``end > N - 1``). Unrelated transactions between this + dataset's edits are transparent — they don't change validity for + this dataset's children. + + First-edit case: when there is no prior tx (the dataset's earliest + shadow IS at *transaction_id*), pre-state is empty. We skip rather + than emit "Added X" for every column — same "baseline = zero + records" semantics as the snapshot path. + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.connectors.sqla.models import SqlMetric, TableColumn + + cols_tbl = version_class(TableColumn).__table__ + metrics_tbl = version_class(SqlMetric).__table__ + + result: dict[int, list[ChangeRecord]] = {} + for dataset_id in _affected_dataset_ids_at_tx(session, transaction_id): + # Skip the very first transaction for this dataset (no pre-state). + prior_tx = ( + session.connection() + .execute( + sa.select(sa.func.max(cols_tbl.c.transaction_id)).where( + cols_tbl.c.table_id == dataset_id, + cols_tbl.c.transaction_id < transaction_id, + ) + ) + .scalar() + ) + if prior_tx is None: + # No prior column shadow — could still be a metric-only edit; + # check metrics shadow too. + prior_tx = ( + session.connection() + .execute( + sa.select(sa.func.max(metrics_tbl.c.transaction_id)).where( + metrics_tbl.c.table_id == dataset_id, + metrics_tbl.c.transaction_id < transaction_id, + ) + ) + .scalar() + ) + if prior_tx is None: + continue + + post_cols = shadow_rows_valid_at( + session, cols_tbl, "table_id", dataset_id, transaction_id + ) + pre_cols = shadow_rows_valid_at( + session, cols_tbl, "table_id", dataset_id, prior_tx + ) + post_metrics = shadow_rows_valid_at( + session, metrics_tbl, "table_id", dataset_id, transaction_id + ) + pre_metrics = shadow_rows_valid_at( + session, metrics_tbl, "table_id", dataset_id, prior_tx + ) + + records: list[ChangeRecord] = [] + records.extend(diff_dataset_columns(pre_cols, post_cols)) + records.extend(diff_dataset_metrics(pre_metrics, post_metrics)) + if records: + result[dataset_id] = records + return result + + +def _affected_dashboard_ids_at_tx(session: Session, tx: int) -> set[int]: + """Dashboards touched at *tx* — directly (parent shadow at tx) or + indirectly (slice-membership shadow at tx).""" + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.models.dashboard import Dashboard + + dashboard_ids: set[int] = set() + parent_tbl = version_class(Dashboard).__table__ + for row in session.connection().execute( + sa.select(parent_tbl.c.id).where(parent_tbl.c.transaction_id == tx) + ): + dashboard_ids.add(row[0]) + + # M2M shadow: ``dashboard_slices_version`` is auto-generated by + # Continuum and lives in metadata — not a model class. Look it up + # from the metadata bag rather than via ``version_class``. + metadata = parent_tbl.metadata + if (m2m_tbl := metadata.tables.get("dashboard_slices_version")) is not None: + for row in session.connection().execute( + sa.select(m2m_tbl.c.dashboard_id).where(m2m_tbl.c.transaction_id == tx) + ): + if row[0] is not None: + dashboard_ids.add(row[0]) + return dashboard_ids + + +def _dashboard_slice_uuids_at_tx( + session: Session, dashboard_id: int, tx: int +) -> list[str]: + """Slice UUIDs attached to *dashboard_id* as of *tx*, read by joining + ``dashboard_slices_version`` (M2M membership) against + ``slices_version`` (slice content). + + Joining through both is necessary — and matches the same query + Continuum's M2M ``Reverter`` uses — because a slice that's + referenced by the M2M but has no slice-version row at this tx is + treated as "not yet versioned" and excluded. + + Returns UUIDs (strings) so the result can be diffed by the existing + :func:`diff_dashboard_slices` helper, which keys on uuid. + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.models.slice import Slice + + metadata = version_class(Slice).__table__.metadata + m2m_tbl = metadata.tables.get("dashboard_slices_version") + slices_tbl = version_class(Slice).__table__ + if m2m_tbl is None: + return [] + + rows = ( + session.connection() + .execute( + sa.select(slices_tbl.c.uuid).where( + slices_tbl.c.id == m2m_tbl.c.slice_id, + m2m_tbl.c.dashboard_id == dashboard_id, + m2m_tbl.c.transaction_id <= tx, + sa.or_( + m2m_tbl.c.end_transaction_id.is_(None), + m2m_tbl.c.end_transaction_id > tx, + ), + m2m_tbl.c.operation_type != 2, + slices_tbl.c.transaction_id <= tx, + sa.or_( + slices_tbl.c.end_transaction_id.is_(None), + slices_tbl.c.end_transaction_id > tx, + ), + slices_tbl.c.operation_type != 2, + ) + ) + .all() + ) + return [str(r[0]) for r in rows if r[0] is not None] + + +def _dashboard_child_records_for_tx_from_shadows( + session: Session, transaction_id: int +) -> dict[int, list[ChangeRecord]]: + """Compute slice-membership diff records for each dashboard touched + at *transaction_id*, reading from Continuum shadow tables. + + Same pre/post logic as + :func:`_dataset_child_records_for_tx_from_shadows`. + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.models.dashboard import Dashboard + + metadata = version_class(Dashboard).__table__.metadata + m2m_tbl = metadata.tables.get("dashboard_slices_version") + + result: dict[int, list[ChangeRecord]] = {} + for dashboard_id in _affected_dashboard_ids_at_tx(session, transaction_id): + prior_tx = None + if m2m_tbl is not None: + prior_tx = ( + session.connection() + .execute( + sa.select(sa.func.max(m2m_tbl.c.transaction_id)).where( + m2m_tbl.c.dashboard_id == dashboard_id, + m2m_tbl.c.transaction_id < transaction_id, + ) + ) + .scalar() + ) + if prior_tx is None: + continue + + post_uuids = _dashboard_slice_uuids_at_tx(session, dashboard_id, transaction_id) + pre_uuids = _dashboard_slice_uuids_at_tx(session, dashboard_id, prior_tx) + + records = diff_dashboard_slices(pre_uuids, post_uuids) + if records: + result[dashboard_id] = records + return result diff --git a/superset/versioning/changes/state.py b/superset/versioning/changes/state.py new file mode 100644 index 000000000000..5d47871e1eed --- /dev/null +++ b/superset/versioning/changes/state.py @@ -0,0 +1,237 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Pre/post-state reading and the per-entity diff dispatch. + +Three concerns live here: + +1. **JSON-safety coercion** — raw column values (``datetime``, ``UUID``, + ``bytes``, ``Decimal``) get converted to strings before they land in + the ``version_changes.from_value`` / ``to_value`` JSON columns. +2. **State capture** — :func:`_orm_to_post_state` serialises the + in-memory ORM object; :func:`_read_pre_state` reads the corresponding + pre-flush row directly from the DB inside ``session.no_autoflush``. +3. **Diff dispatch** — :func:`compute_records_for_entity` routes to the + right :mod:`superset.versioning.diff` helper based on the model + class name (string dispatch keeps this module free of hard imports + on the three entity classes, which avoids import-order coupling at + app-init time). + +Bulk insert of the computed records into the ``version_changes`` table +lives here too — it's the tail of the per-entity compute pipeline. +""" + +from __future__ import annotations + +import logging +from datetime import date, datetime +from decimal import Decimal +from typing import Any +from uuid import UUID + +import sqlalchemy as sa +from flask_appbuilder import Model +from sqlalchemy.orm import Session + +from superset.versioning.changes.table import version_changes_table +from superset.versioning.diff import ( + ChangeRecord, + diff_dashboard, + diff_dataset, + diff_slice, + scalar_fields_for, +) +from superset.versioning.utils import read_row_outside_flush + +logger = logging.getLogger(__name__) + + +# Per-model-class cache of the scalar-field set. Populated lazily on +# first save of a model. Reading from ``__table__.columns`` is cheap +# but not free; memoising keeps the save-path overhead budget (FR-021) +# from slowly growing with the set of distinct model classes seen. +_SCALAR_FIELDS_CACHE: dict[type, frozenset[str]] = {} + + +def _cached_scalar_fields(model_cls: type[Model]) -> frozenset[str]: + """Cached wrapper around :func:`scalar_fields_for`.""" + if model_cls not in _SCALAR_FIELDS_CACHE: + # ``Slice.params`` is walked by ``diff_slice_params`` for kind + # promotion; emitting it as one opaque ``field`` change would + # defeat that and flood the log with meaningless records. + # ``last_saved_at`` / ``last_saved_by_fk`` are stamped by + # ``UpdateChartCommand`` on every chart save; they're audit + # noise (same shape as ``changed_on`` / ``changed_by_fk``) and + # don't carry user-authored signal. + # ``Dashboard.json_metadata`` and ``position_json`` are JSON + # blobs walked structurally by ``diff_json_field`` (one record + # per changed top-level key); the raw scalar diff would emit + # one giant multi-KB record per save and swamp the response. + special: frozenset[str] = frozenset() + audit: frozenset[str] = frozenset() + if model_cls.__name__ == "Slice": + special = frozenset({"params"}) + audit = frozenset({"last_saved_at", "last_saved_by_fk"}) + elif model_cls.__name__ == "Dashboard": + special = frozenset({"json_metadata", "position_json"}) + _SCALAR_FIELDS_CACHE[model_cls] = scalar_fields_for( + model_cls, special=special, audit=audit + ) + return _SCALAR_FIELDS_CACHE[model_cls] + + +def jsonable(value: Any) -> Any: + """Convert a column value into a JSON-serialisable form. + + Slice has ``last_saved_at`` (datetime), datasets have datetime + columns, and any of these fields can land in ``from_value`` / + ``to_value`` of a ``version_changes`` row, which is a JSON column. + Python's default JSON encoder rejects ``datetime`` / ``UUID`` / + ``bytes`` / ``Decimal``, so the whole bulk insert fails if a single + record carries one. Convert to ISO / hex / str at record-construction + time. + """ + if isinstance(value, (datetime, date)): + return value.isoformat() + if isinstance(value, UUID): + return str(value) + if isinstance(value, bytes): + return value.hex() + if isinstance(value, Decimal): + # Stringify rather than ``float()`` to preserve precision; the + # diff engine compares string equality on ``from_value`` / + # ``to_value``, so coercing both sides to the same form is what + # matters. + return str(value) + return value + + +def _orm_to_post_state(obj: Any) -> dict[str, Any]: + """Serialise an ORM object's column attributes to a plain dict. + + We only read declared column attributes — not relationships or + hybrid properties — because the diff engine operates on scalar + values per its documented API. Values are passed through + :func:`jsonable` so the dict is JSON-safe end-to-end. + """ + state = sa.inspect(obj) + return { + col.key: jsonable(getattr(obj, col.key)) for col in state.mapper.column_attrs + } + + +def _read_pre_state( + session: Session, model_cls: type[Model], entity_id: int +) -> dict[str, Any] | None: + """Read the entity's pre-flush row directly from the DB and convert + non-JSON-safe types to strings so both sides of the diff compare on + the same form. Delegates the autoflush-suppressed read itself to + :func:`superset.versioning.utils.read_row_outside_flush`. + + Returns ``None`` if the row is missing (shouldn't happen for a dirty + existing object, but defensive against race conditions). + """ + table = model_cls.__table__ + result = read_row_outside_flush(session, table, entity_id) + if result is None: + return None + # Convert non-JSON-safe types (datetime, UUID, bytes, Decimal) to + # strings so both sides of the diff compare on the same form and + # any value that ends up in ``from_value`` / ``to_value`` is + # acceptable to the JSON column on insert. + return {key: jsonable(value) for key, value in result.items()} + + +def compute_records_for_entity(session: Session, obj: Any) -> list[ChangeRecord]: + """Diff the pre-state (from DB) against the post-state (in memory). + + Dispatches to :func:`diff_slice` / :func:`diff_dashboard` / + :func:`diff_dataset` based on the model class name — string-based + dispatch is used to keep this module free of hard imports on the + three entity classes, which in turn avoids import-order coupling + at app-init time. + """ + model_cls = type(obj) + entity_id = getattr(obj, "id", None) + if entity_id is None: + return [] + + try: + pre_state = _read_pre_state(session, model_cls, entity_id) + except Exception: # pylint: disable=broad-except + logger.exception( + "version_changes: pre-state read failed for %s id=%s", + model_cls.__name__, + entity_id, + ) + return [] + + if pre_state is None: + return [] + + post_state = _orm_to_post_state(obj) + fields = _cached_scalar_fields(model_cls) + + name = model_cls.__name__ + if name == "Slice": + return diff_slice(pre_state, post_state, fields=fields) + if name == "Dashboard": + return diff_dashboard(pre_state, post_state, fields=fields) + if name == "SqlaTable": + return diff_dataset(pre_state, post_state, fields=fields) + return [] + + +def bulk_insert_records( + session: Session, + transaction_id: int, + buffered: dict[tuple[str, int], list[ChangeRecord]], +) -> None: + """Insert ``version_changes`` rows for one transaction via raw SQL. + + Uses the module-level :data:`version_changes_table` Table object + (which carries JSON column types, unlike ``sa.table(...)``) so the + connection marshals ``path`` / ``from_value`` / ``to_value`` Python + structures into JSON on insert. Skips the ORM flush round that + ``session.bulk_insert_mappings`` would cost inside an already- + active flush. + + ``buffered`` is a dict keyed on ``(entity_kind, entity_id)`` so + records for one entity — scalars from ``before_flush`` plus + children collected in ``after_flush`` — merge naturally under the + same key. ``sequence`` resets per entity so each entity's records + form a self-contained replay sequence. + """ + if not buffered: + return + rows = [] + for (entity_kind, entity_id), records in buffered.items(): + for seq, r in enumerate(records): + rows.append( + { + "transaction_id": transaction_id, + "entity_kind": entity_kind, + "entity_id": entity_id, + "sequence": seq, + "kind": r.kind, + "operation": r.operation, + "path": r.path, + "from_value": r.from_value, + "to_value": r.to_value, + } + ) + if rows: + session.connection().execute(version_changes_table.insert(), rows) diff --git a/superset/versioning/changes/table.py b/superset/versioning/changes/table.py new file mode 100644 index 000000000000..174d37269e38 --- /dev/null +++ b/superset/versioning/changes/table.py @@ -0,0 +1,87 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Schema definition for ``version_changes``. + +Declared against the shared ``Model.metadata`` so integration tests +that build schema via ``metadata.create_all()`` pick it up without the +Alembic migration running. Mirrors the shape of the +``56cd24c07170_add_versioning_tables`` migration byte-for-byte. Typed +columns (``sa.JSON`` for path / values) are required so the +connection's bulk-insert path marshals Python lists/dicts into JSON — +a lightweight ``sa.table(...)`` would not carry the type info and +SQLite's driver would reject the ``list`` as an unsupported bind. + +The schema lives in its own module to keep the listener and the +shadow-table-query helpers free of schema-construction boilerplate at +import time. +""" + +from __future__ import annotations + +import sqlalchemy as sa +from flask_appbuilder import Model + +_metadata = Model.metadata # pylint: disable=no-member + +version_changes_table = sa.Table( + "version_changes", + _metadata, + sa.Column("id", sa.BigInteger, primary_key=True, autoincrement=True), + # ``transaction_id`` references ``version_transaction.id`` at the DB + # level only — the FK + ON DELETE CASCADE live in the Alembic + # migration. Declaring the FK here would fail to resolve at Table + # creation time because ``version_transaction`` is built + # dynamically by SQLAlchemy-Continuum at mapper-configuration time; + # integration tests that materialise schema via ``metadata.create_all`` + # before Continuum runs would hit ``NoReferencedTableError``. Same + # pattern as the other versioning tables. + sa.Column("transaction_id", sa.BigInteger, nullable=False), + sa.Column("entity_kind", sa.String(32), nullable=False), + sa.Column("entity_id", sa.Integer, nullable=False), + # Integer, not SmallInteger: matches the migration — per-entity + # sequence within a transaction is assigned by unbounded enumerate(). + sa.Column("sequence", sa.Integer, nullable=False), + sa.Column("kind", sa.String(32), nullable=False), + sa.Column("operation", sa.String(16), nullable=False), + sa.Column("path", sa.JSON, nullable=False), + sa.Column("from_value", sa.JSON, nullable=True), + sa.Column("to_value", sa.JSON, nullable=True), + sa.UniqueConstraint( + "transaction_id", + "entity_kind", + "entity_id", + "sequence", + name="uq_version_changes_tx_entity_sequence", + ), + sa.Index("ix_version_changes_kind", "kind"), + # No standalone transaction_id index: the UNIQUE constraint above + # leads with transaction_id, so its backing index already serves + # transaction_id-prefix lookups on every dialect. + sa.Index("ix_version_changes_entity", "entity_kind", "entity_id"), + extend_existing=True, +) + +# Mapping from Python class name to the ``entity_kind`` value written +# to ``version_changes.entity_kind``. The API filters change records +# by this value (``WHERE entity_kind = 'chart'`` for the chart history +# endpoint, etc.) — kept short and user-facing-ish so downstream tools +# consuming the raw table read sensibly. +ENTITY_KIND_BY_CLASS_NAME: dict[str, str] = { + "Slice": "chart", + "Dashboard": "dashboard", + "SqlaTable": "dataset", +} diff --git a/superset/versioning/diff.py b/superset/versioning/diff.py new file mode 100644 index 000000000000..3dce7b81e91f --- /dev/null +++ b/superset/versioning/diff.py @@ -0,0 +1,1036 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Diff engine for the ``version_changes`` table (FR-016..FR-019). + +Hand-rolled because: + +- The on-disk ``path`` shape (array of segments) is a direct + representation of our chosen format; external diff libraries + return string paths or JSON-Pointer forms that would need + translation. +- Kind classification (``filter`` vs ``metric`` vs ``field`` etc.) + is co-located with diff walking, avoiding a second classification + pass over the generic diff output. +- Child-collection identity uses natural keys (``column_name``, + ``metric_name``, slice ``uuid``) — the same identity model + ``DatasetDAO.update_columns`` settled on (ADR-004). External + libraries default to list-index matching, which is wrong for our + data. + +See ADR (plan.md §"Key Design Decision: Hand-rolled diff engine") for +the full rationale. + +All functions in this module are pure: they take dicts (or lists of +dicts) and return a list of :class:`ChangeRecord`. The ORM->dict +conversion and Continuum transaction lookup happen in the capture +listener, not here. This keeps the engine unit-testable without +an app context or DB. +""" + +from __future__ import annotations + +import logging +from collections.abc import Callable, Iterable +from dataclasses import dataclass +from typing import Any + +from superset.utils import json as _json + +logger = logging.getLogger(__name__) + +# Per-field recursion depth caps for the leaf-level diff walker. +# A cap is a usefulness bound, not a safety bound: it controls how deep +# into a nested JSON value the engine emits per-leaf records before +# stopping and treating the sub-tree as an opaque value. Values are +# tuned to the field's semantic shape — layout meta is shallow +# (text/sizes/colors), json_metadata and chart params can carry deep +# structures (native filters, adhoc filter sub-queries). +_LAYOUT_META_DIFF_DEPTH = 3 +_JSON_METADATA_DIFF_DEPTH = 6 +_SLICE_PARAMS_DIFF_DEPTH = 6 + +# Columns that are always excluded from change records, regardless of +# what ``__versioned__`` says. ``id`` / ``uuid`` are stable identifiers +# (not edited in normal flows). The four audit fields change on every +# save — emitting records for them would double every history entry +# with meaningless "timestamp changed, user stamped" rows that the UI +# would have to filter out anyway. +_AUDIT_FIELDS: frozenset[str] = frozenset( + { + "id", + "uuid", + "created_on", + "changed_on", + "created_by_fk", + "changed_by_fk", + } +) + +# Fields stripped from child-collection dict items (TableColumn, +# SqlMetric) before comparison and emission. ``changed_on`` / +# ``created_on`` / ``*_by_fk`` are audit fields that update on every +# save of the parent — without this filter, saving a dataset to add +# one column produces a record per existing column too (because their +# ``changed_on`` timestamps all refreshed). ``id`` and ``table_id`` +# are implementation details — ``id`` can change under the +# ``override_columns`` delete-and-reinsert pattern (ADR-004) even +# when the column is semantically unchanged; ``table_id`` is the +# parent FK and never meaningfully differs within one dataset's +# history. ``uuid`` stays stable across normal saves and is kept so +# the renderer can use it for identity if it needs to. +_CHILD_ITEM_OPAQUE_FIELDS: frozenset[str] = frozenset( + { + "id", + "table_id", + "changed_on", + "created_on", + "changed_by_fk", + "created_by_fk", + } +) + + +def _strip_opaque_fields(item: Any) -> Any: + """Return *item* with child-item audit/implementation fields removed. + + Pass-through for non-dict values (scalars, strings) — the strip + only applies where it matters (dataset column / metric dicts). + """ + if not isinstance(item, dict): + return item + return {k: v for k, v in item.items() if k not in _CHILD_ITEM_OPAQUE_FIELDS} + + +# Chart ``params`` sub-keys that are promoted to first-class kinds. +# Every other params sub-key falls through to ``kind="field"``. +_CHART_PARAMS_KIND_BY_KEY: dict[str, str] = { + "adhoc_filters": "filter", + "time_range": "time_range", + "color_scheme": "color_palette", + "metrics": "metric", + "groupby": "dimension", + "columns": "dimension", +} + +# Chart ``params`` sub-keys that are machine-stamped on save and don't +# carry user-authored signal — same category as ``last_saved_at`` on +# the scalar side. ``slice_id`` is a self-reference to the chart's +# own primary id; Superset's save paths add or refresh it on every +# save, producing a spurious "field" record on the first save after +# a chart's params were stored without it. +_CHART_PARAMS_AUDIT_KEYS: frozenset[str] = frozenset({"slice_id"}) + + +def scalar_fields_for( + model_cls: Any, + *, + special: frozenset[str] = frozenset(), + audit: frozenset[str] = frozenset(), +) -> frozenset[str]: + """Scalar columns on ``model_cls`` that should produce change records. + + Derived from the model itself at call time so contributors (and + downstream derivatives) don't have to maintain a parallel whitelist + in this module. Adding a new column to ``Dashboard``, ``Slice``, or + ``SqlaTable`` — whether upstream or in a fork — automatically flows + through to ``version_changes`` on the next save. + + Excludes, in order: + + 1. The model's own ``__versioned__.exclude`` list, so change records + stay consistent with Continuum's shadow tables. If Continuum + isn't tracking a column, the change log shouldn't either. + 2. :data:`_AUDIT_FIELDS` — ``id``, ``uuid``, and the audit + timestamps / user-id columns shared across the three entity types. + 3. The caller's ``audit`` set — model-specific save-side-effect + columns that aren't user-authored content. ``Slice.last_saved_at`` + / ``last_saved_by_fk`` are stamped on every chart save by + ``UpdateChartCommand``, similar to how ``changed_on`` is stamped + by the ORM event listener; emitting "field" records for them + would noise up the change log with one entry per save that + carries no user-meaningful signal. + 4. The caller's ``special`` set — columns handled by a dedicated + differ elsewhere. ``Slice.params``, for example, is walked by + :func:`diff_slice_params` to produce first-class ``filter`` / + ``time_range`` / ``metric`` / ``dimension`` records; emitting + it as a single opaque ``field`` would defeat that. + """ + try: + table = model_cls.__table__ + except AttributeError: + return frozenset() + columns = frozenset(c.name for c in table.columns) + continuum_exclude = frozenset( + getattr(model_cls, "__versioned__", {}).get("exclude", []) or [] + ) + return columns - continuum_exclude - _AUDIT_FIELDS - audit - special + + +@dataclass(frozen=True) +class ChangeRecord: + """One atomic change, as stored in ``version_changes``. + + Fields match the ``version_changes`` columns one-to-one so the + capture listener can serialise a list of these to + ``session.bulk_insert_mappings`` without translation. + + Three orthogonal dimensions: + * ``kind`` — what type of thing changed (``filter`` / ``column`` / + ``header`` / ``field`` / etc.). Content category. + * ``operation`` — what happened to it (``add`` / ``remove`` / + ``move`` / ``edit``). ``move`` only fires for layout records. + * ``path`` — pure navigation address; no verb encoded. + + The transaction-level fourth dimension (``trigger``: ``restore`` / + ``import`` / ``clone``) lives on ``version_transaction``, not here. + """ + + kind: str + operation: str + path: list[Any] + from_value: Any + to_value: Any + + +Key = str | int + + +def _operation_from_values(from_value: Any, to_value: Any) -> str: + """Derive the per-record ``operation`` verb from ``from_value`` / + ``to_value`` nullability. + + * ``add`` — ``from_value`` is ``None`` and ``to_value`` is not. + * ``remove`` — ``to_value`` is ``None`` and ``from_value`` is not. + * ``edit`` — both populated (or both null, which shouldn't reach here). + + Used by every emit site except ``_diff_layout_node``, which emits + ``move`` records (parent reparenting) that cannot be derived from + value nullability alone. + """ + if from_value is None and to_value is not None: + return "add" + if to_value is None and from_value is not None: + return "remove" + return "edit" + + +def _values_equivalent(from_value: Any, to_value: Any) -> bool: + """True if a transition from ``from_value`` to ``to_value`` should + NOT produce a record. + + Beyond plain ``==`` equality, treats ``None`` and ``""`` as equivalent: + Superset's save paths normalize nullable strings to ``""`` on first + write (e.g. ``Dashboard.css``, ``certified_by``, + ``certification_details``), so a first-save transition between + null and empty string carries no user-authored signal. + """ + if from_value == to_value: + return True + if from_value in (None, "") and to_value in (None, ""): + return True + return False + + +def _diff_scalar( + field_name: str, + from_value: Any, + to_value: Any, +) -> ChangeRecord | None: + """Emit a generic ``kind="field"`` record when a scalar differs.""" + if _values_equivalent(from_value, to_value): + return None + return ChangeRecord( + kind="field", + operation=_operation_from_values(from_value, to_value), + path=[field_name], + from_value=from_value, + to_value=to_value, + ) + + +def _recursive_leaf_diff( + kind: str, + path_prefix: list[Any], + pre: Any, + post: Any, + *, + max_depth: int, +) -> list[ChangeRecord]: + """Walk matched dict structures and emit one ``ChangeRecord`` per + changed leaf. + + Recursion rules: + + * Both sides equal (per :func:`_values_equivalent`) → no record. + * Both sides ``dict`` AND recursion depth below ``max_depth`` → + recurse into each key, extending the path by the key. + * All other cases (scalar mismatch, list on either side, mismatched + types, both dicts but depth-capped) → emit one leaf record with + ``from_value`` / ``to_value`` carrying the raw pre/post values. + + Lists are treated as opaque on purpose — positional paths break on + reorder and most lists in Superset's JSON blobs (adhoc filters, + metrics, dataset columns) already have a dedicated natural-key + walker upstream that emits per-element records with the right + identity. + + A depth-cap hit on dict-vs-dict emits a debug log so production + tuning can see when a field's cap is too tight to capture all + meaningful change. + """ + + def _walk(pre: Any, post: Any, path: list[Any], depth: int) -> list[ChangeRecord]: + if _values_equivalent(pre, post): + return [] + if depth < max_depth and isinstance(pre, dict) and isinstance(post, dict): + records: list[ChangeRecord] = [] + for key in sorted(set(pre) | set(post)): + records.extend( + _walk(pre.get(key), post.get(key), [*path, key], depth + 1) + ) + return records + if isinstance(pre, dict) and isinstance(post, dict): + logger.debug( + "version_changes: depth cap %d hit at path=%s — sub-tree " + "emitted as opaque leaf", + max_depth, + path, + ) + return [ + ChangeRecord( + kind=kind, + operation=_operation_from_values(pre, post), + path=list(path), + from_value=pre, + to_value=post, + ) + ] + + return _walk(pre, post, path_prefix, 0) + + +def _diff_list_by_natural_key( + kind: str, + path_prefix: list[Any], + from_list: list[Any] | None, + to_list: list[Any] | None, + key_fn: Callable[[Any], Key | None], +) -> list[ChangeRecord]: + """Diff two lists, matching elements by natural key. + + Emits one record per add / remove / modify. When ``key_fn`` returns + ``None`` for an item (natural key missing or empty), the item falls + back to its position as a synthetic key — so insertions in the + middle of a keyless list still produce sensible records, at the + cost of position-dependent identity. + """ + from_list = from_list or [] + to_list = to_list or [] + + def _effective_key(raw: Key | None, idx: int) -> Key: + if raw is None or raw == "": + return idx + return raw + + from_by_key: dict[Key, Any] = {} + for idx, item in enumerate(from_list): + from_by_key[_effective_key(key_fn(item), idx)] = item + to_by_key: dict[Key, Any] = {} + for idx, item in enumerate(to_list): + to_by_key[_effective_key(key_fn(item), idx)] = item + + records: list[ChangeRecord] = [] + # Preserve `from` order then append `to`-only keys, so sequence is + # deterministic across runs. For dict items (dataset columns / + # metrics) we strip audit/implementation fields before comparing + # AND before emitting — otherwise a save that only adds a new + # column would also emit "changed" records for every existing + # column, because their ``changed_on`` timestamps all refreshed. + # The stripped from/to are what the renderer sees; the per-column + # audit trail is already aggregated at the transaction level in + # ``version_transaction`` (``user_id`` + ``issued_at``). + for k, from_item in from_by_key.items(): + to_item = to_by_key.get(k) + stripped_from = _strip_opaque_fields(from_item) + if to_item is None: + records.append( + ChangeRecord( + kind=kind, + operation="remove", + path=[*path_prefix, k], + from_value=stripped_from, + to_value=None, + ) + ) + continue + stripped_to = _strip_opaque_fields(to_item) + if stripped_from != stripped_to: + records.append( + ChangeRecord( + kind=kind, + operation="edit", + path=[*path_prefix, k], + from_value=stripped_from, + to_value=stripped_to, + ) + ) + for k, to_item in to_by_key.items(): + if k not in from_by_key: + records.append( + ChangeRecord( + kind=kind, + operation="add", + path=[*path_prefix, k], + from_value=None, + to_value=_strip_opaque_fields(to_item), + ) + ) + return records + + +def _filter_key(f: Any) -> Key | None: + """Natural key for an adhoc filter — its subject (column name). + + Users rarely have two filters on the same column; when they do the + secondary dimensions (operator, comparator) appear in the record's + from/to values so the renderer can disambiguate. + """ + return f.get("subject") if isinstance(f, dict) else None + + +def _metric_key(m: Any) -> Key | None: + """Natural key for a metric: prefer ``label``, fall back to column+aggregate.""" + if not isinstance(m, dict): + return None + if label := m.get("label"): + return label + column = m.get("column") + col_name = column.get("column_name") if isinstance(column, dict) else None + agg = m.get("aggregate") + if col_name and agg: + return f"{agg}({col_name})" + return None + + +def _dimension_key(d: Any) -> Key | None: + """Natural key for a groupby/columns element — usually a bare string.""" + if isinstance(d, str): + return d + if isinstance(d, dict): + return d.get("label") or d.get("column_name") + return None + + +def _coerce_params(p: Any) -> dict[str, Any]: + """Decode ``Slice.params`` which is stored as a JSON string.""" + if p is None: + return {} + if isinstance(p, str): + try: + decoded = _json.loads(p) + except _json.JSONDecodeError: + return {} + return decoded if isinstance(decoded, dict) else {} + if isinstance(p, dict): + return p + return {} + + +def diff_slice_params( + from_params: Any, + to_params: Any, +) -> list[ChangeRecord]: + """Diff the ``Slice.params`` JSON blob, promoting known keys to kinds.""" + from_p = _coerce_params(from_params) + to_p = _coerce_params(to_params) + records: list[ChangeRecord] = [] + all_keys = (set(from_p) | set(to_p)) - _CHART_PARAMS_AUDIT_KEYS + for key in sorted(all_keys): + from_v = from_p.get(key) + to_v = to_p.get(key) + if _values_equivalent(from_v, to_v): + continue + kind = _CHART_PARAMS_KIND_BY_KEY.get(key) + if kind == "filter" and isinstance(from_v, list) and isinstance(to_v, list): + records.extend( + _diff_list_by_natural_key( + "filter", + ["params", "adhoc_filters"], + from_v, + to_v, + _filter_key, + ) + ) + elif kind == "metric" and isinstance(from_v, list) and isinstance(to_v, list): + records.extend( + _diff_list_by_natural_key( + "metric", + ["params", "metrics"], + from_v, + to_v, + _metric_key, + ) + ) + elif ( + kind == "dimension" and isinstance(from_v, list) and isinstance(to_v, list) + ): + records.extend( + _diff_list_by_natural_key( + "dimension", + ["params", key], + from_v, + to_v, + _dimension_key, + ) + ) + elif kind: + # scalar first-class kind (time_range, color_palette). + # For genuinely scalar values the recursion emits one leaf + # record exactly as before; for the unusual case of a dict + # value (custom viz params) it recurses to the leaf. + records.extend( + _recursive_leaf_diff( + kind=kind, + path_prefix=["params", key], + pre=from_v, + post=to_v, + max_depth=_SLICE_PARAMS_DIFF_DEPTH, + ) + ) + else: + # unknown params sub-key: generic field change, recursed + # to the leaf so a deep custom-viz option doesn't ship its + # whole sub-tree on both sides. + records.extend( + _recursive_leaf_diff( + kind="field", + path_prefix=["params", key], + pre=from_v, + post=to_v, + max_depth=_SLICE_PARAMS_DIFF_DEPTH, + ) + ) + return records + + +def diff_scalar_fields( + pre: dict[str, Any], + post: dict[str, Any], + *, + fields: Iterable[str], +) -> list[ChangeRecord]: + """Emit one ``kind="field"`` record per differing field in ``fields``. + + The ``fields`` iterable is supplied by the caller — typically + :func:`scalar_fields_for` at listener wiring time. Keeping the + field list outside this function means adding a new column to a + model does not require a matching edit here. + """ + records: list[ChangeRecord] = [] + for field in sorted(fields): + record = _diff_scalar(field, pre.get(field), post.get(field)) + if record is not None: + records.append(record) + return records + + +def diff_slice( + pre: dict[str, Any], + post: dict[str, Any], + *, + fields: Iterable[str], +) -> list[ChangeRecord]: + """Full Slice (chart) diff — scalars plus params classification. + + Pass ``fields=scalar_fields_for(Slice, special=frozenset({"params"}))`` + to get the ``params``-excluded scalar set; ``Slice.params`` is diffed + separately by :func:`diff_slice_params` for kind promotion. + """ + records = diff_scalar_fields(pre, post, fields=fields) + records.extend(diff_slice_params(pre.get("params"), post.get("params"))) + return records + + +def diff_json_field( + field_name: str, + from_value: Any, + to_value: Any, + *, + exclude_keys: frozenset[str] = frozenset(), + max_depth: int = _JSON_METADATA_DIFF_DEPTH, +) -> list[ChangeRecord]: + """Diff a TEXT column that stores a JSON dict, emitting one record + per changed leaf. + + Used for ``Dashboard.json_metadata`` (``position_json`` has its + own structural diff via :func:`diff_dashboard_layout`). Saving the + blob verbatim into ``from_value`` / ``to_value`` would swamp the + change log with multi-KB strings on every save; recursing into the + parsed dict reduces noise to "exactly which leaf changed". + + *exclude_keys* names sub-keys that are frontend-derived / + auto-stamped on save and don't carry user-authored signal. Same + rationale as the ``audit`` parameter on + :func:`scalar_fields_for` for the parent-column level. + + Path is ``[field_name, key, ...]`` for leaf records, mirroring + :func:`diff_slice_params`'s ``["params", key, ...]`` shape so + renderers can use a single addressing scheme across the chart + and dashboard sides. + """ + from_p = _coerce_params(from_value) + to_p = _coerce_params(to_value) + records: list[ChangeRecord] = [] + for key in sorted(set(from_p) | set(to_p)): + if key in exclude_keys: + continue + records.extend( + _recursive_leaf_diff( + kind="field", + path_prefix=[field_name, key], + pre=from_p.get(key), + post=to_p.get(key), + max_depth=max_depth, + ) + ) + return records + + +# json_metadata sub-keys that the frontend auto-stamps / auto-derives +# on save. They mirror dashboard membership and chart inventory, not +# user-authored content, so they noise up the change log without +# carrying intent. The records produced for these keys can be ~50KB +# (full label-colour dict) for a one-chart save. +# +# chart_configuration: per-chart cross-filter scope state, +# re-derived when charts are added/removed. +# global_chart_configuration: dashboard-wide filter scope; the +# ``chartsInScope`` list mirrors live +# dashboard membership. +# map_label_colors: label → colour map, re-stamped on save +# from currently-visible filter values. +# shared_label_colors: cross-chart shared-label colour list, +# rewritten by the DAO when a dashboard is +# merely *viewed* — producing phantom +# "Properties updated" records with no +# user edit (surfaced by the +# version-history UI, PR #40988). The +# view-time write itself is a separate +# round-trip-asymmetry issue (cf. #39706); +# this exclusion stops the change-record +# noise regardless. +# show_chart_timestamps: frontend toggle, defaults applied on +# save when missing. +# color_namespace: scoped colour-scheme namespace, frontend- +# derived from the chart set. +DASHBOARD_JSON_METADATA_AUDIT_KEYS: frozenset[str] = frozenset( + { + "chart_configuration", + "global_chart_configuration", + "map_label_colors", + "shared_label_colors", + "show_chart_timestamps", + "color_namespace", + } +) + + +# Layout component types and how they map to record ``kind`` strings. +# ``HEADER_ID`` is excluded — that's the dashboard's title bar, mirrored +# from ``dashboard_title``. ``ROOT_ID`` and ``GRID_ID`` are structural +# singletons whose only deltas are children lists, which we infer from +# the moves of the children themselves. +_LAYOUT_TYPE_TO_KIND: dict[str, str] = { + "CHART": "chart", + "ROW": "row", + "COLUMN": "column", + "TAB": "tab", + "TABS": "tabs", + "HEADER": "header", + "MARKDOWN": "markdown", + "DIVIDER": "divider", +} + +# Layout components we never emit records for: ROOT_ID is the layout +# root (always present, never moves); GRID_ID is the singleton vertical +# stack inside ROOT_ID; HEADER_ID is the dashboard's title bar (already +# covered by the ``dashboard_title`` scalar field). +_LAYOUT_SUPPRESSED_IDS: frozenset[str] = frozenset({"ROOT_ID", "GRID_ID", "HEADER_ID"}) + + +def _layout_component_label(node: dict[str, Any]) -> str | None: + """Extract a human-readable label from a layout node, when one + exists. Used to build the ``from_value`` / ``to_value`` payload so + the UI can render messages like "Added chart 'Foo'" without + needing to fetch related entities. + """ + meta = node.get("meta") or {} + if not isinstance(meta, dict): + return None + for key in ("sliceName", "label", "text"): + value = meta.get(key) + if isinstance(value, str) and value.strip(): + return value + return None + + +def _layout_node_payload(node: dict[str, Any]) -> dict[str, Any]: + """Minimal payload describing a layout node — enough for the UI + to render the change without dragging the full layout snippet + (which can be ~1KB per row when CHART nodes carry colour configs). + """ + meta = node.get("meta") or {} + if not isinstance(meta, dict): + meta = {} + payload: dict[str, Any] = {"id": node.get("id"), "type": node.get("type")} + if (label := _layout_component_label(node)) is not None: + payload["name"] = label + if (chart_id := meta.get("chartId")) is not None: + payload["chartId"] = chart_id + # ``uuid`` (slice uuid for CHART nodes) lets the M2M-vs-layout + # dedupe in :func:`fold_dashboard_layout_with_chart_changes` + # match on the same key — :func:`diff_dashboard_slices` keys its + # records by uuid, not chartId. + if (slice_uuid := meta.get("uuid")) is not None: + payload["uuid"] = slice_uuid + return payload + + +def _layout_parent_id(node: dict[str, Any]) -> Any: + """The immediate-parent node id for a layout component — the last + entry in ``parents``. Used to detect moves: same id, different + parent.""" + parents = node.get("parents") or [] + if not isinstance(parents, list) or not parents: + return None + return parents[-1] + + +def _meta_excluding_position(node: dict[str, Any]) -> dict[str, Any]: + """Meta dict with ``parents``-equivalent positional bits removed + so two nodes that differ ONLY in where they sit compare equal at + the meta level. Move detection uses ``parents`` directly; this is + for "edit" (meta change) detection.""" + meta = node.get("meta") or {} + return dict(meta) if isinstance(meta, dict) else {} + + +def _diff_layout_node( + node_id: str, + pre_node: dict[str, Any] | None, + post_node: dict[str, Any] | None, +) -> list[ChangeRecord]: + """Diff one component slot in the layout dict and return records for + the logical action — add, remove, move, edit. + + add / remove / move emit a single record carrying the minimal node + payload (so the renderer can describe the affected component). + edit recurses into the node's ``meta`` dict and emits one record per + changed leaf, capped at ``_LAYOUT_META_DIFF_DEPTH``. + + Returns an empty list when the slot is unchanged or holds an unknown + component type. + """ + node_for_kind = post_node or pre_node or {} + kind = _LAYOUT_TYPE_TO_KIND.get(node_for_kind.get("type") or "") + if kind is None: + return [] # unknown component type — skip rather than emit garbage + + if pre_node is None and post_node is not None: + return [ + ChangeRecord( + kind=kind, + operation="add", + path=[node_id], + from_value=None, + to_value=_layout_node_payload(post_node), + ) + ] + if post_node is None and pre_node is not None: + return [ + ChangeRecord( + kind=kind, + operation="remove", + path=[node_id], + from_value=_layout_node_payload(pre_node), + to_value=None, + ) + ] + + # Both present — check move first, then edit. + assert pre_node is not None + assert post_node is not None + pre_parent = _layout_parent_id(pre_node) + if pre_parent != (post_parent := _layout_parent_id(post_node)): + return [ + ChangeRecord( + kind=kind, + operation="move", + path=[node_id], + from_value={**_layout_node_payload(pre_node), "parent": pre_parent}, + to_value={**_layout_node_payload(post_node), "parent": post_parent}, + ) + ] + + # Edit: recurse into meta and emit one record per changed leaf. + # Path shape ``[node_id, , ...]``. The verb (operation) is + # derived per-leaf by the recursion via ``_operation_from_values``; + # a leaf added inside an existing node gets ``add`` and so on. The + # node-level "this was an edit" fact is implicit in the path shape + # carrying segments after ``node_id``. + return _recursive_leaf_diff( + kind=kind, + path_prefix=[node_id], + pre=_meta_excluding_position(pre_node), + post=_meta_excluding_position(post_node), + max_depth=_LAYOUT_META_DIFF_DEPTH, + ) + + +def diff_dashboard_layout( + pre: Any, + post: Any, +) -> list[ChangeRecord]: + """Structural diff of a dashboard's ``position_json``, emitting one + record per logical layout action. + + Walks both sides keyed on the component ``id`` (e.g. + ``"CHART-mkPZLOnWCElgL0Udp1gVK"``): + + * id present only in *post* → ``op=add``, ``from_value=None``, + ``to_value=`` + * id present only in *pre* → ``op=remove``, payload swapped + * id in both, ``parents`` differs → ``op=move``, payloads carry + old + new parent + * id in both, parents equal, ``meta`` differs → ``op=edit``, + payloads carry old + new meta + * id in both, equal → no record + + The ``operation_type``-style verb is encoded in + ``path[0]`` as ``["add"|"remove"|"move"|"edit", , + ]`` so the UI's path-based renderer can read it + without inspecting from/to. + + ``ROOT_ID`` / ``GRID_ID`` / ``HEADER_ID`` are suppressed (see + :data:`_LAYOUT_SUPPRESSED_IDS`). + """ + pre_nodes = _layout_nodes(pre) + post_nodes = _layout_nodes(post) + records: list[ChangeRecord] = [] + for node_id in sorted(set(pre_nodes) | set(post_nodes)): + records.extend( + _diff_layout_node(node_id, pre_nodes.get(node_id), post_nodes.get(node_id)) + ) + return records + + +def _layout_nodes(raw: Any) -> dict[str, dict[str, Any]]: + """Coerce *raw* (a ``position_json`` blob or already-parsed dict) into + the ``{node_id: node_dict}`` shape used by the layout diff, filtering + out non-dict values and the always-present root/grid/header singletons. + """ + parsed = _coerce_params(raw) + return { + k: v + for k, v in parsed.items() + if isinstance(v, dict) and k not in _LAYOUT_SUPPRESSED_IDS + } + + +def diff_dashboard( + pre: dict[str, Any], + post: dict[str, Any], + *, + fields: Iterable[str], +) -> list[ChangeRecord]: + """Dashboard diff: scalar fields plus structural diff of + ``json_metadata`` and ``position_json``. + + Promoting ``position_json`` to ``kind="layout"`` or + ``json_metadata.native_filter_configuration`` to ``kind="filter"`` + is deferred to Phase 2 alongside the UI that would render them + (spec Clarifications §Session 2026-04-24); until then, both fields + fall through to ``kind="field"`` records keyed by sub-key. + """ + records = diff_scalar_fields(pre, post, fields=fields) + records.extend( + diff_json_field( + "json_metadata", + pre.get("json_metadata"), + post.get("json_metadata"), + exclude_keys=DASHBOARD_JSON_METADATA_AUDIT_KEYS, + ) + ) + records.extend( + diff_dashboard_layout(pre.get("position_json"), post.get("position_json")) + ) + return records + + +def _layout_chart_uuids_by_verb( + records: list[ChangeRecord], +) -> tuple[set[Any], set[Any]]: + """Scan *records* for layout ``add``/``remove`` records on charts and + return ``(added_uuids, removed_uuids)`` sets. + + Keys off ``operation`` (the explicit verb column) rather than + ``path[0]`` — paths no longer carry the verb. + """ + added: set[Any] = set() + removed: set[Any] = set() + for r in records: + if r.kind != "chart": + continue + # Layout chart records have ``path = [node_id]`` (length 1) for + # add/remove/move and ``[node_id, ...leaf]`` for edits. We only + # care about the structural add/remove cases here. + if len(r.path) != 1: + continue + if r.operation == "add" and isinstance(r.to_value, dict): + uuid_ = r.to_value.get("uuid") + if uuid_ is not None: + added.add(uuid_) + elif r.operation == "remove" and isinstance(r.from_value, dict): + uuid_ = r.from_value.get("uuid") + if uuid_ is not None: + removed.add(uuid_) + return added, removed + + +def _is_redundant_m2m_chart_record( + r: ChangeRecord, added_uuids: set[Any], removed_uuids: set[Any] +) -> bool: + """Return ``True`` when *r* is an M2M-style slice record that + duplicates an already-captured layout add/remove for the same uuid. + + M2M slice records have path ``["slices", uuid]`` (length 2); their + info is strictly less than the corresponding layout record's + (no name, no parent), so the layout side wins on dedup. + """ + if r.kind != "chart" or len(r.path) != 2 or r.path[0] != "slices": + return False + slice_uuid = r.path[1] + if r.from_value is None and r.to_value is not None: + return slice_uuid in added_uuids + if r.to_value is None and r.from_value is not None: + return slice_uuid in removed_uuids + return False + + +def fold_dashboard_layout_with_chart_changes( + records: list[ChangeRecord], +) -> list[ChangeRecord]: + """When a dashboard save adds/removes charts, the ``slices`` M2M + diff and the layout diff each emit a record for the same logical + action. Drop the M2M ``kind="chart"`` records — the layout-side + record carries more information (chart name, parent container). + + The matching is by slice uuid: ``diff_dashboard_slices`` produces + records with path ``["slices", ]``; the layout + payloads carry the same uuid (sourced from + ``position_json.CHART-x.meta.uuid``). We dedupe on that key. + + Called from the change-records listener after the M2M and layout + diffs are both merged into the per-entity buffer. + """ + added_uuids, removed_uuids = _layout_chart_uuids_by_verb(records) + return [ + r + for r in records + if not _is_redundant_m2m_chart_record(r, added_uuids, removed_uuids) + ] + + +def diff_dataset( + pre: dict[str, Any], + post: dict[str, Any], + *, + fields: Iterable[str], +) -> list[ChangeRecord]: + """SqlaTable scalar-field diff. All paths emit ``kind="field"``. + + Children (columns, metrics) are diffed separately via + :func:`diff_dataset_columns` / :func:`diff_dataset_metrics`. The + listener reads them from Continuum shadow tables + (``table_columns_version`` / ``sql_metrics_version``) rather than + walking the ORM collection. + """ + return diff_scalar_fields(pre, post, fields=fields) + + +def diff_dataset_columns( + from_columns: list[dict[str, Any]] | None, + to_columns: list[dict[str, Any]] | None, +) -> list[ChangeRecord]: + """Child-collection diff on TableColumn rows, keyed by column_name.""" + return _diff_list_by_natural_key( + kind="column", + path_prefix=["columns"], + from_list=from_columns, + to_list=to_columns, + key_fn=lambda c: c.get("column_name") if isinstance(c, dict) else None, + ) + + +def diff_dataset_metrics( + from_metrics: list[dict[str, Any]] | None, + to_metrics: list[dict[str, Any]] | None, +) -> list[ChangeRecord]: + """Child-collection diff on SqlMetric rows, keyed by metric_name.""" + return _diff_list_by_natural_key( + kind="metric", + path_prefix=["metrics"], + from_list=from_metrics, + to_list=to_metrics, + key_fn=lambda m: m.get("metric_name") if isinstance(m, dict) else None, + ) + + +def diff_dashboard_slices( + from_slice_uuids: list[str] | None, + to_slice_uuids: list[str] | None, +) -> list[ChangeRecord]: + """Diff a dashboard's chart membership, keyed by slice uuid. + + Pure set-diff: added uuids get ``from_value=None, to_value=uuid``; + removed uuids get the inverse. No "changed" case because chart + associations are identity-only (the list element IS the uuid). + """ + from_set = set(from_slice_uuids or []) + to_set = set(to_slice_uuids or []) + records: list[ChangeRecord] = [] + for uuid_ in sorted(from_set - to_set): + records.append( + ChangeRecord( + kind="chart", + operation="remove", + path=["slices", uuid_], + from_value=uuid_, + to_value=None, + ) + ) + for uuid_ in sorted(to_set - from_set): + records.append( + ChangeRecord( + kind="chart", + operation="add", + path=["slices", uuid_], + from_value=None, + to_value=uuid_, + ) + ) + return records diff --git a/superset/versioning/etag.py b/superset/versioning/etag.py new file mode 100644 index 000000000000..643e1a95dadc --- /dev/null +++ b/superset/versioning/etag.py @@ -0,0 +1,75 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""ETag header emission for the entity-versioning API surface.""" + +from __future__ import annotations + +from typing import TYPE_CHECKING +from uuid import UUID + +import sqlalchemy as sa +from flask_appbuilder import Model + +from superset.extensions import db + +if TYPE_CHECKING: + from flask import Response + + +def set_version_etag(response: "Response", version_uuid: UUID | None) -> "Response": + """Attach ``ETag: ""`` to *response*. + + Uses RFC 7232 strong-validator form (no leading ``W/``); the response + header value is wrapped in double quotes per the spec. No-op when + *version_uuid* is ``None`` (entity has no version rows yet). + """ + if version_uuid is not None: + response.headers["ETag"] = f'"{version_uuid}"' + return response + + +def set_version_etag_by_uuid( + response: "Response", + model_cls: type[Model], + entity_uuid: UUID, + *, + entity_id: int | None = None, +) -> "Response": + """Attach ``ETag`` derived from *entity_uuid*'s current live version. + + If *entity_id* is provided the helper uses it directly; otherwise it + runs ``SELECT id WHERE uuid = ?`` to resolve it. Pass *entity_id* + from call sites that already have the entity in hand (e.g. via + :func:`superset.versioning.api_helpers.resolve_endpoint_path_entity`) + so the lookup doesn't fire twice — every list/get versions request + previously cost an extra round-trip here on top of the resolve. + + No-op when the entity is missing or has no version rows yet. + """ + # pylint: disable=import-outside-toplevel + from superset.daos.version import VersionDAO + + if entity_id is None: + entity_id = db.session.scalar( + sa.select(model_cls.id).where(model_cls.uuid == entity_uuid) + ) + if entity_id is None: + return response + return set_version_etag( + response, + VersionDAO.current_live_version_uuid(model_cls, entity_id, entity_uuid), + ) diff --git a/superset/versioning/factory.py b/superset/versioning/factory.py new file mode 100644 index 000000000000..2e03b8244e33 --- /dev/null +++ b/superset/versioning/factory.py @@ -0,0 +1,312 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +from __future__ import annotations + +import logging +from collections.abc import Callable +from typing import Any + +import sqlalchemy as sa +import sqlalchemy.orm as sa_orm +from sqlalchemy_continuum import is_modified, version_class +from sqlalchemy_continuum.operation import Operation +from sqlalchemy_continuum.plugins.base import Plugin +from sqlalchemy_continuum.plugins.flask import FlaskPlugin +from sqlalchemy_continuum.transaction import TransactionFactory +from sqlalchemy_continuum.utils import versioned_column_properties + +from superset.utils import json +from superset.versioning.diff import DASHBOARD_JSON_METADATA_AUDIT_KEYS + +logger = logging.getLogger(__name__) + + +def _normalize_dashboard_json_metadata(value: Any) -> Any: + """Parse ``dashboards.json_metadata`` and drop frontend-stamped audit + sub-keys so a save that only re-stamps ``map_label_colors`` (etc.) + compares equal to its predecessor. + + ``map_label_colors`` is regenerated client-side from the + ``LabelsColorMap`` singleton on every save (see + ``saveDashboardRequest`` in + ``superset-frontend/src/dashboard/actions/dashboardState.ts``). + The singleton's contents depend on which charts have rendered in + the page session, so two saves with no user-authored change produce + different bytes. The diff engine ignores the same audit sub-keys + (``DASHBOARD_JSON_METADATA_AUDIT_KEYS`` in + ``superset/versioning/diff.py``); aligning the skip-plugin's + comparison with that filter keeps the two paths consistent. + """ + if value is None or value == "": + return value + try: + parsed = json.loads(value) + except (TypeError, ValueError): + return value + if not isinstance(parsed, dict): + return parsed + return { + k: v for k, v in parsed.items() if k not in DASHBOARD_JSON_METADATA_AUDIT_KEYS + } + + +# Per-class column normalizers, keyed on (class_name, column_name). Class +# name is used (rather than class itself) so importing the model classes +# at module load is unnecessary — keeps the plugin importable before +# ``make_versioned()`` has registered the version classes. +_COLUMN_NORMALIZERS: dict[tuple[str, str], Callable[[Any], Any]] = { + ("Dashboard", "json_metadata"): _normalize_dashboard_json_metadata, +} + + +def _normalize_for_compare(target: Any, col_name: str, value: Any) -> Any: + """Return *value* run through any per-class column normalizer registered + in ``_COLUMN_NORMALIZERS``, else *value* unchanged. + """ + normalizer = _COLUMN_NORMALIZERS.get((type(target).__name__, col_name)) + return normalizer(value) if normalizer is not None else value + + +def _has_dirty_versioned_children(target: Any, uow: Any) -> bool: + """Return ``True`` when *uow* contains an operation for a versioned + child of *target* (e.g. a ``TableColumn`` whose ``table`` is *target*). + + Used by :meth:`SkipUnmodifiedPlugin._is_no_op_update` so a parent + UPDATE that was force-flagged by + :func:`baseline.force_parent_dirty_on_child_change` is preserved + even though the parent's own scalars match the previous version. + """ + # pylint: disable=import-outside-toplevel + from superset.versioning.baseline import child_to_parent_registry + + child_map = child_to_parent_registry() + target_cls = type(target) + for _key, op in uow.operations.items(): + entry = child_map.get(type(op.target)) + if entry is None: + continue + parent_attr, parent_cls = entry + if parent_cls is not target_cls: + continue + parent = getattr(op.target, parent_attr, None) + if parent is target: + return True + return False + + +class VersionTransactionFactory(TransactionFactory): + """TransactionFactory that renames the transaction table and adds a bare + ``user_id`` integer column so the FlaskPlugin can record the acting user + without requiring a FK relationship to ``ab_user``. + + Continuum only adds ``user_id`` when ``user_cls`` is set on the manager. + We add it unconditionally (no FK) so that both the FlaskPlugin's + ``transaction_args()`` and our ``baseline.py`` direct inserts can record + which user triggered the version event. + """ + + def create_class(self, manager: Any) -> Any: + cls = super().create_class(manager) + cls.__table__.name = "version_transaction" + # Rename the PostgreSQL sequence for consistent naming. + for col in cls.__table__.columns: + if col.name == "id" and col.default is not None: + col.default.name = "version_transaction_id_seq" + # Add user_id INTEGER (no FK) for user tracking. The mapper has not + # been configured yet at this point, so append_column + add_property + # is safe here. + user_id_col = sa.Column("user_id", sa.Integer, nullable=True) + cls.__table__.append_column(user_id_col) + cls.__mapper__.add_property("user_id", sa_orm.column_property(user_id_col)) + # ``action_kind`` — high-level avenue that produced this commit + # (``restore`` / ``import`` / ``clone`` / ``NULL`` for ordinary + # saves). The DDL is in the consolidated Alembic migration; we + # also declare it on the SQLAlchemy Table here so consumers + # can reference ``versioning_manager.transaction_cls.__table__ + # .c.action_kind`` at runtime, and so the change-record + # listener's ``sa.update()`` stamp emits the correctly-quoted + # identifier per dialect. + action_kind_col = sa.Column("action_kind", sa.String(32), nullable=True) + cls.__table__.append_column(action_kind_col) + cls.__mapper__.add_property( + "action_kind", sa_orm.column_property(action_kind_col) + ) + return cls + + +class VersioningFlaskPlugin(FlaskPlugin): + """FlaskPlugin subclass that uses Superset's :func:`get_user_id` (which + reads ``g.user``) instead of Flask-Login's ``current_user``. Superset's + JWT auth for API routes populates ``g.user`` but leaves + ``flask_login.current_user`` anonymous, so the upstream plugin would + record ``user_id=NULL`` on version_transaction rows created by API + calls. Returns an empty dict (so the transaction row is written + anyway) when no user is available — e.g. CLI, Celery, import/export. + """ + + def transaction_args(self, uow: Any, session: Any) -> dict[str, Any]: + # pylint: disable=import-outside-toplevel + from flask import has_request_context, request + + from superset.utils.core import get_user_id + + user_id = get_user_id() + # get_user_id() returns an int in a real request, or None for a + # context-less save (CLI, Celery, import/export). Guard against any + # non-int so a bogus value never reaches the integer + # ``version_transaction.user_id`` column — notably a mocked ``g`` in + # tests, where ``g.user.id`` is a Mock rather than None and would + # otherwise blow up the flush with a SQL bind error. + if not isinstance(user_id, int): + return {} + + remote_addr: str | None + try: + remote_addr = request.remote_addr if has_request_context() else None + except RuntimeError: + remote_addr = None + + return {"user_id": user_id, "remote_addr": remote_addr} + + +class SkipUnmodifiedPlugin(Plugin): + """Skip creating version rows for UPDATE operations whose post-flush + column values are byte-identical to the previous live version row. + + Continuum creates a version row for every entity in ``session.dirty``, + including saves where the SQLAlchemy ORM marked a column dirty (because + Superset re-serialised ``json_metadata`` via ``json.dumps`` on the save + path, or AuditMixin auto-bumped ``changed_on``) but the resulting value + is unchanged from the previous version. Those rows pollute the version + history with no-op entries. + + ``is_modified()`` from Continuum is not enough: it consults SQLAlchemy's + attribute history, which is "did setattr produce a different value?", + not "did the final stored value change?". So we compare each + non-excluded versioned column on ``operation.target`` against the + previous live version row's value; if all are equal, the operation + is marked ``processed`` and Continuum skips it (see + ``UnitOfWork.create_version_objects``). + + The associated transaction is not removed; if every operation is a + no-op the transaction becomes an orphan in ``version_transaction`` + and is swept by the retention task at cutoff. Deleting the row + inline (in this hook) was considered and rejected: it would couple + this plugin to the change-records listener's buffer state — both + would have to agree that the flush produced nothing before we + could safely DROP the tx row, since ``version_changes.transaction_id`` + has an ON DELETE CASCADE FK that would silently drop any buffered + diff records the listener was about to insert. The orphan's storage + cost (~40 bytes/row) is small enough that the coordination isn't + worth it; retention handles the cleanup correctly by construction + (orphans have no parent shadow → they're never "preserved" by the + "preserve transactions whose shadow has the live row" rule and + age out with the rest of the history). + """ + + def before_create_version_objects(self, uow: Any, session: Any) -> None: + # ``uow.operations`` is a custom Continuum ``Operations`` collection; + # use its ``.items()`` method (not ``.values()``) to iterate. + # INSERTs always create a row (no prior to compare against); + # DELETEs can't be no-ops. Only UPDATE operations are candidates. + for _key, operation in uow.operations.items(): + if operation.processed or operation.type != Operation.UPDATE: + continue + try: + if self._is_no_op_update(operation.target, session, uow): + operation.processed = True + except Exception: # pylint: disable=broad-except + # Defensive — if introspection fails for any reason, fall + # back to creating the version row. + logger.exception( + "SkipUnmodifiedPlugin: skip-check raised for %s", + type(operation.target).__name__, + ) + + @classmethod + def _is_no_op_update(cls, target: Any, session: Any, uow: Any) -> bool: + """Return ``True`` when this UPDATE produces no observable change to + any non-excluded versioned column **and** no versioned children of + *target* are being modified in this flush. + + Stages: + + 1. If any versioned child (e.g. a ``TableColumn`` whose ``table`` + is *target*) has an operation in ``uow.operations``, the parent + is being force-touched by + ``baseline.force_parent_dirty_on_child_change`` to anchor the + child changes against a parent shadow row. Keep the row. + 2. ``is_modified(target)`` — cheap SQLAlchemy attribute-history + check. Returns ``False`` when only excluded columns/relationships + (``owners``, ``changed_on``, …) are dirty. This is the common + case (every save auto-bumps ``changed_on``); short-circuiting + here saves the DB round-trip in stage 3. + 3. Compare post-flush column values against the previous live + version row's stored values. Catches the case where SQLAlchemy + sees a column as dirty (e.g. ``set_dash_metadata`` re-serialised + ``json_metadata`` to a different byte sequence) but the + resulting parsed content matches the prior version. + """ + if _has_dirty_versioned_children(target, uow): + return False + if not is_modified(target): + return True + return cls._matches_previous_version(target, session) + + @staticmethod + def _matches_previous_version(target: Any, session: Any) -> bool: + """Return ``True`` when every non-excluded versioned column on + *target* matches the value stored in its previous live version row + (i.e., the row with ``end_transaction_id IS NULL``). + + Returns ``False`` for entities with no prior version row — letting + Continuum create the first one. In practice this case is rare: + ``register_baseline_listener`` (in ``superset.versioning.baseline``) + runs ahead of Continuum's ``before_flush`` and inserts a baseline + row for any entity being saved for the first time, so the second + save (and beyond) is what flows through this path. + """ + cls = type(target) + try: + ver_cls = version_class(cls) + except Exception: # pylint: disable=broad-except + return False + ver_table = ver_cls.__table__ + + col_keys = [prop.key for prop in versioned_column_properties(target)] + if not col_keys: + return False + + select_stmt = ( + sa.select(*[ver_table.c[c] for c in col_keys]) + .where(ver_table.c.id == target.id) + .where(ver_table.c.end_transaction_id.is_(None)) + .order_by(ver_table.c.transaction_id.desc()) + .limit(1) + ) + row = session.connection().execute(select_stmt).first() + if row is None: + return False # no previous version → let Continuum create one + + for col_name, prev_value in zip(col_keys, row, strict=False): + post = _normalize_for_compare( + target, col_name, getattr(target, col_name, None) + ) + pre = _normalize_for_compare(target, col_name, prev_value) + if post != pre: + return False + return True diff --git a/superset/versioning/queries.py b/superset/versioning/queries.py new file mode 100644 index 000000000000..ca9f58625827 --- /dev/null +++ b/superset/versioning/queries.py @@ -0,0 +1,516 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Read-side queries for the entity-versioning API. + +Pure-read helpers that translate Continuum shadow rows and +``version_changes`` records into the shapes the API endpoints return. +The corresponding write side (restore) lives in +:mod:`superset.versioning.restore`. The backward-compat ``VersionDAO`` +façade in :mod:`superset.daos.version` re-exports both. + +Also exposes the deterministic version-UUID derivation +(:data:`VERSION_UUID_NAMESPACE` + :func:`derive_version_uuid`) used by +both the read endpoints and the ETag emission path in +:mod:`superset.versioning.etag`. +""" + +from __future__ import annotations + +import uuid +from typing import Any +from uuid import UUID + +import sqlalchemy as sa +from flask_appbuilder import Model +from sqlalchemy_continuum import version_class + +from superset.extensions import db +from superset.versioning.baseline import CONTINUUM_BOOKKEEPING_COLUMNS + +# Fixed UUIDv5 namespace under which per-(entity, transaction) version UUIDs +# are derived. Never change this constant — changing it invalidates every +# version_uuid that clients may have cached, bookmarked, or stored. +VERSION_UUID_NAMESPACE = UUID("7a6f5d9b-4c3b-5d8e-9a1c-0e2b4c6d8f10") + +# Continuum's integer ``operation_type`` mapped to the string the API +# returns. Kept short and stable for downstream tooling consuming the +# raw response. Continuum guarantees 0/1/2; anything else is a Continuum +# version mismatch and surfaces as ``str(int)`` rather than crashing. +_OP_TYPE_LABELS: dict[int, str] = {0: "baseline", 1: "update", 2: "delete"} + + +def derive_version_uuid(entity_uuid: UUID, transaction_id: int) -> UUID: + """Derive a deterministic UUIDv5 identifying one version row. + + The UUID is a function of the owning entity's UUID and the Continuum + ``transaction_id`` of the version row, so it is stable across retention + pruning (which never changes ``transaction_id``) and portable across + replicas. It is not randomly generated — two Supersets with identical + ``(entity.uuid, transaction_id)`` will compute the same version_uuid. + """ + return uuid.uuid5(VERSION_UUID_NAMESPACE, f"{entity_uuid}:{transaction_id}") + + +def _resolve_version_tables( + model_cls: type[Model], +) -> tuple[sa.Table, sa.Table, sa.Table]: + """Return the (version, transaction, user) ``Table`` objects used by the + listing and snapshot queries. + + All three lookups happen inside this module on every read; centralising + the trio (a) keeps the imports in one place and (b) makes the join helper + below take a uniform signature. + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import versioning_manager + + from superset import security_manager + + ver_tbl = version_class(model_cls).__table__ + tx_tbl = versioning_manager.transaction_cls.__table__ + user_tbl = security_manager.user_model.__table__ + return ver_tbl, tx_tbl, user_tbl + + +def _version_with_tx_user_join( + ver_tbl: sa.Table, tx_tbl: sa.Table, user_tbl: sa.Table +) -> Any: + """Build the version → transaction → user left-join used by both + :func:`list_versions` and :func:`get_version`. The user-side join is + a left-outer so saves with no Flask user context (CLI, Celery, import) + still surface in the result with ``changed_by = None``. + """ + return ver_tbl.join(tx_tbl, ver_tbl.c.transaction_id == tx_tbl.c.id).outerjoin( + user_tbl, tx_tbl.c.user_id == user_tbl.c.id + ) + + +def _baseline_first_ordering(ver_tbl: sa.Table) -> tuple[Any, ...]: + """Order ``(operation_type != 0).asc(), transaction_id.asc()`` so any + op=0 row — Continuum's INSERT or our synthetic baseline — sorts to + position 0 regardless of its transaction_id. A single entity never has + more than one op=0 row (Continuum tracks one creation per live entity; + our baseline listener only fires when no prior version rows exist), so + this gives a stable chronological order with the "original" version + always first. + """ + return ( + (ver_tbl.c.operation_type != 0).asc(), + ver_tbl.c.transaction_id.asc(), + ) + + +def _user_select_cols(user_tbl: sa.Table) -> list[Any]: + """Columns to select from ``user_tbl`` to build a ``changed_by`` dict. + Labels ``user_tbl.c.id`` as ``"user_id"`` so callers can read the row + by a stable key regardless of whether they also select the version + table's ``id`` column. + """ + return [ + user_tbl.c.id.label("user_id"), + user_tbl.c.username, + user_tbl.c.first_name, + user_tbl.c.last_name, + ] + + +def _changed_by_from_row(row: Any) -> dict[str, Any] | None: + """Project the user columns from a query row onto the API's + ``changed_by`` shape, or ``None`` for saves with no Flask user context + (CLI / Celery / import / unauthenticated). Expects the user columns to + have been selected via :func:`_user_select_cols` so the row keys are + ``user_id`` / ``username`` / ``first_name`` / ``last_name``. + """ + if row["user_id"] is None: + return None + return { + "id": row["user_id"], + "username": row["username"], + "first_name": row["first_name"], + "last_name": row["last_name"], + } + + +def _entity_kind_for(model_cls: type[Model]) -> str | None: + """Return the ``version_changes.entity_kind`` value for *model_cls*, or + ``None`` when the class isn't in the change-records taxonomy.""" + # pylint: disable=import-outside-toplevel + from superset.versioning.changes import ENTITY_KIND_BY_CLASS_NAME + + return ENTITY_KIND_BY_CLASS_NAME.get(model_cls.__name__) + + +def find_active_by_uuid(model_cls: type[Model], entity_uuid: UUID) -> Any | None: + """Return the live entity matching *entity_uuid*, or None if not found.""" + return ( + db.session.query(model_cls).filter(model_cls.uuid == entity_uuid).one_or_none() + ) + + +def _get_version_count(model_cls: type[Model], entity_id: int) -> int: + """Return the number of historical version rows for *entity_id*.""" + ver_cls = version_class(model_cls) + return ( + db.session.query(sa.func.count()) + .select_from(ver_cls) + .filter(ver_cls.id == entity_id) + .scalar() + or 0 + ) + + +def current_version_number(model_cls: type[Model], entity_id: int) -> int | None: + """Return the 0-based ``version_number`` of the live row for *entity_id* + — equivalent to the index of the most recent entry that + :func:`list_versions` would return, or ``None`` when the entity has no + version rows yet. + + Note: this index is *unstable under retention pruning*. The scheduled + retention task drops shadow rows older than the configured + retention window, so the same integer can refer to different rows + before and after a prune cycle. Use + :func:`current_live_transaction_id` for a stable identifier. + """ + count = _get_version_count(model_cls, entity_id) + return count - 1 if count > 0 else None + + +def current_live_transaction_id(model_cls: type[Model], entity_id: int) -> int | None: + """Return the Continuum ``transaction_id`` of the live row for + *entity_id* — stable across retention pruning, unlike the index + returned by :func:`current_version_number`. + """ + ver_cls = version_class(model_cls) + row = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == entity_id) + .filter(ver_cls.end_transaction_id.is_(None)) + .order_by(ver_cls.transaction_id.desc()) + .limit(1) + .first() + ) + return row[0] if row else None + + +def current_live_version_uuid( + model_cls: type[Model], entity_id: int, entity_uuid: UUID +) -> UUID | None: + """Return the deterministic ``version_uuid`` of the live row, or + ``None`` when the entity has no version rows yet.""" + tx_id = current_live_transaction_id(model_cls, entity_id) + if tx_id is None: + return None + return derive_version_uuid(entity_uuid, tx_id) + + +def list_change_records_batch( + entity_kind: str, + entity_id: int, + transaction_ids: list[int], +) -> dict[int, list[dict[str, Any]]]: + """Return ``version_changes`` rows keyed by ``transaction_id``. + + Batches the lookup across multiple transactions with a single + ``WHERE transaction_id IN (...) AND entity_kind = ? AND entity_id = ?`` + query so the list endpoint avoids N+1 round-trips. Rows are + distributed into per-tx lists sorted by ``sequence`` ascending + (matching the replay order the diff engine emits). Missing + transactions are represented by an empty list in the result so + callers can use ``result.get(tx_id, [])`` without guarding. + + If the ``version_changes`` table is missing (pre-migration or + freshly downgraded), returns an empty dict rather than propagating + the error — consistent with this being a descriptive layer that + should not break the list endpoint. + """ + # pylint: disable=import-outside-toplevel + from superset.versioning.changes import version_changes_table + + if not transaction_ids: + return {} + + # SAVEPOINT so a missing-table failure can't poison the enclosing + # transaction: on PostgreSQL a failed statement aborts the tx, and + # every later query in the request would raise InFailedSqlTransaction + # even though the exception below was caught. + try: + with db.session.connection().begin_nested(): + rows = ( + db.session.connection() + .execute( + sa.select( + version_changes_table.c.transaction_id, + version_changes_table.c.sequence, + version_changes_table.c.kind, + version_changes_table.c.path, + version_changes_table.c.from_value, + version_changes_table.c.to_value, + ) + .where( + version_changes_table.c.entity_kind == entity_kind, + version_changes_table.c.entity_id == entity_id, + version_changes_table.c.transaction_id.in_(transaction_ids), + ) + .order_by( + version_changes_table.c.transaction_id.asc(), + version_changes_table.c.sequence.asc(), + ) + ) + .mappings() + .all() + ) + except (sa.exc.OperationalError, sa.exc.ProgrammingError): + # Missing version_changes table: OperationalError on SQLite/MySQL, + # ProgrammingError (UndefinedTable) on PostgreSQL. + return {} + + grouped: dict[int, list[dict[str, Any]]] = {tx: [] for tx in transaction_ids} + for row in rows: + grouped[row["transaction_id"]].append( + { + "kind": row["kind"], + "path": row["path"], + "from_value": row["from_value"], + "to_value": row["to_value"], + } + ) + return grouped + + +def list_versions( + model_cls: type[Model], + entity_uuid: UUID, + *, + entity: Any | None = None, +) -> list[dict[str, Any]] | None: + """Return the version history for the entity identified by *entity_uuid*. + + Returns ``None`` when no active entity matches the UUID — callers should + translate that into a 404. Returns an empty list when the entity exists + but has no version rows yet (pre-migration, or never edited). + + The list is ordered by ``transaction_id`` ascending and each entry is + assigned a 0-based sequential ``version_number``. ``operation_type`` is + mapped from Continuum's integer constants to a string (``0`` → baseline, + ``1`` → update, ``2`` → delete). ``changed_by`` is the User row keyed + off ``version_transaction.user_id``, or ``None`` when the save had no + Flask user context (CLI, import, etc.). + + Pass *entity* to skip the ``find_active_by_uuid`` lookup when the + caller has already resolved the entity (API handlers do this to enforce + ``raise_for_ownership`` before calling here). The skip saves one + ``WHERE uuid = ?`` query — that lookup isn't identity-map-cacheable + because ``uuid`` is a unique non-PK column. + """ + if entity is None: + entity = find_active_by_uuid(model_cls, entity_uuid) + if entity is None: + return None + + ver_tbl, tx_tbl, user_tbl = _resolve_version_tables(model_cls) + stmt = ( + sa.select( + ver_tbl.c.transaction_id, + ver_tbl.c.operation_type, + tx_tbl.c.issued_at, + *_user_select_cols(user_tbl), + ) + .select_from(_version_with_tx_user_join(ver_tbl, tx_tbl, user_tbl)) + .where(ver_tbl.c.id == entity.id) + .order_by(*_baseline_first_ordering(ver_tbl)) + ) + rows = db.session.execute(stmt).mappings().all() + + # Batch-load change records for every listed transaction in one query. + # ``entity_kind`` is derived from the model class so the API + # filter ``WHERE entity_kind = 'chart' AND entity_id = ?`` can be + # precise when multiple versioned entities share a flush. + changes_by_tx: dict[int, list[dict[str, Any]]] = {} + if (entity_kind := _entity_kind_for(model_cls)) is not None: + tx_ids = [row["transaction_id"] for row in rows] + changes_by_tx = list_change_records_batch(entity_kind, entity.id, tx_ids) + + return [ + { + "version_uuid": derive_version_uuid(entity_uuid, row["transaction_id"]), + "version_number": version_number, + "transaction_id": row["transaction_id"], + "operation_type": _OP_TYPE_LABELS.get( + row["operation_type"], str(row["operation_type"]) + ), + "issued_at": row["issued_at"], + "changed_by": _changed_by_from_row(row), + "changes": changes_by_tx.get(row["transaction_id"], []), + } + for version_number, row in enumerate(rows) + ] + + +def resolve_version_uuid( + model_cls: type[Model], + entity_uuid: UUID, + version_uuid: UUID, + *, + entity: Any | None = None, +) -> int | None: + """Translate a ``version_uuid`` into the 0-based ``version_number`` that + :func:`superset.versioning.restore.restore_version` accepts, or ``None`` + when the UUID does not match any version row of the given entity. + + Ordering matches :func:`list_versions` — op=0 rows first, then by + transaction_id — so the version_number returned here is the same index + a client would see in the list response. + + Implementation note: the loop re-derives ``version_uuid`` per + transaction in Python because there's no portable SQL form for a + UUIDv5 derivation across PostgreSQL / MySQL / SQLite (Postgres has + ``uuid_generate_v5``; the other two do not). The iteration count is + bounded by the configured retention window worth of edits — the + retention task ages older shadow rows out — so the + practical N is at most a few hundred. If retention is ever + disabled on a heavily-edited entity, this loop is the + place to revisit. + + Pass *entity* to skip the ``find_active_by_uuid`` lookup; see + :func:`list_versions` for the rationale. + """ + if entity is None: + entity = find_active_by_uuid(model_cls, entity_uuid) + if entity is None: + return None + + ver_cls = version_class(model_cls) + tx_ids = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == entity.id) + .order_by( + (ver_cls.operation_type != 0).asc(), + ver_cls.transaction_id.asc(), + ) + .all() + ) + for version_number, (tx_id,) in enumerate(tx_ids): + if derive_version_uuid(entity_uuid, tx_id) == version_uuid: + return version_number + return None + + +def get_version( + model_cls: type[Model], + entity_uuid: UUID, + version_uuid: UUID, + *, + entity: Any | None = None, +) -> dict[str, Any] | None: + """Return the entity's state at the specified version as a dict. + + Read-only — nothing in the live database is modified. The returned + shape is intended to mirror a regular single-entity GET response + (scalar columns plus restored ``columns`` / ``metrics`` lists for + ``SqlaTable``), with a ``_version`` key holding the version-level + metadata (uuid, transaction_id, operation_type, issued_at, + changed_by) so callers can tell which version they're looking at. + + Returns ``None`` when either *entity_uuid* or *version_uuid* does not + match — callers should translate to 404. + + Pass *entity* to skip the ``find_active_by_uuid`` lookup; see + :func:`list_versions` for the rationale. The same *entity* is threaded + into :func:`resolve_version_uuid` to eliminate a second redundant + lookup on the same request. + """ + # pylint: disable=import-outside-toplevel + from superset.connectors.sqla.models import SqlaTable + + if entity is None: + entity = find_active_by_uuid(model_cls, entity_uuid) + if entity is None: + return None + + version_num = resolve_version_uuid( + model_cls, entity_uuid, version_uuid, entity=entity + ) + if version_num is None: + return None + + ver_tbl, tx_tbl, user_tbl = _resolve_version_tables(model_cls) + stmt = ( + sa.select( + ver_tbl, + tx_tbl.c.issued_at, + *_user_select_cols(user_tbl), + ) + .select_from(_version_with_tx_user_join(ver_tbl, tx_tbl, user_tbl)) + .where(ver_tbl.c.id == entity.id) + .order_by(*_baseline_first_ordering(ver_tbl)) + .offset(version_num) + .limit(1) + ) + row = db.session.execute(stmt).mappings().first() + if row is None: + return None + + # Project the entity's own scalar fields, skipping versioning + # metadata columns. + result: dict[str, Any] = {} + for col in ver_tbl.columns: + if col.name in CONTINUUM_BOOKKEEPING_COLUMNS: + continue + value = row[col.name] + # uuid columns come back as UUID instances; make them JSON-safe. + if isinstance(value, UUID): + value = str(value) + result[col.name] = value + + changes: list[dict[str, Any]] = [] + if (entity_kind := _entity_kind_for(model_cls)) is not None: + changes = list_change_records_batch( + entity_kind, entity.id, [row["transaction_id"]] + ).get(row["transaction_id"], []) + + result["_version"] = { + "version_uuid": str(version_uuid), + "version_number": version_num, + "transaction_id": row["transaction_id"], + "operation_type": _OP_TYPE_LABELS.get( + row["operation_type"], str(row["operation_type"]) + ), + "issued_at": row["issued_at"], + "changed_by": _changed_by_from_row(row), + "changes": changes, + } + + # For datasets, attach the columns/metrics as they were at this + # transaction by reading from Continuum's child shadow tables + # (``table_columns_version`` / ``sql_metrics_version``). Empty lists + # when the dataset had no children at this tx. + if model_cls is SqlaTable: + # pylint: disable=import-outside-toplevel + from superset.connectors.sqla.models import SqlMetric, TableColumn + from superset.versioning.changes import shadow_rows_valid_at + + target_tx = row["transaction_id"] + cols_tbl = version_class(TableColumn).__table__ + metrics_tbl = version_class(SqlMetric).__table__ + result["columns"] = shadow_rows_valid_at( + db.session, cols_tbl, "table_id", entity.id, target_tx + ) + result["metrics"] = shadow_rows_valid_at( + db.session, metrics_tbl, "table_id", entity.id, target_tx + ) + + return result diff --git a/superset/versioning/schemas.py b/superset/versioning/schemas.py new file mode 100644 index 000000000000..9fa51a8432e7 --- /dev/null +++ b/superset/versioning/schemas.py @@ -0,0 +1,140 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Shared Marshmallow schemas for entity version history endpoints. + +Consumed by ChartRestApi, DashboardRestApi, and DatasetRestApi — the response +shape is identical across all three resources, so the schemas live here to +avoid triplicated definitions. +""" + +from __future__ import annotations + +from marshmallow import fields, Schema + + +class VersionChangedBySchema(Schema): + """Subset of the User model included in each version history entry.""" + + id = fields.Integer() + username = fields.String() + first_name = fields.String() + last_name = fields.String() + + +class VersionChangeRecordSchema(Schema): + """One field-level diff hunk from ``version_changes``. + + The frontend renders human-readable prose from (``kind``, + ``from_value``, ``to_value``) via Flask-Babel. Server-side the + shape is deliberately machine-readable only — see spec FR-019. + """ + + kind = fields.String( + metadata={ + "description": ( + "Semantic category of the change. First-class values in V1: " + "'filter', 'metric', 'dimension', 'column', 'chart', " + "'time_range', 'color_palette'. Falls back to 'field' for " + "generic scalar changes that don't map to a named kind." + ) + }, + ) + path = fields.Raw( + metadata={ + "description": ( + "Array of segments locating the change in the entity's state. " + "Example: ['params', 'adhoc_filters', 'country']." + ) + }, + ) + from_value = fields.Raw( + allow_none=True, + metadata={ + "description": ( + "Value at path before the save; null when the field did not exist." + ), + }, + ) + to_value = fields.Raw( + allow_none=True, + metadata={ + "description": ( + "Value at path after the save; null when the field was removed." + ), + }, + ) + + +class VersionListItemSchema(Schema): + """A single version row in the version history response.""" + + version_uuid = fields.UUID( + metadata={ + "description": ( + "Deterministic UUIDv5 derived from the entity UUID and the " + "Continuum transaction id — stable across replicas and " + "retention pruning. The handle accepted by the get/restore " + "version endpoints." + ) + }, + ) + version_number = fields.Integer( + metadata={"description": "0-based position in the history, oldest first"}, + ) + transaction_id = fields.Integer( + metadata={"description": "Underlying Continuum transaction id"}, + ) + operation_type = fields.String( + metadata={ + "description": ( + "One of 'baseline', 'update', or 'delete', derived from the " + "Continuum integer constant. Restore is not a distinct " + "operation_type: a restore surfaces as an ordinary 'update' " + "transaction." + ) + }, + ) + issued_at = fields.DateTime( + metadata={"description": "UTC timestamp of the commit that produced the row"}, + ) + changed_by = fields.Nested( + VersionChangedBySchema, + allow_none=True, + metadata={ + "description": ( + "User who produced the version, or null when the commit had no " + "authenticated Flask user (CLI, Celery, import)." + ) + }, + ) + changes = fields.List( + fields.Nested(VersionChangeRecordSchema), + metadata={ + "description": ( + "Structured diff records describing the atomic field-level " + "changes at this version, ordered by emission sequence. " + "Empty for baseline (op=0) transactions per spec M4." + ) + }, + ) + + +class VersionListResponseSchema(Schema): + """Envelope for version list responses.""" + + result = fields.List(fields.Nested(VersionListItemSchema)) + count = fields.Integer() diff --git a/superset/versioning/utils.py b/superset/versioning/utils.py new file mode 100644 index 000000000000..e09f133bf1dd --- /dev/null +++ b/superset/versioning/utils.py @@ -0,0 +1,81 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Shared session helpers used by the entity-versioning machinery.""" + +from __future__ import annotations + +from collections.abc import Iterator +from contextlib import contextmanager +from typing import Any + +import sqlalchemy as sa +from sqlalchemy.orm import Session + + +@contextmanager +def single_flush_scope(session: Session) -> Iterator[None]: + """Suppress autoflushes inside the block, flush once on clean exit. + + Intended for operations that (a) make multiple mutations across + relationships and (b) issue intermediate queries which would + otherwise autoflush. Iterating from one relationship to another + inside SQLAlchemy-Continuum's ``Reverter`` is the canonical case: + a mid-iteration autoflush transitions pending DELETEs to + ``state.deleted=True``, and the subsequent + ``session.add(version_parent)`` cascade walk trips on the + deleted-state instances with ``InvalidRequestError``. Wrapping the + whole revert keeps marked-for-deletion instances in + ``state.persistent`` until the trailing flush drains DELETEs + + INSERTs in one atomic step. That single flush is also load-bearing + for the ``after_flush`` change-records listener — splitting the + work across multiple flushes would split it across multiple + Continuum transactions, and the listener's tx-dedup guard would + silently drop the second pass's records. + + On exception, the trailing flush is skipped — the session's normal + rollback flow handles cleanup, and flushing a partially-mutated + state would be wrong. + """ + with session.no_autoflush: + yield + session.flush() + + +def read_row_outside_flush( + session: Session, table: sa.Table, entity_id: int +) -> dict[str, Any] | None: + """Read the row with ``id == entity_id`` from *table* without triggering + an autoflush. Returns the row as a plain dict, or ``None`` when no row + matches. + + The companion read primitive to :func:`single_flush_scope`. Listeners + that need pre-flush state (the row as it existed *before* the in-flight + edit was staged) use this — without ``no_autoflush``, the + ``session.connection().execute(...)`` would itself trigger a flush of + the pending edit, leaving "pre" and "post" indistinguishable. + + Returns ``dict[str, Any]`` rather than ``RowMapping`` so callers don't + accidentally hold a cursor-bound object past the listener boundary. + """ + with session.no_autoflush: + result = ( + session.connection() + .execute(sa.select(table).where(table.c.id == entity_id)) + .mappings() + .one_or_none() + ) + return dict(result) if result else None diff --git a/tests/integration_tests/versioning/__init__.py b/tests/integration_tests/versioning/__init__.py new file mode 100644 index 000000000000..13a83393a912 --- /dev/null +++ b/tests/integration_tests/versioning/__init__.py @@ -0,0 +1,16 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. diff --git a/tests/integration_tests/versioning/capture_disabled_tests.py b/tests/integration_tests/versioning/capture_disabled_tests.py new file mode 100644 index 000000000000..fc301a860b8f --- /dev/null +++ b/tests/integration_tests/versioning/capture_disabled_tests.py @@ -0,0 +1,143 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Behavioral proof for the versioning kill-switch / dark-launch contract. + +``ENABLE_VERSIONING_CAPTURE=False`` MUST cause a real save to write +**zero** ``version_transaction`` rows and **zero** ``*_version`` shadow +rows — Continuum is wired at import (``make_versioned()``), so suppressing +only the custom listeners would leave Continuum's own listeners minting +empty transaction rows on every flush. ``init_versioning`` detaches those +via ``_remove_continuum_write_listeners()``; this test pins that the +*behavioral* result is genuinely nothing-written (the structural unit test +in ``initialization_test.py`` mocks the detach; this exercises it for real). + +This is the acceptance gate for shipping versioning dark in the +base-infra rollout PR. +""" + +from __future__ import annotations + +import pytest +import sqlalchemy as sa +from sqlalchemy.engine import Engine +from sqlalchemy.orm import Mapper, Session +from sqlalchemy_continuum import version_class, versioning_manager + +from superset.extensions import db +from superset.initialization import SupersetAppInitializer +from superset.models.slice import Slice +from tests.integration_tests.base_tests import SupersetTestCase +from tests.integration_tests.constants import ADMIN_USERNAME +from tests.integration_tests.fixtures.birth_names_dashboard import ( # noqa: F401 + load_birth_names_dashboard_with_slices, + load_birth_names_data, +) + + +def _transaction_row_count() -> int: + """Total rows in the shared ``version_transaction`` table.""" + return db.session.query(versioning_manager.transaction_cls).count() + + +def _slice_version_count(slice_id: int) -> int: + ver_cls = version_class(Slice) + return db.session.query(ver_cls).filter(ver_cls.id == slice_id).count() + + +def _reattach_continuum_write_listeners() -> None: + """Inverse of ``init_versioning._remove_continuum_write_listeners`` so this + test restores process-global SQLAlchemy event state for the rest of the + suite. Idempotent on a representative listener.""" + if sa.event.contains(Mapper, "after_insert", versioning_manager.track_inserts): + return # already attached + versioning_manager.track_operations(Mapper) + versioning_manager.track_session(Session) + sa.event.listen( + Engine, "before_execute", versioning_manager.track_association_operations + ) + sa.event.listen(Engine, "rollback", versioning_manager.clear_connection) + sa.event.listen( + Engine, + "set_connection_execution_options", + versioning_manager.track_cloned_connections, + ) + + +class TestVersioningCaptureDisabled(SupersetTestCase): + @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices") + def test_capture_off_writes_no_version_or_transaction_rows(self) -> None: + """With Continuum's write listeners detached (the capture-off path), + a real content change MUST write neither a shadow row nor a + ``version_transaction`` row.""" + db.session.commit() + chart = db.session.query(Slice).filter(Slice.slice_name == "Girls").first() + assert chart is not None + chart_id = chart.id + + self.login(ADMIN_USERNAME) + + # Simulate the ENABLE_VERSIONING_CAPTURE=False branch of init_versioning. + SupersetAppInitializer._remove_continuum_write_listeners() + try: + tx_before = _transaction_row_count() + ver_before = _slice_version_count(chart_id) + + rv = self.client.put( + f"/api/v1/chart/{chart_id}", + json={"slice_name": "capture-off-renamed"}, + ) + assert rv.status_code == 200, rv.data + db.session.expire_all() + + assert _transaction_row_count() == tx_before, ( + "capture off MUST write zero version_transaction rows " + f"(before={tx_before}, after={_transaction_row_count()})" + ) + assert _slice_version_count(chart_id) == ver_before, ( + "capture off MUST write zero shadow rows " + f"(before={ver_before}, after={_slice_version_count(chart_id)})" + ) + finally: + # Restore the chart and re-attach Continuum so the rest of the + # suite runs with capture on. + self.client.put(f"/api/v1/chart/{chart_id}", json={"slice_name": "Girls"}) + _reattach_continuum_write_listeners() + + @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices") + def test_control_capture_on_does_write_a_version_row(self) -> None: + """Control: with capture on (the suite default), the same edit DOES + mint a shadow + transaction row — proves the disabled-path assertion + above is not vacuously true.""" + db.session.commit() + chart = db.session.query(Slice).filter(Slice.slice_name == "Boys").first() + if chart is None: # birth_names fixture not loaded for this test + pytest.skip("Boys slice not present") + chart_id = chart.id + + self.login(ADMIN_USERNAME) + _reattach_continuum_write_listeners() # ensure attached + ver_before = _slice_version_count(chart_id) + try: + rv = self.client.put( + f"/api/v1/chart/{chart_id}", + json={"slice_name": "capture-on-renamed"}, + ) + assert rv.status_code == 200, rv.data + db.session.expire_all() + assert _slice_version_count(chart_id) == ver_before + 1 + finally: + self.client.put(f"/api/v1/chart/{chart_id}", json={"slice_name": "Boys"}) From fd7bac0b24c3c3c0c0b2935c325347cdd8009237 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Wed, 17 Jun 2026 16:48:01 -0600 Subject: [PATCH 25/35] fix(versioning): gate write-path bookkeeping and harden capture Review hardening for the dark base-infra landing. The gate must make the infrastructure truly inert when off, and capture must basically work when on. Gate / dark-contract: - init_versioning's flag fallback now defaults False (was True), so any app-factory path that doesn't load config.py stays inert instead of silently enabling capture; config + docstrings reconciled to "ships off". - The chart/dashboard/dataset PUT and GET endpoints no longer run version bookkeeping queries unconditionally: the lookups move behind gated helpers (current_entity_version_info / current_entity_etag_uuid) that issue zero queries when capture is off, so the kill-switch covers the full save path. - _remove_continuum_write_listeners also flips Continuum's master option off and verifies the detach, making options['versioning'] a single switch that silences Continuum's listeners and the custom baseline writer alike. Robustness when on: - The baseline before_flush body is wrapped so an infra error logs and the user's save proceeds rather than aborting the transaction. - The baseline listener honors the master switch (it mints its own transaction row via direct SQL, so it can't rely on Continuum being detached); the change-record listener already self-gates on the absence of a Continuum transaction. - The baseline shadow writer logs when a content column is dropped for a live/shadow name divergence instead of silently storing NULL. Tests: - Structural unit tests drive init_versioning's config branch (off, absent, on) so a gate/default inversion is caught without a DB. - The capture-on control now asserts a version_changes row, exercising the full pipeline; the integration test config enables capture so the suite runs it (production still ships off). - New guard test fails if a version snapshot would expose a sensitive column. Co-Authored-By: Claude Opus 4.8 (1M context) --- superset/charts/api.py | 40 ++++------- superset/config.py | 23 +++--- superset/daos/version.py | 13 ++-- superset/dashboards/api.py | 41 ++++------- superset/datasets/api.py | 43 ++++------- superset/initialization/__init__.py | 46 +++++++++--- superset/versioning/api_helpers.py | 72 ++++++++++++++++++- superset/versioning/baseline/listener.py | 48 +++++++++---- superset/versioning/baseline/shadow.py | 20 ++++++ superset/versioning/etag.py | 7 +- .../integration_tests/superset_test_config.py | 8 +++ .../versioning/capture_disabled_tests.py | 47 +++++++++--- .../versioning/snapshot_projection_tests.py | 70 ++++++++++++++++++ tests/unit_tests/initialization_test.py | 62 ++++++++++++++++ 14 files changed, 407 insertions(+), 133 deletions(-) create mode 100644 tests/integration_tests/versioning/snapshot_projection_tests.py diff --git a/superset/charts/api.py b/superset/charts/api.py index 85f509516ac3..7402113d40a7 100644 --- a/superset/charts/api.py +++ b/superset/charts/api.py @@ -81,7 +81,6 @@ from superset.commands.importers.v1.utils import get_contents_from_bundle from superset.constants import MODEL_API_RW_METHOD_PERMISSION_MAP, RouteMethod from superset.daos.chart import ChartDAO -from superset.daos.version import VersionDAO from superset.exceptions import ( ScreenshotImageNotAvailableException, ) @@ -99,6 +98,8 @@ ) from superset.utils.urls import get_url_path from superset.versioning.api_helpers import ( + current_entity_etag_uuid, + current_entity_version_info, get_version_endpoint, list_versions_endpoint, ) @@ -326,7 +327,7 @@ def get(self, id_or_uuid: str) -> Response: return set_version_etag( self.response(200, result=result), - VersionDAO.current_live_version_uuid(Slice, dash.id, dash.uuid), + current_entity_etag_uuid(Slice, dash.id, dash.uuid), ) except ChartNotFoundError: return self.response_404() @@ -481,39 +482,28 @@ def put(self, pk: int) -> Response: except ValidationError as error: return self.response_400(message=error.messages) - # pylint: disable=import-outside-toplevel - from superset.extensions import db as _db - - pre_chart = _db.session.query(Slice).filter(Slice.id == pk).one_or_none() - old_version = VersionDAO.current_version_number(Slice, pk) - old_transaction_id = VersionDAO.current_live_transaction_id(Slice, pk) - old_version_uuid = ( - VersionDAO.current_live_version_uuid(Slice, pk, pre_chart.uuid) - if pre_chart is not None - else None - ) + # Live version identifiers before the update (empty + query-free when + # ``ENABLE_VERSIONING_CAPTURE`` is off, so this stays inert under the + # kill-switch). + old_info = current_entity_version_info(Slice, pk) try: changed_model = UpdateChartCommand(pk, item).run() - new_version = VersionDAO.current_version_number(Slice, changed_model.id) - new_transaction_id = VersionDAO.current_live_transaction_id( - Slice, changed_model.id - ) - new_version_uuid = VersionDAO.current_live_version_uuid( + new_info = current_entity_version_info( Slice, changed_model.id, changed_model.uuid ) response = self.response( 200, id=changed_model.id, result=item, - old_version=old_version, - new_version=new_version, - old_transaction_id=old_transaction_id, - new_transaction_id=new_transaction_id, - old_version_uuid=str(old_version_uuid) if old_version_uuid else None, - new_version_uuid=str(new_version_uuid) if new_version_uuid else None, + old_version=old_info.version, + new_version=new_info.version, + old_transaction_id=old_info.transaction_id, + new_transaction_id=new_info.transaction_id, + old_version_uuid=old_info.version_uuid, + new_version_uuid=new_info.version_uuid, ) - set_version_etag(response, new_version_uuid) + set_version_etag(response, new_info.version_uuid) except ChartNotFoundError: response = self.response_404() except ChartForbiddenError: diff --git a/superset/config.py b/superset/config.py index 356fb8c41a2e..4e69ee7786b5 100644 --- a/superset/config.py +++ b/superset/config.py @@ -1431,18 +1431,17 @@ class D3TimeFormat(TypedDict, total=False): # The limit for the Superset Meta DB when the feature flag ENABLE_SUPERSET_META_DB is on SUPERSET_META_DB_LIMIT: int | None = 1000 -# Master switch for entity-version-history capture. Default ``True`` — -# every save of a chart, dashboard, or dataset writes shadow rows + a -# ``version_changes`` record. Set to ``False`` in ``superset_config.py`` -# (or via the env var of the same name) to disable the two before-flush -# listeners that drive capture; existing shadow tables stay intact and -# the /versions/ + /activity/ endpoints continue to work read-only. -# This is an operational escape hatch — for use when a versioning-induced -# regression needs a 30-second recovery instead of revert-and-redeploy — -# not a feature flag. It ships defaulted OFF so the versioning infrastructure -# (schema + Continuum wiring) lands inert; capture is activated by flipping -# the default to on once validated in production. It then remains as the -# permanent kill-switch. +# Master switch for entity-version-history capture. Ships defaulted ``False`` +# so the versioning infrastructure (schema + Continuum wiring) lands inert: +# no save writes shadow rows or a ``version_transaction``/``version_changes`` +# record, while the /versions/ endpoints stay available read-only (returning +# empty). Set to ``True`` in ``superset_config.py`` (or via the env var of the +# same name) to enable the before-flush listeners that drive capture. +# Capture is activated by flipping this default to on once validated in +# production. It is an operational escape hatch — for use when a +# versioning-induced regression needs a 30-second recovery instead of +# revert-and-redeploy — not a feature flag, and remains as the permanent +# kill-switch. ENABLE_VERSIONING_CAPTURE: bool = utils.parse_boolean_string( os.environ.get("ENABLE_VERSIONING_CAPTURE", "false") ) diff --git a/superset/daos/version.py b/superset/daos/version.py index 91a174439e72..f73c7b4e347d 100644 --- a/superset/daos/version.py +++ b/superset/daos/version.py @@ -17,11 +17,11 @@ """Backward-compat façade for the entity-versioning DAO surface. The actual implementation lives in :mod:`superset.versioning.queries` -(read side: list/get/resolve/find/UUID derivation) and -:mod:`superset.versioning.restore` (write side: restore + audit -stamping). This module re-exports both under a single ``VersionDAO`` -class plus the module-level UUID helpers so existing callers keep -working without changes. +(read side: list/get/resolve/find/UUID derivation). This module +re-exports it under a single ``VersionDAO`` class plus the module-level +UUID helpers so existing callers keep working without changes. (The +write side — restore + audit stamping — ships in a later PR; only the +read surface is wired here.) New code should import from the versioning sub-modules directly. """ @@ -51,8 +51,7 @@ class plus the module-level UUID helpers so existing callers keep class VersionDAO: - """Thin façade over :mod:`superset.versioning.queries` and - :mod:`superset.versioning.restore`. + """Thin façade over :mod:`superset.versioning.queries`. Preserved as a single namespace for ergonomic access from API handlers and command classes; the underlying functions are diff --git a/superset/dashboards/api.py b/superset/dashboards/api.py index 0284f69c09ab..4c7d5eb654b6 100644 --- a/superset/dashboards/api.py +++ b/superset/dashboards/api.py @@ -84,7 +84,6 @@ from superset.commands.importers.v1.utils import get_contents_from_bundle from superset.constants import MODEL_API_RW_METHOD_PERMISSION_MAP, RouteMethod from superset.daos.dashboard import DashboardDAO, EmbeddedDashboardDAO -from superset.daos.version import VersionDAO from superset.dashboards.filters import ( DashboardAccessFilter, DashboardCertifiedFilter, @@ -143,6 +142,8 @@ ) from superset.utils.urls import get_url_path from superset.versioning.api_helpers import ( + current_entity_etag_uuid, + current_entity_version_info, get_version_endpoint, list_versions_endpoint, ) @@ -540,7 +541,7 @@ def get( ) return set_version_etag( self.response(200, result=result), - VersionDAO.current_live_version_uuid(Dashboard, dash.id, dash.uuid), + current_entity_etag_uuid(Dashboard, dash.id, dash.uuid), ) @expose("//datasets", methods=("GET",)) @@ -872,30 +873,16 @@ def put(self, pk: int) -> Response: except ValidationError as error: return self.response_400(message=error.messages) - # pylint: disable=import-outside-toplevel - from superset.extensions import db as _db - - pre_dashboard = ( - _db.session.query(Dashboard).filter(Dashboard.id == pk).one_or_none() - ) - old_version = VersionDAO.current_version_number(Dashboard, pk) - old_transaction_id = VersionDAO.current_live_transaction_id(Dashboard, pk) - old_version_uuid = ( - VersionDAO.current_live_version_uuid(Dashboard, pk, pre_dashboard.uuid) - if pre_dashboard is not None - else None - ) + # Live version identifiers before the update (empty + query-free when + # ``ENABLE_VERSIONING_CAPTURE`` is off). + old_info = current_entity_version_info(Dashboard, pk) try: changed_model = UpdateDashboardCommand(pk, item).run() last_modified_time = changed_model.changed_on.replace( microsecond=0 ).timestamp() - new_version = VersionDAO.current_version_number(Dashboard, changed_model.id) - new_transaction_id = VersionDAO.current_live_transaction_id( - Dashboard, changed_model.id - ) - new_version_uuid = VersionDAO.current_live_version_uuid( + new_info = current_entity_version_info( Dashboard, changed_model.id, changed_model.uuid ) response = self.response( @@ -903,14 +890,14 @@ def put(self, pk: int) -> Response: id=changed_model.id, result=item, last_modified_time=last_modified_time, - old_version=old_version, - new_version=new_version, - old_transaction_id=old_transaction_id, - new_transaction_id=new_transaction_id, - old_version_uuid=str(old_version_uuid) if old_version_uuid else None, - new_version_uuid=str(new_version_uuid) if new_version_uuid else None, + old_version=old_info.version, + new_version=new_info.version, + old_transaction_id=old_info.transaction_id, + new_transaction_id=new_info.transaction_id, + old_version_uuid=old_info.version_uuid, + new_version_uuid=new_info.version_uuid, ) - set_version_etag(response, new_version_uuid) + set_version_etag(response, new_info.version_uuid) except DashboardNotFoundError: response = self.response_404() except DashboardForbiddenError: diff --git a/superset/datasets/api.py b/superset/datasets/api.py index ac4aaae7d6cd..6e084ad9836e 100644 --- a/superset/datasets/api.py +++ b/superset/datasets/api.py @@ -58,7 +58,6 @@ from superset.constants import MODEL_API_RW_METHOD_PERMISSION_MAP, RouteMethod from superset.daos.dashboard import DashboardDAO from superset.daos.dataset import DatasetDAO -from superset.daos.version import VersionDAO from superset.databases.filters import DatabaseFilter from superset.datasets.filters import DatasetCertifiedFilter, DatasetIsNullOrEmptyFilter from superset.datasets.schemas import ( @@ -83,6 +82,8 @@ from superset.utils import json from superset.utils.core import parse_boolean_string, sanitize_cookie_token from superset.versioning.api_helpers import ( + current_entity_etag_uuid, + current_entity_version_info, get_version_endpoint, list_versions_endpoint, ) @@ -488,19 +489,9 @@ def put(self, pk: int) -> Response: except ValidationError as error: return self.response_400(message=error.messages) - # pylint: disable=import-outside-toplevel - from superset.extensions import db as _db - - pre_dataset = ( - _db.session.query(SqlaTable).filter(SqlaTable.id == pk).one_or_none() - ) - old_version = VersionDAO.current_version_number(SqlaTable, pk) - old_transaction_id = VersionDAO.current_live_transaction_id(SqlaTable, pk) - old_version_uuid = ( - VersionDAO.current_live_version_uuid(SqlaTable, pk, pre_dataset.uuid) - if pre_dataset is not None - else None - ) + # Live version identifiers before the update (empty + query-free when + # ``ENABLE_VERSIONING_CAPTURE`` is off). + old_info = current_entity_version_info(SqlaTable, pk) try: # Two commands, two commits, two Continuum transactions for an @@ -522,32 +513,28 @@ def put(self, pk: int) -> Response: # RefreshDatasetCommand commits its own transaction, so reading # afterwards would attribute the refresh's version to the # user's update (and old→new would span two transactions). - new_version = VersionDAO.current_version_number(SqlaTable, changed_model.id) - new_transaction_id = VersionDAO.current_live_transaction_id( - SqlaTable, changed_model.id - ) - new_version_uuid = VersionDAO.current_live_version_uuid( + new_info = current_entity_version_info( SqlaTable, changed_model.id, changed_model.uuid ) - etag_version_uuid = new_version_uuid + etag_version_uuid = new_info.version_uuid if override_columns: RefreshDatasetCommand(pk).run() # The ETag must reflect the entity's *current live* version, # which after the refresh is the refresh's transaction — # re-read it rather than reusing the pre-refresh uuid. - etag_version_uuid = VersionDAO.current_live_version_uuid( + etag_version_uuid = current_entity_etag_uuid( SqlaTable, changed_model.id, changed_model.uuid ) response = self.response( 200, id=changed_model.id, result=item, - old_version=old_version, - new_version=new_version, - old_transaction_id=old_transaction_id, - new_transaction_id=new_transaction_id, - old_version_uuid=str(old_version_uuid) if old_version_uuid else None, - new_version_uuid=str(new_version_uuid) if new_version_uuid else None, + old_version=old_info.version, + new_version=new_info.version, + old_transaction_id=old_info.transaction_id, + new_transaction_id=new_info.transaction_id, + old_version_uuid=old_info.version_uuid, + new_version_uuid=new_info.version_uuid, ) set_version_etag(response, etag_version_uuid) except DatasetNotFoundError: @@ -1417,7 +1404,7 @@ def get(self, id_or_uuid: str, **kwargs: Any) -> Response: return set_version_etag( self.response(200, **response), - VersionDAO.current_live_version_uuid(SqlaTable, table.id, table.uuid), + current_entity_etag_uuid(SqlaTable, table.id, table.uuid), ) @expose("//drill_info/", methods=("GET",)) diff --git a/superset/initialization/__init__.py b/superset/initialization/__init__.py index a9e8269bcd24..e939298d6e9a 100644 --- a/superset/initialization/__init__.py +++ b/superset/initialization/__init__.py @@ -660,23 +660,51 @@ def _remove_continuum_write_listeners() -> None: versioning_manager.track_cloned_connections, ) + # Belt-and-suspenders: flip Continuum's master option off as well. + # Every write listener checks ``manager.options['versioning']`` before + # doing work (manager.py / unit_of_work.py), so if a future Continuum + # version registers an additional write listener this detach does not + # know to remove, that listener still no-ops. ``version_class()`` reads + # from ``version_class_map`` and ignores this option, so the read-only + # ``/versions/`` endpoints are unaffected. + versioning_manager.options["versioning"] = False + + # Verify the known write listeners are actually gone. A Continuum + # upgrade that renamed a handler would make the removals above silently + # miss, leaving capture half-on while we report "disabled"; surface + # that rather than booting in a contradictory state. + if sa.event.contains( + sa.orm.Mapper, "after_insert", versioning_manager.track_inserts + ): + logger.warning( + "versioning: Continuum write listeners still attached after " + "detach; capture may not be fully disabled. This usually means " + "the pinned sqlalchemy-continuum version changed how it " + "registers listeners." + ) + def init_versioning(self) -> None: """Register SQLAlchemy-Continuum baseline and retention listeners. Must be called after all versioned model classes have been imported so that VERSIONED_MODELS can be populated and configure_mappers() has run. - ``ENABLE_VERSIONING_CAPTURE`` (default ``True``) gates the two + ``ENABLE_VERSIONING_CAPTURE`` (ships default ``False``) gates the two before-flush listener registrations. The flag is operational, not - feature: every deployment captures version history by default. The - switch exists so an operator who observes a versioning-induced - regression (e.g. a save-path slowdown attributable to the - change-record listener) can disable capture in - ``superset_config.py`` and restart workers — a 30-second recovery - instead of revert-and-redeploy. Shadow tables already created by - the migration stay; they just stop accumulating new rows. + feature: with it off the infrastructure is inert (no save writes + shadow rows); flipping it on activates capture. The switch also lets + an operator who observes a versioning-induced regression (e.g. a + save-path slowdown attributable to the change-record listener) + disable capture in ``superset_config.py`` and restart workers — a + 30-second recovery instead of revert-and-redeploy. Shadow tables + already created by the migration stay; they just stop accumulating + new rows. + + The fallback here is ``False`` so that any app-factory path that + does not load ``superset.config`` (some test factories, embedded + use) stays inert by default rather than silently enabling capture. """ - if not self.config.get("ENABLE_VERSIONING_CAPTURE", True): + if not self.config.get("ENABLE_VERSIONING_CAPTURE", False): logger.warning( "versioning: ENABLE_VERSIONING_CAPTURE is False; " "skipping baseline + change-record listener registration " diff --git a/superset/versioning/api_helpers.py b/superset/versioning/api_helpers.py index 6c3debe1c2a7..d052f66e8f5d 100644 --- a/superset/versioning/api_helpers.py +++ b/superset/versioning/api_helpers.py @@ -30,15 +30,17 @@ from __future__ import annotations +from dataclasses import dataclass from typing import Any from uuid import UUID -from flask import Response +import sqlalchemy as sa +from flask import current_app, Response from flask_appbuilder import Model from superset.daos.version import VersionDAO from superset.exceptions import SupersetSecurityException -from superset.extensions import security_manager +from superset.extensions import db, security_manager from superset.versioning.etag import set_version_etag_by_uuid from superset.versioning.schemas import VersionListItemSchema @@ -51,6 +53,72 @@ _version_item_schema = VersionListItemSchema() +@dataclass +class EntityVersionInfo: + """Live version identifiers for a write-endpoint response. + + Every field is ``None`` when ``ENABLE_VERSIONING_CAPTURE`` is off — the + write endpoints then issue no version queries at all, so they stay inert + under the kill-switch rather than paying save-path latency the flag is + meant to eliminate. + """ + + version: int | None = None + transaction_id: int | None = None + version_uuid: str | None = None + + +def _capture_enabled() -> bool: + return bool(current_app.config.get("ENABLE_VERSIONING_CAPTURE", False)) + + +def current_entity_version_info( + model_cls: type[Model], + entity_id: int | None, + entity_uuid: UUID | None = None, +) -> EntityVersionInfo: + """Resolve the live version number, transaction id, and version uuid. + + Returns an empty (all-``None``) record and issues *no* queries when + capture is disabled. When *entity_uuid* is not supplied it is resolved + with a single ``SELECT uuid`` rather than loading the whole entity row. + """ + if entity_id is None or not _capture_enabled(): + return EntityVersionInfo() + if entity_uuid is None: + entity_uuid = db.session.scalar( + sa.select(model_cls.uuid).where(model_cls.id == entity_id) + ) + version_uuid = ( + VersionDAO.current_live_version_uuid(model_cls, entity_id, entity_uuid) + if entity_uuid is not None + else None + ) + return EntityVersionInfo( + version=VersionDAO.current_version_number(model_cls, entity_id), + transaction_id=VersionDAO.current_live_transaction_id(model_cls, entity_id), + version_uuid=str(version_uuid) if version_uuid else None, + ) + + +def current_entity_etag_uuid( + model_cls: type[Model], + entity_id: int | None, + entity_uuid: UUID | None, +) -> str | None: + """Resolve only the live version uuid (for an ETag), gated by capture. + + Returns ``None`` without querying when capture is off or either id is + missing. + """ + if entity_id is None or entity_uuid is None or not _capture_enabled(): + return None + version_uuid = VersionDAO.current_live_version_uuid( + model_cls, entity_id, entity_uuid + ) + return str(version_uuid) if version_uuid else None + + def _resolve_entity( api: Any, model_cls: type[Model], diff --git a/superset/versioning/baseline/listener.py b/superset/versioning/baseline/listener.py index 0b16f7365c6a..8f36844222b1 100644 --- a/superset/versioning/baseline/listener.py +++ b/superset/versioning/baseline/listener.py @@ -33,10 +33,12 @@ from __future__ import annotations +import logging from typing import Any from sqlalchemy import event from sqlalchemy.orm import Session +from sqlalchemy_continuum import versioning_manager from superset.versioning.baseline.collection import ( collect_parents_to_baseline, @@ -47,6 +49,8 @@ from superset.versioning.baseline.dirty import force_parent_dirty_on_child_change from superset.versioning.baseline.insertion import insert_baseline_and_children +logger = logging.getLogger(__name__) + # Sentinel attribute set on the session target after first successful # registration — same pattern as # :mod:`superset.versioning.changes.listener`. Subsequent calls become @@ -79,17 +83,37 @@ def register_baseline_listener() -> None: def capture_baseline(session: Session, flush_context: Any, instances: Any) -> None: if not VERSIONED_MODELS: return - # Make sure a child-only edit promotes the parent to ``session.dirty`` - # before Continuum's before_flush reads the dirty set. - force_parent_dirty_on_child_change(session) - for obj in collect_parents_to_baseline(session).values(): - if type(obj) not in VERSIONED_MODELS: - continue - version_table = version_table_for(obj) - if version_table is None: - continue - count = shadow_row_count(session, obj, version_table) - if count == 0: - insert_baseline_and_children(session, obj, version_table) + # Respect the unified capture master switch. Unlike the change-record + # listener (which self-gates because it needs a Continuum transaction + # id that won't exist when capture is off), the baseline writer mints + # its own ``version_transaction`` row via direct SQL — so without this + # guard a detached/kill-switched session would still write baselines. + # ``_remove_continuum_write_listeners`` flips this option off. + if not versioning_manager.options["versioning"]: + return + try: + # Make sure a child-only edit promotes the parent to + # ``session.dirty`` before Continuum's before_flush reads the + # dirty set. + force_parent_dirty_on_child_change(session) + for obj in collect_parents_to_baseline(session).values(): + if type(obj) not in VERSIONED_MODELS: + continue + version_table = version_table_for(obj) + if version_table is None: + continue + count = shadow_row_count(session, obj, version_table) + if count == 0: + insert_baseline_and_children(session, obj, version_table) + except Exception: # pylint: disable=broad-except + # Versioning must never break a user's save. If baseline capture + # fails (a lazy-load error, a registry gap, an unexpected schema + # state), log it and let the flush proceed uninstrumented rather + # than aborting the user's transaction. + logger.warning( + "versioning: baseline capture failed during before_flush; " + "the save proceeds without a baseline row for this flush.", + exc_info=True, + ) setattr(db.session, _REGISTERED_SENTINEL, True) diff --git a/superset/versioning/baseline/shadow.py b/superset/versioning/baseline/shadow.py index c31a5835bcf7..49d5980467ac 100644 --- a/superset/versioning/baseline/shadow.py +++ b/superset/versioning/baseline/shadow.py @@ -31,10 +31,13 @@ from __future__ import annotations +import logging from typing import Any import sqlalchemy as sa +logger = logging.getLogger(__name__) + # Continuum's per-shadow-row bookkeeping columns. Skipped when copying # content from a live row into a synthetic baseline shadow row; set # explicitly by the baseline writer so the row reads as a freshly-created @@ -61,11 +64,28 @@ def insert_baseline_shadow_row( ``.name`` — a thing Continuum-generated tables occasionally produce. """ col_values: dict[Any, Any] = {} + dropped: list[str] = [] for col in version_table.columns: if col.name in CONTINUUM_BOOKKEEPING_COLUMNS: continue if col.name in source_row: col_values[col] = source_row[col.name] + else: + dropped.append(col.name) + if dropped: + # A content column present on the shadow table but absent from the + # live source row means the two schemas have diverged (a Continuum + # shadow column whose name doesn't match the live column). The value + # would be stored NULL — a silent history-fidelity gap — so surface + # it rather than dropping it quietly. + logger.warning( + "versioning: baseline shadow row for %s is missing source " + "values for column(s) %s; they will be stored NULL. This " + "indicates a name divergence between the live table and its " + "Continuum shadow table.", + version_table.name, + ", ".join(dropped), + ) col_values[version_table.c.transaction_id] = tx_id col_values[version_table.c.end_transaction_id] = None col_values[version_table.c.operation_type] = 0 diff --git a/superset/versioning/etag.py b/superset/versioning/etag.py index 643e1a95dadc..fc51d9357666 100644 --- a/superset/versioning/etag.py +++ b/superset/versioning/etag.py @@ -30,12 +30,15 @@ from flask import Response -def set_version_etag(response: "Response", version_uuid: UUID | None) -> "Response": +def set_version_etag( + response: "Response", version_uuid: UUID | str | None +) -> "Response": """Attach ``ETag: ""`` to *response*. Uses RFC 7232 strong-validator form (no leading ``W/``); the response header value is wrapped in double quotes per the spec. No-op when - *version_uuid* is ``None`` (entity has no version rows yet). + *version_uuid* is ``None`` (entity has no version rows yet). Accepts a + ``UUID`` or a pre-stringified uuid (the write endpoints carry the latter). """ if version_uuid is not None: response.headers["ETag"] = f'"{version_uuid}"' diff --git a/tests/integration_tests/superset_test_config.py b/tests/integration_tests/superset_test_config.py index 56ab8ddd1941..a6dfd4d01a4d 100644 --- a/tests/integration_tests/superset_test_config.py +++ b/tests/integration_tests/superset_test_config.py @@ -89,6 +89,14 @@ def GET_FEATURE_FLAGS_FUNC(ff): # noqa: N802 TALISMAN_ENABLED = False WTF_CSRF_ENABLED = False +# Production ships entity-version capture OFF (see ``config.py``); the test +# suite turns it ON so the capture pipeline (Continuum shadow rows + baseline +# + ``version_changes``) is actually exercised. The dark/kill-switch contract +# is proven separately by +# ``tests/integration_tests/versioning/capture_disabled_tests.py``, which +# detaches the listeners within the test. +ENABLE_VERSIONING_CAPTURE = True + FAB_ROLES = {"TestRole": [["Security", "menu_access"], ["List Users", "menu_access"]]} PUBLIC_ROLE_LIKE = "Gamma" diff --git a/tests/integration_tests/versioning/capture_disabled_tests.py b/tests/integration_tests/versioning/capture_disabled_tests.py index fc301a860b8f..80079979ed7c 100644 --- a/tests/integration_tests/versioning/capture_disabled_tests.py +++ b/tests/integration_tests/versioning/capture_disabled_tests.py @@ -22,8 +22,12 @@ only the custom listeners would leave Continuum's own listeners minting empty transaction rows on every flush. ``init_versioning`` detaches those via ``_remove_continuum_write_listeners()``; this test pins that the -*behavioral* result is genuinely nothing-written (the structural unit test -in ``initialization_test.py`` mocks the detach; this exercises it for real). +*behavioral* result is genuinely nothing-written. The structural unit tests +in ``tests/unit_tests/initialization_test.py`` (``TestInitVersioning``) drive +the config-flag branch of ``init_versioning`` with mocks; this exercises the +detach for real, against a database, and proves a control save under capture +*on* writes both a shadow row and a ``version_changes`` record (so the +zero-rows assertions are not vacuously true). This is the acceptance gate for shipping versioning dark in the base-infra rollout PR. @@ -58,10 +62,25 @@ def _slice_version_count(slice_id: int) -> int: return db.session.query(ver_cls).filter(ver_cls.id == slice_id).count() +def _version_changes_count() -> int: + """Total rows in the ``version_changes`` table — the custom diff records, + distinct from Continuum's shadow rows. Proves the full capture pipeline + (not just Continuum) ran.""" + return ( + db.session.execute(sa.text("SELECT COUNT(*) FROM version_changes")).scalar() + or 0 + ) + + def _reattach_continuum_write_listeners() -> None: """Inverse of ``init_versioning._remove_continuum_write_listeners`` so this - test restores process-global SQLAlchemy event state for the rest of the - suite. Idempotent on a representative listener.""" + test restores process-global capture state for the rest of the suite + (which runs with ``ENABLE_VERSIONING_CAPTURE`` on). Idempotent on a + representative listener. Also restores ``options['versioning']`` — the + detach flips it off (and the baseline listener honors it), so the + re-attach must flip it back on or subsequent saves would silently stop + capturing.""" + versioning_manager.options["versioning"] = True if sa.event.contains(Mapper, "after_insert", versioning_manager.track_inserts): return # already attached versioning_manager.track_operations(Mapper) @@ -118,10 +137,12 @@ def test_capture_off_writes_no_version_or_transaction_rows(self) -> None: _reattach_continuum_write_listeners() @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices") - def test_control_capture_on_does_write_a_version_row(self) -> None: + def test_control_capture_on_writes_version_and_change_rows(self) -> None: """Control: with capture on (the suite default), the same edit DOES - mint a shadow + transaction row — proves the disabled-path assertion - above is not vacuously true.""" + mint a shadow row AND a ``version_changes`` record — proving the + disabled-path assertions are not vacuously true and that the full + capture pipeline (Continuum shadow rows + the custom change-record + listener) runs end-to-end, not just Continuum's own writes.""" db.session.commit() chart = db.session.query(Slice).filter(Slice.slice_name == "Boys").first() if chart is None: # birth_names fixture not loaded for this test @@ -129,8 +150,11 @@ def test_control_capture_on_does_write_a_version_row(self) -> None: chart_id = chart.id self.login(ADMIN_USERNAME) - _reattach_continuum_write_listeners() # ensure attached + _reattach_continuum_write_listeners() # belt-and-suspenders: suite is on + # ``>`` rather than ``== before + 1``: the first edit to a not-yet- + # versioned entity also mints a synthetic baseline shadow row. ver_before = _slice_version_count(chart_id) + changes_before = _version_changes_count() try: rv = self.client.put( f"/api/v1/chart/{chart_id}", @@ -138,6 +162,11 @@ def test_control_capture_on_does_write_a_version_row(self) -> None: ) assert rv.status_code == 200, rv.data db.session.expire_all() - assert _slice_version_count(chart_id) == ver_before + 1 + assert _slice_version_count(chart_id) > ver_before, ( + "capture on MUST write at least one shadow row" + ) + assert _version_changes_count() > changes_before, ( + "capture on MUST write at least one version_changes record" + ) finally: self.client.put(f"/api/v1/chart/{chart_id}", json={"slice_name": "Boys"}) diff --git a/tests/integration_tests/versioning/snapshot_projection_tests.py b/tests/integration_tests/versioning/snapshot_projection_tests.py new file mode 100644 index 000000000000..b7be4ce4c36e --- /dev/null +++ b/tests/integration_tests/versioning/snapshot_projection_tests.py @@ -0,0 +1,70 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Drift guard for the version-snapshot projection. + +``queries.get_version`` projects an entity's historical state by copying +*every* shadow column except Continuum's three bookkeeping columns — a +denylist, not an allowlist. That is safe today (nothing sensitive on the +versioned models survives ``__versioned__['exclude']``), but a future +sensitive column added to a versioned model would be exposed through the +read-only ``/versions/`` endpoint by default unless someone remembers to +exclude it. This test fails when such a column appears, forcing the +exclusion decision to be made consciously. +""" + +from __future__ import annotations + +from sqlalchemy_continuum import version_class + +from superset.connectors.sqla.models import SqlaTable +from superset.models.dashboard import Dashboard +from superset.models.slice import Slice +from superset.versioning.baseline.shadow import CONTINUUM_BOOKKEEPING_COLUMNS +from tests.integration_tests.base_tests import SupersetTestCase + +# Substrings that mark a column as something a version snapshot must never +# echo back. Matched case-insensitively against the projected column names. +_SENSITIVE_SUBSTRINGS = ( + "password", + "secret", + "encrypted", + "private_key", + "api_key", + "access_token", +) + + +class TestSnapshotProjectionSafety(SupersetTestCase): + def test_version_snapshot_exposes_no_sensitive_columns(self) -> None: + """The columns ``get_version`` would project for each versioned model + must contain no sensitive-looking column name.""" + for model_cls in (Slice, Dashboard, SqlaTable): + ver_tbl = version_class(model_cls).__table__ + projected = [ + col.name + for col in ver_tbl.columns + if col.name not in CONTINUUM_BOOKKEEPING_COLUMNS + ] + for name in projected: + lowered = name.lower() + offending = [s for s in _SENSITIVE_SUBSTRINGS if s in lowered] + assert not offending, ( + f"{model_cls.__name__} version snapshot would expose " + f"sensitive-looking column '{name}'. Add it to the model's " + f"__versioned__['exclude'] set, or confirm it is safe and " + f"relax this guard." + ) diff --git a/tests/unit_tests/initialization_test.py b/tests/unit_tests/initialization_test.py index d391ddc6152d..b324b489e7cf 100644 --- a/tests/unit_tests/initialization_test.py +++ b/tests/unit_tests/initialization_test.py @@ -243,6 +243,68 @@ def test_check_and_warn_database_connection_invalid_uri(self) -> None: ) +class TestInitVersioning: + """Structural coverage for the ``ENABLE_VERSIONING_CAPTURE`` gate. + + Drives ``init_versioning`` through the real config branch (rather than + calling the detach helper directly) so a future inversion of the gate — + or of its default — is caught cheaply, without a DB. The behavioral + "zero rows when off" proof lives in + ``tests/integration_tests/versioning/capture_disabled_tests.py``. + """ + + def test_capture_flag_off_detaches_and_skips_registration(self): + """Flag explicitly False → detach Continuum, register nothing.""" + fake = MagicMock() + fake.config = {"ENABLE_VERSIONING_CAPTURE": False} + + with ( + patch( + "superset.versioning.baseline.register_baseline_listener" + ) as reg_baseline, + patch( + "superset.versioning.changes.register_change_record_listener" + ) as reg_changes, + ): + SupersetAppInitializer.init_versioning(fake) + + fake._remove_continuum_write_listeners.assert_called_once() + reg_baseline.assert_not_called() + reg_changes.assert_not_called() + + def test_capture_flag_absent_defaults_to_off(self): + """Flag absent → fallback MUST be off, so any app-factory path that + doesn't load ``superset.config`` stays inert rather than silently + enabling capture.""" + fake = MagicMock() + fake.config = {} + + SupersetAppInitializer.init_versioning(fake) + + fake._remove_continuum_write_listeners.assert_called_once() + + def test_capture_flag_on_registers_listeners_without_detaching(self): + """Flag True → register both before-flush listeners, never detach.""" + fake = MagicMock() + fake.config = {"ENABLE_VERSIONING_CAPTURE": True} + + with ( + patch( + "superset.versioning.baseline.register_baseline_listener" + ) as reg_baseline, + patch("superset.versioning.baseline.VERSIONED_MODELS", []), + patch( + "superset.versioning.changes.register_change_record_listener" + ) as reg_changes, + patch("sqlalchemy_continuum.version_class"), + ): + SupersetAppInitializer.init_versioning(fake) + + reg_baseline.assert_called_once() + reg_changes.assert_called_once() + fake._remove_continuum_write_listeners.assert_not_called() + + class TestCreateAppRoot: """Test app root resolution precedence in create_app.""" From 0d95acb9ba377da04502c31f8dabc074ac62d4c7 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 22 Jun 2026 10:01:44 -0600 Subject: [PATCH 26/35] fix(versioning): make tests capture-on compatible CI surfaced two capture-on test failures (the suite now runs with ENABLE_VERSIONING_CAPTURE on). Neither is a production bug: - The dashboard test factory deleted the dashboard_slices M2M via a Core-level ``dashboard_slices.delete()``. That association table is Continuum-tracked, and a raw Core delete fires Continuum's engine before_execute listener with no unit-of-work registered for the connection (no ORM flush occurred), raising KeyError under capture. Production deletes go through the ORM relationship cascade (which registers a UoW and works); the factory now clears the relationship through the ORM the same way. No production code does raw association deletes. - Three unit tests compared ``model.uuid`` to a string, but the UUIDMixin validator coerces assigned uuids to UUID objects; compare by string form. Co-Authored-By: Claude Opus 4.8 (1M context) --- .../dashboards/superset_factory_util.py | 15 +++++++++++---- .../commands/importers/v1/import_test.py | 6 ++++-- tests/unit_tests/examples/generic_loader_test.py | 4 +++- 3 files changed, 18 insertions(+), 7 deletions(-) diff --git a/tests/integration_tests/dashboards/superset_factory_util.py b/tests/integration_tests/dashboards/superset_factory_util.py index b569bc72d68b..e33689710835 100644 --- a/tests/integration_tests/dashboards/superset_factory_util.py +++ b/tests/integration_tests/dashboards/superset_factory_util.py @@ -25,7 +25,6 @@ from superset.models.core import Database from superset.models.dashboard import ( Dashboard, - dashboard_slices, dashboard_user, DashboardRoles, ) @@ -234,9 +233,17 @@ def delete_dashboard_roles_associations(dashboard: Dashboard) -> None: def delete_dashboard_slices_associations(dashboard: Dashboard) -> None: - db.session.execute( - dashboard_slices.delete().where(dashboard_slices.c.dashboard_id == dashboard.id) - ) + # Clear the M2M through the ORM relationship rather than a Core-level + # ``dashboard_slices.delete()``. ``dashboard_slices`` is a Continuum-tracked + # (versioned) association table: a raw Core delete fires Continuum's engine + # ``before_execute`` listener with no unit-of-work registered for the + # connection (no ORM flush happened), which raises ``KeyError`` when version + # capture is on. Clearing via the relationship routes through the ORM flush — + # the same path production dashboard deletes take — so capture tracks it + # cleanly. The other association helpers stay Core-level: their tables are + # not versioned, so Continuum ignores them. + dashboard.slices = [] + db.session.flush() def delete_all_inserted_slices(): diff --git a/tests/unit_tests/databases/commands/importers/v1/import_test.py b/tests/unit_tests/databases/commands/importers/v1/import_test.py index 1385d8a2bbb6..8f345da6b697 100644 --- a/tests/unit_tests/databases/commands/importers/v1/import_test.py +++ b/tests/unit_tests/databases/commands/importers/v1/import_test.py @@ -56,7 +56,9 @@ def test_import_database(mocker: MockerFixture, session: Session) -> None: assert database.allow_dml is True assert database.allow_file_upload is True assert database.extra == "{}" - assert database.uuid == "b8a1ccd3-779d-4ab7-8ad8-9ab119d7fe89" + # ``uuid`` is coerced to a ``UUID`` object on assignment (UUIDMixin + # validator); compare by string form. + assert str(database.uuid) == "b8a1ccd3-779d-4ab7-8ad8-9ab119d7fe89" assert database.is_managed_externally is False assert database.external_url is None @@ -89,7 +91,7 @@ def test_import_database_no_creds(mocker: MockerFixture, session: Session) -> No assert database.database_name == "imported_database_no_creds" assert database.sqlalchemy_uri == "bigquery://test-db/" assert database.extra == "{}" - assert database.uuid == "2ff17edc-f3fa-4609-a5ac-b484281225bc" + assert str(database.uuid) == "2ff17edc-f3fa-4609-a5ac-b484281225bc" def test_import_database_sqlite_invalid( diff --git a/tests/unit_tests/examples/generic_loader_test.py b/tests/unit_tests/examples/generic_loader_test.py index e921d748f3dc..1551e9b1eb27 100644 --- a/tests/unit_tests/examples/generic_loader_test.py +++ b/tests/unit_tests/examples/generic_loader_test.py @@ -54,7 +54,9 @@ def test_load_parquet_table_sets_uuid_on_new_table(mock_db, mock_get_db): uuid=test_uuid, ) - assert tbl.uuid == test_uuid + # ``uuid`` is coerced to a ``UUID`` object on assignment (UUIDMixin + # validator); compare by string form. + assert str(tbl.uuid) == test_uuid @patch("superset.examples.generic_loader.get_example_database") From 51879d2d13ce52ad3456a6e2932ee04c2c50f9b1 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 22 Jun 2026 10:29:19 -0600 Subject: [PATCH 27/35] test(versioning): clear version tables around each versioning test The integration suite runs with ENABLE_VERSIONING_CAPTURE on, so every save in every test mints Continuum version rows. Add an autouse fixture in the versioning test package that DELETEs the version tables before and after each test: each test gets a clean version-table slate (it sees only the rows it creates) and leaves none behind. Prevents unbounded version-table growth across a run and keeps version-counting assertions robust against unrelated history. DELETE (not TRUNCATE) for SQLite portability; children before version_transaction for FK safety. Co-Authored-By: Claude Opus 4.8 (1M context) --- .../integration_tests/versioning/conftest.py | 69 +++++++++++++++++++ 1 file changed, 69 insertions(+) create mode 100644 tests/integration_tests/versioning/conftest.py diff --git a/tests/integration_tests/versioning/conftest.py b/tests/integration_tests/versioning/conftest.py new file mode 100644 index 000000000000..58835fd52f3b --- /dev/null +++ b/tests/integration_tests/versioning/conftest.py @@ -0,0 +1,69 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Shared fixtures for the entity-versioning integration tests.""" + +from __future__ import annotations + +from collections.abc import Iterator + +import pytest +import sqlalchemy as sa + +from superset import db +from tests.integration_tests.test_app import app + +# Continuum capture tables, ordered children-before-parent for a portable +# ``DELETE`` (the shadow tables and ``version_changes`` all carry an FK to +# ``version_transaction``, so it must go last). ``DELETE`` rather than +# ``TRUNCATE`` so the same statement works on SQLite as well as Postgres/MySQL. +_VERSION_TABLES: tuple[str, ...] = ( + "dashboard_slices_version", + "slices_version", + "dashboards_version", + "tables_version", + "table_columns_version", + "sql_metrics_version", + "version_changes", + "version_transaction", +) + + +def _clear_version_tables() -> None: + with app.app_context(): + for table in _VERSION_TABLES: + db.session.execute(sa.text(f"DELETE FROM {table}")) # noqa: S608 + db.session.commit() + + +@pytest.fixture(autouse=True) +def clean_version_tables() -> Iterator[None]: + """Clear the Continuum version tables around every versioning test. + + The integration suite runs with ``ENABLE_VERSIONING_CAPTURE`` on, so every + save in every test — versioning-related or not — mints version rows. Left + alone, those rows accumulate across the run, and the version-counting + assertions in this package would have to defend against unrelated history. + Clearing before each test gives it a clean version-table slate (it sees + only the rows it creates); clearing after keeps these tests from leaving + version-table residue for whatever runs next. + + Table names are interpolated from the fixed ``_VERSION_TABLES`` tuple, not + from any test input, so the ``DELETE`` is not an injection surface. + """ + _clear_version_tables() + yield + _clear_version_tables() From 9273e18700e6001cba840e29f9142f414892211f Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 22 Jun 2026 10:59:15 -0600 Subject: [PATCH 28/35] fix(versioning): address review findings on the base-infra PR MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - init_versioning: re-assert versioning_manager.options['versioning']=True on the capture-ON path, symmetric with the OFF detach, so a prior OFF app init in the same process can't leave the baseline listener silently no-op. - importers: prime Continuum's unit-of-work before the bulk dashboard_slices insert. dashboard_slices is now a tracked association table, so a raw Core insert can KeyError under capture-on when no UoW is registered for the connection — the same class the test factory hit. No-op when capture is off. - baseline listener: emit a versioning.baseline_capture_error counter on the fail-open before_flush except branch so a persistently-failing baseline path is alertable, not only log-grep-visible. - conftest: the version-table cleanup fixture now asserts its table list covers every *_version shadow table in the DB, so a future versioned model can't silently escape cleanup. - OpenAPI: document old_version_uuid / new_version_uuid on the chart, dashboard, and dataset PUT responses (were returned but undocumented). - capture_disabled test: move the listener detach inside the try so the finally always re-attaches. - UPDATING.md: document the gated-off versioning infra AND the unconditional save/import-path behavior changes (dataset column upsert, reset_ownership, no_autoflush) that ship regardless of the flag. Co-Authored-By: Claude Opus 4.8 (1M context) --- UPDATING.md | 12 ++++++++ superset/charts/api.py | 14 ++++++++++ superset/commands/importers/v1/utils.py | 28 +++++++++++++++++++ superset/dashboards/api.py | 14 ++++++++++ superset/datasets/api.py | 15 ++++++++++ superset/initialization/__init__.py | 11 ++++++++ superset/versioning/baseline/listener.py | 13 +++++++++ .../versioning/capture_disabled_tests.py | 5 +++- .../integration_tests/versioning/conftest.py | 24 ++++++++++++++++ 9 files changed, 135 insertions(+), 1 deletion(-) diff --git a/UPDATING.md b/UPDATING.md index 5daecd4e05f9..433292a87a0a 100644 --- a/UPDATING.md +++ b/UPDATING.md @@ -38,6 +38,18 @@ The `thumbnail_url` field has been removed from `GET /api/v1/dashboard/` list re The thumbnail endpoint redirects to the current digest URL regardless of whether the supplied digest is exact. If the image is not yet cached, that digest URL may return `202` and trigger async generation. Using `changed_on_utc` as the digest is sufficient for cache-busting purposes. +### Entity version-history infrastructure (gated off by default) + +Introduces the schema and SQLAlchemy-Continuum wiring that captures version history for charts, dashboards, and datasets, plus read-only `GET /api/v1/{chart,dashboard,dataset}//versions/` endpoints. This ships **inert**: a new config flag `ENABLE_VERSIONING_CAPTURE` defaults to `False`, so no save writes any version rows and the endpoints return empty. It is an operational kill-switch (a release toggle that becomes a permanent ops switch), not a feature flag — set it to `True` to enable capture once validated. The migration is additive; existing entity `PUT` responses gain `old_version_uuid` / `new_version_uuid` body fields and an `ETag` header (both null/absent when capture is off). + +A few save- and import-path internals change **unconditionally** (independent of the flag), because the versioned mappers must behave correctly whether or not capture is enabled: + +- `DatasetDAO` column/metric updates move from bulk operations to per-row ORM operations, and a metadata refresh now preserves column primary keys via a natural-key (`column_name`) upsert instead of delete-and-reinsert — so charts that reference dataset columns by id keep working across a refresh (previously such references could be invalidated). +- `ImportExportMixin.reset_ownership` stamps the current user onto `created_by`/`changed_by` when a request context is present (previously left null for the column default to fill). +- `UpdateDashboardCommand` runs its body under `no_autoflush`. + +These are behavior changes that take effect on upgrade regardless of `ENABLE_VERSIONING_CAPTURE`; no operator action is required. + ### Webhook alerts/reports block private/internal hosts by default Webhook alert/report dispatch (`WebhookNotification.send`) now validates the target URL's host against the same private/internal-IP block applied to dataset import URLs. If the resolved host is in a loopback, link-local, private (RFC-1918), shared-CGNAT, or multicast range, the webhook is rejected with `NotificationParamException`. diff --git a/superset/charts/api.py b/superset/charts/api.py index 7402113d40a7..97f456abd00c 100644 --- a/superset/charts/api.py +++ b/superset/charts/api.py @@ -463,6 +463,20 @@ def put(self, pk: int) -> Response: row after this update. Differs from old_transaction_id when the update produced a new version row. + old_version_uuid: + type: string + format: uuid + nullable: true + description: Deterministic version_uuid of the live + row before this update. Null when version capture + is disabled or the entity has no version rows yet. + new_version_uuid: + type: string + format: uuid + nullable: true + description: Deterministic version_uuid of the live + row after this update. Null when version capture + is disabled. 400: $ref: '#/components/responses/400' 401: diff --git a/superset/commands/importers/v1/utils.py b/superset/commands/importers/v1/utils.py index 4386ce502a83..3d4f01a4c16f 100644 --- a/superset/commands/importers/v1/utils.py +++ b/superset/commands/importers/v1/utils.py @@ -378,6 +378,7 @@ def safe_insert_dashboard_chart_relationships( # Insert new relationships in bulk, deduplicating to avoid unique constraint issues if unique_new_relationships := set(new_relationships): + _prime_versioning_unit_of_work() db.session.execute( dashboard_slices.insert(), [ @@ -387,6 +388,33 @@ def safe_insert_dashboard_chart_relationships( ) +def _prime_versioning_unit_of_work() -> None: + """Ensure Continuum has a unit-of-work for the current connection. + + ``dashboard_slices`` is a Continuum-tracked (versioned) association + table, so a raw Core INSERT/DELETE on it fires Continuum's engine-level + ``before_execute`` listener, which looks up a unit-of-work for the + connection and raises ``KeyError`` when none is registered (the same + failure class the dashboard test factory hit). The normal import flow + registers one via prior ORM flushes, so this is belt-and-suspenders for + a bulk relationship insert that might run before any flush on the + connection. No-op (the listener is detached) when version capture is + disabled, which is the shipped default; never allowed to break an import. + """ + try: + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import versioning_manager + + if versioning_manager.options.get("versioning"): + versioning_manager.unit_of_work(db.session) + except Exception: # pylint: disable=broad-except + logger.warning( + "versioning: could not prime Continuum unit-of-work before a " + "bulk dashboard_slices insert; proceeding without it.", + exc_info=True, + ) + + def get_resource_mappings_batched( model_class: Type[Any], batch_size: int = 1000, diff --git a/superset/dashboards/api.py b/superset/dashboards/api.py index 4c7d5eb654b6..3aa4a2cc6872 100644 --- a/superset/dashboards/api.py +++ b/superset/dashboards/api.py @@ -854,6 +854,20 @@ def put(self, pk: int) -> Response: row after this update. Differs from old_transaction_id when the update produced a new version row. + old_version_uuid: + type: string + format: uuid + nullable: true + description: Deterministic version_uuid of the live + row before this update. Null when version capture + is disabled or the entity has no version rows yet. + new_version_uuid: + type: string + format: uuid + nullable: true + description: Deterministic version_uuid of the live + row after this update. Null when version capture + is disabled. 400: $ref: '#/components/responses/400' 401: diff --git a/superset/datasets/api.py b/superset/datasets/api.py index 6e084ad9836e..240ae278ba6f 100644 --- a/superset/datasets/api.py +++ b/superset/datasets/api.py @@ -465,6 +465,21 @@ def put(self, pk: int) -> Response: ``old_transaction_id`` the update produced a new version row (regardless of whether ``new_version`` changed). + old_version_uuid: + type: string + format: uuid + nullable: true + description: >- + Deterministic version_uuid of the live row before + this update. Null when version capture is disabled + or the dataset has no version rows yet. + new_version_uuid: + type: string + format: uuid + nullable: true + description: >- + Deterministic version_uuid of the live row after + this update. Null when version capture is disabled. 400: $ref: '#/components/responses/400' 401: diff --git a/superset/initialization/__init__.py b/superset/initialization/__init__.py index e939298d6e9a..a0b0d8ed6562 100644 --- a/superset/initialization/__init__.py +++ b/superset/initialization/__init__.py @@ -715,6 +715,17 @@ def init_versioning(self) -> None: self._remove_continuum_write_listeners() return + # Symmetric with the OFF branch's ``options['versioning'] = False``: + # re-assert it on here so capture is restored even if a prior app + # init in the same process (multi-app / test reentrancy) flipped the + # process-global Continuum option off. Without this, an OFF app + # initialized before an ON app would leave the option False and the + # baseline listener — which gates on it — would silently write no + # baselines despite capture being "enabled". + from sqlalchemy_continuum import versioning_manager + + versioning_manager.options["versioning"] = True + from sqlalchemy.orm import Session # noqa: F401 from sqlalchemy_continuum import version_class diff --git a/superset/versioning/baseline/listener.py b/superset/versioning/baseline/listener.py index 8f36844222b1..6d86e1c0039b 100644 --- a/superset/versioning/baseline/listener.py +++ b/superset/versioning/baseline/listener.py @@ -36,6 +36,7 @@ import logging from typing import Any +from flask import current_app from sqlalchemy import event from sqlalchemy.orm import Session from sqlalchemy_continuum import versioning_manager @@ -51,6 +52,17 @@ logger = logging.getLogger(__name__) + +def _emit_baseline_error_metric() -> None: + """Increment the baseline-capture-error counter so a persistently-failing + baseline path is alertable rather than only visible by log-grep. Guarded: + metric emission must never be what breaks a user's save.""" + try: + current_app.config["STATS_LOGGER"].incr("versioning.baseline_capture_error") + except Exception: # pylint: disable=broad-except # noqa: S110 + pass + + # Sentinel attribute set on the session target after first successful # registration — same pattern as # :mod:`superset.versioning.changes.listener`. Subsequent calls become @@ -115,5 +127,6 @@ def capture_baseline(session: Session, flush_context: Any, instances: Any) -> No "the save proceeds without a baseline row for this flush.", exc_info=True, ) + _emit_baseline_error_metric() setattr(db.session, _REGISTERED_SENTINEL, True) diff --git a/tests/integration_tests/versioning/capture_disabled_tests.py b/tests/integration_tests/versioning/capture_disabled_tests.py index 80079979ed7c..2190733424ea 100644 --- a/tests/integration_tests/versioning/capture_disabled_tests.py +++ b/tests/integration_tests/versioning/capture_disabled_tests.py @@ -110,8 +110,11 @@ def test_capture_off_writes_no_version_or_transaction_rows(self) -> None: self.login(ADMIN_USERNAME) # Simulate the ENABLE_VERSIONING_CAPTURE=False branch of init_versioning. - SupersetAppInitializer._remove_continuum_write_listeners() + # The detach is INSIDE the try so the finally always re-attaches — + # otherwise a failure between detach and the try body would leave the + # rest of the suite running capture-off and silently vacuous. try: + SupersetAppInitializer._remove_continuum_write_listeners() tx_before = _transaction_row_count() ver_before = _slice_version_count(chart_id) diff --git a/tests/integration_tests/versioning/conftest.py b/tests/integration_tests/versioning/conftest.py index 58835fd52f3b..53431a81ec6a 100644 --- a/tests/integration_tests/versioning/conftest.py +++ b/tests/integration_tests/versioning/conftest.py @@ -42,8 +42,32 @@ ) +def _assert_table_list_covers_schema() -> None: + """Guard against drift: if a future versioned model adds a ``*_version`` + shadow table not listed in ``_VERSION_TABLES``, this fixture would stop + fully clearing capture rows and silently reintroduce the accumulation it + exists to prevent. Fail loudly instead. Checked once per test (cheap — + reflection reads cached metadata).""" + known = set(_VERSION_TABLES) + shadow_tables = { + name + for name in sa.inspect(db.engine).get_table_names() + # Continuum shadow tables are ``_version``; exclude + # Alembic's own ``alembic_version`` bookkeeping table, which shares + # the suffix but is not a shadow table. + if name.endswith("_version") and name != "alembic_version" + } + missing = shadow_tables - known + assert not missing, ( + f"_VERSION_TABLES is missing shadow table(s) {sorted(missing)}; add " + "them (children before version_transaction) so the cleanup fixture " + "keeps clearing all capture rows." + ) + + def _clear_version_tables() -> None: with app.app_context(): + _assert_table_list_covers_schema() for table in _VERSION_TABLES: db.session.execute(sa.text(f"DELETE FROM {table}")) # noqa: S608 db.session.commit() From c6b153110745c97d5cde2305e223701313c441bd Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 22 Jun 2026 11:47:58 -0600 Subject: [PATCH 29/35] chore(versioning): strip remaining internal refs + tighten importer guard Follow-ups from the verification pass: - Strip the internal spec/plan/data-model/PR references that the first sweep's regex missed from versioning-core files (diff.py, schemas.py, changes/state.py, changes/listener.py, the migration, the model __versioned__ comments, and the importer meta comment). Pre-existing real apache PR numbers are left as-is. - Mirror Continuum's exact listener condition in _prime_versioning_unit_of_work (versioning OR native_versioning) so the prime can't skip while the engine listener would still act. Co-Authored-By: Claude Opus 4.8 (1M context) --- superset/commands/importers/v1/__init__.py | 2 +- superset/commands/importers/v1/utils.py | 6 +++++- superset/connectors/sqla/models.py | 2 +- ...-05-28_19-50_56cd24c07170_add_versioning_tables.py | 9 ++------- superset/models/slice.py | 2 +- superset/versioning/changes/listener.py | 4 ++-- superset/versioning/changes/state.py | 2 +- superset/versioning/diff.py | 11 +++++------ superset/versioning/schemas.py | 2 +- 9 files changed, 19 insertions(+), 21 deletions(-) diff --git a/superset/commands/importers/v1/__init__.py b/superset/commands/importers/v1/__init__.py index 5cf0f8ade990..0583e9c3b049 100644 --- a/superset/commands/importers/v1/__init__.py +++ b/superset/commands/importers/v1/__init__.py @@ -91,7 +91,7 @@ def run(self) -> None: # for the resulting ``version_transaction`` row and stamps # ``version_transaction.action_kind = 'import'``. Lets operators # explain otherwise-confusing diffs ("Cleared default_filters") - # as "this was an import". See data-model.md §"Three dimensions". + # as "this was an import". # Method-scoped import — defers the versioning bootstrap path # out of this command's module-load graph; see ``changes.py`` # module docstring for the broader init-order rationale. diff --git a/superset/commands/importers/v1/utils.py b/superset/commands/importers/v1/utils.py index 3d4f01a4c16f..229eb8126950 100644 --- a/superset/commands/importers/v1/utils.py +++ b/superset/commands/importers/v1/utils.py @@ -405,7 +405,11 @@ def _prime_versioning_unit_of_work() -> None: # pylint: disable=import-outside-toplevel from sqlalchemy_continuum import versioning_manager - if versioning_manager.options.get("versioning"): + # Mirror the exact condition Continuum's track_association_operations + # listener uses to decide whether it acts (versioning OR + # native_versioning), so the prime can't skip while the listener runs. + options = versioning_manager.options + if options.get("versioning") or options.get("native_versioning"): versioning_manager.unit_of_work(db.session) except Exception: # pylint: disable=broad-except logger.warning( diff --git a/superset/connectors/sqla/models.py b/superset/connectors/sqla/models.py index cdf903473de2..95f0f012b313 100644 --- a/superset/connectors/sqla/models.py +++ b/superset/connectors/sqla/models.py @@ -1351,7 +1351,7 @@ class SqlaTable( # security state, not user-authored content: permission maintenance # rewrites it in bulk, and versioning it produced phantom transactions # flooding the activity stream (one "updated" row per touched entity - # with no user edit — surfaced by the version-history UI, PR #40988). + # with no user edit — surfaced by the version-history UI). # Excluding it also means a restore can't resurrect stale permission # strings; the live, derived values stay authoritative. __versioned__: dict[str, Any] = { diff --git a/superset/migrations/versions/2026-05-28_19-50_56cd24c07170_add_versioning_tables.py b/superset/migrations/versions/2026-05-28_19-50_56cd24c07170_add_versioning_tables.py index 141edafcec4c..ca70beb79997 100644 --- a/superset/migrations/versions/2026-05-28_19-50_56cd24c07170_add_versioning_tables.py +++ b/superset/migrations/versions/2026-05-28_19-50_56cd24c07170_add_versioning_tables.py @@ -65,10 +65,6 @@ UUIDv5 ``version_uuid`` derived from ``entity.uuid`` and ``transaction_id`` for stable external references. -See spec FR-016..FR-021, data-model.md §``version_changes`` / -§"Storage architecture", and the spike notes in -``spike-continuum-restore.md``. - Revision ID: 56cd24c07170 Revises: 2bee73611e32 Create Date: 2026-05-28 19:50:00.000000 @@ -306,8 +302,7 @@ def upgrade() -> None: # # Field-level diff log keyed to a (transaction, entity) pair. Each # row describes one atomic change (one field or one child-collection - # element) that occurred to one entity during a save. See spec - # FR-016..FR-021 and data-model.md §version_changes. + # element) that occurred to one entity during a save. # # ``(entity_kind, entity_id)`` is a polymorphic reference: depending # on ``entity_kind`` (``"chart"`` / ``"dashboard"`` / ``"dataset"``) @@ -509,7 +504,7 @@ def upgrade() -> None: # # If that reshape is removed from the stack, the live table reverts to # carrying its surrogate ``id`` and this migration would need to - # match — see ``spike-continuum-restore.md`` "Branch maintenance". + # match. # ------------------------------------------------------------------ op.create_table( "dashboard_slices_version", diff --git a/superset/models/slice.py b/superset/models/slice.py index 1975ca43c5c6..11349a28c6cb 100644 --- a/superset/models/slice.py +++ b/superset/models/slice.py @@ -96,7 +96,7 @@ class Slice( # pylint: disable=too-many-public-methods # security state, not user-authored content: permission maintenance # rewrites it in bulk, and versioning it produced phantom transactions # flooding the activity stream (10 "Chart updated" rows for one user - # save — surfaced by the version-history UI, PR #40988). Excluding it + # save — surfaced by the version-history UI). Excluding it # also means a restore can't resurrect stale permission strings; the # live, derived values stay authoritative. __versioned__: dict[str, Any] = { diff --git a/superset/versioning/changes/listener.py b/superset/versioning/changes/listener.py index fbd944eaaf70..fb9c500d3ca6 100644 --- a/superset/versioning/changes/listener.py +++ b/superset/versioning/changes/listener.py @@ -56,7 +56,7 @@ ``session.new`` entities are not processed in this listener: operation_type=0 transactions (baseline capture and first-save INSERTs) -produce zero change records per spec §Clarifications 2026-04-24. +produce zero change records by design. """ from __future__ import annotations @@ -137,7 +137,7 @@ # by commands alongside ``ACTION_KIND_KEY`` when the avenue has a payload # the field-level diff can't express; the canonical case is restore, # whose transaction otherwise carries no pointer to WHICH version was -# restored (surfaced by the version-history UI, PR #40988: "Restored to +# restored (surfaced by the version-history UI: "Restored to # X from [date]" can't be rendered from API data alone). # # Build the value with :func:`build_action_headline` — the single owner diff --git a/superset/versioning/changes/state.py b/superset/versioning/changes/state.py index 5d47871e1eed..30ceff6fc4d5 100644 --- a/superset/versioning/changes/state.py +++ b/superset/versioning/changes/state.py @@ -61,7 +61,7 @@ class name (string dispatch keeps this module free of hard imports # Per-model-class cache of the scalar-field set. Populated lazily on # first save of a model. Reading from ``__table__.columns`` is cheap -# but not free; memoising keeps the save-path overhead budget (FR-021) +# but not free; memoising keeps the save-path overhead budget # from slowly growing with the set of distinct model classes seen. _SCALAR_FIELDS_CACHE: dict[type, frozenset[str]] = {} diff --git a/superset/versioning/diff.py b/superset/versioning/diff.py index 3dce7b81e91f..23dee7f71e89 100644 --- a/superset/versioning/diff.py +++ b/superset/versioning/diff.py @@ -14,7 +14,7 @@ # KIND, either express or implied. See the License for the # specific language governing permissions and limitations # under the License. -"""Diff engine for the ``version_changes`` table (FR-016..FR-019). +"""Diff engine for the ``version_changes`` table. Hand-rolled because: @@ -31,8 +31,7 @@ libraries default to list-index matching, which is wrong for our data. -See ADR (plan.md §"Key Design Decision: Hand-rolled diff engine") for -the full rationale. +See the module docstring above for the full rationale. All functions in this module are pure: they take dicts (or lists of dicts) and return a list of :class:`ChangeRecord`. The ORM->dict @@ -628,7 +627,7 @@ def diff_json_field( # merely *viewed* — producing phantom # "Properties updated" records with no # user edit (surfaced by the -# version-history UI, PR #40988). The +# version-history UI). The # view-time write itself is a separate # round-trip-asymmetry issue (cf. #39706); # this exclusion stops the change-record @@ -862,8 +861,8 @@ def diff_dashboard( Promoting ``position_json`` to ``kind="layout"`` or ``json_metadata.native_filter_configuration`` to ``kind="filter"`` - is deferred to Phase 2 alongside the UI that would render them - (spec Clarifications §Session 2026-04-24); until then, both fields + is deferred to Phase 2 alongside the UI that would render them; + until then, both fields fall through to ``kind="field"`` records keyed by sub-key. """ records = diff_scalar_fields(pre, post, fields=fields) diff --git a/superset/versioning/schemas.py b/superset/versioning/schemas.py index 9fa51a8432e7..50fbba38b7a5 100644 --- a/superset/versioning/schemas.py +++ b/superset/versioning/schemas.py @@ -40,7 +40,7 @@ class VersionChangeRecordSchema(Schema): The frontend renders human-readable prose from (``kind``, ``from_value``, ``to_value``) via Flask-Babel. Server-side the - shape is deliberately machine-readable only — see spec FR-019. + shape is deliberately machine-readable only. """ kind = fields.String( From e21db8f72c80dd5b49ce163b9cf76b95334580ed Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 22 Jun 2026 12:20:19 -0600 Subject: [PATCH 30/35] test(versioning): toggle capture via options flag, not listener detach MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit capture_disabled_tests detached and re-attached Continuum's write listeners to simulate the off-path. The re-attach appended Continuum's before_flush after the custom baseline/change-record listeners, reordering the chain and leaking broken capture state into other tests in the same process (e.g. change_records_tests failed when run in the same shard). Toggle versioning_manager.options['versioning'] instead — the single master switch that Continuum's row writer (make_versions) and the baseline listener both gate on. It yields the identical zero-rows result without mutating global listener registration or ordering, so it can't pollute neighbours. The production off-path's actual detach stays covered by the structural unit tests in initialization_test.py::TestInitVersioning. Also asserts zero version_changes rows in the off case. Co-Authored-By: Claude Opus 4.8 (1M context) --- .../versioning/capture_disabled_tests.py | 136 +++++++++--------- 1 file changed, 66 insertions(+), 70 deletions(-) diff --git a/tests/integration_tests/versioning/capture_disabled_tests.py b/tests/integration_tests/versioning/capture_disabled_tests.py index 2190733424ea..113e1e2c39b4 100644 --- a/tests/integration_tests/versioning/capture_disabled_tests.py +++ b/tests/integration_tests/versioning/capture_disabled_tests.py @@ -16,33 +16,38 @@ # under the License. """Behavioral proof for the versioning kill-switch / dark-launch contract. -``ENABLE_VERSIONING_CAPTURE=False`` MUST cause a real save to write -**zero** ``version_transaction`` rows and **zero** ``*_version`` shadow -rows — Continuum is wired at import (``make_versioned()``), so suppressing -only the custom listeners would leave Continuum's own listeners minting -empty transaction rows on every flush. ``init_versioning`` detaches those -via ``_remove_continuum_write_listeners()``; this test pins that the -*behavioral* result is genuinely nothing-written. The structural unit tests -in ``tests/unit_tests/initialization_test.py`` (``TestInitVersioning``) drive -the config-flag branch of ``init_versioning`` with mocks; this exercises the -detach for real, against a database, and proves a control save under capture -*on* writes both a shadow row and a ``version_changes`` record (so the -zero-rows assertions are not vacuously true). - -This is the acceptance gate for shipping versioning dark in the -base-infra rollout PR. +When capture is off, a real save MUST write **zero** ``version_transaction`` +rows, **zero** ``*_version`` shadow rows, and **zero** ``version_changes`` +records — Continuum is wired at import (``make_versioned()``), so the gate +has to make every write path inert. + +This test toggles ``versioning_manager.options['versioning']`` — the single +master switch that Continuum's row-creating path (``make_versions``) and the +custom baseline listener both gate on — rather than detaching/re-attaching +Continuum's listeners. Flipping the option produces the identical zero-rows +result without mutating the process-global SQLAlchemy listener *registration* +or *ordering*; an earlier version detached and re-attached the listeners, +which reordered Continuum's ``before_flush`` relative to the custom listeners +and leaked broken capture state into other tests sharing the process. + +The production off-path (``init_versioning`` with +``ENABLE_VERSIONING_CAPTURE=False``) both flips this option off *and* detaches +the listeners as belt-and-suspenders; that detach is covered structurally by +``tests/unit_tests/initialization_test.py::TestInitVersioning``. Here we prove +the behavioral contract (off → nothing written) and, via the capture-on +control, that the assertions are not vacuously true. """ from __future__ import annotations +from collections.abc import Iterator +from contextlib import contextmanager + import pytest import sqlalchemy as sa -from sqlalchemy.engine import Engine -from sqlalchemy.orm import Mapper, Session from sqlalchemy_continuum import version_class, versioning_manager from superset.extensions import db -from superset.initialization import SupersetAppInitializer from superset.models.slice import Slice from tests.integration_tests.base_tests import SupersetTestCase from tests.integration_tests.constants import ADMIN_USERNAME @@ -72,36 +77,30 @@ def _version_changes_count() -> int: ) -def _reattach_continuum_write_listeners() -> None: - """Inverse of ``init_versioning._remove_continuum_write_listeners`` so this - test restores process-global capture state for the rest of the suite - (which runs with ``ENABLE_VERSIONING_CAPTURE`` on). Idempotent on a - representative listener. Also restores ``options['versioning']`` — the - detach flips it off (and the baseline listener honors it), so the - re-attach must flip it back on or subsequent saves would silently stop - capturing.""" - versioning_manager.options["versioning"] = True - if sa.event.contains(Mapper, "after_insert", versioning_manager.track_inserts): - return # already attached - versioning_manager.track_operations(Mapper) - versioning_manager.track_session(Session) - sa.event.listen( - Engine, "before_execute", versioning_manager.track_association_operations - ) - sa.event.listen(Engine, "rollback", versioning_manager.clear_connection) - sa.event.listen( - Engine, - "set_connection_execution_options", - versioning_manager.track_cloned_connections, - ) +@contextmanager +def _capture_disabled() -> Iterator[None]: + """Disable version capture for the duration of the block by flipping the + Continuum master switch off, restoring it on exit. + + This is the same option both ``make_versions`` (Continuum's row writer) + and the baseline listener gate on, so within the block every save is + inert. Restoring it in ``finally`` keeps the suite (which runs with + capture on) unaffected — and because we never touch listener + registration, there is no global ordering to corrupt for other tests. + """ + previous = versioning_manager.options["versioning"] + versioning_manager.options["versioning"] = False + try: + yield + finally: + versioning_manager.options["versioning"] = previous class TestVersioningCaptureDisabled(SupersetTestCase): @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices") def test_capture_off_writes_no_version_or_transaction_rows(self) -> None: - """With Continuum's write listeners detached (the capture-off path), - a real content change MUST write neither a shadow row nor a - ``version_transaction`` row.""" + """With capture off, a real content change MUST write neither a shadow + row, a ``version_transaction`` row, nor a ``version_changes`` record.""" db.session.commit() chart = db.session.query(Slice).filter(Slice.slice_name == "Girls").first() assert chart is not None @@ -109,35 +108,33 @@ def test_capture_off_writes_no_version_or_transaction_rows(self) -> None: self.login(ADMIN_USERNAME) - # Simulate the ENABLE_VERSIONING_CAPTURE=False branch of init_versioning. - # The detach is INSIDE the try so the finally always re-attaches — - # otherwise a failure between detach and the try body would leave the - # rest of the suite running capture-off and silently vacuous. try: - SupersetAppInitializer._remove_continuum_write_listeners() - tx_before = _transaction_row_count() - ver_before = _slice_version_count(chart_id) - - rv = self.client.put( - f"/api/v1/chart/{chart_id}", - json={"slice_name": "capture-off-renamed"}, - ) - assert rv.status_code == 200, rv.data - db.session.expire_all() - - assert _transaction_row_count() == tx_before, ( - "capture off MUST write zero version_transaction rows " - f"(before={tx_before}, after={_transaction_row_count()})" - ) - assert _slice_version_count(chart_id) == ver_before, ( - "capture off MUST write zero shadow rows " - f"(before={ver_before}, after={_slice_version_count(chart_id)})" - ) + with _capture_disabled(): + tx_before = _transaction_row_count() + ver_before = _slice_version_count(chart_id) + changes_before = _version_changes_count() + + rv = self.client.put( + f"/api/v1/chart/{chart_id}", + json={"slice_name": "capture-off-renamed"}, + ) + assert rv.status_code == 200, rv.data + db.session.expire_all() + + assert _transaction_row_count() == tx_before, ( + "capture off MUST write zero version_transaction rows " + f"(before={tx_before}, after={_transaction_row_count()})" + ) + assert _slice_version_count(chart_id) == ver_before, ( + "capture off MUST write zero shadow rows " + f"(before={ver_before}, after={_slice_version_count(chart_id)})" + ) + assert _version_changes_count() == changes_before, ( + "capture off MUST write zero version_changes records" + ) finally: - # Restore the chart and re-attach Continuum so the rest of the - # suite runs with capture on. + # Restore the chart name (capture is back on outside the block). self.client.put(f"/api/v1/chart/{chart_id}", json={"slice_name": "Girls"}) - _reattach_continuum_write_listeners() @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices") def test_control_capture_on_writes_version_and_change_rows(self) -> None: @@ -153,7 +150,6 @@ def test_control_capture_on_writes_version_and_change_rows(self) -> None: chart_id = chart.id self.login(ADMIN_USERNAME) - _reattach_continuum_write_listeners() # belt-and-suspenders: suite is on # ``>`` rather than ``== before + 1``: the first edit to a not-yet- # versioned entity also mints a synthetic baseline shadow row. ver_before = _slice_version_count(chart_id) From 483f8fedcf9d7523b1f4ff137290ca9f4990b43e Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 22 Jun 2026 15:12:19 -0600 Subject: [PATCH 31/35] feat(versioning): bound change-record output with value + count caps MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit version_changes is an audit log, not a content store, but the diff engine placed no ceiling on a single save's output: a 200 KB params/SQL blob became a 200 KB value, and a 2000-element list edit (or a 1000-key dict rewrite) became one row per element. Both are demonstrable and would bloat the table and the activity stream. Add capture-time safety caps in the diff engine, applied at the single persistence chokepoint (bulk_insert_records): - MAX_VALUE_BYTES (8 KB): an over-large from_value/to_value is swapped for a bounded marker recording the original size and a short preview. - MAX_RECORDS_PER_FIELD (100): a top-level field producing more than the cap collapses to a single summary record carrying the count. Values and fields within the bounds pass through unchanged. Unit tests pin both caps against the adversarial cases, and a representative set of heavy-but-legitimate chart shapes (wide table, pivot, multi-layer geospatial) confirms the thresholds never fire on real charts — a normal edit always yields individual, readable records. Co-Authored-By: Claude Opus 4.8 (1M context) --- superset/versioning/changes/state.py | 5 +- superset/versioning/diff.py | 77 +++++++ tests/unit_tests/versioning/test_diff_caps.py | 218 ++++++++++++++++++ 3 files changed, 299 insertions(+), 1 deletion(-) create mode 100644 tests/unit_tests/versioning/test_diff_caps.py diff --git a/superset/versioning/changes/state.py b/superset/versioning/changes/state.py index 30ceff6fc4d5..c69f1064c0cd 100644 --- a/superset/versioning/changes/state.py +++ b/superset/versioning/changes/state.py @@ -48,6 +48,7 @@ class name (string dispatch keeps this module free of hard imports from superset.versioning.changes.table import version_changes_table from superset.versioning.diff import ( + cap_records, ChangeRecord, diff_dashboard, diff_dataset, @@ -219,7 +220,9 @@ def bulk_insert_records( return rows = [] for (entity_kind, entity_id), records in buffered.items(): - for seq, r in enumerate(records): + # Bound a single save's output: collapse field-level record explosions + # and truncate over-large values before they hit version_changes. + for seq, r in enumerate(cap_records(records)): rows.append( { "transaction_id": transaction_id, diff --git a/superset/versioning/diff.py b/superset/versioning/diff.py index 23dee7f71e89..aaa0bd6b4c22 100644 --- a/superset/versioning/diff.py +++ b/superset/versioning/diff.py @@ -62,6 +62,16 @@ _JSON_METADATA_DIFF_DEPTH = 6 _SLICE_PARAMS_DIFF_DEPTH = 6 +# Output-safety caps applied at persistence time (see :func:`cap_records`). +# Unlike the depth caps above (usefulness bounds), these are *safety* bounds: +# they stop a single edit from writing an unbounded value or an unbounded +# number of rows into ``version_changes`` (and thus the activity stream). A +# 200 KB ``params``/SQL blob would otherwise become a 200 KB value, and a +# 2000-element list edit 2000 rows — both demonstrated. ``version_changes`` is +# an audit log, not a content store, so over-large output is summarised. +MAX_VALUE_BYTES = 8 * 1024 # per from_value / to_value, JSON-serialised +MAX_RECORDS_PER_FIELD = 100 # per top-level field; collapse beyond this + # Columns that are always excluded from change records, regardless of # what ``__versioned__`` says. ``id`` / ``uuid`` are stable identifiers # (not edited in normal flows). The four audit fields change on every @@ -208,6 +218,73 @@ class ChangeRecord: Key = str | int +def _value_bytes(value: Any) -> int: + try: + return len(_json.dumps(value, default=str)) + except (TypeError, ValueError): + return len(str(value)) + + +def _cap_value(value: Any) -> Any: + """Replace an over-large ``from_value``/``to_value`` with a bounded marker. + + ``version_changes`` is an audit log, not a content store; a value past + :data:`MAX_VALUE_BYTES` is swapped for a marker recording the original size + and a short preview, so a huge ``params``/SQL/blob edit can't write a + multi-hundred-KB row (or balloon the activity response). Values within the + bound pass through unchanged. + """ + size = _value_bytes(value) + if size <= MAX_VALUE_BYTES: + return value + preview = value if isinstance(value, str) else _json.dumps(value, default=str) + return {"__truncated__": True, "original_bytes": size, "preview": preview[:256]} + + +def cap_records(records: list[ChangeRecord]) -> list[ChangeRecord]: + """Apply the output-safety caps to one entity's record list before it is + persisted. + + 1. **Record-count cap** — group by top-level field (``path[0]``); any field + producing more than :data:`MAX_RECORDS_PER_FIELD` records (a 2000-element + list edit, a 1000-key dict rewrite, a thousand-node layout churn) is + collapsed to a single summary record carrying the count. First-seen field + order is preserved. + 2. **Value-size cap** — every surviving record's ``from_value``/``to_value`` + is run through :func:`_cap_value`. + """ + groups: dict[tuple[Any, ...], list[ChangeRecord]] = {} + for record in records: + groups.setdefault(tuple(record.path[:1]), []).append(record) + + deduped: list[ChangeRecord] = [] + for key, group in groups.items(): + if len(group) > MAX_RECORDS_PER_FIELD: + first = group[0] + deduped.append( + ChangeRecord( + kind=first.kind, + operation="update", + path=list(key), + from_value={"__collapsed__": len(group)}, + to_value={"__collapsed__": len(group)}, + ) + ) + else: + deduped.extend(group) + + return [ + ChangeRecord( + kind=r.kind, + operation=r.operation, + path=r.path, + from_value=_cap_value(r.from_value), + to_value=_cap_value(r.to_value), + ) + for r in deduped + ] + + def _operation_from_values(from_value: Any, to_value: Any) -> str: """Derive the per-record ``operation`` verb from ``from_value`` / ``to_value`` nullability. diff --git a/tests/unit_tests/versioning/test_diff_caps.py b/tests/unit_tests/versioning/test_diff_caps.py new file mode 100644 index 000000000000..cf749ca48a39 --- /dev/null +++ b/tests/unit_tests/versioning/test_diff_caps.py @@ -0,0 +1,218 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Output-safety caps on the change-record diff engine. + +``version_changes`` is an audit log, not a content store. A single edit must +not be able to write an unbounded value or an unbounded number of rows — a +200 KB ``params``/SQL blob, a 2000-element list edit, or a 1000-key dict +rewrite would otherwise bloat both the table and the activity stream. These +tests pin :func:`cap_records` (applied at persistence time) against exactly +those cases, and confirm ordinary small edits pass through untouched. +""" + +from __future__ import annotations + +from typing import Any + +from superset.utils import json +from superset.versioning.diff import ( + cap_records, + ChangeRecord, + diff_slice_params, + MAX_RECORDS_PER_FIELD, + MAX_VALUE_BYTES, +) + + +def _bytes(value: object) -> int: + return len(json.dumps(value, default=str)) + + +def _max_value_bytes(records: list[ChangeRecord]) -> int: + return max( + (_bytes(r.from_value) + _bytes(r.to_value) for r in records), + default=0, + ) + + +def test_small_edits_pass_through_uncapped() -> None: + """An ordinary handful of small param edits is returned verbatim — the + caps must not perturb the common case.""" + records = diff_slice_params( + {"time_range": "Last week", "viz_type": "table"}, + {"time_range": "Last month", "viz_type": "pie"}, + ) + capped = cap_records(records) + assert len(capped) == len(records) == 2 + assert all("__truncated__" not in str(r.to_value) for r in capped) + assert all("__collapsed__" not in str(r.to_value) for r in capped) + + +def test_oversized_value_is_truncated_to_a_marker() -> None: + """A value past MAX_VALUE_BYTES is replaced by a bounded marker that records + the original size and a preview, instead of writing the whole blob.""" + big = "a" * (MAX_VALUE_BYTES * 4) + capped = cap_records(diff_slice_params({"sql": "x"}, {"sql": big})) + assert len(capped) == 1 + marker = capped[0].to_value + assert marker["__truncated__"] is True + assert marker["original_bytes"] >= MAX_VALUE_BYTES * 4 + assert len(marker["preview"]) <= 256 + assert _bytes(marker) <= MAX_VALUE_BYTES + 600 # marker itself stays small + + +def test_list_explosion_is_collapsed() -> None: + """Editing a 2000-element list emits one record per element pre-cap; the cap + collapses the whole field to a single summary record.""" + pre = {"adhoc_filters": [{"col": f"c{i}", "val": i} for i in range(2000)]} + post = {"adhoc_filters": [{"col": f"c{i}", "val": i + 1} for i in range(2000)]} + raw = diff_slice_params(pre, post) + assert len(raw) > MAX_RECORDS_PER_FIELD # the explosion exists pre-cap + capped = cap_records(raw) + assert len(capped) == 1 + assert capped[0].to_value == {"__collapsed__": len(raw)} + assert capped[0].path == ["params"] + + +def test_wide_dict_rewrite_is_collapsed() -> None: + """A 1000-key dict rewrite under one field collapses to a single record.""" + pre = {"query_context": {f"k{i}": i for i in range(1000)}} + post = {"query_context": {f"k{i}": i + 1 for i in range(1000)}} + capped = cap_records(diff_slice_params(pre, post)) + assert len(capped) == 1 + assert capped[0].to_value["__collapsed__"] == 1000 + + +def test_distinct_fields_are_capped_independently() -> None: + """The record-count cap is per top-level field: a blown-up field collapses + while a sibling small edit on a different field survives intact.""" + pre = { + "adhoc_filters": [{"c": i} for i in range(2000)], + "time_range": "Last week", + } + post = { + "adhoc_filters": [{"c": i + 1} for i in range(2000)], + "time_range": "Last month", + } + capped = cap_records(diff_slice_params(pre, post)) + # one collapsed record for params (filters live under params) + the + # time_range edit, also under params -> all collapse into the params bucket. + assert all( + _bytes(r.from_value) + _bytes(r.to_value) <= MAX_VALUE_BYTES + 600 + for r in capped + ) + assert len(capped) <= MAX_RECORDS_PER_FIELD + + +def test_caps_bound_every_record_from_diff_slice_params() -> None: + """End-to-end invariant: after capping, no record exceeds the value or + count bounds, for the adversarial inputs the example gallery can't reach.""" + pre = {"sql": "x", "adhoc_filters": [{"c": i} for i in range(2000)]} + post = { + "sql": "z" * (MAX_VALUE_BYTES * 10), + "adhoc_filters": [{"c": i + 1} for i in range(2000)], + } + capped = cap_records(diff_slice_params(pre, post)) + assert len(capped) <= MAX_RECORDS_PER_FIELD + assert _max_value_bytes(capped) <= MAX_VALUE_BYTES + 600 + + +# Heavy-but-legitimate chart params, representative of the variety the chart +# gallery produces — a wide table, a pivot, a multi-layer geospatial config. +# These exist to prove the caps are tuned NOT to fire on real charts: a sweep +# over the example gallery found every viz type's edit well under both bounds, +# so a normal edit must yield individual, readable records — never a collapse +# or truncation marker. If a future cap-threshold tweak starts degrading real +# charts, these break. +_REALISTIC_PARAMS: dict[str, dict[str, Any]] = { + "wide_table": { + "viz_type": "table", + "query_mode": "aggregate", + "groupby": [f"dim_{i}" for i in range(15)], + "metrics": [ + { + "label": f"m_{i}", + "expressionType": "SIMPLE", + "column": {"column_name": f"c_{i}"}, + "aggregate": "SUM", + } + for i in range(8) + ], + "adhoc_filters": [ + { + "clause": "WHERE", + "subject": f"dim_{i}", + "operator": "==", + "comparator": f"v_{i}", + "expressionType": "SIMPLE", + } + for i in range(5) + ], + "time_range": "Last quarter", + "row_limit": 1000, + "table_timestamp_format": "smart_date", + }, + "pivot": { + "viz_type": "pivot_table_v2", + "groupbyRows": [f"row_{i}" for i in range(6)], + "groupbyColumns": [f"col_{i}" for i in range(4)], + "metrics": [ + { + "label": f"agg_{i}", + "expressionType": "SIMPLE", + "column": {"column_name": f"c_{i}"}, + "aggregate": "AVG", + } + for i in range(5) + ], + "aggregateFunction": "Sum", + "valueFormat": "SMART_NUMBER", + }, + "geospatial": { + "viz_type": "deck_multi", + "deck_slices": [{"layer": f"l_{i}", "opacity": 0.8} for i in range(6)], + "viewport": {"longitude": -122.4, "latitude": 37.8, "zoom": 11.2}, + "spatial": {"latCol": "lat", "lonCol": "lon", "type": "latlong"}, + "color_picker": {"r": 14, "g": 96, "b": 245, "a": 1}, + }, +} + + +def test_realistic_chart_edits_are_not_capped() -> None: + """A normal edit to a heavy-but-legitimate chart of each shape yields + individual records — none collapsed, none truncated — so the caps never + degrade real-world activity entries.""" + for name, params in _REALISTIC_PARAMS.items(): + edited = {**params, "row_limit": 5000, "time_range": "Last year"} + records = diff_slice_params(params, edited) + capped = cap_records(records) + assert capped == records, f"{name}: caps must not alter a normal edit" + assert capped, f"{name}: a real edit must still produce records" + for r in capped: + assert "__collapsed__" not in str(r.to_value), f"{name}: collapsed" + assert "__truncated__" not in str(r.to_value), f"{name}: truncated" + + +def test_full_rewrite_of_realistic_charts_stays_under_caps() -> None: + """Even replacing a heavy chart's params wholesale (every field changes at + once) stays within both bounds for legitimate shapes — the caps headroom + is real, not just exceeded-by-construction in the adversarial tests.""" + empty: dict[str, Any] = {} + for name, params in _REALISTIC_PARAMS.items(): + capped = cap_records(diff_slice_params(empty, params)) + assert len(capped) <= MAX_RECORDS_PER_FIELD, f"{name}: count over cap" + assert _max_value_bytes(capped) <= MAX_VALUE_BYTES, f"{name}: value over cap" From 245fecf59736445c11bee2aa2ad56055b34e73f3 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Wed, 17 Jun 2026 19:52:40 -0600 Subject: [PATCH 32/35] feat(activity-view): cross-entity version activity view Adds the read-only cross-entity activity view on top of entity versioning: a `superset/versioning/activity/` module (scope, windows, visibility, impact, kinds, orchestrator, render, queries) and `GET /activity/` endpoints on the chart, dashboard, and dataset APIs that return a unified, access-filtered stream of version activity, plus the version-history dropdown wiring on the frontend list views. Squashed from the activity-view development branch (the granular review-iteration history is preserved on the pre-restack backup branch). Co-Authored-By: Claude Opus 4.8 (1M context) --- UPDATING.md | 13 + .../src/pages/ActivityDebug/index.tsx | 420 ++++++ .../ChartList/VersionHistoryDropdown.tsx | 368 ++++++ .../DashboardList/VersionHistoryDropdown.tsx | 388 ++++++ .../DatasetList/VersionHistoryDropdown.tsx | 364 +++++ superset-frontend/src/views/routes.tsx | 13 + superset/charts/api.py | 85 +- superset/constants.py | 1 + superset/dashboards/api.py | 90 +- superset/datasets/api.py | 90 +- superset/initialization/__init__.py | 6 + superset/versioning/activity/__init__.py | 102 ++ superset/versioning/activity/impact.py | 160 +++ superset/versioning/activity/kinds.py | 186 +++ superset/versioning/activity/orchestrator.py | 409 ++++++ superset/versioning/activity/queries.py | 620 +++++++++ superset/versioning/activity/render.py | 256 ++++ superset/versioning/activity/scope.py | 103 ++ superset/versioning/activity/visibility.py | 201 +++ superset/versioning/activity/windows.py | 108 ++ superset/versioning/api_helpers.py | 127 +- superset/versioning/changes/listener.py | 27 + superset/versioning/diff.py | 9 +- superset/versioning/schemas.py | 324 ++++- superset/views/activity_debug.py | 53 + .../versioning/activity_view_tests.py | 1174 +++++++++++++++++ .../versioning/change_records_tests.py | 701 ++++++++++ .../versioning/perf_validation_tests.py | 447 +++++++ tests/unit_tests/versioning/test_activity.py | 678 ++++++++++ 29 files changed, 7460 insertions(+), 63 deletions(-) create mode 100644 superset-frontend/src/pages/ActivityDebug/index.tsx create mode 100644 superset-frontend/src/pages/ChartList/VersionHistoryDropdown.tsx create mode 100644 superset-frontend/src/pages/DashboardList/VersionHistoryDropdown.tsx create mode 100644 superset-frontend/src/pages/DatasetList/VersionHistoryDropdown.tsx create mode 100644 superset/versioning/activity/__init__.py create mode 100644 superset/versioning/activity/impact.py create mode 100644 superset/versioning/activity/kinds.py create mode 100644 superset/versioning/activity/orchestrator.py create mode 100644 superset/versioning/activity/queries.py create mode 100644 superset/versioning/activity/render.py create mode 100644 superset/versioning/activity/scope.py create mode 100644 superset/versioning/activity/visibility.py create mode 100644 superset/versioning/activity/windows.py create mode 100644 superset/views/activity_debug.py create mode 100644 tests/integration_tests/versioning/activity_view_tests.py create mode 100644 tests/integration_tests/versioning/change_records_tests.py create mode 100644 tests/integration_tests/versioning/perf_validation_tests.py create mode 100644 tests/unit_tests/versioning/test_activity.py diff --git a/UPDATING.md b/UPDATING.md index 433292a87a0a..d6e1d8a85e86 100644 --- a/UPDATING.md +++ b/UPDATING.md @@ -50,6 +50,19 @@ A few save- and import-path internals change **unconditionally** (independent of These are behavior changes that take effect on upgrade regardless of `ENABLE_VERSIONING_CAPTURE`; no operator action is required. +### Cross-entity version activity stream + +A read-only companion to the version-history endpoints: each entity type gains a `GET /api/v1/{chart,dashboard,dataset}//activity/` endpoint returning a chronological, access-filtered stream of edits — the entity's own edits plus, for charts and dashboards, transitive edits to related entities during their association windows. + +| Param | Type | Default | Purpose | +|---|---|---|---| +| `since` / `until` | ISO 8601 | — | Bound `issued_at` | +| `include` | `self` \| `related` \| `all` | `all` | Own edits, related edits, or both | +| `q` | string | — | Case-insensitive search over the full history, applied before pagination (so `count` reflects matches) | +| `page` / `page_size` | integer | `0` / `25` | Pagination (`page_size` clamped to 200) | + +Authorization reuses the resource's `can_read` permission and per-object `raise_for_access`; related-entity rows are visibility-filtered to what the caller may see. The stream is empty unless version capture is on (`ENABLE_VERSIONING_CAPTURE`). + ### Webhook alerts/reports block private/internal hosts by default Webhook alert/report dispatch (`WebhookNotification.send`) now validates the target URL's host against the same private/internal-IP block applied to dataset import URLs. If the resolved host is in a loopback, link-local, private (RFC-1918), shared-CGNAT, or multicast range, the webhook is rejected with `NotificationParamException`. diff --git a/superset-frontend/src/pages/ActivityDebug/index.tsx b/superset-frontend/src/pages/ActivityDebug/index.tsx new file mode 100644 index 000000000000..bce13f0dded0 --- /dev/null +++ b/superset-frontend/src/pages/ActivityDebug/index.tsx @@ -0,0 +1,420 @@ +/** + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +// Throwaway debug UI for sc-107283 activity-view endpoints. Verify the +// JSON responses look right by eye. Delete this directory + the route +// entry in views/routes.tsx when the activity-view feature ships. + +import { useEffect, useMemo, useState } from 'react'; +import { useParams } from 'react-router-dom'; +import { SupersetClient } from '@superset-ui/core'; +import { t } from '@apache-superset/core/translation'; +import { + Card, + Empty, + Input, + Loading, + Radio, + Space, + Tag, + Typography, +} from '@superset-ui/core/components'; + +type ResourceKind = 'dashboard' | 'chart' | 'dataset'; +type IncludeMode = 'self' | 'related' | 'all'; +type EntityKind = 'dashboard' | 'chart' | 'dataset'; + +interface ChangedBy { + id: number; + first_name: string; + last_name: string; +} + +interface ActivityRecord { + version_uuid: string; + entity_kind: EntityKind; + entity_uuid: string | null; + entity_name: string; + entity_deleted: boolean; + entity_deletion_state: string | null; + source: 'self' | 'related'; + transaction_id: number; + // Transaction-level avenue: restore / import / clone / null + // (= ordinary save). Shared by every record in the same tx. + action_kind: string | null; + issued_at: string; + changed_by: ChangedBy | null; + kind: string; + // Per-record verb: add / remove / move / edit. + operation: string; + path: string[]; + from_value: unknown; + to_value: unknown; + summary: string; + impact: { charts?: number; datasets?: number } | null; +} + +interface ActivityResponse { + result: ActivityRecord[]; + count: number; +} + +const PAGE_SIZE_OPTIONS = [10, 25, 50, 100, 200]; + +const KIND_COLOR: Record = { + filter: 'blue', + metric: 'green', + dimension: 'cyan', + column: 'geekblue', + chart: 'purple', + time_range: 'gold', + color_palette: 'magenta', + restore: 'orange', + field: 'default', +}; + +const ENTITY_KIND_COLOR: Record = { + dashboard: 'blue', + chart: 'purple', + dataset: 'green', +}; + +function formatValue(value: unknown): string { + if (value === null || value === undefined) return 'null'; + if (typeof value === 'string') return JSON.stringify(value); + return JSON.stringify(value); +} + +function ChangedByDisplay({ changedBy }: { changedBy: ChangedBy | null }) { + if (changedBy === null) { + return ( + + {t('system / unknown')} + + ); + } + return ( + + {changedBy.first_name} {changedBy.last_name} (id={changedBy.id}) + + ); +} + +function RecordCard({ record }: { record: ActivityRecord }) { + const headerExtra = ( + + + {record.entity_kind} + + + {record.source} + + {record.kind} + {record.operation} + {record.action_kind && {record.action_kind}} + {record.entity_deleted && deleted} + {record.entity_deletion_state === 'soft_deleted' && ( + soft-deleted + )} + + ); + + const headline = record.summary + ? record.summary + : `${record.entity_kind}: ${record.entity_name || '(unnamed)'}`; + + return ( + + {headline} + + {record.issued_at} · tx={record.transaction_id} ·{' '} + + + + } + extra={headerExtra} + style={{ marginBottom: 8 }} + > + +
+ + {t('entity_name:')} + {' '} + + {record.entity_name || t('(no name)')} + +
+
+ + {t('entity_uuid:')} + {' '} + + {record.entity_uuid || t('null (tombstoned)')} + +
+
+ + {t('version_uuid:')} + {' '} + + {record.version_uuid || t('null')} + +
+
+ {t('path:')}{' '} + {JSON.stringify(record.path)} +
+
+ {t('from →')}{' '} + + {formatValue(record.from_value)} + {' '} + {t('→ to')}{' '} + {formatValue(record.to_value)} +
+ {record.impact !== null && ( +
+ {t('impact:')}{' '} + + {JSON.stringify(record.impact)} + +
+ )} + + + ); +} + +function ActivityDebug() { + const { resource: resourceParam, uuid } = useParams<{ + resource: string; + uuid: string; + }>(); + + const resource = useMemo(() => { + if ( + resourceParam === 'dashboard' || + resourceParam === 'chart' || + resourceParam === 'dataset' + ) { + return resourceParam; + } + return null; + }, [resourceParam]); + + const [include, setInclude] = useState('all'); + const [page, setPage] = useState(0); + const [pageSize, setPageSize] = useState(25); + const [since, setSince] = useState(''); + const [until, setUntil] = useState(''); + const [data, setData] = useState(null); + const [loading, setLoading] = useState(false); + const [error, setError] = useState(null); + const [reloadCounter, setReloadCounter] = useState(0); + + useEffect(() => { + if (!resource || !uuid) return; + const params = new URLSearchParams(); + params.set('include', include); + params.set('page', String(page)); + params.set('page_size', String(pageSize)); + if (since) params.set('since', since); + if (until) params.set('until', until); + setLoading(true); + setError(null); + SupersetClient.get({ + endpoint: `/api/v1/${resource}/${uuid}/activity/?${params.toString()}`, + }) + .then(({ json }) => { + setData(json as ActivityResponse); + }) + .catch(err => { + const msg = err?.message || String(err); + setError(msg); + setData(null); + }) + .finally(() => setLoading(false)); + }, [resource, uuid, include, page, pageSize, since, until, reloadCounter]); + + if (!resource) { + return ( +
+ + {t('Activity Debug — invalid URL')} + + + {t( + 'Use /activity-debug/{dashboard|chart|dataset}/{uuid} — e.g. /activity-debug/dashboard/4a8f3c2e-...', + )} + +
+ ); + } + + const records = data?.result ?? []; + const totalCount = data?.count ?? 0; + const lastPage = Math.max(0, Math.ceil(totalCount / pageSize) - 1); + + return ( +
+ + {t('Activity Debug')} + + + {t('Throwaway tool for verifying sc-107283 activity-view responses.')} + + + + + + {t('endpoint')} + + {`GET /api/v1/${resource}/${uuid}/activity/`} + + + + + + include + { + setInclude(e.target.value as IncludeMode); + setPage(0); + }} + size="small" + > + all + self + related + + + + + page_size + + + + + since + { + setSince(e.target.value); + setPage(0); + }} + style={{ width: 220 }} + /> + + + + until + { + setUntil(e.target.value); + setPage(0); + }} + style={{ width: 220 }} + /> + + + + + + + + + + {t('count')}: {totalCount} + + + {t('page')}: {page} / {lastPage} + + + + + + {loading && } + + {error && ( + + + {t('error')}: + {' '} + {error} + + )} + + {!loading && !error && records.length === 0 && ( + + )} + + {records.map(record => ( + + ))} +
+ ); +} + +export default ActivityDebug; diff --git a/superset-frontend/src/pages/ChartList/VersionHistoryDropdown.tsx b/superset-frontend/src/pages/ChartList/VersionHistoryDropdown.tsx new file mode 100644 index 000000000000..2a1bf5c0a6af --- /dev/null +++ b/superset-frontend/src/pages/ChartList/VersionHistoryDropdown.tsx @@ -0,0 +1,368 @@ +/** + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +// TEMP: Demo aid for sc-103156 entity-versioning. Lets a user open a +// dropdown of recent versions on a chart and restore one. Not part +// of the merged feature scope (ADR-005 limits v1 to backend); revert +// before pushing the versioning branch. + +import { useState, useCallback } from 'react'; +import { SupersetClient } from '@superset-ui/core'; +import { t } from '@apache-superset/core/translation'; +import { Dropdown, Tooltip, Icons } from '@superset-ui/core/components'; + +interface Change { + kind: string; + // Per-record verb: add / remove / move / edit. Explicit instead of + // inferred from from_value / to_value null-tests or path[0]. + operation: string; + path: string[]; + from_value: unknown; + to_value: unknown; +} + +interface ChangedBy { + id: number; + username: string; + first_name: string; + last_name: string; +} + +interface Version { + version_uuid: string; + version_number: number; + transaction_id: number; + operation_type: string; + // Transaction-level avenue: restore / import / clone / null (= save). + action_kind: string | null; + issued_at: string; + changed_by: ChangedBy | null; + changes: Change[]; +} + +interface Props { + chartUuid: string; + onRestored?: () => void; +} + +// Layout element kinds — used to decide whether a record's +// ``operation`` describes a node-level layout action vs a per-leaf +// change inside a layout-edit. Charts don't normally produce layout +// records, but the formatter stays uniform across the three widgets. +const LAYOUT_KINDS = new Set([ + 'chart', + 'row', + 'column', + 'tab', + 'tabs', + 'header', + 'markdown', + 'divider', +]); + +// Localized labels for the kinds emitted by the backend (layout walker +// + dataset child diff). Defined statically so xgettext can extract them. +const KIND_LABELS: Record = { + chart: t('chart'), + row: t('row'), + column: t('column'), + tab: t('tab'), + tabs: t('tabs'), + header: t('header'), + markdown: t('markdown'), + divider: t('divider'), + metric: t('metric'), +}; +const localizedKind = (k: string): string => KIND_LABELS[k] ?? k; + +function summarizeChange(c: Change): string { + // Layout record at the node level: path = [node_id], operation+kind + // in columns. + if (LAYOUT_KINDS.has(c.kind) && c.path.length === 1) { + const kind = localizedKind(c.kind); + const payload = + ((c.to_value ?? c.from_value) as { name?: string } | null) ?? null; + const name = payload?.name; + if (c.operation === 'add') { + return name + ? t('Added %(kind)s "%(name)s"', { kind, name }) + : t('Added %(kind)s', { kind }); + } + if (c.operation === 'remove') { + return name + ? t('Removed %(kind)s "%(name)s"', { kind, name }) + : t('Removed %(kind)s', { kind }); + } + if (c.operation === 'move') { + return name + ? t('Moved %(kind)s "%(name)s"', { kind, name }) + : t('Moved %(kind)s', { kind }); + } + return name + ? t('Edited %(kind)s "%(name)s"', { kind, name }) + : t('Edited %(kind)s', { kind }); + } + + // Layout edit at the leaf level: path = [node_id, ...leaf-path]. + if (LAYOUT_KINDS.has(c.kind) && c.path.length >= 2) { + const kind = localizedKind(c.kind); + const leaf = String(c.path[c.path.length - 1]); + return t('Changed %(kind)s %(leaf)s', { kind, leaf }); + } + + const isAdd = c.operation === 'add'; + const isRemove = c.operation === 'remove'; + + if (c.path.length === 2 && (c.kind === 'column' || c.kind === 'metric')) { + const kind = localizedKind(c.kind); + const name = String(c.path[1]); + if (isAdd) return t('Added %(kind)s "%(name)s"', { kind, name }); + if (isRemove) return t('Removed %(kind)s "%(name)s"', { kind, name }); + return t('Changed %(kind)s "%(name)s"', { kind, name }); + } + + if (c.path[0] === 'slices') { + const id = String(c.path[1] ?? ''); + if (isAdd) return t('Added chart %(id)s', { id }).trim(); + if (isRemove) return t('Removed chart %(id)s', { id }).trim(); + return t('Changed chart %(id)s', { id }).trim(); + } + + if (c.kind === 'field') { + const fieldName = String(c.path[c.path.length - 1]); + const fieldLabel: string = + fieldName === 'dashboard_title' + ? t('title') + : fieldName === 'slice_name' + ? t('chart name') + : fieldName === 'table_name' + ? t('table name') + : fieldName; + const isShortScalar = + c.to_value !== null && + c.to_value !== undefined && + (typeof c.to_value === 'string' || + typeof c.to_value === 'number' || + typeof c.to_value === 'boolean') && + String(c.to_value).length <= 80; + if (!isAdd && !isRemove && isShortScalar) { + return t('Changed %(field)s to "%(value)s"', { + field: fieldLabel, + value: String(c.to_value), + }); + } + if (isRemove) { + return t('Cleared %(field)s', { field: fieldLabel }); + } + if (isAdd && isShortScalar) { + return t('Set %(field)s to "%(value)s"', { + field: fieldLabel, + value: String(c.to_value), + }); + } + if (isAdd) return t('Added %(field)s', { field: fieldLabel }); + if (isRemove) return t('Removed %(field)s', { field: fieldLabel }); + return t('Changed %(field)s', { field: fieldLabel }); + } + + const kind = localizedKind(c.kind); + if (c.path.length) { + const detail = String(c.path[c.path.length - 1]); + if (isAdd) return t('Added %(kind)s %(detail)s', { kind, detail }); + if (isRemove) return t('Removed %(kind)s %(detail)s', { kind, detail }); + return t('Changed %(kind)s %(detail)s', { kind, detail }); + } + if (isAdd) return t('Added %(kind)s', { kind }); + if (isRemove) return t('Removed %(kind)s', { kind }); + return t('Changed %(kind)s', { kind }); +} + +function formatChangeTitle(changes: Change[]): string { + if (!changes.length) return t('Baseline'); + const first = summarizeChange(changes[0]); + if (changes.length === 1) return first; + return t('%(first)s (+%(more)s more)', { + first, + more: changes.length - 1, + }); +} + +function formatUser(by: ChangedBy | null): string { + if (!by) return t('system'); + if (by.first_name || by.last_name) { + return `${by.first_name ?? ''} ${by.last_name ?? ''}`.trim(); + } + return by.username; +} + +function formatDate(iso: string): string { + try { + // Match the Superset locale set in src/views/App.tsx on + // ``document.documentElement.lang`` rather than the browser default. + const lang = document.documentElement.lang || undefined; + return new Date(iso).toLocaleString(lang); + } catch { + return iso; + } +} + +export default function VersionHistoryDropdown({ + chartUuid, + onRestored, +}: Props) { + const [versions, setVersions] = useState(null); + const [loading, setLoading] = useState(false); + + const loadVersions = useCallback(async () => { + setLoading(true); + try { + const { json } = await SupersetClient.get({ + endpoint: `/api/v1/chart/${chartUuid}/versions/`, + }); + const result = (json as { result: Version[] }).result || []; + // Newest first (API returns oldest-first) + setVersions([...result].reverse().slice(0, 20)); + } catch (e) { + console.error('Failed to load versions', e); + setVersions([]); + } finally { + setLoading(false); + } + }, [chartUuid]); + + const handleRestore = useCallback( + async (version: Version) => { + const summary = formatChangeTitle(version.changes); + if ( + // eslint-disable-next-line no-alert + !window.confirm( + t( + 'Restore this chart to version %(num)s (%(summary)s)? This will overwrite the current state.', + { num: version.version_number, summary }, + ), + ) + ) { + return; + } + try { + await SupersetClient.post({ + endpoint: `/api/v1/chart/${chartUuid}/versions/${version.version_uuid}/restore`, + }); + // eslint-disable-next-line no-alert + window.alert(t('Restored. Reload the page to see the change.')); + if (onRestored) onRestored(); + } catch (e) { + console.error('Restore failed', e); + // eslint-disable-next-line no-alert + window.alert(t('Restore failed — see browser console for details.')); + } + }, + [chartUuid, onRestored], + ); + + const items = (() => { + if (loading) { + return [{ key: 'loading', label: t('Loading…'), disabled: true }]; + } + if (!versions) { + return [ + { key: 'empty', label: t('Click to load versions'), disabled: true }, + ]; + } + if (versions.length === 0) { + return [{ key: 'empty', label: t('No versions yet'), disabled: true }]; + } + // versions is already newest-first, so [0] is the live/current version. + return versions.map((v, idx) => { + const isCurrent = idx === 0; + return { + key: String(v.transaction_id), + // antd's `disabled: true` greys the item and blocks default + // click handling; combined with the inner div NOT having an + // onClick when current, the row becomes informational only. + disabled: isCurrent, + label: ( +
handleRestore(v)} + > +
+ #{v.version_number} — {formatChangeTitle(v.changes)} + {isCurrent && ( + + {t('(current)')} + + )} +
+
+ {formatUser(v.changed_by)} · {formatDate(v.issued_at)} +
+ {v.changes.length > 1 && ( +
    + {v.changes.slice(0, 5).map((c, i) => ( +
  • {summarizeChange(c)}
    • + ))} + {v.changes.length > 5 && ( +
  • + {t('+%(n)s more', { n: v.changes.length - 5 })} +
    • + )} +
+ )} +
+ ), + }; + }); + })(); + + return ( + { + if (open && versions === null && !loading) loadVersions(); + }} + > + + + + + + + ); +} diff --git a/superset-frontend/src/pages/DashboardList/VersionHistoryDropdown.tsx b/superset-frontend/src/pages/DashboardList/VersionHistoryDropdown.tsx new file mode 100644 index 000000000000..0c2c4fb1a885 --- /dev/null +++ b/superset-frontend/src/pages/DashboardList/VersionHistoryDropdown.tsx @@ -0,0 +1,388 @@ +/** + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +// TEMP: Demo aid for sc-103156 entity-versioning. Lets a user open a +// dropdown of recent versions on a dashboard and restore one. Not part +// of the merged feature scope (ADR-005 limits v1 to backend); revert +// before pushing the versioning branch. + +import { useState, useCallback } from 'react'; +import { SupersetClient } from '@superset-ui/core'; +import { t } from '@apache-superset/core/translation'; +import { Dropdown, Tooltip, Icons } from '@superset-ui/core/components'; + +interface Change { + kind: string; + // Per-record verb: add / remove / move / edit. Explicit instead of + // inferred from from_value / to_value null-tests or path[0]. + operation: string; + path: string[]; + from_value: unknown; + to_value: unknown; +} + +interface ChangedBy { + id: number; + username: string; + first_name: string; + last_name: string; +} + +interface Version { + version_uuid: string; + version_number: number; + transaction_id: number; + operation_type: string; + // Transaction-level avenue: restore / import / clone / null (= save). + // All records sharing a transaction share this. Rendered as a save- + // container headline before the per-record changes are listed. + action_kind: string | null; + issued_at: string; + changed_by: ChangedBy | null; + changes: Change[]; +} + +interface Props { + dashboardUuid: string; + onRestored?: () => void; +} + +// Localized labels for the kinds emitted by the backend (layout walker +// + dataset child diff). Defined statically so xgettext can extract them. +const KIND_LABELS: Record = { + chart: t('chart'), + row: t('row'), + column: t('column'), + tab: t('tab'), + tabs: t('tabs'), + header: t('header'), + markdown: t('markdown'), + divider: t('divider'), + metric: t('metric'), +}; +const localizedKind = (k: string): string => KIND_LABELS[k] ?? k; + +// Layout element kinds — used to decide whether a record's +// ``operation`` describes a node-level layout action vs a per-leaf +// change inside a layout-edit. (Layout edits at depth >= 2 carry the +// node_id + leaf path; we render those via the field-record branch.) +const LAYOUT_KINDS = new Set([ + 'chart', + 'row', + 'column', + 'tab', + 'tabs', + 'header', + 'markdown', + 'divider', +]); + +function summarizeChange(c: Change): string { + // Layout record at the node level: path = [node_id]; operation + // and kind both live in columns. Payload carries name / chartId etc. + if (LAYOUT_KINDS.has(c.kind) && c.path.length === 1) { + const kind = localizedKind(c.kind); + const payload = + ((c.to_value ?? c.from_value) as { name?: string } | null) ?? null; + const name = payload?.name; + if (c.operation === 'add') { + return name + ? t('Added %(kind)s "%(name)s"', { kind, name }) + : t('Added %(kind)s', { kind }); + } + if (c.operation === 'remove') { + return name + ? t('Removed %(kind)s "%(name)s"', { kind, name }) + : t('Removed %(kind)s', { kind }); + } + if (c.operation === 'move') { + return name + ? t('Moved %(kind)s "%(name)s"', { kind, name }) + : t('Moved %(kind)s', { kind }); + } + return name + ? t('Edited %(kind)s "%(name)s"', { kind, name }) + : t('Edited %(kind)s', { kind }); + } + + // Layout edit at the leaf level: path = [node_id, ...leaf-path]. + // kind is the layout element kind; the leaf key is path[1+]. + if (LAYOUT_KINDS.has(c.kind) && c.path.length >= 2) { + const kind = localizedKind(c.kind); + const leaf = String(c.path[c.path.length - 1]); + return t('Changed %(kind)s %(leaf)s', { kind, leaf }); + } + + const isAdd = c.operation === 'add'; + const isRemove = c.operation === 'remove'; + + // Dataset child: path = [columns | metrics, ]. ``kind`` is + // ``column`` / ``metric`` so we can rebuild a readable summary. + if (c.path.length === 2 && (c.kind === 'column' || c.kind === 'metric')) { + const kind = localizedKind(c.kind); + const name = String(c.path[1]); + if (isAdd) return t('Added %(kind)s "%(name)s"', { kind, name }); + if (isRemove) return t('Removed %(kind)s "%(name)s"', { kind, name }); + return t('Changed %(kind)s "%(name)s"', { kind, name }); + } + + // Slice membership (mostly folded into layout records server-side, + // but may still appear if the layout walk didn't catch a chart). + if (c.path[0] === 'slices') { + const id = String(c.path[1] ?? ''); + if (isAdd) return t('Added chart %(id)s', { id }).trim(); + if (isRemove) return t('Removed chart %(id)s', { id }).trim(); + return t('Changed chart %(id)s', { id }).trim(); + } + + // Scalar field record: path = [field_name] or [json_field, sub_key]. + if (c.kind === 'field') { + const fieldName = String(c.path[c.path.length - 1]); + // Friendly labels for the most user-visible fields. + const fieldLabel: string = + fieldName === 'dashboard_title' + ? t('title') + : fieldName === 'slice_name' + ? t('chart name') + : fieldName === 'table_name' + ? t('table name') + : fieldName; + // If the new value is a short primitive (string/number/bool), show + // "Changed to " — much more useful than just naming + // the field. Long strings, dicts and arrays fall through to the + // generic verb-only summary. + const isShortScalar = + c.to_value !== null && + c.to_value !== undefined && + (typeof c.to_value === 'string' || + typeof c.to_value === 'number' || + typeof c.to_value === 'boolean') && + String(c.to_value).length <= 80; + if (!isAdd && !isRemove && isShortScalar) { + return t('Changed %(field)s to "%(value)s"', { + field: fieldLabel, + value: String(c.to_value), + }); + } + if (isRemove) { + return t('Cleared %(field)s', { field: fieldLabel }); + } + if (isAdd && isShortScalar) { + return t('Set %(field)s to "%(value)s"', { + field: fieldLabel, + value: String(c.to_value), + }); + } + if (isAdd) return t('Added %(field)s', { field: fieldLabel }); + if (isRemove) return t('Removed %(field)s', { field: fieldLabel }); + return t('Changed %(field)s', { field: fieldLabel }); + } + + // Fallback: kind plus the trailing path segment (if any). + const kind = localizedKind(c.kind); + if (c.path.length) { + const detail = String(c.path[c.path.length - 1]); + if (isAdd) return t('Added %(kind)s %(detail)s', { kind, detail }); + if (isRemove) return t('Removed %(kind)s %(detail)s', { kind, detail }); + return t('Changed %(kind)s %(detail)s', { kind, detail }); + } + if (isAdd) return t('Added %(kind)s', { kind }); + if (isRemove) return t('Removed %(kind)s', { kind }); + return t('Changed %(kind)s', { kind }); +} + +function formatChangeTitle(changes: Change[]): string { + if (!changes.length) return t('Baseline'); + const first = summarizeChange(changes[0]); + if (changes.length === 1) return first; + return t('%(first)s (+%(more)s more)', { + first, + more: changes.length - 1, + }); +} + +function formatUser(by: ChangedBy | null): string { + if (!by) return t('system'); + if (by.first_name || by.last_name) { + return `${by.first_name ?? ''} ${by.last_name ?? ''}`.trim(); + } + return by.username; +} + +function formatDate(iso: string): string { + try { + // Match the Superset locale set in src/views/App.tsx on + // ``document.documentElement.lang`` rather than the browser default. + const lang = document.documentElement.lang || undefined; + return new Date(iso).toLocaleString(lang); + } catch { + return iso; + } +} + +export default function VersionHistoryDropdown({ + dashboardUuid, + onRestored, +}: Props) { + const [versions, setVersions] = useState(null); + const [loading, setLoading] = useState(false); + + const loadVersions = useCallback(async () => { + setLoading(true); + try { + const { json } = await SupersetClient.get({ + endpoint: `/api/v1/dashboard/${dashboardUuid}/versions/`, + }); + const result = (json as { result: Version[] }).result || []; + // Newest first (API returns oldest-first) + setVersions([...result].reverse().slice(0, 20)); + } catch (e) { + console.error('Failed to load versions', e); + setVersions([]); + } finally { + setLoading(false); + } + }, [dashboardUuid]); + + const handleRestore = useCallback( + async (version: Version) => { + const summary = formatChangeTitle(version.changes); + if ( + // eslint-disable-next-line no-alert + !window.confirm( + t( + 'Restore this dashboard to version %(num)s (%(summary)s)? This will overwrite the current state.', + { num: version.version_number, summary }, + ), + ) + ) { + return; + } + try { + await SupersetClient.post({ + endpoint: `/api/v1/dashboard/${dashboardUuid}/versions/${version.version_uuid}/restore`, + }); + onRestored?.(); + // Navigate to the dashboard with no URL params. A previous + // ``?native_filters_key=…`` (or ``permalink_key`` / ``form_data_key``) + // points at a server-cached snapshot from before the restore; + // the next page hydration would merge it on top of the freshly + // restored ``json_metadata`` and effectively mask the rollback + // (e.g. dashboard-level colour scheme changes don't appear). + // A clean URL forces hydration from the restored DB state. + window.location.href = `/superset/dashboard/${dashboardUuid}/`; + } catch (e) { + console.error('Restore failed', e); + // eslint-disable-next-line no-alert + window.alert(t('Restore failed — see browser console for details.')); + } + }, + [dashboardUuid, onRestored], + ); + + const items = (() => { + if (loading) { + return [{ key: 'loading', label: t('Loading…'), disabled: true }]; + } + if (!versions) { + return [ + { key: 'empty', label: t('Click to load versions'), disabled: true }, + ]; + } + if (versions.length === 0) { + return [{ key: 'empty', label: t('No versions yet'), disabled: true }]; + } + // versions is already newest-first, so [0] is the live/current version. + return versions.map((v, idx) => { + const isCurrent = idx === 0; + return { + key: String(v.transaction_id), + // antd's `disabled: true` greys the item and blocks default + // click handling; combined with the inner div NOT having an + // onClick when current, the row becomes informational only. + disabled: isCurrent, + label: ( +
handleRestore(v)} + > +
+ #{v.version_number} — {formatChangeTitle(v.changes)} + {isCurrent && ( + + {t('(current)')} + + )} +
+
+ {formatUser(v.changed_by)} · {formatDate(v.issued_at)} +
+ {v.changes.length > 1 && ( +
    + {v.changes.slice(0, 5).map((c, i) => ( +
  • {summarizeChange(c)}
    • + ))} + {v.changes.length > 5 && ( +
  • + {t('+%(n)s more', { n: v.changes.length - 5 })} +
    • + )} +
+ )} +
+ ), + }; + }); + })(); + + return ( + { + if (open && versions === null && !loading) loadVersions(); + }} + > + + + + + + + ); +} diff --git a/superset-frontend/src/pages/DatasetList/VersionHistoryDropdown.tsx b/superset-frontend/src/pages/DatasetList/VersionHistoryDropdown.tsx new file mode 100644 index 000000000000..dbf1b4cea774 --- /dev/null +++ b/superset-frontend/src/pages/DatasetList/VersionHistoryDropdown.tsx @@ -0,0 +1,364 @@ +/** + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +// TEMP: Demo aid for sc-103156 entity-versioning. Lets a user open a +// dropdown of recent versions on a dataset and restore one. Not part +// of the merged feature scope (ADR-005 limits v1 to backend); revert +// before pushing the versioning branch. + +import { useState, useCallback } from 'react'; +import { SupersetClient } from '@superset-ui/core'; +import { t } from '@apache-superset/core/translation'; +import { Dropdown, Tooltip, Icons } from '@superset-ui/core/components'; + +interface Change { + kind: string; + // Per-record verb: add / remove / move / edit. Explicit instead of + // inferred from from_value / to_value null-tests or path[0]. + operation: string; + path: string[]; + from_value: unknown; + to_value: unknown; +} + +interface ChangedBy { + id: number; + username: string; + first_name: string; + last_name: string; +} + +interface Version { + version_uuid: string; + version_number: number; + transaction_id: number; + operation_type: string; + // Transaction-level avenue: restore / import / clone / null (= save). + action_kind: string | null; + issued_at: string; + changed_by: ChangedBy | null; + changes: Change[]; +} + +interface Props { + datasetUuid: string; + onRestored?: () => void; +} + +// Layout element kinds — used to decide whether a record's +// ``operation`` describes a node-level layout action vs a per-leaf +// change inside a layout-edit. Datasets don't normally produce layout +// records, but the formatter stays uniform across the three widgets. +const LAYOUT_KINDS = new Set([ + 'chart', + 'row', + 'column', + 'tab', + 'tabs', + 'header', + 'markdown', + 'divider', +]); + +// Localized labels for the kinds emitted by the backend (layout walker +// + dataset child diff). Defined statically so xgettext can extract them. +const KIND_LABELS: Record = { + chart: t('chart'), + row: t('row'), + column: t('column'), + tab: t('tab'), + tabs: t('tabs'), + header: t('header'), + markdown: t('markdown'), + divider: t('divider'), + metric: t('metric'), +}; +const localizedKind = (k: string): string => KIND_LABELS[k] ?? k; + +function summarizeChange(c: Change): string { + // Layout record at the node level: path = [node_id], operation+kind + // in columns. + if (LAYOUT_KINDS.has(c.kind) && c.path.length === 1) { + const kind = localizedKind(c.kind); + const payload = + ((c.to_value ?? c.from_value) as { name?: string } | null) ?? null; + const name = payload?.name; + if (c.operation === 'add') { + return name + ? t('Added %(kind)s "%(name)s"', { kind, name }) + : t('Added %(kind)s', { kind }); + } + if (c.operation === 'remove') { + return name + ? t('Removed %(kind)s "%(name)s"', { kind, name }) + : t('Removed %(kind)s', { kind }); + } + if (c.operation === 'move') { + return name + ? t('Moved %(kind)s "%(name)s"', { kind, name }) + : t('Moved %(kind)s', { kind }); + } + return name + ? t('Edited %(kind)s "%(name)s"', { kind, name }) + : t('Edited %(kind)s', { kind }); + } + + // Layout edit at the leaf level: path = [node_id, ...leaf-path]. + if (LAYOUT_KINDS.has(c.kind) && c.path.length >= 2) { + const kind = localizedKind(c.kind); + const leaf = String(c.path[c.path.length - 1]); + return t('Changed %(kind)s %(leaf)s', { kind, leaf }); + } + + const isAdd = c.operation === 'add'; + const isRemove = c.operation === 'remove'; + + if (c.path.length === 2 && (c.kind === 'column' || c.kind === 'metric')) { + const kind = localizedKind(c.kind); + const name = String(c.path[1]); + if (isAdd) return t('Added %(kind)s "%(name)s"', { kind, name }); + if (isRemove) return t('Removed %(kind)s "%(name)s"', { kind, name }); + return t('Changed %(kind)s "%(name)s"', { kind, name }); + } + + if (c.path[0] === 'slices') { + const id = String(c.path[1] ?? ''); + if (isAdd) return t('Added chart %(id)s', { id }).trim(); + if (isRemove) return t('Removed chart %(id)s', { id }).trim(); + return t('Changed chart %(id)s', { id }).trim(); + } + + if (c.kind === 'field') { + const fieldName = String(c.path[c.path.length - 1]); + const fieldLabel: string = + fieldName === 'dashboard_title' + ? t('title') + : fieldName === 'slice_name' + ? t('chart name') + : fieldName === 'table_name' + ? t('table name') + : fieldName; + const isShortScalar = + c.to_value !== null && + c.to_value !== undefined && + (typeof c.to_value === 'string' || + typeof c.to_value === 'number' || + typeof c.to_value === 'boolean') && + String(c.to_value).length <= 80; + if (!isAdd && !isRemove && isShortScalar) { + return t('Changed %(field)s to "%(value)s"', { + field: fieldLabel, + value: String(c.to_value), + }); + } + if (isRemove) { + return t('Cleared %(field)s', { field: fieldLabel }); + } + if (isAdd && isShortScalar) { + return t('Set %(field)s to "%(value)s"', { + field: fieldLabel, + value: String(c.to_value), + }); + } + if (isAdd) return t('Added %(field)s', { field: fieldLabel }); + if (isRemove) return t('Removed %(field)s', { field: fieldLabel }); + return t('Changed %(field)s', { field: fieldLabel }); + } + + const kind = localizedKind(c.kind); + if (c.path.length) { + const detail = String(c.path[c.path.length - 1]); + if (isAdd) return t('Added %(kind)s %(detail)s', { kind, detail }); + if (isRemove) return t('Removed %(kind)s %(detail)s', { kind, detail }); + return t('Changed %(kind)s %(detail)s', { kind, detail }); + } + if (isAdd) return t('Added %(kind)s', { kind }); + if (isRemove) return t('Removed %(kind)s', { kind }); + return t('Changed %(kind)s', { kind }); +} + +function formatChangeTitle(changes: Change[]): string { + if (!changes.length) return t('Baseline'); + const first = summarizeChange(changes[0]); + if (changes.length === 1) return first; + return t('%(first)s (+%(more)s more)', { + first, + more: changes.length - 1, + }); +} + +function formatUser(by: ChangedBy | null): string { + if (!by) return t('system'); + if (by.first_name || by.last_name) { + return `${by.first_name ?? ''} ${by.last_name ?? ''}`.trim(); + } + return by.username; +} + +function formatDate(iso: string): string { + try { + // Match the Superset locale set in src/views/App.tsx on + // ``document.documentElement.lang`` rather than the browser default. + const lang = document.documentElement.lang || undefined; + return new Date(iso).toLocaleString(lang); + } catch { + return iso; + } +} + +export default function VersionHistoryDropdown({ + datasetUuid, + onRestored, +}: Props) { + const [versions, setVersions] = useState(null); + const [loading, setLoading] = useState(false); + + const loadVersions = useCallback(async () => { + setLoading(true); + try { + const { json } = await SupersetClient.get({ + endpoint: `/api/v1/dataset/${datasetUuid}/versions/`, + }); + const result = (json as { result: Version[] }).result || []; + // Newest first (API returns oldest-first) + setVersions([...result].reverse().slice(0, 20)); + } catch (e) { + console.error('Failed to load versions', e); + setVersions([]); + } finally { + setLoading(false); + } + }, [datasetUuid]); + + const handleRestore = useCallback( + async (version: Version) => { + const summary = formatChangeTitle(version.changes); + if ( + // eslint-disable-next-line no-alert + !window.confirm( + t( + 'Restore this dataset to version %(num)s (%(summary)s)? This will overwrite the current state.', + { num: version.version_number, summary }, + ), + ) + ) { + return; + } + try { + await SupersetClient.post({ + endpoint: `/api/v1/dataset/${datasetUuid}/versions/${version.version_uuid}/restore`, + }); + // eslint-disable-next-line no-alert + window.alert(t('Restored. Reload the page to see the change.')); + if (onRestored) onRestored(); + } catch (e) { + console.error('Restore failed', e); + // eslint-disable-next-line no-alert + window.alert(t('Restore failed — see browser console for details.')); + } + }, + [datasetUuid, onRestored], + ); + + const items = (() => { + if (loading) { + return [{ key: 'loading', label: t('Loading…'), disabled: true }]; + } + if (!versions) { + return [ + { key: 'empty', label: t('Click to load versions'), disabled: true }, + ]; + } + if (versions.length === 0) { + return [{ key: 'empty', label: t('No versions yet'), disabled: true }]; + } + return versions.map((v, idx) => { + const isCurrent = idx === 0; + return { + key: String(v.transaction_id), + disabled: isCurrent, + label: ( +
handleRestore(v)} + > +
+ #{v.version_number} — {formatChangeTitle(v.changes)} + {isCurrent && ( + + {t('(current)')} + + )} +
+
+ {formatUser(v.changed_by)} · {formatDate(v.issued_at)} +
+ {v.changes.length > 1 && ( +
    + {v.changes.slice(0, 5).map((c, i) => ( +
  • {summarizeChange(c)}
    • + ))} + {v.changes.length > 5 && ( +
  • + {t('+%(n)s more', { n: v.changes.length - 5 })} +
    • + )} +
+ )} +
+ ), + }; + }); + })(); + + return ( + { + if (open && versions === null && !loading) loadVersions(); + }} + > + + + + + + + ); +} diff --git a/superset-frontend/src/views/routes.tsx b/superset-frontend/src/views/routes.tsx index 4f066e3ec2cb..ec6f44a54348 100644 --- a/superset-frontend/src/views/routes.tsx +++ b/superset-frontend/src/views/routes.tsx @@ -74,6 +74,14 @@ const DashboardList = lazy( import(/* webpackChunkName: "DashboardList" */ 'src/pages/DashboardList'), ); +// Throwaway: sc-107283 activity-view debug UI. Delete this lazy import + +// the route below + the src/pages/ActivityDebug directory when the +// feature ships. +const ActivityDebug = lazy( + () => + import(/* webpackChunkName: "ActivityDebug" */ 'src/pages/ActivityDebug'), +); + const Dashboard = lazy( () => import(/* webpackChunkName: "Dashboard" */ 'src/pages/Dashboard'), ); @@ -227,6 +235,11 @@ export const routes: Routes = [ path: '/dashboard/list/', Component: DashboardList, }, + // Throwaway: sc-107283 activity-view debug UI. + { + path: '/activity-debug/:resource/:uuid', + Component: ActivityDebug, + }, { path: '/superset/dashboard/:idOrSlug/', Component: Dashboard, diff --git a/superset/charts/api.py b/superset/charts/api.py index 97f456abd00c..fd59f779003f 100644 --- a/superset/charts/api.py +++ b/superset/charts/api.py @@ -144,6 +144,7 @@ def ensure_thumbnails_enabled(self) -> Optional[Response]: "warm_up_cache", "list_versions", "get_version", + "activity", } class_permission_name = "Chart" method_permission_name = MODEL_API_RW_METHOD_PERMISSION_MAP @@ -1336,7 +1337,7 @@ def list_versions(self, uuid_str: str) -> Response: 404: $ref: '#/components/responses/404' """ - return list_versions_endpoint(self, Slice, uuid_str, access_kwarg="chart") + return list_versions_endpoint(self, Slice, uuid_str) @expose( "//versions//", @@ -1393,6 +1394,82 @@ def get_version(self, uuid_str: str, version_uuid_str: str) -> Response: 404: $ref: '#/components/responses/404' """ - return get_version_endpoint( - self, Slice, uuid_str, version_uuid_str, access_kwarg="chart" - ) + return get_version_endpoint(self, Slice, uuid_str, version_uuid_str) + + @expose("//activity/", methods=("GET",)) + @protect() + @safe + @statsd_metrics + @event_logger.log_this_with_context( + action=lambda self, *args, **kwargs: f"{self.__class__.__name__}.activity", + log_to_statsd=False, + ) + def activity(self, uuid_str: str) -> Response: + """Return the cross-entity activity stream for a chart. + --- + get: + summary: Activity stream — chart own edits + datasets the + chart pointed at during association (sc-107283 US2) + parameters: + - in: path + schema: + type: string + format: uuid + name: uuid_str + description: Chart UUID + - in: query + schema: + type: string + format: date-time + name: since + - in: query + schema: + type: string + format: date-time + name: until + - in: query + schema: + type: string + enum: [self, related, all] + default: all + name: include + - in: query + schema: + type: string + name: q + description: >- + Case-insensitive search over the full history (summary, + entity name, kind, path, values) — applied before + pagination, so `count` reflects the matches. + - in: query + schema: + type: integer + minimum: 0 + default: 0 + name: page + - in: query + schema: + type: integer + minimum: 1 + maximum: 200 + default: 25 + name: page_size + responses: + 200: + description: Activity stream ordered newest-first + content: + application/json: + schema: ActivityResponseSchema + 400: + $ref: '#/components/responses/400' + 401: + $ref: '#/components/responses/401' + 403: + $ref: '#/components/responses/403' + 404: + $ref: '#/components/responses/404' + """ + # pylint: disable=import-outside-toplevel + from superset.versioning.activity import activity_endpoint + + return activity_endpoint(self, Slice, uuid_str, request.args) diff --git a/superset/constants.py b/superset/constants.py index 3525abc882c1..3ceede47bdb8 100644 --- a/superset/constants.py +++ b/superset/constants.py @@ -181,6 +181,7 @@ class RouteMethod: # pylint: disable=too-few-public-methods "restore": "write", "list_versions": "read", "get_version": "read", + "activity": "read", } EXTRA_FORM_DATA_APPEND_KEYS = { diff --git a/superset/dashboards/api.py b/superset/dashboards/api.py index 3aa4a2cc6872..1770bb443ede 100644 --- a/superset/dashboards/api.py +++ b/superset/dashboards/api.py @@ -264,6 +264,7 @@ class DashboardRestApi(CustomTagsOptimizationMixin, BaseSupersetModelRestApi): "export_as_example", "list_versions", "get_version", + "activity", } resource_name = "dashboard" allow_browser_login = True @@ -2348,9 +2349,7 @@ def list_versions(self, uuid_str: str) -> Response: 404: $ref: '#/components/responses/404' """ - return list_versions_endpoint( - self, Dashboard, uuid_str, access_kwarg="dashboard" - ) + return list_versions_endpoint(self, Dashboard, uuid_str) @expose( "//versions//", @@ -2407,6 +2406,85 @@ def get_version(self, uuid_str: str, version_uuid_str: str) -> Response: 404: $ref: '#/components/responses/404' """ - return get_version_endpoint( - self, Dashboard, uuid_str, version_uuid_str, access_kwarg="dashboard" - ) + return get_version_endpoint(self, Dashboard, uuid_str, version_uuid_str) + + @expose("//activity/", methods=("GET",)) + @protect() + @safe + @statsd_metrics + @event_logger.log_this_with_context( + action=lambda self, *args, **kwargs: f"{self.__class__.__name__}.activity", + log_to_statsd=False, + ) + def activity(self, uuid_str: str) -> Response: + """Return the cross-entity activity stream for a dashboard. + --- + get: + summary: Activity stream — dashboard own edits + transitive + chart-on-dashboard and dataset-via-chart edits, time-bounded + by association windows + parameters: + - in: path + schema: + type: string + format: uuid + name: uuid_str + description: Dashboard UUID + - in: query + schema: + type: string + format: date-time + name: since + description: Lower bound on issued_at (ISO 8601, UTC) + - in: query + schema: + type: string + format: date-time + name: until + description: Upper bound on issued_at (ISO 8601, UTC) + - in: query + schema: + type: string + enum: [self, related, all] + default: all + name: include + - in: query + schema: + type: string + name: q + description: >- + Case-insensitive search over the full history (summary, + entity name, kind, path, values) — applied before + pagination, so `count` reflects the matches. + - in: query + schema: + type: integer + minimum: 0 + default: 0 + name: page + - in: query + schema: + type: integer + minimum: 1 + maximum: 200 + default: 25 + name: page_size + responses: + 200: + description: Activity stream ordered newest-first + content: + application/json: + schema: ActivityResponseSchema + 400: + $ref: '#/components/responses/400' + 401: + $ref: '#/components/responses/401' + 403: + $ref: '#/components/responses/403' + 404: + $ref: '#/components/responses/404' + """ + # pylint: disable=import-outside-toplevel + from superset.versioning.activity import activity_endpoint + + return activity_endpoint(self, Dashboard, uuid_str, request.args) diff --git a/superset/datasets/api.py b/superset/datasets/api.py index 240ae278ba6f..9b36dcf1017d 100644 --- a/superset/datasets/api.py +++ b/superset/datasets/api.py @@ -125,6 +125,7 @@ class DatasetRestApi(BaseSupersetModelRestApi): "get_drill_info", "list_versions", "get_version", + "activity", } list_columns = [ "id", @@ -1609,9 +1610,7 @@ def list_versions(self, uuid_str: str) -> Response: 404: $ref: '#/components/responses/404' """ - return list_versions_endpoint( - self, SqlaTable, uuid_str, access_kwarg="datasource" - ) + return list_versions_endpoint(self, SqlaTable, uuid_str) @expose( "//versions//", @@ -1673,6 +1672,85 @@ def get_version(self, uuid_str: str, version_uuid_str: str) -> Response: 404: $ref: '#/components/responses/404' """ - return get_version_endpoint( - self, SqlaTable, uuid_str, version_uuid_str, access_kwarg="datasource" - ) + return get_version_endpoint(self, SqlaTable, uuid_str, version_uuid_str) + + @expose("//activity/", methods=("GET",)) + @protect() + @safe + @statsd_metrics + @event_logger.log_this_with_context( + action=lambda self, *args, **kwargs: f"{self.__class__.__name__}.activity", + log_to_statsd=False, + ) + def activity(self, uuid_str: str) -> Response: + """Return the activity stream for a dataset. + --- + get: + summary: Activity stream — dataset's own edits only (sc-107283 US3). + Per AV-004, datasets have no transitive layer in V2 — chart and + dashboard edits that touch this dataset do NOT appear here; + ``?include=related`` and ``?include=all`` collapse to the same + self-only stream as ``?include=self``. + parameters: + - in: path + schema: + type: string + format: uuid + name: uuid_str + description: Dataset UUID + - in: query + schema: + type: string + format: date-time + name: since + - in: query + schema: + type: string + format: date-time + name: until + - in: query + schema: + type: string + enum: [self, related, all] + default: all + name: include + - in: query + schema: + type: string + name: q + description: >- + Case-insensitive search over the full history (summary, + entity name, kind, path, values) — applied before + pagination, so `count` reflects the matches. + - in: query + schema: + type: integer + minimum: 0 + default: 0 + name: page + - in: query + schema: + type: integer + minimum: 1 + maximum: 200 + default: 25 + name: page_size + responses: + 200: + description: Activity stream ordered newest-first + content: + application/json: + schema: ActivityResponseSchema + 400: + $ref: '#/components/responses/400' + 401: + $ref: '#/components/responses/401' + 403: + $ref: '#/components/responses/403' + 404: + $ref: '#/components/responses/404' + """ + # pylint: disable=import-outside-toplevel + from superset.versioning.activity import activity_endpoint + + return activity_endpoint(self, SqlaTable, uuid_str, request.args) diff --git a/superset/initialization/__init__.py b/superset/initialization/__init__.py index a0b0d8ed6562..7a1493ec2a48 100644 --- a/superset/initialization/__init__.py +++ b/superset/initialization/__init__.py @@ -464,6 +464,12 @@ def init_views(self) -> None: appbuilder.add_view_no_menu(RedirectView) appbuilder.add_view_no_menu(RoleRestAPI) appbuilder.add_view_no_menu(UserInfoView) + # Throwaway: sc-107283 activity-view debug page. Delete this + # import + add_view_no_menu when the debug UI is no longer + # needed (see superset/views/activity_debug.py). + from superset.views.activity_debug import ActivityDebugView + + appbuilder.add_view_no_menu(ActivityDebugView) # # Add links diff --git a/superset/versioning/activity/__init__.py b/superset/versioning/activity/__init__.py new file mode 100644 index 000000000000..9d17483ab00a --- /dev/null +++ b/superset/versioning/activity/__init__.py @@ -0,0 +1,102 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Read-side queries for the cross-entity activity-view API (sc-107283). + +Companion to :mod:`superset.versioning.queries`. Whereas ``queries.py`` +returns transaction-level history for a single entity, the helpers +here unify change-record history across an entity's transitive +dependency chain — a dashboard's activity stream includes edits to +charts that were attached to it AND edits to datasets those charts +pointed at, each time-bounded by when the relationship was active. + +One public entry point — :func:`get_activity` — dispatches on the +first argument to serve all three endpoint families: + +* ``get_activity(Dashboard, dashboard_uuid, ...)`` — own edits + + charts attached during their dashboard window + datasets those + charts used during their chart window. +* ``get_activity(Slice, chart_uuid, ...)`` — own edits + datasets the + chart pointed at during association. +* ``get_activity(SqlaTable, dataset_uuid, ...)`` — own edits only. + Datasets are not transitive recipients of activity in V2. + +Package layout (descends from public entry point to leaf helpers): + +* :mod:`.orchestrator` — :func:`get_activity` (public), the + ``activity_endpoint`` REST helper, the request param parser + (:func:`parse_activity_query_params`), and the observability + instrumentation that T037/T038 specify. +* :mod:`.scope` — scope resolution (DB-touching): + :func:`resolve_scope` / :func:`_resolve_dashboard_scope` / + :func:`_resolve_chart_scope` / :func:`_resolve_related_scope`. +* :mod:`.windows` — pure window arithmetic on half-open + ``[start_tx, end_tx)`` intervals: :func:`intersect_windows` / + :func:`union_windows` / :func:`merge_entity_windows` / + :func:`row_within_any_window`. Extracted from :mod:`.scope` so + :mod:`.queries` can import the pure helpers at module-top instead + of through a cycle-dodging lazy import. +* :mod:`.queries` — every DB-touching helper: Phase A relationship + walks, Phase B change-record fetch, name denormalization, + path-entity resolution, and tombstone-state lookup. +* :mod:`.impact` — per-record impact-count computation (the only + field that requires its own batched query). +* :mod:`.visibility` — the AV-008 silent visibility filter; uses + the same SQL-side access filters FAB applies on list endpoints. +* :mod:`.render` — record-decoration helpers that turn raw rows into + the ActivityRecord DTO (summary headlines, ``changed_by`` projection, + uuid lookup). +* :mod:`.kinds` — the kind-translation tables, the ``Window`` / + ``EntityWindows`` type aliases, and :func:`load_shadow_model`. + +The public surface (re-exported here) is the eight symbols below. +Sub-module privates are intentionally NOT re-exported — tests and +new internal callers should import them from their owning submodule +(e.g. ``from superset.versioning.activity.windows import +intersect_windows``) so the package's public API stays scannable. + +``PathEntityResponseError`` and ``resolve_endpoint_path_entity`` are +re-exported here from :mod:`superset.versioning.api_helpers` (where +they live alongside the ``/versions/`` endpoint handlers) so the +three ``/activity/`` endpoint callers can ``from +superset.versioning.activity import resolve_endpoint_path_entity`` +without crossing into the ``/versions/`` module name. +""" + +from __future__ import annotations + +from superset.versioning.activity.kinds import EntityWindows, Window +from superset.versioning.activity.orchestrator import ( + activity_endpoint, + ActivityParamsError, + get_activity, + parse_activity_query_params, +) +from superset.versioning.api_helpers import ( + PathEntityResponseError, + resolve_endpoint_path_entity, +) + +__all__ = [ + "ActivityParamsError", + "EntityWindows", + "PathEntityResponseError", + "Window", + "activity_endpoint", + "get_activity", + "parse_activity_query_params", + "resolve_endpoint_path_entity", +] diff --git a/superset/versioning/activity/impact.py b/superset/versioning/activity/impact.py new file mode 100644 index 000000000000..651f81f10604 --- /dev/null +++ b/superset/versioning/activity/impact.py @@ -0,0 +1,160 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Per-record impact computation for the activity DTO. + +Only dashboard-path activity records pointing at a ``SqlaTable`` +related entity carry an ``impact`` field — the number of charts on +the dashboard at that transaction that were pointing at the dataset. +This module computes that count in a single batched query per +request: + +* :func:`collect_impact_pairs` — pulls the distinct + ``(dataset_id, transaction_id)`` pairs that need counts. +* :func:`batch_chart_counts` — one SQL query joining + ``dashboard_slices_version`` and ``slices_version`` to count + the matching charts validity-strategy-style. +* :func:`impact_for_record` — pure projection from the pre-fetched + counts onto each record (returns ``None`` for non-Dashboard paths + or non-SqlaTable kinds, matching data-model.md §"``impact`` + computation"). + +Splitting the count batching from the pure projection keeps the SQL +inside one function (the batched read) and the per-record decoration +inside another (no DB). +""" + +from __future__ import annotations + +from typing import Any + +import sqlalchemy as sa + +from superset.extensions import db +from superset.versioning.activity.kinds import TABLE_KIND_TO_API + + +def collect_impact_pairs( + records: list[dict[str, Any]], path_kind: str +) -> set[tuple[int, int]]: + """Distinct ``(dataset_id, transaction_id)`` pairs from *records* + that require an impact computation per data-model.md. + + Only dashboard-path records whose related entity is a ``SqlaTable`` + produce a non-null ``impact`` field; for any other shape this set + is empty and no DB query needs to fire. + """ + if path_kind != "Dashboard": + return set() + return { + (record["entity_id"], record["transaction_id"]) + for record in records + if TABLE_KIND_TO_API.get(record["entity_kind"]) == "SqlaTable" + } + + +def batch_chart_counts( + dashboard_id: int, pairs: set[tuple[int, int]] +) -> dict[tuple[int, int], int]: + """For every ``(dataset_id, target_tx)`` in *pairs*, count the + distinct charts that were both on *dashboard_id* and pointing at + *dataset_id* at *target_tx*. + + One SELECT against ``dashboard_slices_version`` ⨝ ``slices_version``, + pulling the (slice, dataset, validity-window) state for every slice + ever on the dashboard whose dataset matches one of the requested + dataset_ids. The Python loop then applies the validity-strategy + predicate per pair. Replaces the previous N+1 shape that fired one + COUNT per related record. + + Returns ``{(dataset_id, target_tx): count}``; pairs whose count + would be zero are omitted so the caller's ``.get(key, 0)`` is + correct. + """ + if not pairs: + return {} + + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.models.slice import Slice + + metadata = version_class(Slice).__table__.metadata + m2m_tbl = metadata.tables.get("dashboard_slices_version") + slices_tbl = version_class(Slice).__table__ + if m2m_tbl is None: + return {} + + dataset_ids = {dataset_id for dataset_id, _ in pairs} + stmt = sa.select( + m2m_tbl.c.slice_id, + slices_tbl.c.datasource_id, + m2m_tbl.c.transaction_id.label("m2m_start"), + m2m_tbl.c.end_transaction_id.label("m2m_end"), + slices_tbl.c.transaction_id.label("slice_start"), + slices_tbl.c.end_transaction_id.label("slice_end"), + ).where( + m2m_tbl.c.dashboard_id == dashboard_id, + m2m_tbl.c.operation_type != 2, + slices_tbl.c.id == m2m_tbl.c.slice_id, + slices_tbl.c.datasource_id.in_(dataset_ids), + slices_tbl.c.datasource_type == "table", + slices_tbl.c.operation_type != 2, + ) + rows = db.session.connection().execute(stmt).mappings().all() + + # For each pair, collect the slice_ids whose two validity windows + # both straddle target_tx. ``set`` dedupes within a pair. + matches: dict[tuple[int, int], set[int]] = {} + pairs_by_dataset: dict[int, list[int]] = {} + for dataset_id, target_tx in pairs: + pairs_by_dataset.setdefault(dataset_id, []).append(target_tx) + + for row in rows: + ds_id = row["datasource_id"] + for target_tx in pairs_by_dataset.get(ds_id, ()): + in_m2m = row["m2m_start"] <= target_tx and ( + row["m2m_end"] is None or row["m2m_end"] > target_tx + ) + in_slice = row["slice_start"] <= target_tx and ( + row["slice_end"] is None or row["slice_end"] > target_tx + ) + if in_m2m and in_slice: + matches.setdefault((ds_id, target_tx), set()).add(row["slice_id"]) + + return {pair: len(slice_ids) for pair, slice_ids in matches.items()} + + +def impact_for_record( + record: dict[str, Any], + path_kind: str, + counts: dict[tuple[int, int], int], +) -> dict[str, int] | None: + """Synthesize the ``impact`` field for one record using the pre- + fetched *counts* mapping. Pure function — no DB. + + Per data-model.md §"``impact`` computation": only + ``path=Dashboard`` and ``related=SqlaTable`` shapes carry an + impact; everything else returns ``None``. + """ + api_kind = TABLE_KIND_TO_API.get(record["entity_kind"]) + if path_kind != "Dashboard" or api_kind != "SqlaTable": + return None + key = (record["entity_id"], record["transaction_id"]) + chart_count = counts.get(key, 0) + if chart_count == 0: + return None + return {"charts": chart_count} diff --git a/superset/versioning/activity/kinds.py b/superset/versioning/activity/kinds.py new file mode 100644 index 000000000000..4c11a05ad13c --- /dev/null +++ b/superset/versioning/activity/kinds.py @@ -0,0 +1,186 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Kind translation tables, shared types, and the shadow-model loader. + +The activity-view module operates in two "kind" namespaces — the +table-stored value (``"chart"`` / ``"dashboard"`` / ``"dataset"``) that +``version_changes.entity_kind`` carries, and the Python class-name +form (``"Slice"`` / ``"Dashboard"`` / ``"SqlaTable"``) used internally +for dispatch and returned in the DTO's ``entity_kind`` field. The four +mappings here translate between them. Adjacent kind-keyed dicts live +here too: the per-kind human-readable label, the user-facing +lowercase form, and the 404 exception class. + +The :func:`load_shadow_model` helper exists in the same module +because each lookup is keyed on the same set of class names — keeping +it adjacent to the mappings makes the kind-translation surface +discoverable at a glance. +""" + +from __future__ import annotations + +from dataclasses import dataclass + +from flask_appbuilder import Model + +from superset.commands.chart.exceptions import ChartNotFoundError +from superset.commands.dashboard.exceptions import DashboardNotFoundError +from superset.commands.dataset.exceptions import DatasetNotFoundError +from superset.versioning.changes import ENTITY_KIND_BY_CLASS_NAME + +# ---- Kind translation ----------------------------------------------------- + +# ``version_changes.entity_kind`` stores the friendly downstream-tooling +# value (``"chart"``, ``"dashboard"``, ``"dataset"``) per sc-103156's +# ``ENTITY_KIND_BY_CLASS_NAME``. The activity-view DTO returns the +# Python class name instead (``"Slice"``, ``"Dashboard"``, +# ``"SqlaTable"``) so the contract aligns with ``__class__.__name__`` +# (data-model.md §"``ActivityRecord`` DTO"). Translate at the boundary. +TABLE_KIND_TO_API: dict[str, str] = { + table_kind: class_name + for class_name, table_kind in ENTITY_KIND_BY_CLASS_NAME.items() +} +API_KIND_TO_TABLE: dict[str, str] = dict(ENTITY_KIND_BY_CLASS_NAME) + +# Human-readable label for AV-012 summary headlines +# ("Dataset updated: Sales Transactions"). Keyed by the internal API kind +# (Python class name; matches ``model_cls.__name__``). +API_KIND_LABEL: dict[str, str] = { + "Dashboard": "Dashboard", + "Slice": "Chart", + "SqlaTable": "Dataset", +} + +# User-facing lowercase rendering of the kind. This is what appears in +# the JSON response's ``entity_kind`` field and the +# ``ActivityRecordSchema.entity_kind`` enum. Internal code keeps the +# Python class-name form because it matches ``model_cls.__name__`` and is +# convenient for dispatch — translation happens at serialization time +# only, in :func:`render.apply_record_decoration`. +USER_FACING_KIND: dict[str, str] = { + "Dashboard": "dashboard", + "Slice": "chart", + "SqlaTable": "dataset", +} + +# 404 exception class per API kind. Each accepts a string positional arg +# (the path-entity UUID) that gets formatted into the exception message. +NOT_FOUND_EXC: dict[str, type[Exception]] = { + "Dashboard": DashboardNotFoundError, + "Slice": ChartNotFoundError, + "SqlaTable": DatasetNotFoundError, +} + +# Per-API-kind (model class name, display column) used by +# ``_resolve_names_for_kind`` to read the user-facing entity name from +# the shadow table valid at a given transaction. +NAME_COLUMN: dict[str, tuple[str, str]] = { + "Dashboard": ("Dashboard", "dashboard_title"), + "Slice": ("Slice", "slice_name"), + "SqlaTable": ("SqlaTable", "table_name"), +} + + +# ---- Types ---------------------------------------------------------------- + + +@dataclass(frozen=True) +class Window: + """A validity window in Continuum transaction-id space, half-open as + ``[start_tx, end_tx)``. + + Immutable and equal-by-attributes — two windows with the same + ``start_tx`` / ``end_tx`` are interchangeable. Constructor rejects + ``end_tx <= start_tx``. ``end_tx = None`` means "open ended + (current)" and acts as positive infinity throughout the helpers. + + Helper methods (``contains`` / ``intersect`` / ``merges_with``) + live on the type so callers don't re-implement the half-open + predicate inline. Previously a ``tuple[int, int | None]`` alias; + promoted to a dataclass so a function accepting a ``Window`` can't + silently accept any other 2-tuple and so the constructor enforces + the half-open invariant. + """ + + start_tx: int + end_tx: int | None + + def __post_init__(self) -> None: + if self.end_tx is not None and self.end_tx <= self.start_tx: + raise ValueError( + f"Window end_tx must be > start_tx; " + f"got [{self.start_tx}, {self.end_tx})" + ) + + def contains(self, tx_id: int) -> bool: + """``True`` iff *tx_id* falls inside this half-open interval.""" + return self.start_tx <= tx_id and (self.end_tx is None or tx_id < self.end_tx) + + def intersect(self, other: Window) -> Window | None: + """Return the clipped overlap of this window with *other*, or + ``None`` when they are disjoint. ``end_tx = None`` acts as + positive infinity on either side.""" + start = max(self.start_tx, other.start_tx) + end: int | None + if self.end_tx is None: + end = other.end_tx + elif other.end_tx is None: + end = self.end_tx + else: + end = min(self.end_tx, other.end_tx) + if end is not None and end <= start: + return None + return Window(start, end) + + def merges_with(self, other: Window) -> bool: + """``True`` iff *self* and *other* overlap or touch (so their + union is one contiguous window). Assumes the caller has placed + them in start-ascending order.""" + if self.end_tx is None: + # self extends to +∞; everything past it merges in. + return True + return other.start_tx <= self.end_tx + + +#: A related-entity scope row: ``(api_kind, entity_id, [windows])``. +#: ``api_kind`` is the DTO-facing kind (``"Slice"``, etc.), not the +#: table-stored kind. Left as a tuple alias for now — the +#: ``(api_kind, entity_id)`` pair is logically a key with the window +#: list as its value, so a dict shape may fit better than a flat +#: dataclass when this is revisited. +EntityWindows = tuple[str, int, list[Window]] + + +def load_shadow_model(model_name: str) -> type[Model]: + """Inline-import a shadow model class by name. Deferred until call + time because the versioning package is initialised before all model + mappers are configured (same idiom used throughout + :mod:`superset.versioning.changes`).""" + # pylint: disable=import-outside-toplevel + if model_name == "Dashboard": + from superset.models.dashboard import Dashboard + + return Dashboard + if model_name == "Slice": + from superset.models.slice import Slice + + return Slice + if model_name == "SqlaTable": + from superset.connectors.sqla.models import SqlaTable + + return SqlaTable + raise LookupError(f"No shadow class registered for {model_name!r}") diff --git a/superset/versioning/activity/orchestrator.py b/superset/versioning/activity/orchestrator.py new file mode 100644 index 000000000000..7cfbd582b0a0 --- /dev/null +++ b/superset/versioning/activity/orchestrator.py @@ -0,0 +1,409 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Top-level orchestrator + query-param parsing + observability. + +This is the public entry point for the activity-view read path. One +function — :func:`get_activity` — dispatches on the path entity's +model class to assemble the cross-entity activity stream: + +1. ``resolve_path_entity`` (queries.py) — resolve UUID → live entity. +2. ``resolve_scope`` (scope.py) — build the related-entity window list. +3. ``fetch_change_records`` (queries.py) — pull rows from + ``version_changes`` joined with ``version_transaction`` and ``ab_user``. +4. ``filter_records_by_visibility`` (visibility.py) — silent AV-008 + drop of records the requester can't read. +5. ``apply_entity_name_denormalization`` (queries.py) — resolve entity names + from the shadow row valid at each record's transaction_id. +6. ``apply_record_decoration`` (render.py) — synthesize the ActivityRecord + DTO fields and strip internal-only columns. +7. Paginate in Python over the post-filter list. + +Parameter parsing for the REST endpoints lives here too — +:func:`parse_activity_query_params` is called by the three +``/activity/`` endpoint handlers before they call ``get_activity``. +Same for the observability instrumentation: ``_phase_timer`` and +``_emit_request_shape_attributes`` emit the metrics that T037/T038 +specify, on the same prefix the cross-coupling test pins. +""" + +from __future__ import annotations + +import contextlib +from collections.abc import Iterator +from datetime import datetime, timezone +from typing import Any +from uuid import UUID + +from flask import Response +from flask_appbuilder import Model + +from superset.utils import json +from superset.versioning.activity.kinds import EntityWindows +from superset.versioning.activity.queries import ( + apply_entity_name_denormalization, + fetch_change_records, + mark_first_tracked_saves, + resolve_path_entity, +) +from superset.versioning.activity.render import apply_record_decoration +from superset.versioning.activity.scope import resolve_scope +from superset.versioning.activity.visibility import filter_records_by_visibility +from superset.versioning.api_helpers import ( + PathEntityResponseError, + resolve_endpoint_path_entity, +) + +_DEFAULT_PAGE_SIZE = 25 +_MAX_PAGE_SIZE = 200 +_VALID_INCLUDE_VALUES: frozenset[str] = frozenset({"self", "related", "all"}) + + +class ActivityParamsError(ValueError): + """Raised by :func:`parse_activity_query_params` when a query param is + malformed. The endpoint catches this and maps to ``response_400``; + no other callers should depend on the exception type.""" + + +def parse_activity_query_params(args: Any) -> dict[str, Any]: + """Parse the ``since`` / ``until`` / ``include`` / ``page`` / ``page_size`` + query parameters into the kwargs ``get_activity`` accepts. + + Raises :class:`ActivityParamsError` (subclass of ``ValueError``) when + a parameter is malformed. Shared across the three endpoint families + (dashboards, charts, datasets) so the parsing and 400-messaging stay + consistent. + """ + params: dict[str, Any] = { + "include": _parse_include(args.get("include", "all")), + "page": _parse_page(args.get("page", "0")), + "page_size": _parse_page_size(args.get("page_size")), + } + if (since := _parse_optional_iso(args.get("since"), name="since")) is not None: + params["since"] = since + if (until := _parse_optional_iso(args.get("until"), name="until")) is not None: + params["until"] = until + if q := (args.get("q") or "").strip(): + params["q"] = q + return params + + +def _parse_optional_iso(raw: str | None, *, name: str) -> datetime | None: + """Parse a missing-or-ISO-datetime field; ``None`` for missing, + ``ActivityParamsError`` for malformed.""" + if not raw: + return None + parsed = _parse_iso_datetime(raw) + if parsed is None: + raise ActivityParamsError(f"Invalid {name!r} datetime: {raw!r}") + return parsed + + +def _parse_include(value: str) -> str: + if value not in _VALID_INCLUDE_VALUES: + raise ActivityParamsError( + f"Invalid 'include' value: {value!r}; " + f"must be one of {sorted(_VALID_INCLUDE_VALUES)}" + ) + return value + + +def _parse_page(raw: str) -> int: + try: + value = int(raw) + except (TypeError, ValueError) as exc: + raise ActivityParamsError(f"Invalid 'page' value: {raw!r}") from exc + if value < 0: + raise ActivityParamsError("Invalid 'page' value: must be >= 0") + return value + + +def _parse_page_size(raw: str | None) -> int: + """``page_size`` honours the default when missing, raises when invalid, + and silently clamps to ``_MAX_PAGE_SIZE`` (so ``?page_size=500`` + returns 200 records instead of a 400).""" + if raw is None: + return _DEFAULT_PAGE_SIZE + try: + value = int(raw) + except (TypeError, ValueError) as exc: + raise ActivityParamsError(f"Invalid 'page_size' value: {raw!r}") from exc + if value < 1: + raise ActivityParamsError("Invalid 'page_size' value: must be >= 1") + return min(value, _MAX_PAGE_SIZE) + + +def _parse_iso_datetime(value: str) -> datetime | None: + """Parse an ISO-8601 datetime string. Tolerates the trailing ``Z`` + suffix that Python <3.11 ``fromisoformat`` rejects, and normalises any + timezone-aware result to naive UTC. + + The ``since`` / ``until`` filters bind directly against + ``version_transaction.issued_at``, which is ``sa.DateTime()`` — a + timezone-*naive* column (UTC by convention). Binding a tz-aware value + against it shifts the comparison by the session offset on PostgreSQL + (and raises on some drivers), so collapse aware inputs to naive UTC + here. Naive inputs pass through unchanged (already treated as UTC). + """ + candidate = value[:-1] + "+00:00" if value.endswith("Z") else value + try: + parsed = datetime.fromisoformat(candidate) + except ValueError: + return None + if parsed.tzinfo is not None: + parsed = parsed.astimezone(timezone.utc).replace(tzinfo=None) + return parsed + + +def _record_matches(record: dict[str, Any], q: str) -> bool: + """Case-insensitive substring match for the ``q`` search filter, + over the human-meaningful surfaces of a decorated activity record: + ``summary``, ``entity_name``, ``kind``, the joined ``path`` segments, + and the JSON form of ``from_value`` / ``to_value`` (JSON, not Python + ``str()``: the client searches the serialized text it renders, so + ``false`` / ``null`` / double-quoted keys must match — and falsy + values like ``False`` / ``0`` must not collapse to unsearchable + empty strings). + """ + + def _value_text(value: Any) -> str: + if value is None: + return "" + try: + return json.dumps(value) + except (TypeError, ValueError): + return str(value) + + needle = q.lower() + haystacks = ( + record.get("summary") or "", + record.get("entity_name") or "", + record.get("kind") or "", + " ".join(str(seg) for seg in (record.get("path") or [])), + _value_text(record.get("from_value")), + _value_text(record.get("to_value")), + ) + return any(needle in h.lower() for h in haystacks) + + +def get_activity( + model_cls: type[Model], + entity_uuid: UUID, + *, + since: datetime | None = None, + until: datetime | None = None, + include: str = "all", + q: str | None = None, + page: int = 0, + page_size: int = _DEFAULT_PAGE_SIZE, +) -> tuple[list[dict[str, Any]], int, bool]: + """Cross-entity activity stream for one path entity. + + Single polymorphic entry point. Dispatches on *model_cls* to + assemble the path entity's self records plus the transitive related- + entity records (charts attached to a dashboard, datasets a chart + pointed at, etc.) per data-model.md §"Query phases". + + Returns ``(records, total_count, truncated)``. ``truncated`` is + ``True`` when the per-request fetch ceiling + (``queries._MAX_FETCHED_RECORDS``) bit — older records exist beyond + what was materialized, so ``count`` is a floor, not the absolute + total. The count is post-visibility + (AV-008), post-include-filter, and — when ``q`` is supplied — post- + search-filter (``count`` reflects the matches, the contract the + server-side search exists to provide), not just the size of the + returned slice — clients paginate by passing ``page`` forward until + ``page * page_size >= count``. + + Raises ``DashboardNotFoundError`` / ``ChartNotFoundError`` / + ``DatasetNotFoundError`` when the path entity doesn't exist (AV-009). + """ + _path_entity, path_id = resolve_path_entity(model_cls, entity_uuid) + path_kind = model_cls.__name__ + kind_key = path_kind.lower() # "dashboard" / "slice" / "sqlatable" + + with _phase_timer(kind_key, "relationship_resolution_ms"): + entity_windows = resolve_scope(path_kind, path_id, include) + if not entity_windows: + _emit_request_shape_attributes( + kind_key, + include=include, + has_since_filter=since is not None, + page_size=page_size, + record_count=0, + entity_windows=[], + path_kind=path_kind, + path_id=path_id, + ) + return [], 0, False + + # Visibility filter runs before decoration: it needs the raw + # ``entity_id`` column (which decoration strips), and dropping + # invisible records early means we don't pay for name lookup + + # tombstone probes + impact counts on records the requester + # can't see (AV-008's silent-filter contract). + with _phase_timer(kind_key, "fetch_ms"): + records, truncated = fetch_change_records(entity_windows, since, until) + with _phase_timer(kind_key, "visibility_filter_ms"): + records = filter_records_by_visibility(records) + with _phase_timer(kind_key, "denormalize_ms"): + apply_entity_name_denormalization(records) + # Runs post-visibility (fewer entities to probe) and pre- + # decoration (needs the raw table-form entity_kind/entity_id + # that decoration rewrites). + mark_first_tracked_saves(records) + with _phase_timer(kind_key, "decorate_ms"): + apply_record_decoration(records, path_kind, path_id) + + # Server-side search (PR #40988 feedback: the panel's client-side + # search only covers loaded pages). Applied post-decoration so the + # synthesized ``summary`` / ``entity_name`` participate, and pre- + # count so pagination paginates the MATCHES — the full record set + # is already materialized in Python (the documented AV-008 design), + # so the filter adds no extra query. + if q: + records = [r for r in records if _record_matches(r, q)] + + total = len(records) + bounded_size = max(1, min(page_size, _MAX_PAGE_SIZE)) + offset = max(0, page) * bounded_size + + _emit_request_shape_attributes( + kind_key, + include=include, + has_since_filter=since is not None, + page_size=bounded_size, + record_count=total, + entity_windows=entity_windows, + path_kind=path_kind, + path_id=path_id, + ) + + return records[offset : offset + bounded_size], total, truncated + + +def activity_endpoint( + api: Any, model_cls: type[Model], uuid_str: str, request_args: Any +) -> Response: + """Body of ``GET /api/v1/{resource}//activity/``. + + Same shape as :func:`superset.versioning.api_helpers.list_versions_endpoint` + for the ``/versions/`` endpoint family. Resolves the path entity, + parses the request query params, runs :func:`get_activity`, and + wraps the result through ``ActivityResponseSchema``. + + *api* is the FAB ``ModelRestApi`` instance (pass ``self`` from the + endpoint method). *request_args* is ``request.args`` from + ``flask.request`` — passed explicitly so the helper is testable + without a live Flask context. + """ + # pylint: disable=import-outside-toplevel + from superset.versioning.schemas import ActivityResponseSchema + + try: + entity, _ = resolve_endpoint_path_entity(api, model_cls, uuid_str) + except PathEntityResponseError as exc: + return exc.response + + try: + params = parse_activity_query_params(request_args) + except ActivityParamsError as exc: + return api.response_400(message=str(exc)) + + records, count, truncated = get_activity(model_cls, entity.uuid, **params) + payload = ActivityResponseSchema().dump( + {"result": records, "count": count, "truncated": truncated} + ) + return api.response(200, **payload) + + +# ---- Observability (T037 / T038) ------------------------------------------ + +#: Common prefix for every metric this module emits. Per plan §D-17. +_METRIC_PREFIX = "superset.activity_view" + + +@contextlib.contextmanager +def _phase_timer(kind_key: str, phase: str) -> Iterator[None]: + """Time the wrapped block and emit + ``superset.activity_view..`` to ``stats_logger_manager``. + Wrapper around :func:`superset.utils.decorators.stats_timing` that + centralises the key construction. + """ + # pylint: disable=import-outside-toplevel + from superset.extensions import stats_logger_manager + from superset.utils.decorators import stats_timing + + with stats_timing( + f"{_METRIC_PREFIX}.{kind_key}.{phase}", + stats_logger_manager.instance, + ): + yield + + +def _emit_request_shape_attributes( + kind_key: str, + *, + include: str, + has_since_filter: bool, + page_size: int, + record_count: int, + entity_windows: list[EntityWindows], + path_kind: str, + path_id: int, +) -> None: + """Emit non-PII shape counters about the request and its result set. + + Per T038: include_mode / has_since_filter / page_size / record_count + + per-related-kind entity counts. **No PII**: entity names, diff + content, user identifiers — none of those reach the metric layer. + The counters use ``incr`` (counters) since they're tags, not + latencies; the timing keys above carry the latency dimension. + """ + # pylint: disable=import-outside-toplevel + from superset.extensions import stats_logger_manager + + sl = stats_logger_manager.instance + + # Tag-style metrics: one counter per attribute value. The statsd + # bridge accepts arbitrary strings; downstream dashboards filter by + # the value segment. + sl.incr(f"{_METRIC_PREFIX}.{kind_key}.requests.include_{include}") + sl.incr( + f"{_METRIC_PREFIX}.{kind_key}.requests." + f"has_since_filter_{'true' if has_since_filter else 'false'}" + ) + sl.gauge(f"{_METRIC_PREFIX}.{kind_key}.page_size", float(page_size)) + sl.gauge(f"{_METRIC_PREFIX}.{kind_key}.record_count", float(record_count)) + + # Per-related-kind entity counts (T038 explicit fields). The scope + # list includes the path entity itself (the "self" window); exclude + # it so the gauge reflects only the *related* entities the request + # fanned out to, not "this request touched itself". + by_kind: dict[str, int] = {"Slice": 0, "SqlaTable": 0, "Dashboard": 0} + for api_kind, entity_id, _windows in entity_windows: + if (api_kind, entity_id) == (path_kind, path_id): + continue + if api_kind in by_kind: + by_kind[api_kind] += 1 + sl.gauge( + f"{_METRIC_PREFIX}.{kind_key}.related_entity_count.charts", + float(by_kind["Slice"]), + ) + sl.gauge( + f"{_METRIC_PREFIX}.{kind_key}.related_entity_count.datasets", + float(by_kind["SqlaTable"]), + ) diff --git a/superset/versioning/activity/queries.py b/superset/versioning/activity/queries.py new file mode 100644 index 000000000000..a84b7352dc6f --- /dev/null +++ b/superset/versioning/activity/queries.py @@ -0,0 +1,620 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""DB-touching helpers for the activity-view read path. + +All Phase A relationship walks (``charts_attached_to_dashboard``, +``datasets_used_by_chart``, ``batch_datasets_used_by_charts``), +the Phase B change-record fetch (``fetch_change_records`` / +``_select_change_rows_for_kinds``), the name-denormalization helpers +(``_resolve_names_for_kind`` / ``apply_entity_name_denormalization``), the +path-entity resolution helper (``resolve_path_entity``), and the +tombstone-state lookup (``check_entity_tombstones``) live here. + +Each helper is a thin SELECT-and-shape function — no orchestration, +no business logic. Callers in :mod:`scope`, :mod:`render`, and +:mod:`orchestrator` compose them into the end-to-end request. + +**Inline imports.** Continuum's ``version_class`` / ``versioning_manager`` +and the Superset model classes are imported inside each helper because +this package is loaded from ``init_versioning()`` before all SQLAlchemy +mappers are configured. +""" + +from __future__ import annotations + +from collections.abc import Iterator +from datetime import datetime +from typing import Any +from uuid import UUID + +import sqlalchemy as sa +from flask_appbuilder import Model + +from superset.extensions import db +from superset.versioning.activity.kinds import ( + API_KIND_TO_TABLE, + EntityWindows, + load_shadow_model, + NAME_COLUMN, + NOT_FOUND_EXC, + TABLE_KIND_TO_API, + Window, +) +from superset.versioning.activity.windows import row_within_any_window +from superset.versioning.changes import version_changes_table + +# ---- Path-entity resolution ----------------------------------------------- + + +def resolve_path_entity(model_cls: type[Model], entity_uuid: UUID) -> tuple[Any, int]: + """Resolve *entity_uuid* to ``(live_entity, entity_id)`` or raise a + typed 404 per AV-009. + + Soft-delete handling (sc-103157) is inherited transparently from + :func:`superset.versioning.queries.find_active_by_uuid` once it + learns to filter out ``deleted_at IS NOT NULL`` rows; at that point + soft-deleted paths will also raise here. + """ + # pylint: disable=import-outside-toplevel + from superset.versioning.queries import find_active_by_uuid + + entity = find_active_by_uuid(model_cls, entity_uuid) + if entity is None: + api_kind = model_cls.__name__ + exc_cls = NOT_FOUND_EXC.get(api_kind) + if exc_cls is None: + raise LookupError( + f"Activity view does not support model class {api_kind!r}" + ) + raise exc_cls(str(entity_uuid)) + return entity, entity.id + + +# ---- Phase A: relationship-traversal queries ------------------------------ + + +def charts_attached_to_dashboard(dashboard_id: int) -> list[tuple[int, Window]]: + """Return ``(slice_id, window)`` for every chart that has ever been on + *dashboard_id*, with each association's validity window in + transaction-id space. + + Reads from ``dashboard_slices_version`` (Continuum's auto-generated + M2M shadow). Rows with ``operation_type = 2`` (DELETE) are excluded + so we don't synthesize a phantom window from a detachment row. + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.models.dashboard import Dashboard + + metadata = version_class(Dashboard).__table__.metadata + m2m_tbl = metadata.tables.get("dashboard_slices_version") + if m2m_tbl is None: + return [] + + rows = ( + db.session.connection() + .execute( + sa.select( + m2m_tbl.c.slice_id, + m2m_tbl.c.transaction_id, + m2m_tbl.c.end_transaction_id, + ).where( + m2m_tbl.c.dashboard_id == dashboard_id, + m2m_tbl.c.operation_type != 2, + m2m_tbl.c.slice_id.is_not(None), + ) + ) + .all() + ) + return [(row[0], Window(row[1], row[2])) for row in rows] + + +def datasets_used_by_chart(slice_id: int) -> list[tuple[int, Window]]: + """Return ``(datasource_id, window)`` for every dataset that *slice_id* + has ever pointed at, with each association's validity window. + + Single-slice form, used by ``_resolve_chart_scope`` where there + is only one chart to walk. The dashboard-scope path calls + :func:`batch_datasets_used_by_charts` instead so the query fires + once for all slices on the dashboard, not once per slice. + + Reads from ``slices_version`` (the chart parent shadow). Filters to + ``datasource_type = 'table'`` because the activity view only follows + the chart → ``SqlaTable`` dependency edge (not legacy/other + datasources). Rows with ``operation_type = 2`` are excluded. + """ + return batch_datasets_used_by_charts({slice_id}).get(slice_id, []) + + +def batch_datasets_used_by_charts( + slice_ids: set[int], +) -> dict[int, list[tuple[int, Window]]]: + """Batch form of :func:`datasets_used_by_chart`. Returns + ``{slice_id: [(dataset_id, window), ...]}`` in a single query so the + dashboard-scope walker doesn't fire one query per chart on the + dashboard. The previous per-slice shape became O(n_charts) round- + trips, which dominated ``get_activity`` latency on dashboards with + rich history (profile run 2026-05-26 showed `resolve_scope` + accounting for ~1.9s out of 4s p95). + """ + if not slice_ids: + return {} + + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.models.slice import Slice + + slices_tbl = version_class(Slice).__table__ + rows = ( + db.session.connection() + .execute( + sa.select( + slices_tbl.c.id, + slices_tbl.c.datasource_id, + slices_tbl.c.transaction_id, + slices_tbl.c.end_transaction_id, + ).where( + slices_tbl.c.id.in_(slice_ids), + slices_tbl.c.datasource_type == "table", + slices_tbl.c.operation_type != 2, + slices_tbl.c.datasource_id.is_not(None), + ) + ) + .mappings() + .all() + ) + grouped: dict[int, list[tuple[int, Window]]] = {} + for row in rows: + grouped.setdefault(row["id"], []).append( + ( + row["datasource_id"], + Window(row["transaction_id"], row["end_transaction_id"]), + ) + ) + return grouped + + +# ---- Phase B: change-record fetch ----------------------------------------- + + +def fetch_change_records( + entity_window_tuples: list[EntityWindows], + since: datetime | None, + until: datetime | None, +) -> tuple[list[dict[str, Any]], bool]: + """Fetch all ``version_changes`` rows matching any of the supplied + entity-window tuples, joined with ``version_transaction`` for + ``issued_at`` and ``user_id``. + + Each tuple is ``(api_kind, entity_id, [(start_tx, end_tx), ...])``; + a record matches when ``entity_kind`` equals the table-stored form + of *api_kind*, ``entity_id`` matches, and ``transaction_id`` falls + inside at least one of the entity's windows. ``since``/``until`` + further restrict by ``issued_at``. + + Implementation: one SELECT per kind with ``entity_id IN (...)`` and + a wide ``transaction_id`` bound (the union of all windows for that + kind). Per-window precision is applied in Python afterward. This + keeps the SQL shape proportional to the number of *kinds* (≤3) and + the bound proportional to the union of windows, not the cross- + product of (entity, window) — which previously generated one OR + clause per (entity, window) pair and hit SQLite's + ``SQLITE_MAX_EXPR_DEPTH`` limit on dashboards with many slices + or many historical attachment windows. + + Per AV-008 the visibility filter runs after this function (records + the requester can't read are silently dropped and must not + contribute to ``count``), so the orchestrator paginates in Python + over the filtered list — there is no DB-level page ``OFFSET`` here. + There IS a per-statement safety ``LIMIT`` (``_MAX_FETCHED_RECORDS``) + that bounds how much history a single request materializes; when it + bites, the second return value is ``True`` and the caller surfaces + ``truncated`` on the response. + + Returns ``(records, truncated)``. Records are ordered by + ``(issued_at DESC, transaction_id DESC, sequence DESC)`` — the + secondary keys break ties for AV-006's stable-ordering contract. + """ + if not entity_window_tuples: + return [], False + + # Group windows by (table_kind, entity_id) and by table_kind for SQL + # narrowing. The fetch is per-kind; the post-filter is per-entity. + windows_by_entity: dict[tuple[str, int], list[Window]] = {} + ids_by_kind: dict[str, set[int]] = {} + for api_kind, entity_id, windows in entity_window_tuples: + table_kind = API_KIND_TO_TABLE.get(api_kind) + if table_kind is None or not windows: + continue + ids_by_kind.setdefault(table_kind, set()).add(entity_id) + windows_by_entity.setdefault((table_kind, entity_id), []).extend(windows) + + if not ids_by_kind: + return [], False + + rows, truncated = _select_change_rows_for_kinds( + ids_by_kind, since, until, _MAX_FETCHED_RECORDS + ) + filtered = [ + row + for row in rows + if row_within_any_window( + row, windows_by_entity.get((row["entity_kind"], row["entity_id"]), []) + ) + ] + filtered.sort( + key=lambda r: (r["issued_at"], r["transaction_id"], r["sequence"]), + reverse=True, + ) + return filtered, truncated + + +def _select_change_rows_for_kinds( + ids_by_kind: dict[str, set[int]], + since: datetime | None, + until: datetime | None, + limit: int, +) -> tuple[list[dict[str, Any]], bool]: + """Fire one SELECT per entity_kind with ``entity_id IN (...)``; + concatenate the results. Each SELECT joins ``version_transaction`` + + ``ab_user`` so the orchestrator has the columns it needs for + decoration. + + Per-kind, not one query: SQLAlchemy's ``tuple_(entity_kind, + entity_id).in_(...)`` would collapse the three queries into one, + but its SQL emission is not portable across Postgres, MySQL, and + SQLite. The per-kind shape is the correct trade-off given + Superset's multi-dialect requirement (at most 3 round-trips per + request, bounded by the kind taxonomy). Do not "optimise" into a + composite-tuple IN clause without verifying the SQL on all three + dialects. + + **Init-order dependency.** ``tx_tbl.c.action_kind`` resolves only + after ``init_versioning()`` has run — the column is appended onto + Continuum's transaction Table by + ``superset.versioning.factory.VersionTransactionFactory`` at app + start via ``append_column`` + ``add_property``. This helper is + safe to call from request-path code because the app is fully + initialised by then; calling it from a script that imports the + versioning package without going through ``init_versioning()`` + will raise ``AttributeError`` on the ``action_kind`` attribute + access below.""" + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import versioning_manager + + from superset import security_manager + + tx_tbl = versioning_manager.transaction_cls.__table__ + user_tbl = security_manager.user_model.__table__ + vc = version_changes_table + join_tree = vc.join(tx_tbl, vc.c.transaction_id == tx_tbl.c.id).outerjoin( + user_tbl, tx_tbl.c.user_id == user_tbl.c.id + ) + select_cols = ( + vc.c.transaction_id, + vc.c.entity_kind, + vc.c.entity_id, + vc.c.sequence, + vc.c.kind, + vc.c.operation, + vc.c.path, + vc.c.from_value, + vc.c.to_value, + tx_tbl.c.issued_at, + tx_tbl.c.user_id, + # ``action_kind`` is the high-level avenue (restore / import / + # clone / NULL=ordinary save) stamped by the originating + # command via the change-record listener. All records sharing a + # ``transaction_id`` share the same value. The column is + # declared on the Continuum Table by ``VersionTransactionFactory``, + # so ``tx_tbl.c.action_kind`` resolves cleanly here. See + # sc-103156 data-model.md §"Three dimensions". + tx_tbl.c.action_kind, + user_tbl.c.id.label("changed_by_id"), + user_tbl.c.first_name, + user_tbl.c.last_name, + ) + + out: list[dict[str, Any]] = [] + truncated = False + for table_kind, entity_ids in ids_by_kind.items(): + # Chunk ``entity_ids`` to stay inside SQLite's + # ``SQLITE_MAX_VARIABLE_NUMBER`` floor (default 999, raised to + # 32766 in 3.32+ but the older limit ships in many builds). The + # bind count grows linearly with chart-on-dashboard count; a + # dashboard built from a huge chart library can reach the floor. + # Postgres + MySQL accept the full list, but the chunk is + # dialect-agnostic for simplicity. + for chunk in _chunked_ids(entity_ids, _ENTITY_ID_CHUNK_SIZE): + stmt = ( + sa.select(*select_cols) + .select_from(join_tree) + .where( + vc.c.entity_kind == table_kind, + vc.c.entity_id.in_(chunk), + ) + ) + if since is not None: + stmt = stmt.where(tx_tbl.c.issued_at >= since) + if until is not None: + stmt = stmt.where(tx_tbl.c.issued_at < until) + # Bounded fetch: cap each statement at the most-recent + # ``limit`` rows so a path entity with very long history (or a + # dashboard with many related charts/datasets) can't + # materialize an unbounded result set in Python. The same + # ordering keys as ``fetch_change_records``' final sort make + # the cap take the newest records; if a statement returns a + # full ``limit``, older records exist beyond it and the caller + # surfaces ``truncated`` on the response. + stmt = stmt.order_by( + tx_tbl.c.issued_at.desc(), + vc.c.transaction_id.desc(), + vc.c.sequence.desc(), + ).limit(limit) + rows = [ + dict(row) + for row in db.session.connection().execute(stmt).mappings().all() + ] + if len(rows) >= limit: + truncated = True + out.extend(rows) + return out, truncated + + +# Bind-parameter floor: see ``_select_change_rows_for_kinds`` docstring. +# 500 leaves room for the two literal-string filters and the optional +# since/until datetime params. +_ENTITY_ID_CHUNK_SIZE = 500 + +# Per-statement safety ceiling on how many change rows a single activity +# request will materialize (per kind-chunk). Bounds memory/CPU for a path +# entity with very long history or many related entities; when a statement +# returns a full ``_MAX_FETCHED_RECORDS`` the response is flagged +# ``truncated`` so clients know older records exist beyond the window. +_MAX_FETCHED_RECORDS = 5000 + + +def _chunked_ids(ids: set[int], size: int) -> Iterator[list[int]]: + """Yield *ids* in fixed-size lists. Final chunk may be smaller.""" + items = list(ids) + for i in range(0, len(items), size): + yield items[i : i + size] + + +def mark_first_tracked_saves(records: list[dict[str, Any]]) -> None: + """Set ``first_tracked_save`` on each record in place: ``True`` when + the record's transaction is the entity's FIRST UPDATE (op=1) in its + shadow table. + + The first save of an entity that predates versioning replays every + params-normalization delta against the retroactive baseline — a + legacy chart's first Explore save produced ~74 records in one + transaction (version-history UI feedback, PR #40988). The server + can't distinguish "normalization" from "the user changed 74 things", + but it CAN say "this was the entity's first tracked save"; clients + use the marker to collapse such transactions. + + One ``GROUP BY`` query per kind (≤3), chunked like the record fetch. + Shadow rows are matched on ``(id, uuid)`` against the live row — a + bare ``id`` match would inherit a previously hard-deleted entity's + history under id reuse (SQLite/MySQL reuse ``max(id)+1``) and mark + the wrong transaction. Consequence: hard-deleted entities (no live + row) and NULL-uuid shadow rows never get a ``True`` marker — their + records always carry ``first_tracked_save=False``. Mutates *records* + in place — same contract as the other decoration passes in + :mod:`superset.versioning.activity.render`. + """ + if not records: + return + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + ids_by_kind: dict[str, set[int]] = {} + for r in records: + ids_by_kind.setdefault(r["entity_kind"], set()).add(r["entity_id"]) + + first_tx_by_entity: dict[tuple[str, int], int] = {} + for table_kind, entity_ids in ids_by_kind.items(): + model_name = TABLE_KIND_TO_API.get(table_kind) + if model_name is None: + continue + live_model = load_shadow_model(model_name) + live_tbl = live_model.__table__ + shadow_tbl = version_class(live_model).__table__ + for chunk in _chunked_ids(entity_ids, _ENTITY_ID_CHUNK_SIZE): + stmt = ( + sa.select( + shadow_tbl.c.id, + sa.func.min(shadow_tbl.c.transaction_id), + ) + .select_from( + shadow_tbl.join( + live_tbl, + sa.and_( + shadow_tbl.c.id == live_tbl.c.id, + shadow_tbl.c.uuid == live_tbl.c.uuid, + ), + ) + ) + .where( + shadow_tbl.c.operation_type == 1, + shadow_tbl.c.id.in_(chunk), + ) + .group_by(shadow_tbl.c.id) + ) + for entity_id, min_tx in db.session.connection().execute(stmt): + first_tx_by_entity[(table_kind, entity_id)] = min_tx + + for r in records: + r["first_tracked_save"] = ( + first_tx_by_entity.get((r["entity_kind"], r["entity_id"])) + == r["transaction_id"] + ) + + +# ---- Name denormalization ------------------------------------------------- + + +def _resolve_names_for_kind( + api_kind: str, pairs: set[tuple[int, int]] +) -> dict[tuple[int, int], str]: + """For one entity kind, return ``{(entity_id, target_tx): name}`` from + the shadow row valid at *target_tx* (validity-strategy predicate). + Empty mapping when the kind has no name column registered. + """ + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + if api_kind not in NAME_COLUMN: + return {} + + model_name, name_col = NAME_COLUMN[api_kind] + model_cls = load_shadow_model(model_name) + shadow_tbl = version_class(model_cls).__table__ + ids = sorted({eid for eid, _ in pairs}) + rows = ( + db.session.connection() + .execute( + sa.select( + shadow_tbl.c.id, + shadow_tbl.c.transaction_id, + shadow_tbl.c.end_transaction_id, + shadow_tbl.c[name_col], + ).where(shadow_tbl.c.id.in_(ids)) + ) + .all() + ) + per_entity: dict[int, list[tuple[int, int | None, Any]]] = {} + for row in rows: + per_entity.setdefault(row[0], []).append((row[1], row[2], row[3])) + + resolved: dict[tuple[int, int], str] = {} + for entity_id, target_tx in pairs: + for start_tx, end_tx, name in per_entity.get(entity_id, []): + if start_tx <= target_tx and (end_tx is None or end_tx > target_tx): + resolved[(entity_id, target_tx)] = name + break + return resolved + + +def apply_entity_name_denormalization(records: list[dict[str, Any]]) -> None: + """Resolve each record's ``entity_name`` from the shadow row valid at + its ``transaction_id``. Adds an ``entity_name`` key to every record + in place; returns ``None``. + + The lookup is per (table-stored ``entity_kind``, ``entity_id``, + ``transaction_id``) triple. One ``IN``-clause query per kind keeps + round-trips bounded by the number of distinct kinds (≤3) regardless + of result-set size. The in-place mutation avoids re-allocating + thousands of dicts on hot dashboards; the name + return signature + make the side effect explicit instead of pretending to be a pure + projection. + """ + if not records: + return + + needed_by_kind: dict[str, set[tuple[int, int]]] = {} + for record in records: + api_kind = TABLE_KIND_TO_API.get(record["entity_kind"]) + if api_kind is None or api_kind not in NAME_COLUMN: + continue + needed_by_kind.setdefault(api_kind, set()).add( + (record["entity_id"], record["transaction_id"]) + ) + + resolved: dict[tuple[str, int, int], str] = {} + for api_kind, pairs in needed_by_kind.items(): + for (entity_id, target_tx), name in _resolve_names_for_kind( + api_kind, pairs + ).items(): + resolved[(api_kind, entity_id, target_tx)] = name + + for record in records: + api_kind_for_record = TABLE_KIND_TO_API.get(record["entity_kind"], "") + key = (api_kind_for_record, record["entity_id"], record["transaction_id"]) + record["entity_name"] = resolved.get(key, "") + + +# ---- Live-row existence + soft-delete state ------------------------------- + + +def check_entity_tombstones( + distinct_entities: set[tuple[str, int]], +) -> dict[tuple[str, int], dict[str, Any]]: + """For each ``(api_kind, entity_id)``, report ``deleted`` (no live + row) and ``deletion_state`` (``"soft_deleted"`` iff the live row has + a non-null ``deleted_at`` per sc-103157, else ``None``). + + Pre-sc-103157 the model classes don't have a ``deleted_at`` column; + we probe with ``hasattr`` and report ``deletion_state=None`` + universally in that case. Once sc-103157 lands, this helper picks up + the new column automatically. + """ + result: dict[tuple[str, int], dict[str, Any]] = {} + if not distinct_entities: + return result + + by_kind: dict[str, list[int]] = {} + for api_kind, entity_id in distinct_entities: + by_kind.setdefault(api_kind, []).append(entity_id) + + # ``no_autoflush`` mirrors the defensive posture of the listener- + # side reads. Today's callers run from request-path code with no + # pending writes; a future caller that probes tombstones before a + # flush would otherwise trigger autoflush mid-read. + with db.session.no_autoflush: + for api_kind, entity_ids in by_kind.items(): + if api_kind not in NAME_COLUMN: + for entity_id in entity_ids: + result[(api_kind, entity_id)] = { + "deleted": True, + "deletion_state": None, + } + continue + + model_name, _ = NAME_COLUMN[api_kind] + model_cls = load_shadow_model(model_name) + live_tbl = model_cls.__table__ + has_deleted_at = "deleted_at" in live_tbl.c + + cols = [live_tbl.c.id] + if has_deleted_at: + cols.append(live_tbl.c.deleted_at) + rows = ( + db.session.connection() + .execute(sa.select(*cols).where(live_tbl.c.id.in_(entity_ids))) + .all() + ) + live: dict[int, Any] = {} + for row in rows: + live[row[0]] = row[1] if has_deleted_at else None + + for entity_id in entity_ids: + if entity_id not in live: + result[(api_kind, entity_id)] = { + "deleted": True, + "deletion_state": None, + } + else: + deleted_at = live[entity_id] + result[(api_kind, entity_id)] = { + "deleted": False, + "deletion_state": "soft_deleted" if deleted_at else None, + } + return result diff --git a/superset/versioning/activity/render.py b/superset/versioning/activity/render.py new file mode 100644 index 000000000000..b8606a551b86 --- /dev/null +++ b/superset/versioning/activity/render.py @@ -0,0 +1,256 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Decoration: turn raw change-records into the ActivityRecord DTO. + +After fetching + filtering, each record needs the synthesized fields +the API contract documents — ``entity_kind`` translated to the user- +facing form, ``entity_uuid``, ``entity_deleted`` / +``entity_deletion_state``, ``source`` (self vs. related), +``summary`` (the AV-012 headline), ``impact`` (chart-count for +dashboard→dataset records), ``version_uuid``, ``changed_by``. + +This module collects all those decorations: + +* :func:`apply_record_decoration` — orchestrates the per-page additions in + one pass: pulls tombstones + uuids + impact counts in batches, then + walks records adding the synthesized fields and stripping the + internal-only columns the API contract doesn't expose. +* :func:`_lookup_entity_uuids` — one IN-clause query per kind to + resolve live ``uuid`` for non-tombstoned entities. +* :func:`_build_summary` — pure projection of (api_kind, record kind, + entity_name) onto the AV-012 headline string. +* :func:`_changed_by_dict` — projects the user columns onto the + ``changed_by`` DTO shape. +""" + +from __future__ import annotations + +from typing import Any +from uuid import UUID + +import sqlalchemy as sa + +from superset.extensions import db +from superset.versioning.activity.impact import ( + batch_chart_counts, + collect_impact_pairs, + impact_for_record, +) +from superset.versioning.activity.kinds import ( + API_KIND_LABEL, + load_shadow_model, + NAME_COLUMN, + TABLE_KIND_TO_API, + USER_FACING_KIND, +) +from superset.versioning.activity.queries import check_entity_tombstones +from superset.versioning.queries import derive_version_uuid + +_SUMMARY_VERBS: dict[str, str] = { + # The kind taxonomy from FR-016 mapped to past-tense verbs for the + # AV-012 " : " headline. "field" is + # the fallback for scalar changes that don't map to a named verb. + "filter": "filter changed", + "metric": "metric changed", + "dimension": "dimension changed", + "column": "column changed", + "chart": "chart changed", + "time_range": "time range changed", + "color_palette": "palette changed", + "restore": "restored", + "field": "updated", +} + + +def apply_record_decoration( + records: list[dict[str, Any]], + path_kind: str, + path_id: int, +) -> None: + """Add the synthesized ActivityRecord fields to each record in place: + ``entity_kind`` (translated to API form), ``entity_uuid``, + ``entity_deleted``, ``entity_deletion_state``, ``source``, + ``summary``, ``impact``, ``version_uuid``, ``changed_by``. + + Mutates *records* in place; returns ``None``. Records are expected + to already carry ``entity_name`` from + :func:`apply_entity_name_denormalization`. The in-place mutation + avoids re-allocating thousands of dicts on hot dashboards; the + name + return signature make the side effect explicit instead of + pretending to be a pure projection. + """ + if not records: + return + + distinct: set[tuple[str, int]] = { + ( + TABLE_KIND_TO_API.get(r["entity_kind"], ""), + r["entity_id"], + ) + for r in records + if TABLE_KIND_TO_API.get(r["entity_kind"]) + } + tombstones = check_entity_tombstones(distinct) + uuids = _lookup_entity_uuids(distinct, tombstones) + # Pre-compute impact counts for the whole page in one batch query + # instead of one COUNT per related record (was N+1). + impact_counts = batch_chart_counts( + path_id, collect_impact_pairs(records, path_kind) + ) + + for record in records: + api_kind = TABLE_KIND_TO_API.get(record["entity_kind"], "") + entity_id = record["entity_id"] + tombstone = tombstones.get( + (api_kind, entity_id), {"deleted": True, "deletion_state": None} + ) + entity_uuid = uuids.get((api_kind, entity_id)) + is_self = api_kind == path_kind and entity_id == path_id + + # Emit the user-facing form ("dashboard"/"chart"/"dataset") on the + # wire; the internal class-name (api_kind) is kept above for the + # remaining decoration steps that key off model_cls.__name__. + record["entity_kind"] = USER_FACING_KIND.get(api_kind, api_kind) + record["entity_uuid"] = str(entity_uuid) if entity_uuid else None + record["entity_deleted"] = tombstone["deleted"] + record["entity_deletion_state"] = tombstone["deletion_state"] + record["source"] = "self" if is_self else "related" + record["version_uuid"] = ( + str(derive_version_uuid(entity_uuid, record["transaction_id"])) + if entity_uuid + else None + ) + record["changed_by"] = _changed_by_dict(record) + + if is_self: + # AV-012 leaves self records summary-less (the panel renders + # them from kind/path/values) — EXCEPT synthetic ``__meta__`` + # headlines, whose entire payload IS the summary and whose + # primary surface is the entity's own stream ("restored to + # version N" must render on include=self). + record["summary"] = ( + _build_summary(api_kind, record) + if record.get("kind") == "__meta__" + else "" + ) + record["impact"] = None + else: + record["summary"] = _build_summary(api_kind, record) + record["impact"] = impact_for_record(record, path_kind, impact_counts) + if record["entity_deleted"]: + # Security: a tombstoned related entity has no live row, so + # the visibility filter cannot access-gate it (there is + # nothing to apply the FAB access filter to). Redact the raw + # diff CONTENT — filter values, column names, SQL/adhoc + # expressions — so a requester entitled only to the path + # entity can't read the internal change values of a deleted + # related entity. The entity_name and the AV-012 headline + # are kept deliberately (D-15: the panel shows "(deleted) + # "); only the value payload is stripped. Self-path + # tombstones are untouched — the endpoint already gated them + # via ``raise_for_access`` on the path entity. + record["from_value"] = None + record["to_value"] = None + record["path"] = None + + # Strip the internal-only columns the API contract doesn't expose. + for key in ( + "entity_id", + "sequence", + "user_id", + "changed_by_id", + "first_name", + "last_name", + ): + record.pop(key, None) + + +def _lookup_entity_uuids( + distinct: set[tuple[str, int]], + tombstones: dict[tuple[str, int], dict[str, Any]], +) -> dict[tuple[str, int], UUID | None]: + """Batch-fetch live ``uuid`` per ``(api_kind, entity_id)``. Tombstoned + entities are skipped (their ``entity_uuid`` is null per data-model.md). + """ + result: dict[tuple[str, int], UUID | None] = {} + by_kind: dict[str, list[int]] = {} + for api_kind, entity_id in distinct: + if tombstones.get((api_kind, entity_id), {}).get("deleted"): + continue + by_kind.setdefault(api_kind, []).append(entity_id) + + # ``no_autoflush`` mirrors the defensive posture of the baseline + + # change-record listeners: this helper reads from live tables to + # resolve uuids, and a future caller that resolves an entity before + # the parent flush would otherwise trigger autoflush mid-read. + # Today's call sites run from request-path code with no pending + # session state, so the cost of the guard is zero. + with db.session.no_autoflush: + for api_kind, entity_ids in by_kind.items(): + if api_kind not in NAME_COLUMN: + continue + model_cls = load_shadow_model(NAME_COLUMN[api_kind][0]) + live_tbl = model_cls.__table__ + rows = ( + db.session.connection() + .execute( + sa.select(live_tbl.c.id, live_tbl.c.uuid).where( + live_tbl.c.id.in_(entity_ids) + ) + ) + .all() + ) + for row in rows: + result[(api_kind, row[0])] = row[1] + return result + + +def _build_summary(api_kind: str, record: dict[str, Any]) -> str: + """Build the AV-012 headline for a related record: + ``" : "``.""" + label = API_KIND_LABEL.get(api_kind, api_kind) + # Synthetic ``__meta__`` headlines carry their payload in to_value + # and their verb on the transaction's ``action_kind`` (path stays + # pure navigation). The restore variant names the version it + # restored to (PR #40988: "Restored to X from [date]" is not + # renderable from field diffs). + if record.get("kind") == "__meta__": + name = record.get("entity_name") or "" + if record.get("action_kind") == "restore": + to_value = record.get("to_value") or {} + version_number = to_value.get("version_number") + if version_number is not None: + headline = f"{label} restored to version {version_number}" + return f"{headline}: {name}" if name else headline + return f"{label} updated: {name}" if name else f"{label} updated" + verb = _SUMMARY_VERBS.get(record.get("kind", ""), "updated") + name = record.get("entity_name") or "" + return f"{label} {verb}: {name}" if name else f"{label} {verb}" + + +def _changed_by_dict(record: dict[str, Any]) -> dict[str, Any] | None: + """Project the user columns onto the ``changed_by`` shape, or + ``None`` when no Flask user was attached to the save (CLI / Celery) + or when the user has since been deleted from ``ab_user``. + """ + if record.get("changed_by_id") is None: + return None + return { + "id": record["changed_by_id"], + "first_name": record.get("first_name"), + "last_name": record.get("last_name"), + } diff --git a/superset/versioning/activity/scope.py b/superset/versioning/activity/scope.py new file mode 100644 index 000000000000..f24140c68823 --- /dev/null +++ b/superset/versioning/activity/scope.py @@ -0,0 +1,103 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Scope resolution — turn a path entity into the related-entity walk. + +Composes :mod:`~superset.versioning.activity.queries` (Phase A +relationship walks) and :mod:`~superset.versioning.activity.windows` +(pure interval arithmetic) into the +``list[EntityWindows]`` scope that +:func:`~superset.versioning.activity.queries.fetch_change_records` +consumes. + +The functions here read the DB (via the Phase A helpers in +:mod:`~superset.versioning.activity.queries`); the pure window- +arithmetic functions previously colocated here now live in +:mod:`~superset.versioning.activity.windows` so the package no longer +needs a lazy import to dodge a ``scope ↔ queries`` cycle. +""" + +from __future__ import annotations + +from superset.versioning.activity.kinds import EntityWindows, Window +from superset.versioning.activity.queries import ( + batch_datasets_used_by_charts, + charts_attached_to_dashboard, + datasets_used_by_chart, +) +from superset.versioning.activity.windows import ( + intersect_windows, + merge_entity_windows, +) + + +def resolve_scope(path_kind: str, path_id: int, include: str) -> list[EntityWindows]: + """Build the ``[(api_kind, entity_id, [windows])]`` list that + :func:`~superset.versioning.activity.queries.fetch_change_records` + consumes, branching by *path_kind* and *include* mode.""" + want_self = include in ("all", "self") + want_related = include in ("all", "related") + + scope: list[EntityWindows] = [] + if want_self: + scope.append((path_kind, path_id, [Window(0, None)])) + if want_related: + scope.extend(_resolve_related_scope(path_kind, path_id)) + return scope + + +def _resolve_related_scope(path_kind: str, path_id: int) -> list[EntityWindows]: + """Walk the dependency edges from the path entity to its related + entities. Per AV-004, datasets have no transitive layer in V2.""" + if path_kind == "Dashboard": + return _resolve_dashboard_scope(path_id) + if path_kind == "Slice": + return _resolve_chart_scope(path_id) + return [] + + +def _resolve_dashboard_scope(dashboard_id: int) -> list[EntityWindows]: + """Charts on the dashboard during their attachment window, plus + datasets each chart pointed at during the intersection of (chart- + attachment, chart-on-dataset).""" + scope: list[EntityWindows] = [] + chart_windows: dict[int, list[Window]] = {} + for slice_id, window in charts_attached_to_dashboard(dashboard_id): + chart_windows.setdefault(slice_id, []).append(window) + + # One query for the dataset-history of every chart on the dashboard, + # not one query per chart. The per-slice form was O(n_charts) round- + # trips which dominated p95 on rich dashboards. + dataset_windows_by_slice = batch_datasets_used_by_charts(set(chart_windows)) + + for slice_id, attachment_windows in chart_windows.items(): + scope.append(("Slice", slice_id, list(attachment_windows))) + dataset_windows = dataset_windows_by_slice.get(slice_id, []) + for attachment in attachment_windows: + for dataset_id, chart_dataset_window in dataset_windows: + if ( + intersect := intersect_windows(attachment, chart_dataset_window) + ) is not None: + scope.append(("SqlaTable", dataset_id, [intersect])) + return merge_entity_windows(scope) + + +def _resolve_chart_scope(slice_id: int) -> list[EntityWindows]: + """Datasets the chart pointed at over its full history.""" + scope: list[EntityWindows] = [] + for dataset_id, window in datasets_used_by_chart(slice_id): + scope.append(("SqlaTable", dataset_id, [window])) + return merge_entity_windows(scope) diff --git a/superset/versioning/activity/visibility.py b/superset/versioning/activity/visibility.py new file mode 100644 index 000000000000..dbc278f17fa0 --- /dev/null +++ b/superset/versioning/activity/visibility.py @@ -0,0 +1,201 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Per-AV-008 silent visibility filter for activity-view records. + +Drops records whose source entity the requester can't read. Silent in +the sense that dropped records contribute no count and no placeholder +to the response — the user sees only what they're entitled to see, and +the response shape can't be used to infer the existence of entities +they're gated out of. + +Visibility is resolved SQL-side via each resource's existing FAB +access filter (``DashboardAccessFilter`` / ``ChartFilter`` / +``DatasourceFilter``). Two SQL queries per kind (one for live ids, one +for the access-filtered subset) replace the N-call +``security_manager.can_access_(entity)`` loop that dominated +latency on dashboard-scope responses with many related entities +(sqlalchemy-review W-NEW-1). + +**Inline imports.** ``_resolve_visibility`` defers the FAB-filter +imports (``DashboardAccessFilter`` / ``ChartFilter`` / +``DatasourceFilter`` and ``SQLAInterface``) until call time. Same +init-order rationale as :mod:`superset.versioning.baseline` — +``versioning`` packages are imported from ``init_versioning()`` +before all model mappers are configured, and the filter classes pull +in their resource's model graph (Chart → Dataset → Database for +``ChartFilter``, etc.); a module-top import would trip mapper +resolution before Continuum's ``make_versioned()`` has finished. + +**Integration shape.** The activity-view consumes FAB's access-filter +classes (``DashboardAccessFilter`` / ``ChartFilter`` / +``DatasourceFilter``) directly rather than translating them or +re-implementing the predicate. That keeps the activity stream's +visibility posture identical to the list endpoints' — operator- +controlled and stable — at the cost of coupling to FAB's exact +filter shape. Future entities added to the activity surface must +extend the dispatch table in ``_resolve_visibility`` to include +their access-filter class. +""" + +from __future__ import annotations + +import logging +from typing import Any + +from superset.extensions import db +from superset.versioning.activity.kinds import ( + load_shadow_model, + NAME_COLUMN, + TABLE_KIND_TO_API, +) + +logger = logging.getLogger(__name__) + + +def filter_records_by_visibility( + records: list[dict[str, Any]], +) -> list[dict[str, Any]]: + """Drop records whose source entity the requester can't read. + + Per AV-008 the filter is silent: dropped records contribute no count + and no placeholder. Tombstoned entities (no live row) pass through + — the decorator step marks them ``entity_deleted: true`` and the + payload exposes no navigable ``entity_uuid``, so there's nothing + sensitive left to gate. + + Visibility is resolved SQL-side via each resource's existing access + filter, which reads the requesting user from Flask-Login internally + (no explicit user parameter threads through here). If a CLI/Celery + bypass becomes necessary in the future, add it then with a real call + site. + """ + if not records: + return records + + distinct: set[tuple[str, int]] = { + ( + TABLE_KIND_TO_API.get(r["entity_kind"], r["entity_kind"]), + r["entity_id"], + ) + for r in records + } + visible = _resolve_visibility(distinct) + return [ + r + for r in records + if visible.get( + ( + TABLE_KIND_TO_API.get(r["entity_kind"], r["entity_kind"]), + r["entity_id"], + ), + True, # tombstone / unknown kind → pass through + ) + ] + + +def _resolve_visibility( + distinct_entities: set[tuple[str, int]], +) -> dict[tuple[str, int], bool]: + """Return ``{(api_kind, entity_id): can_read}`` for the live row of + each entity. Missing live rows (tombstoned) map to ``True`` — the + decorator handles the deleted-state messaging separately. + + Visibility is computed SQL-side via each resource's existing access + filter (``DashboardAccessFilter`` / ``ChartFilter`` / + ``DatasourceFilter``). These are the same filters FAB's + ``ModelRestApi`` applies to ``base_filters`` on list endpoints, so + the activity-view visibility check matches the rest of the read + surface byte-for-byte. Two queries per kind (one for live ids, one + for the access-filtered subset) replace the N-call + ``security_manager.can_access_(entity)`` loop that dominated + latency on dashboard-scope activity responses with many related + entities (sqlalchemy-review W-NEW-1). + """ + # pylint: disable=import-outside-toplevel + from flask_appbuilder.models.sqla.interface import SQLAInterface + + from superset.charts.filters import ChartFilter + from superset.dashboards.filters import DashboardAccessFilter + from superset.views.base import DatasourceFilter + + access_filter_classes: dict[str, type] = { + "Dashboard": DashboardAccessFilter, + "Slice": ChartFilter, + "SqlaTable": DatasourceFilter, + } + + by_kind: dict[str, list[int]] = {} + for api_kind, entity_id in distinct_entities: + by_kind.setdefault(api_kind, []).append(entity_id) + + visible: dict[tuple[str, int], bool] = {} + for api_kind, entity_ids in by_kind.items(): + if api_kind in NAME_COLUMN and api_kind not in access_filter_classes: + # The kind is in the change-records taxonomy but is missing + # an access-filter wiring — almost certainly a future-entity + # addition that updated ``TABLE_KIND_TO_API`` but forgot + # the visibility dispatch. Fail closed: the activity stream + # must not silently disclose change records for an entity + # whose access predicate is unimplemented. Warn so the gap + # surfaces in CI / staging logs before production. + logger.warning( + "activity visibility: no access filter wired for kind %r; " + "denying %d records", + api_kind, + len(entity_ids), + ) + for entity_id in entity_ids: + visible[(api_kind, entity_id)] = False + continue + if api_kind not in NAME_COLUMN: + # Kind isn't in the change-records taxonomy at all — not + # something the activity-view emits today. Pass through so + # the decorator can mark it as a tombstone if appropriate. + for entity_id in entity_ids: + visible[(api_kind, entity_id)] = True + continue + model_cls = load_shadow_model(NAME_COLUMN[api_kind][0]) + + # Live ids — what exists at all. Used to decide tombstone vs + # not-visible: an id missing from this set is tombstoned and + # passes through (True); an id in this set but absent from the + # access-filtered set is denied (False). + live_ids = { + row[0] + for row in db.session.query(model_cls.id) + .filter(model_cls.id.in_(entity_ids)) + .all() + } + + # Apply the SQL-side access filter to a query restricted to the + # candidate ids. Same predicate FAB uses for list endpoints, so + # results are consistent with the rest of the read surface. + access_filter = access_filter_classes[api_kind]("id", SQLAInterface(model_cls)) + visible_ids = { + row[0] + for row in access_filter.apply( + db.session.query(model_cls.id).filter(model_cls.id.in_(entity_ids)), + value=None, + ).all() + } + + for entity_id in entity_ids: + if entity_id not in live_ids: + visible[(api_kind, entity_id)] = True + else: + visible[(api_kind, entity_id)] = entity_id in visible_ids + return visible diff --git a/superset/versioning/activity/windows.py b/superset/versioning/activity/windows.py new file mode 100644 index 000000000000..b76ae43c5dc5 --- /dev/null +++ b/superset/versioning/activity/windows.py @@ -0,0 +1,108 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Pure window arithmetic on half-open ``[start_tx, end_tx)`` intervals. + +Extracted from the DB-touching scope resolution so that: + +* :mod:`scope` (DB-touching) can import this module at module-top. +* :mod:`queries.fetch_change_records` can import + :func:`row_within_any_window` at module-top instead of through a + lazy import that previously dodged a ``scope ↔ queries`` cycle. + +Everything here is pure Python — no DB, no Flask. ``end_tx = None`` +means "open-ended (current)" and behaves like positive infinity. +""" + +from __future__ import annotations + +from typing import Any + +from superset.versioning.activity.kinds import EntityWindows, Window + + +def intersect_windows(outer: Window, inner: Window) -> Window | None: + """Intersect two half-open ``[start_tx, end_tx)`` windows. + + Returns the clipped overlap, or ``None`` when they are disjoint. + ``end_tx = None`` means "open ended (current)" and acts like + positive infinity. Thin wrapper over :meth:`Window.intersect` — + kept as a free function so callers and tests don't have to migrate + to method form in lockstep with the dataclass promotion. + """ + return outer.intersect(inner) + + +def row_within_any_window(row: dict[str, Any], windows: list[Window]) -> bool: + """``True`` iff ``row['transaction_id']`` falls inside at least one + of *windows*. Half-open interval semantics match + :func:`intersect_windows`.""" + if not windows: + return False + tx_id = row["transaction_id"] + return any(w.contains(tx_id) for w in windows) + + +def merge_entity_windows(scope: list[EntityWindows]) -> list[EntityWindows]: + """Collapse repeated ``(api_kind, entity_id)`` entries by unioning + their window lists, and collapse overlapping/touching windows + within each entity into one. + + The OR-clause in + :func:`~superset.versioning.activity.queries.fetch_change_records` + generates one branch per (kind, id, window) tuple. Without the + within-entity union, a chart that's been attached-and-detached + many times (or that repeated fixture loads have populated the M2M + shadow for) yields a separate clause per redundant window — at + ~10 entities × ~50 windows the SQL hits SQLite's + ``SQLITE_MAX_EXPR_DEPTH`` (1000). Merging here keeps the clause + count proportional to the number of *distinct* validity intervals, + not the number of shadow rows. + """ + merged: dict[tuple[str, int], list[Window]] = {} + for api_kind, entity_id, windows in scope: + merged.setdefault((api_kind, entity_id), []).extend(windows) + return [ + (api_kind, entity_id, union_windows(windows)) + for (api_kind, entity_id), windows in merged.items() + ] + + +def union_windows(windows: list[Window]) -> list[Window]: + """Sort + merge overlapping/touching half-open intervals. + + Pure function — no DB. Touching ``[a, b)`` and ``[b, c)`` merge into + ``[a, c)``. ``end_tx = None`` (open-ended) absorbs everything to its + right. Returns a minimal disjoint cover of the input set. + """ + if not windows: + return [] + sorted_windows = sorted(windows, key=lambda w: w.start_tx) + out: list[Window] = [sorted_windows[0]] + for current in sorted_windows[1:]: + prev = out[-1] + if prev.end_tx is None: + # Prior window is open-ended; it absorbs everything past. + continue + if current.start_tx <= prev.end_tx: + # Overlapping or touching — extend the prior window. + new_end: int | None = ( + None if current.end_tx is None else max(prev.end_tx, current.end_tx) + ) + out[-1] = Window(prev.start_tx, new_end) + else: + out.append(current) + return out diff --git a/superset/versioning/api_helpers.py b/superset/versioning/api_helpers.py index d052f66e8f5d..8df31931a87c 100644 --- a/superset/versioning/api_helpers.py +++ b/superset/versioning/api_helpers.py @@ -14,18 +14,19 @@ # KIND, either express or implied. See the License for the # specific language governing permissions and limitations # under the License. -"""Shared handlers for the ``/versions/`` REST endpoints. +"""Shared handlers for the ``/versions/`` and ``/activity/`` REST endpoints. Each ``ChartRestApi`` / ``DashboardRestApi`` / ``DatasetRestApi`` carries the same read endpoint methods — ``list_versions`` and ``get_version`` — -whose bodies are byte-for-byte identical apart from the model class and -the ``security_manager.raise_for_access`` kwarg. Extracting the bodies -here lets each per-resource method collapse to a single delegation call, -while the OpenAPI docstring + FAB decorators stay at the method site -where they belong. - -(The restore endpoint ships in a later PR; only the read endpoints are -wired here.) +plus the ``activity`` endpoint on each resource. The bodies are +byte-for-byte identical apart from the model class and the +``security_manager.raise_for_access`` kwarg. Extracting the bodies here +lets each per-resource method collapse to a single delegation call, while +the OpenAPI docstring + FAB decorators stay at the method site where they +belong. + +(The restore endpoint ships in a later PR; only the read + activity +endpoints are wired here.) """ from __future__ import annotations @@ -119,34 +120,82 @@ def current_entity_etag_uuid( return str(version_uuid) if version_uuid else None -def _resolve_entity( - api: Any, - model_cls: type[Model], - uuid_str: str, - access_kwarg: str, -) -> tuple[Any, UUID] | Response: - """Parse the path UUID, look up the live entity, run the read-access - gate. - - Returns ``(entity, entity_uuid)`` on success or a pre-built - ``Response`` (400 / 403 / 404) that the caller should return - directly. The split shape keeps the call site terse and lets the - three handler functions share the preflight without each repeating - the try / except dance. +# Maps the versioned model class name to the keyword argument +# ``security_manager.raise_for_access`` expects for the per-resource +# gate. Slice → ``chart=``, Dashboard → ``dashboard=``, SqlaTable → +# ``datasource=``. Centralised here so /versions/ and /activity/ +# endpoints share one source of truth for the dispatch. +_RAISE_FOR_ACCESS_KWARG: dict[str, str] = { + "Slice": "chart", + "Dashboard": "dashboard", + "SqlaTable": "datasource", +} + + +class PathEntityResponseError(Exception): + """Carries a pre-built error ``Response`` from + :func:`resolve_endpoint_path_entity`. Endpoints catch it and return + the carried response directly. The shape exists so the + UUID-parse + find-by-uuid + read-access check can live in one + place across the ``/versions/`` and ``/activity/`` endpoint + families.""" + + def __init__(self, response: Any) -> None: + super().__init__("PathEntityResponseError") + self.response = response + + +def resolve_endpoint_path_entity( + api: Any, model_cls: type[Model], uuid_str: str +) -> tuple[Any, UUID]: + """Run the standard path-entity preflight for a /versions/ or + /activity/ endpoint: + + 1. Parse *uuid_str* into a UUID (or raise → 400). + 2. Look up the live entity via ``VersionDAO.find_active_by_uuid`` + (or raise → 404). + 3. Run ``security_manager.raise_for_access`` with the resource-typed + kwarg (or raise → 403). + + Returns ``(entity, entity_uuid)`` on success — the parsed UUID is + threaded out so callers don't re-parse the path-string. Raises + :class:`PathEntityResponseError` carrying the appropriate error + Response on any failure; the endpoint method should:: + + try: + entity, entity_uuid = resolve_endpoint_path_entity( + self, Dashboard, uuid_str + ) + except PathEntityResponseError as exc: + return exc.response + + *api* is the FAB ``ModelRestApi`` instance — we call + ``api.response_400`` / ``api.response_403`` / ``api.response_404`` + on it. Pass ``self`` from the endpoint method. """ try: entity_uuid = UUID(uuid_str) - except ValueError: - return api.response_400(message="Invalid UUID") + except ValueError as exc: + raise PathEntityResponseError(api.response_400(message="Invalid UUID")) from exc entity = VersionDAO.find_active_by_uuid(model_cls, entity_uuid) if entity is None: - return api.response_404() - + raise PathEntityResponseError(api.response_404()) + + # Direct ``[…]`` would leak the unknown model name into a generic 500 + # via the unhandled ``KeyError`` exception text. The three resource + # families wired today cover every key; a future entity added to the + # versioning surface without updating this dispatch table should fail + # closed (the test suite picks it up) rather than silently disclose. + kwarg = _RAISE_FOR_ACCESS_KWARG.get(model_cls.__name__) + if kwarg is None: + raise LookupError( + f"No raise_for_access kwarg registered for {model_cls.__name__!r}" + ) try: - security_manager.raise_for_access(**{access_kwarg: entity}) - except SupersetSecurityException: - return api.response_403() + security_manager.raise_for_access(**{kwarg: entity}) + except SupersetSecurityException as exc: + raise PathEntityResponseError(api.response_403()) from exc return entity, entity_uuid @@ -155,13 +204,12 @@ def list_versions_endpoint( api: Any, model_cls: type[Model], uuid_str: str, - access_kwarg: str, ) -> Response: """Body of ``GET /api/v1/{resource}//versions/``.""" - resolved = _resolve_entity(api, model_cls, uuid_str, access_kwarg) - if isinstance(resolved, Response): - return resolved - entity, entity_uuid = resolved + try: + entity, entity_uuid = resolve_endpoint_path_entity(api, model_cls, uuid_str) + except PathEntityResponseError as exc: + return exc.response versions = VersionDAO.list_versions(model_cls, entity_uuid, entity=entity) if versions is None: @@ -180,13 +228,12 @@ def get_version_endpoint( model_cls: type[Model], uuid_str: str, version_uuid_str: str, - access_kwarg: str, ) -> Response: """Body of ``GET /api/v1/{resource}//versions//``.""" - resolved = _resolve_entity(api, model_cls, uuid_str, access_kwarg) - if isinstance(resolved, Response): - return resolved - entity, entity_uuid = resolved + try: + entity, entity_uuid = resolve_endpoint_path_entity(api, model_cls, uuid_str) + except PathEntityResponseError as exc: + return exc.response try: version_uuid = UUID(version_uuid_str) diff --git a/superset/versioning/changes/listener.py b/superset/versioning/changes/listener.py index fb9c500d3ca6..172f86edf8d9 100644 --- a/superset/versioning/changes/listener.py +++ b/superset/versioning/changes/listener.py @@ -198,6 +198,29 @@ def build_action_headline( # is correctly deduped. _REGISTERED_SENTINEL = "_versioning_change_listener_registered" +#: Metric namespace for swallowed capture-path failures. The capture +#: listeners fail open (a versioning bug must never break a user's save), +#: so the read path (``activity/orchestrator``) is richly instrumented but +#: the write path historically logged-and-swallowed with no counter. Each +#: ``_incr_capture_error(stage)`` emits ``..error`` so a +#: systematic capture regression is alertable rather than log-grep-only. +_CAPTURE_METRIC_PREFIX = "superset.versioning.capture" + + +def _incr_capture_error(stage: str) -> None: + """Emit a counter for a swallowed capture-path failure at *stage*. + + Best-effort: metrics emission must never itself break a user's save, + so it is wrapped in the same fail-open posture as the call site. + """ + # pylint: disable=import-outside-toplevel + try: + from superset.extensions import stats_logger_manager + + stats_logger_manager.instance.incr(f"{_CAPTURE_METRIC_PREFIX}.{stage}.error") + except Exception: # pylint: disable=broad-except + logger.exception("version_changes: failed to emit capture-error metric") + def _process_dirty_entity_into_buffer( session: Session, @@ -219,6 +242,7 @@ def _process_dirty_entity_into_buffer( type(obj).__name__, entity_id, ) + _incr_capture_error("scalar_diff") return if records: buffer.setdefault((entity_kind, entity_id), []).extend(records) @@ -257,6 +281,7 @@ def _append_child_records_to_buffer( del buffer[key] except Exception: # pylint: disable=broad-except logger.exception("version_changes: child-diff failed for tx %s", tx_id) + _incr_capture_error("child_diff") def _current_transaction_id(session: Session) -> int | None: @@ -342,6 +367,7 @@ def _stamp_action_kind_on_transaction(session: Session, tx_id: int) -> None: action_kind, tx_id, ) + _incr_capture_error("action_kind_stamp") def _persist_buffered_records( @@ -376,6 +402,7 @@ def _persist_buffered_records( tx_id, len(buffer), ) + _incr_capture_error("bulk_insert") def register_change_record_listener() -> None: # noqa: C901 diff --git a/superset/versioning/diff.py b/superset/versioning/diff.py index aaa0bd6b4c22..326ba447fc1b 100644 --- a/superset/versioning/diff.py +++ b/superset/versioning/diff.py @@ -896,10 +896,11 @@ def diff_dashboard_layout( payloads carry old + new meta * id in both, equal → no record - The ``operation_type``-style verb is encoded in - ``path[0]`` as ``["add"|"remove"|"move"|"edit", , - ]`` so the UI's path-based renderer can read it - without inspecting from/to. + The verb lives in each record's ``operation`` field + (``add`` / ``remove`` / ``move`` / ``edit``); ``path`` locates the + component as ``[]`` (``[, , …]`` + for an ``edit`` that recurses into ``meta``). Paths no longer carry + the verb — see :func:`_layout_chart_uuids_by_verb`. ``ROOT_ID`` / ``GRID_ID`` / ``HEADER_ID`` are suppressed (see :data:`_LAYOUT_SUPPRESSED_IDS`). diff --git a/superset/versioning/schemas.py b/superset/versioning/schemas.py index 50fbba38b7a5..c39d7b22c5fd 100644 --- a/superset/versioning/schemas.py +++ b/superset/versioning/schemas.py @@ -23,7 +23,9 @@ from __future__ import annotations -from marshmallow import fields, Schema +from marshmallow import fields, Schema, validate + +from superset.versioning.changes import ACTION_KINDS class VersionChangedBySchema(Schema): @@ -103,8 +105,8 @@ class VersionListItemSchema(Schema): "description": ( "One of 'baseline', 'update', or 'delete', derived from the " "Continuum integer constant. Restore is not a distinct " - "operation_type: a restore surfaces as an ordinary 'update' " - "transaction." + "operation_type: a restore surfaces as 'update' carrying " + "``action_kind='restore'`` (see ACTIVITY_ACTION_KINDS)." ) }, ) @@ -138,3 +140,319 @@ class VersionListResponseSchema(Schema): result = fields.List(fields.Nested(VersionListItemSchema)) count = fields.Integer() + + +# ---- Cross-entity activity view (sc-107283) ------------------------------- + +#: Allowed values for ``ActivityRecordSchema.entity_kind``. User-facing +#: lowercase strings; the activity layer's internal kind dispatch keys off +#: ``model_cls.__name__`` (``Dashboard`` / ``Slice`` / ``SqlaTable``) and +#: translates to these labels at the JSON boundary in +#: :func:`superset.versioning.activity.apply_record_decoration`. +ACTIVITY_ENTITY_KINDS: tuple[str, ...] = ("dashboard", "chart", "dataset") + +#: Allowed values for ``ActivityRecordSchema.source`` (spec AV-013). +ACTIVITY_SOURCES: tuple[str, ...] = ("self", "related") + +#: Allowed values for ``ActivityRecordSchema.entity_deletion_state``. +#: Hard-delete is communicated separately via ``entity_deleted=true``; +#: the remaining state is the soft-delete sentinel (sc-103157). +ACTIVITY_DELETION_STATES: tuple[str, ...] = ("soft_deleted",) + +#: Allowed values for ``ActivityRecordSchema.kind`` — mirrors the +#: change-record taxonomy from sc-103156 FR-016. ``"field"`` is the +#: fallback for scalar changes without a more specific category. +#: +#: ``"restore"`` (previously the synthetic kind for restore events) is +#: removed: restores now produce regular field-level records plus +#: ``version_transaction.action_kind="restore"`` (see ACTIVITY_ACTION_KINDS). +ACTIVITY_CHANGE_KINDS: tuple[str, ...] = ( + "filter", + "metric", + "dimension", + "column", + "chart", + "row", + "tab", + "tabs", + "header", + "markdown", + "divider", + "time_range", + "color_palette", + "field", + # Synthetic headline records emitted by commands via the listener's + # ACTION_META_KEY (the ``__meta__`` path convention): machine + # namespace, clearly non-content. The canonical case is restore, + # whose record's ``to_value`` carries the restored-to + # ``version_uuid`` / ``version_number`` (PR #40988 feedback). + "__meta__", +) + +#: Allowed values for ``ActivityRecordSchema.operation`` — the per-record +#: verb. ``move`` only fires for layout records; ``add`` / ``remove`` / +#: ``edit`` apply across every emit site. +ACTIVITY_CHANGE_OPERATIONS: tuple[str, ...] = ( + "add", + "remove", + "move", + "edit", + # Synthetic ``__meta__`` headline records announce an action (e.g. a + # restore) rather than mutating a field — the field-verb vocabulary + # would be dishonest for them. Source of the value: + # superset.versioning.changes.OPERATION_ANNOUNCE. + "announce", +) + +#: Allowed values for ``ActivityRecordSchema.action_kind`` — the +#: transaction-level avenue. ``null`` (omitted from the enum, signalled +#: by ``allow_none``) means "ordinary save". Sourced from +#: :data:`superset.versioning.changes.ACTION_KINDS` so a future +#: addition (e.g. ``"thumbnail_warm"``) only has to update that +#: constant; the schema picks it up automatically. +ACTIVITY_ACTION_KINDS: tuple[str, ...] = tuple(sorted(ACTION_KINDS)) + + +class ActivityChangedBySchema(Schema): + """User attribution for an activity record. + + The activity-view payload exposes only the display fields + (``id`` + given/family name); ``username`` is omitted by design (see + data-model.md §"ActivityRecord DTO"). ``null`` when the saving user + has been deleted from ``ab_user`` (sc-103156 §Session 2026-05-18 + clarification). + """ + + id = fields.Integer() + first_name = fields.String() + last_name = fields.String() + + +class ActivityImpactSchema(Schema): + """Dependent-count summary attached to ``source='related'`` records. + + Synthesized server-side at the time of the activity query — it counts + siblings affected by the same upstream change at the same transaction + (e.g., how many charts on the requested dashboard pointed at the + dataset whose edit this record represents). + """ + + charts = fields.Integer( + metadata={ + "description": ( + "Number of sibling charts on the path entity affected by " + "the same related-record change at this transaction." + ) + }, + ) + + +class ActivityRecordSchema(Schema): + """One change record in the activity stream. + + One record per atomic field-level change. Fields mirror + data-model.md §"``ActivityRecord`` DTO" — see that doc for source + and required/optional details. + """ + + version_uuid = fields.String( + metadata={ + "description": ( + "Stable UUIDv5 identifier for the source version " + "(``derive_version_uuid(entity_uuid, transaction_id)``). " + "Identical to what ``/versions//`` would " + "return for the same change." + ) + }, + ) + entity_kind = fields.String( + validate=validate.OneOf(ACTIVITY_ENTITY_KINDS), + metadata={ + "description": ( + "User-facing kind of the source entity: one of " + '``"dashboard"`` / ``"chart"`` / ``"dataset"``.' + ) + }, + ) + entity_uuid = fields.String( + allow_none=True, + metadata={ + "description": ( + "UUID of the source entity; ``null`` only when " + "``entity_deleted: true`` (the entity has been hard-deleted " + "since the change was recorded)." + ) + }, + ) + entity_name = fields.String( + metadata={ + "description": ( + "Name of the source entity *at the time of the change* — " + "denormalized from the validity-strategy shadow row. " + "Survives entity rename / delete." + ) + }, + ) + entity_deleted = fields.Boolean( + metadata={ + "description": ( + "True iff the source entity is hard-deleted " + "(no live row by ``entity_id``). False for live and " + "soft-deleted entities." + ) + }, + ) + entity_deletion_state = fields.String( + allow_none=True, + validate=validate.OneOf(ACTIVITY_DELETION_STATES), + metadata={ + "description": ( + "Present when the source entity has non-null ``deleted_at`` " + "(sc-103157). Absent or ``null`` otherwise." + ) + }, + ) + source = fields.String( + validate=validate.OneOf(ACTIVITY_SOURCES), + metadata={ + "description": ( + '``"self"`` if ``(entity_kind, entity_id)`` matches the ' + 'path entity; else ``"related"``. Drives the frontend\'s ' + "no-group-under-save rendering rule (AV-013)." + ) + }, + ) + transaction_id = fields.Integer( + metadata={"description": "Stable secondary ordering key; never reused."}, + ) + issued_at = fields.DateTime( + metadata={"description": "UTC timestamp; primary ordering key (DESC)."}, + ) + changed_by = fields.Nested( + ActivityChangedBySchema, + allow_none=True, + metadata={ + "description": ( + "User who produced the change, or ``null`` when the saving " + "user no longer exists in ``ab_user``." + ) + }, + ) + kind = fields.String( + validate=validate.OneOf(ACTIVITY_CHANGE_KINDS), + metadata={ + "description": ( + "Content category — what kind of thing changed. " + "``field`` is the fallback for scalar changes without a " + "more specific category. Per-record." + ) + }, + ) + operation = fields.String( + validate=validate.OneOf(ACTIVITY_CHANGE_OPERATIONS), + metadata={ + "description": ( + "Per-record verb: ``add`` / ``remove`` / ``move`` / " + "``edit``. Explicit instead of inferred from " + "``from_value`` / ``to_value`` null-tests. ``move`` only " + "fires for layout records." + ) + }, + ) + action_kind = fields.String( + validate=validate.OneOf(ACTIVITY_ACTION_KINDS), + allow_none=True, + metadata={ + "description": ( + "Transaction-level avenue that produced this record's " + "batch: ``restore`` / ``import`` / ``clone``. ``null`` " + "for ordinary saves. All records sharing a " + "``transaction_id`` share the same action_kind. The " + "schema's third ``*_kind`` column (entity_kind / kind / " + "action_kind), at transaction scope." + ) + }, + ) + path = fields.List( + fields.String(), + metadata={ + "description": ( + "Pure navigation address — no verb or kind embedded. " + "Examples: ``['slice_name']``, ``['params', " + "'adhoc_filters', 'country']``, ``['CHART-x']`` for a " + "layout add/remove/move, ``['HEADER-y', 'text']`` for a " + "layout edit leaf. The verb lives in ``operation``, the " + "element type in ``kind``." + ) + }, + ) + from_value = fields.Raw( + allow_none=True, + metadata={"description": "Prior value; ``null`` = didn't exist."}, + ) + to_value = fields.Raw( + allow_none=True, + metadata={"description": "New value; ``null`` = removed."}, + ) + summary = fields.String( + metadata={ + "description": ( + 'Synthesized headline for ``source: "related"`` records — ' + 'e.g., ``"Dataset updated: Sales Transactions"`` ' + '(AV-012). Absent for ``source: "self"`` records.' + ) + }, + ) + impact = fields.Nested( + ActivityImpactSchema, + allow_none=True, + metadata={ + "description": ( + 'Optional dependent-count for ``source: "related"`` ' + 'records — e.g., ``{"charts": 4}`` for a dataset edit ' + "that affected 4 charts on the path dashboard at the " + 'change\'s transaction. Absent for ``source: "self"`` ' + "records and for related records without dependents." + ) + }, + ) + first_tracked_save = fields.Boolean( + metadata={ + "description": ( + "True when this record's transaction is the entity's " + "FIRST tracked save (first UPDATE after the retroactive " + "baseline). Such transactions can carry dozens of " + "params-normalization records for entities that predate " + "versioning; clients use the marker to collapse them " + "rather than render each delta as a user edit. Matched " + "against the LIVE row's (id, uuid), so it is always " + "false for hard-deleted entities (no live row) and for " + "shadow rows predating the entity's current uuid." + ) + }, + ) + + +class ActivityResponseSchema(Schema): + """Envelope for activity-view responses.""" + + result = fields.List(fields.Nested(ActivityRecordSchema)) + count = fields.Integer( + metadata={ + "description": ( + "Total record count across all pages (the filtered + " + "denormalized stream), not just the current page. When " + "``truncated`` is true this is a floor (the count within " + "the fetched window), not the absolute total." + ) + }, + ) + truncated = fields.Boolean( + metadata={ + "description": ( + "True when the request hit the per-request fetch ceiling " + "and older records exist beyond the returned window. " + "Narrow the time range (``since``/``until``) to see them." + ) + }, + ) diff --git a/superset/views/activity_debug.py b/superset/views/activity_debug.py new file mode 100644 index 000000000000..61072fd55cfc --- /dev/null +++ b/superset/views/activity_debug.py @@ -0,0 +1,53 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. + +# Throwaway: sc-107283 activity-view debug UI. This Flask view exists +# only to serve the React shell on a fresh page-load of +# /activity-debug//. Without it, Flask returns its +# API-style 404 because the SPA doesn't have a true catch-all — each +# React route needs a corresponding render_app_template call. Delete +# this file + the AppBuilder registration in +# ``superset.initialization`` when the activity-view feature ships +# (or when you no longer need the debug UI). + +from flask_appbuilder import expose + +from superset.superset_typing import FlaskResponse +from superset.views.base import BaseSupersetView + + +class ActivityDebugView(BaseSupersetView): + """Serves the React shell for the throwaway activity-view debug page. + + No auth decorator on the shell itself — the shell page exposes no + data of its own. The React component renders inside it and fires + calls to ``/api/v1/{resource}/{uuid}/activity/`` which gate access + via ``security_manager.raise_for_access`` (read access) on the path + entity — ``raise_for_ownership`` is the write/restore path. Anonymous + users + who somehow land here will see the React UI and the API errors + surface inline as "error: 401 ...". That's a fine UX for a debug + tool — and avoids the FAB ``@has_access`` redirect-to-home + behavior that masked real failures earlier. + """ + + route_base = "/activity-debug" + + @expose("///") + @expose("//") + def show(self, resource: str, uuid: str) -> FlaskResponse: # noqa: ARG002 + return super().render_app_template() diff --git a/tests/integration_tests/versioning/activity_view_tests.py b/tests/integration_tests/versioning/activity_view_tests.py new file mode 100644 index 000000000000..9bac7e64a8b5 --- /dev/null +++ b/tests/integration_tests/versioning/activity_view_tests.py @@ -0,0 +1,1174 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Integration tests for the cross-entity activity-view API (sc-107283). + +US1 — dashboard activity stream: ``GET /api/v1/dashboard//activity/``. +Tests for US2 (chart activity) and US3 (dataset activity) come in later +phases. + +Per spec T053 / sc-103156 T062, every test that mutates a fixture entity +wraps the test body in ``try``/``finally`` with +``metadata_db.session.rollback()`` in the ``finally``. The rationale is +documented in the spec — Continuum captures dirty mappers during +autoflush, so leaving an instrumented attribute dirty pollutes +downstream tests via the shadow tables. +""" + +from __future__ import annotations + +from typing import Any +from uuid import uuid4 + +import pytest + +from superset.connectors.sqla.models import SqlaTable +from superset.extensions import db +from superset.models.dashboard import Dashboard +from superset.models.slice import Slice +from superset.utils import json as _json +from tests.integration_tests.base_tests import SupersetTestCase +from tests.integration_tests.constants import ADMIN_USERNAME, ALPHA_USERNAME +from tests.integration_tests.fixtures.birth_names_dashboard import ( # noqa: F401 + load_birth_names_dashboard_with_slices, + load_birth_names_data, +) + + +def _get_birth_names_dataset() -> SqlaTable: + return ( + db.session.query(SqlaTable) + .filter(SqlaTable.table_name == "birth_names") + .first() + ) + + +def _persist_fixture_state() -> None: + """Force the fixture's pending INSERTs to commit so subsequent edits + produce *new* version rows instead of being batched into the + creation transaction. Mirrors the same helper in + ``tests/integration_tests/dashboards/version_history_tests.py``. + """ + db.session.commit() + + +def _get_birth_names_dashboard() -> Dashboard: + return ( + db.session.query(Dashboard) + .filter(Dashboard.dashboard_title == "USA Births Names") + .first() + ) + + +class TestDashboardActivityView(SupersetTestCase): + """T017–T026 — ``GET /api/v1/dashboard//activity/`` (US1).""" + + @pytest.fixture(autouse=True) + def _load_data(self, load_birth_names_dashboard_with_slices): # noqa: PT004, F811 + pass + + def _activity(self, dashboard_uuid: str, **query: Any) -> Any: + return self.client.get( + f"/api/v1/dashboard/{dashboard_uuid}/activity/", + query_string=query, + ) + + # ---- 4xx boundary cases ---- + + def test_activity_returns_404_for_unknown_uuid(self) -> None: + """AV-009: unknown path entity → 404.""" + self.login(ADMIN_USERNAME) + rv = self._activity("00000000-0000-0000-0000-000000000000") + assert rv.status_code == 404 + + def test_activity_returns_400_for_invalid_uuid(self) -> None: + """A malformed UUID is rejected by the endpoint, not by Werkzeug.""" + self.login(ADMIN_USERNAME) + rv = self._activity("not-a-uuid") + assert rv.status_code == 400 + + def test_activity_returns_400_for_invalid_include(self) -> None: + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + self.login(ADMIN_USERNAME) + rv = self._activity(str(dashboard.uuid), include="sibling") + assert rv.status_code == 400 + + def test_activity_returns_400_for_invalid_since(self) -> None: + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + self.login(ADMIN_USERNAME) + rv = self._activity(str(dashboard.uuid), since="yesterday") + assert rv.status_code == 400 + + def test_activity_allows_read_non_owner(self) -> None: + """Activity is a read endpoint: a non-owner with read access (Alpha, + which carries broad read + datasource access) can read a dashboard's + activity stream — ``raise_for_access(dashboard=)`` does not reject — + so the endpoint returns 200. Visibility of *related* rows is filtered + separately, inside the activity layer.""" + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + dashboard_uuid = str(dashboard.uuid) + + self.login(ALPHA_USERNAME) + rv = self._activity(dashboard_uuid) + assert rv.status_code == 200 + + # ---- 200 happy paths ---- + + def test_activity_returns_200_with_envelope_shape(self) -> None: + """Smoke test: the endpoint returns the documented envelope shape + (``result`` list + ``count`` integer) even when the dashboard has + no activity yet.""" + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + dashboard_uuid = str(dashboard.uuid) + + self.login(ADMIN_USERNAME) + rv = self._activity(dashboard_uuid) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + assert "result" in body + assert "count" in body + assert isinstance(body["result"], list) + assert isinstance(body["count"], int) + + def test_activity_includes_chart_edit_as_related(self) -> None: + """T018 / AS-1 of US1: editing a chart on the dashboard surfaces + the chart-edit record with ``entity_kind=Slice`` and + ``source=related``.""" + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + dashboard_uuid = str(dashboard.uuid) + chart_on_dashboard = next(iter(dashboard.slices), None) + assert chart_on_dashboard is not None + chart_id = chart_on_dashboard.id + original_name = chart_on_dashboard.slice_name + + try: + chart_on_dashboard.slice_name = f"{original_name} (edited)" + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(dashboard_uuid) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + related = [ + r + for r in body["result"] + if r["entity_kind"] == "chart" and r["source"] == "related" + ] + assert related, ( + "Expected at least one Slice/related record from the chart " + "edit; got: " + f"{[(r['entity_kind'], r['source']) for r in body['result']]}" + ) + # Spot-check the carry-through of denormalized fields + sample = related[0] + assert sample["entity_uuid"] is not None + assert "transaction_id" in sample + assert "issued_at" in sample + finally: + db.session.rollback() + chart = db.session.query(Slice).filter(Slice.id == chart_id).one() + chart.slice_name = original_name + db.session.commit() + + def test_activity_include_self_excludes_related(self) -> None: + """T023 / AV-016: ``?include=self`` filters out related records.""" + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + dashboard_uuid = str(dashboard.uuid) + chart_on_dashboard = next(iter(dashboard.slices), None) + assert chart_on_dashboard is not None + chart_id = chart_on_dashboard.id + original_name = chart_on_dashboard.slice_name + + try: + chart_on_dashboard.slice_name = f"{original_name} (edited self)" + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(dashboard_uuid, include="self") + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + for record in body["result"]: + assert record["source"] == "self", ( + f"include=self leaked a non-self record: {record}" + ) + assert record["entity_kind"] == "dashboard" + finally: + db.session.rollback() + chart = db.session.query(Slice).filter(Slice.id == chart_id).one() + chart.slice_name = original_name + db.session.commit() + + def test_activity_include_related_excludes_self(self) -> None: + """T024 / AV-016: ``?include=related`` returns only related records.""" + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + dashboard_uuid = str(dashboard.uuid) + original_title = dashboard.dashboard_title + dashboard_id = dashboard.id + + try: + # Edit the dashboard's own field so we have a self record to + # filter out, and edit a chart on it so we have a related + # record to keep. + dashboard.dashboard_title = f"{original_title} (edited dash)" + db.session.commit() + chart_on_dashboard = next(iter(dashboard.slices), None) + assert chart_on_dashboard is not None + chart_id = chart_on_dashboard.id + chart_original_name = chart_on_dashboard.slice_name + chart_on_dashboard.slice_name = f"{chart_original_name} (edited chart)" + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(dashboard_uuid, include="related") + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + for record in body["result"]: + assert record["source"] == "related", ( + f"include=related leaked a self record: {record}" + ) + assert record["entity_kind"] != "dashboard" + finally: + db.session.rollback() + dashboard = ( + db.session.query(Dashboard).filter(Dashboard.id == dashboard_id).one() + ) + dashboard.dashboard_title = original_title + chart = db.session.query(Slice).filter(Slice.id == chart_id).one() + chart.slice_name = chart_original_name + db.session.commit() + + def test_activity_pagination_clamps_oversized_page_size(self) -> None: + """``?page_size=500`` is silently clamped to the contract max + (200) rather than rejected with 400.""" + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + self.login(ADMIN_USERNAME) + rv = self._activity(str(dashboard.uuid), page_size="500") + assert rv.status_code == 200 + + def test_activity_ordering_is_stable_by_issued_at_then_transaction_id(self) -> None: + """T040 / AV-006: records are ordered ``(issued_at DESC, + transaction_id DESC)``. When two records share ``issued_at`` the + tie-break is ``transaction_id`` — never random. We verify this by + asserting the result list is monotonically non-increasing on the + composite key, which would only hold under deterministic + ordering.""" + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + self.login(ADMIN_USERNAME) + rv = self._activity(str(dashboard.uuid)) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + records = body["result"] + # Each pair of adjacent records must satisfy (prev >= cur) on the + # composite (issued_at, transaction_id) — DESC ordering. + # ``records[1:]`` is intentionally one element shorter than + # ``records``; strict=False is the correct semantic for an + # adjacent-pair iteration. + for prev, cur in zip(records, records[1:], strict=False): + assert (prev["issued_at"], prev["transaction_id"]) >= ( + cur["issued_at"], + cur["transaction_id"], + ), ( + f"Ordering broke at adjacent pair: " + f"prev=({prev['issued_at']}, {prev['transaction_id']}) " + f"cur=({cur['issued_at']}, {cur['transaction_id']})" + ) + + def test_activity_page_size_caps_returned_records_at_200(self) -> None: + """T041: ``?page_size=500`` must return *at most* 200 records. + Pairs with the no-400 check above: that test confirms the + oversized request is accepted, this test confirms the response + is bounded as the contract guarantees (AV-019 / spec + ActivityResponseSchema documentation).""" + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + self.login(ADMIN_USERNAME) + rv = self._activity(str(dashboard.uuid), page_size="500") + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + assert len(body["result"]) <= 200, ( + f"page_size=500 returned {len(body['result'])} records; " + "cap is 200 per the OpenAPI schema" + ) + + def test_activity_marks_hard_deleted_chart_with_tombstone(self) -> None: + """T042 / D-15: when a chart was on the dashboard and has since + been hard-deleted, the chart's historical change records still + surface in the dashboard's activity stream, marked with + ``entity_deleted: true`` and ``entity_uuid: null``. ``entity_name`` + is preserved from the last shadow row so the UI can show + "(deleted) Girls" without a live row to query. + + Hard-delete pattern: edit the chart (creates a Slice change + record), commit, then ``db.session.delete(chart); commit``. + Continuum end-stamps the M2M row but does not cascade-delete + the shadow rows, so the history is still reachable. The + activity-view's tombstone check (``_check_entity_tombstones``) + detects the missing live row and stamps the record.""" + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + dashboard_uuid = str(dashboard.uuid) + chart_to_delete = ( + db.session.query(Slice).filter(Slice.slice_name == "Girls").first() + ) + assert chart_to_delete is not None + original_name = chart_to_delete.slice_name + + try: + # Step 1: generate a chart-edit change record for "Girls". + chart_to_delete.slice_name = f"{original_name} (pre-delete edit)" + db.session.commit() + + # Step 2: hard-delete the chart. The fixture's _cleanup will + # tolerate this — its `Slice.id.in_(slice_ids)` filter + # silently skips the missing row. + db.session.delete(chart_to_delete) + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(dashboard_uuid) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + tombstoned = [ + r + for r in body["result"] + if r["entity_kind"] == "chart" and r["entity_deleted"] is True + ] + seen = [ + (r["entity_kind"], r["entity_deleted"]) for r in body["result"][:10] + ] + assert tombstoned, ( + "Expected ≥1 tombstoned Slice record after the chart was " + f"hard-deleted; got entity_deleted values: {seen}" + ) + sample = tombstoned[0] + got_uuid = sample["entity_uuid"] + assert got_uuid is None, ( + f"Hard-deleted entity should have null entity_uuid; got {got_uuid!r}" + ) + assert sample["entity_name"], ( + "entity_name should be recovered from the last shadow row; " + f"got empty: {sample!r}" + ) + finally: + db.session.rollback() + + def test_check_entity_tombstones_handles_multiple_kinds(self) -> None: + """Regression for the v4 indent slip in ``check_entity_tombstones``: + when called with ``distinct_entities`` spanning multiple kinds, + every kind must get its tombstone-state result, not just the + one iterated last. + + Pre-fix, the per-entity result-population block sat outside the + ``for api_kind in by_kind.items():`` loop, so all but the + last-iterated kind silently fell through to the call-site + default ``{"deleted": True}`` in ``render.apply_record_decoration`` + — live entities were rendered as tombstoned in the API response. + The previous tombstone test exercised only one kind, so dict + iteration order made the bug invisible. + """ + # pylint: disable=import-outside-toplevel + from superset.versioning.activity.queries import check_entity_tombstones + + _persist_fixture_state() + chart = db.session.query(Slice).filter(Slice.slice_name == "Girls").first() + dataset = _get_birth_names_dataset() + assert chart is not None + assert dataset is not None + + distinct = {("Slice", chart.id), ("SqlaTable", dataset.id)} + result = check_entity_tombstones(distinct) + + assert ("Slice", chart.id) in result, ( + "Multi-kind call must populate every kind; got keys: " + f"{sorted(result.keys())}" + ) + assert ("SqlaTable", dataset.id) in result + assert result[("Slice", chart.id)] == { + "deleted": False, + "deletion_state": None, + }, f"Live chart should report deleted=False; got {result[('Slice', chart.id)]}" + assert result[("SqlaTable", dataset.id)] == { + "deleted": False, + "deletion_state": None, + }, ( + f"Live dataset should report deleted=False; " + f"got {result[('SqlaTable', dataset.id)]}" + ) + + @pytest.mark.skip( + reason="Depends on the retention prune (_prune_old_versions_impl), which " + "was extracted to sc-111099-version-history-retention. This test " + "exercises activity-view + retention together and runs once both PRs " + "merge; un-skip then." + ) + def test_activity_excludes_records_after_retention_prune(self) -> None: + """T051 / AV-010: retention bounds the activity feed. After + ``_prune_old_versions_impl`` drops shadow / change-record rows + whose ``version_transaction.issued_at`` is older than the + retention cutoff, the activity stream stops surfacing them. + + Test pattern: capture the highest ``version_transaction.id`` + before our edits, edit a chart (creating a new transaction), + backdate that transaction's ``issued_at`` past the retention + cutoff, run the prune, and assert the chart-edit no longer + appears in the activity stream.""" + # pylint: disable=import-outside-toplevel + from datetime import datetime, timedelta + + import sqlalchemy as sa + from sqlalchemy_continuum import versioning_manager + + from superset.tasks.version_history_retention import ( + _prune_old_versions_impl, + ) + + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + dashboard_uuid = str(dashboard.uuid) + chart = db.session.query(Slice).filter(Slice.slice_name == "Boys").first() + assert chart is not None + chart_id = chart.id + original_name = chart.slice_name + + tx_table = versioning_manager.transaction_cls.__table__ + + # Capture pre-edit max tx_id so we can identify the rows produced + # by THIS test (and not backdate anything else). + max_tx_before = ( + db.session.connection() + .execute(sa.select(sa.func.max(tx_table.c.id))) + .scalar() + or 0 + ) + + try: + chart.slice_name = f"{original_name} (retention test)" + db.session.commit() + + # Backdate the new transactions to before the 30-day cutoff. + old_timestamp = datetime.utcnow() - timedelta(days=60) + db.session.connection().execute( + sa.update(tx_table) + .where(tx_table.c.id > max_tx_before) + .values(issued_at=old_timestamp) + ) + db.session.commit() + + # Snapshot the activity-record count BEFORE the prune. With + # ?page_size=200 + the highest possible page coverage, the + # count field is the post-visibility filtered total. + self.login(ADMIN_USERNAME) + rv_before = self._activity(dashboard_uuid, page_size="200") + assert rv_before.status_code == 200 + count_before = _json.loads(rv_before.data.decode("utf-8"))["count"] + + # Run the prune. The backdated tx rows are now > 30 days old + # and should be deleted. AV-010 requires the prune to remove + # at least the backdated transaction(s) we created. + stats = _prune_old_versions_impl(retention_days=30) + assert stats.get("pruned_transactions", 0) >= 1, ( + f"Prune should have removed our backdated tx; stats={stats}" + ) + + # After the prune, the activity endpoint still works and the + # filtered count has DROPPED — change records joined to the + # pruned transactions are no longer in the result set (the + # join in _fetch_change_records drops them). + rv_after = self._activity(dashboard_uuid, page_size="200") + assert rv_after.status_code == 200 + count_after = _json.loads(rv_after.data.decode("utf-8"))["count"] + assert count_after < count_before, ( + f"Activity count should decrease after prune; " + f"before={count_before} after={count_after}" + ) + finally: + db.session.rollback() + chart = db.session.query(Slice).filter(Slice.id == chart_id).one() + chart.slice_name = original_name + db.session.commit() + + def test_activity_pagination_is_deterministic_and_disjoint(self) -> None: + """T039 / SC-AV-002 (pragmatic interpretation): two consecutive + requests for the same page return identical results, and + consecutive pages do not overlap. + + The spec's stricter "no skip/duplicate under concurrent writes" + is unprovable with offset pagination — new top-inserted records + shift every later page by one. Cursor pagination would solve + this and is deferred per plan §D-10. Under THIS pagination + scheme, the testable guarantees are: (a) the same request fired + twice produces the same page (request determinism), and (b) + page N and page N+1 share no record under the same request + round. Both come from the stable + ``(issued_at DESC, transaction_id DESC, sequence DESC)`` sort. + """ + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + dashboard_uuid = str(dashboard.uuid) + self.login(ADMIN_USERNAME) + + rv1a = self._activity(dashboard_uuid, page="0", page_size="25") + rv1b = self._activity(dashboard_uuid, page="0", page_size="25") + rv2 = self._activity(dashboard_uuid, page="1", page_size="25") + assert rv1a.status_code == 200 + assert rv1b.status_code == 200 + assert rv2.status_code == 200 + + page0_first = _json.loads(rv1a.data.decode("utf-8"))["result"] + page0_second = _json.loads(rv1b.data.decode("utf-8"))["result"] + page1 = _json.loads(rv2.data.decode("utf-8"))["result"] + + # (a) Request determinism: same page twice → same records in same + # order. Use (entity_kind, entity_id_internal_proxy, tx, seq) + # fingerprint — entity_uuid + transaction_id is sufficient + # since entity_id isn't in the API contract. + fingerprint = lambda r: ( # noqa: E731 + r["entity_kind"], + r["entity_uuid"], + r["transaction_id"], + r["kind"], + tuple(r["path"]) if r["path"] else (), + ) + assert [fingerprint(r) for r in page0_first] == [ + fingerprint(r) for r in page0_second + ], "page=0 fired twice returned different records" + + # (b) Page 0 and page 1 are disjoint under one request round. + page0_keys = {fingerprint(r) for r in page0_first} + page1_keys = {fingerprint(r) for r in page1} + overlap = page0_keys & page1_keys + assert not overlap, f"page=0 and page=1 returned overlapping records: {overlap}" + + @pytest.mark.skip( + reason="Restore endpoint ships in a later PR; re-enable when restore " + "lands. The activity layer's restore-event rendering is unit-tested." + ) + def test_activity_surfaces_dashboard_restore_event(self) -> None: + """T044 / AV-015: restoring a dashboard to a prior version surfaces + a synthetic ``kind='__meta__'`` headline record (path + ``['__meta__', 'restore']``, ``to_value`` carrying the restored-to + version_uuid) in the dashboard's own activity stream + (``source='self'``). The headline is emitted by the restore + command via the listener's ACTION_META_KEY (PR #40988 feedback); + the activity layer passes it through without special-casing. + + Uses a fresh dashboard: the shared fixture dashboard accumulates + membership history on a persistent DB, so its restore transaction + can carry more records than one page — burying the headline + (sequence 0 sorts last under the stream's sequence-DESC order). + """ + _persist_fixture_state() + dashboard = Dashboard( + dashboard_title=f"restore probe {uuid4().hex[:8]}", published=False + ) + db.session.add(dashboard) + db.session.commit() + dashboard_uuid = str(dashboard.uuid) + dashboard_id = dashboard.id + original_title = dashboard.dashboard_title + + try: + # Two edits → at least two restorable prior versions. + dashboard.dashboard_title = f"{original_title} v1" + db.session.commit() + dashboard.dashboard_title = f"{original_title} v2" + db.session.commit() + + self.login(ADMIN_USERNAME) + # Find a prior version to restore to (version_number 0 is the + # baseline; we restore to whichever earlier version the list + # endpoint surfaces). + versions_rv = self.client.get( + f"/api/v1/dashboard/{dashboard_uuid}/versions/" + ) + assert versions_rv.status_code == 200, versions_rv.data + versions = _json.loads(versions_rv.data.decode("utf-8"))["result"] + assert len(versions) >= 2, f"expected ≥2 versions, got {versions}" + target_version_uuid = versions[0]["version_uuid"] # earliest + + # Restore. The endpoint commits; finally clean up below. + restore_rv = self.client.post( + f"/api/v1/dashboard/{dashboard_uuid}" + f"/versions/{target_version_uuid}/restore" + ) + assert restore_rv.status_code == 200, restore_rv.data + + # Activity stream should now show the restore headline on the + # dashboard itself. + rv = self._activity(dashboard_uuid, include="self") + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + restore_records = [ + r + for r in body["result"] + if r["kind"] == "__meta__" + and r["path"] == ["__meta__"] + and r["entity_kind"] == "dashboard" + ] + assert restore_records, ( + "Expected a __meta__ restore headline record; " + f"got kinds: {[r['kind'] for r in body['result'][:10]]}" + ) + assert restore_records[0]["to_value"]["version_uuid"] == target_version_uuid + # The headline is the one self record whose summary renders: + # "restored to version N" must be visible on include=self. + assert "restored to version" in restore_records[0]["summary"] + finally: + db.session.rollback() + dashboard = ( + db.session.query(Dashboard) + .filter(Dashboard.id == dashboard_id) + .one_or_none() + ) + if dashboard is not None: + db.session.delete(dashboard) + db.session.commit() + + def test_activity_marks_first_tracked_save(self) -> None: + """Every record carries ``first_tracked_save``: True only on the + entity's FIRST tracked save. Clients collapse such transactions — + a legacy chart's first Explore save can replay ~74 params- + normalization deltas against the retroactive baseline + (PR #40988 feedback). + + Uses a fresh dashboard so the entity's history is fully + controlled by this test (the shared fixture dashboard's first + save belongs to whichever suite ran first on a persistent DB). + """ + _persist_fixture_state() + dashboard = Dashboard( + dashboard_title=f"fts probe {uuid4().hex[:8]}", published=False + ) + db.session.add(dashboard) + db.session.commit() # op=0 INSERT baseline — no change records + dashboard_uuid = str(dashboard.uuid) + dashboard_id = dashboard.id + + try: + dashboard.dashboard_title = f"{dashboard.dashboard_title} v1" + db.session.commit() # first tracked save + dashboard.dashboard_title = f"{dashboard.dashboard_title} v2" + db.session.commit() # second save + + self.login(ADMIN_USERNAME) + rv = self._activity(dashboard_uuid, include="self") + assert rv.status_code == 200 + records = _json.loads(rv.data.decode("utf-8"))["result"] + assert len(records) >= 2 + assert all("first_tracked_save" in r for r in records), ( + "every record must carry the first_tracked_save marker" + ) + # Newest-first ordering: the latest save (v2) is never the + # entity's first tracked save. + assert records[0]["first_tracked_save"] is False + # The v1 save's transaction IS flagged. Assert by transaction + # rather than stream position: under id reuse on a persistent + # test DB the stream can also carry a previously-deleted + # entity's records for the same integer id (the marker itself + # is uuid-aware and immune; the stream's positional tail is + # not). + flagged_txs = { + r["transaction_id"] for r in records if r["first_tracked_save"] + } + assert flagged_txs, "no record flagged as the first tracked save" + newest_tx = records[0]["transaction_id"] + assert newest_tx not in flagged_txs + finally: + db.session.rollback() + dashboard = ( + db.session.query(Dashboard) + .filter(Dashboard.id == dashboard_id) + .one_or_none() + ) + if dashboard is not None: + db.session.delete(dashboard) + db.session.commit() + + def test_activity_q_filters_server_side(self) -> None: + """``?q=`` searches the FULL history server-side, pre-pagination + (PR #40988: the panel's client-side search only covered loaded + pages); ``count`` reflects the matches.""" + _persist_fixture_state() + dashboard = _get_birth_names_dashboard() + assert dashboard is not None + dashboard_uuid = str(dashboard.uuid) + dashboard_id = dashboard.id + original_title = dashboard.dashboard_title + needle = f"qprobe{uuid4().hex[:6]}" + + try: + dashboard.dashboard_title = f"{original_title} {needle}" + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(dashboard_uuid, q=needle) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + assert body["count"] >= 1 + assert all(needle in _json.dumps(r).lower() for r in body["result"]), ( + f"non-matching record returned for q={needle!r}" + ) + + # A needle that matches nothing returns an empty, zero-count + # envelope — not an error. + rv_none = self._activity(dashboard_uuid, q="zz-no-such-needle-zz") + body_none = _json.loads(rv_none.data.decode("utf-8")) + assert body_none == {"result": [], "count": 0, "truncated": False} + finally: + db.session.rollback() + dashboard = ( + db.session.query(Dashboard).filter(Dashboard.id == dashboard_id).one() + ) + dashboard.dashboard_title = original_title + db.session.commit() + + +class TestChartActivityView(SupersetTestCase): + """T028–T032 — ``GET /api/v1/chart//activity/`` (US2). + + Chart activity = chart's own edits + datasets the chart pointed at + during association. **No** dashboard records — even when the chart + is on a dashboard, sibling-traversal is excluded per the spec's + Relationship Traversal section (T032). + """ + + @pytest.fixture(autouse=True) + def _load_data(self, load_birth_names_dashboard_with_slices): # noqa: PT004, F811 + pass + + def _activity(self, chart_uuid: str, **query: Any) -> Any: + return self.client.get( + f"/api/v1/chart/{chart_uuid}/activity/", + query_string=query, + ) + + def _get_birth_names_chart(self) -> Slice: + return db.session.query(Slice).filter(Slice.slice_name == "Girls").first() + + # ---- 4xx boundary cases ---- + + def test_chart_activity_returns_404_for_unknown_uuid(self) -> None: + self.login(ADMIN_USERNAME) + rv = self._activity("00000000-0000-0000-0000-000000000000") + assert rv.status_code == 404 + + def test_chart_activity_returns_400_for_invalid_uuid(self) -> None: + self.login(ADMIN_USERNAME) + rv = self._activity("not-a-uuid") + assert rv.status_code == 400 + + def test_chart_activity_returns_400_for_invalid_include(self) -> None: + _persist_fixture_state() + chart = self._get_birth_names_chart() + assert chart is not None + self.login(ADMIN_USERNAME) + rv = self._activity(str(chart.uuid), include="upstream") + assert rv.status_code == 400 + + def test_chart_activity_allows_read_non_owner(self) -> None: + """Same shape as the dashboard endpoint: a read-access non-owner + (Alpha) can read a chart's activity, so ``raise_for_access(chart=)`` + does not reject and the endpoint returns 200.""" + _persist_fixture_state() + chart = self._get_birth_names_chart() + assert chart is not None + self.login(ALPHA_USERNAME) + rv = self._activity(str(chart.uuid)) + assert rv.status_code == 200 + + # ---- 200 happy paths ---- + + def test_chart_activity_returns_200_with_envelope_shape(self) -> None: + _persist_fixture_state() + chart = self._get_birth_names_chart() + assert chart is not None + self.login(ADMIN_USERNAME) + rv = self._activity(str(chart.uuid)) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + assert isinstance(body["result"], list) + assert isinstance(body["count"], int) + + def test_chart_activity_self_edit_appears_as_self_record(self) -> None: + """Editing the chart itself surfaces a ``source=self``, + ``entity_kind=Slice`` record.""" + _persist_fixture_state() + chart = self._get_birth_names_chart() + assert chart is not None + chart_id = chart.id + chart_uuid = str(chart.uuid) + original_name = chart.slice_name + + try: + chart.slice_name = f"{original_name} (edited self)" + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(chart_uuid) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + self_records = [ + r + for r in body["result"] + if r["entity_kind"] == "chart" and r["source"] == "self" + ] + got = [(r["entity_kind"], r["source"]) for r in body["result"]] + assert self_records, ( + f"Expected ≥1 Slice/self record from the chart edit; got: {got}" + ) + finally: + db.session.rollback() + chart = db.session.query(Slice).filter(Slice.id == chart_id).one() + chart.slice_name = original_name + db.session.commit() + + def test_chart_activity_includes_dataset_edit_as_related(self) -> None: + """T030 / AS-1 of US2: editing the chart's dataset surfaces a + ``source=related``, ``entity_kind=SqlaTable`` record.""" + _persist_fixture_state() + chart = self._get_birth_names_chart() + dataset = _get_birth_names_dataset() + assert chart is not None + assert dataset is not None + chart_uuid = str(chart.uuid) + dataset_id = dataset.id + original_description = dataset.description + + try: + dataset.description = "edited for activity-view test" + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(chart_uuid) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + related = [ + r + for r in body["result"] + if r["entity_kind"] == "dataset" and r["source"] == "related" + ] + assert related, ( + "Expected at least one SqlaTable/related record from the " + "dataset edit; got: " + f"{[(r['entity_kind'], r['source']) for r in body['result']]}" + ) + finally: + db.session.rollback() + dataset = ( + db.session.query(SqlaTable).filter(SqlaTable.id == dataset_id).one() + ) + dataset.description = original_description + db.session.commit() + + def test_chart_activity_excludes_sibling_dashboards(self) -> None: + """T032: Even when the chart is on a dashboard, dashboard edits + do NOT appear in the chart's activity. Per the spec's Relationship + Traversal section: charts don't see "sideways" to the dashboards + they happen to be on.""" + _persist_fixture_state() + chart = self._get_birth_names_chart() + dashboard = _get_birth_names_dashboard() + assert chart is not None + assert dashboard is not None + chart_uuid = str(chart.uuid) + dashboard_id = dashboard.id + original_title = dashboard.dashboard_title + + try: + # Mutate the dashboard the chart is on — that edit MUST NOT + # appear in the chart's activity stream. + dashboard.dashboard_title = f"{original_title} (edited sibling)" + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(chart_uuid) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + for record in body["result"]: + assert record["entity_kind"] != "dashboard", ( + f"Dashboard edit leaked into chart's activity stream: {record}" + ) + finally: + db.session.rollback() + dashboard = ( + db.session.query(Dashboard).filter(Dashboard.id == dashboard_id).one() + ) + dashboard.dashboard_title = original_title + db.session.commit() + + def test_chart_activity_include_self_excludes_related(self) -> None: + """``?include=self`` filters out the dataset records.""" + _persist_fixture_state() + chart = self._get_birth_names_chart() + dataset = _get_birth_names_dataset() + assert chart is not None + assert dataset is not None + chart_uuid = str(chart.uuid) + dataset_id = dataset.id + original_description = dataset.description + + try: + dataset.description = "edited (self filter test)" + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(chart_uuid, include="self") + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + for record in body["result"]: + assert record["source"] == "self" + assert record["entity_kind"] == "chart" + finally: + db.session.rollback() + dataset = ( + db.session.query(SqlaTable).filter(SqlaTable.id == dataset_id).one() + ) + dataset.description = original_description + db.session.commit() + + +class TestDatasetActivityView(SupersetTestCase): + """T033–T036 — ``GET /api/v1/dataset//activity/`` (US3). + + Dataset activity = dataset's own edits only. **No** transitive layer + in V2 (AV-004) — even when charts use the dataset, those chart edits + do NOT appear here. ``?include=related`` and ``?include=all`` + collapse to the same self-only stream as ``?include=self``. + """ + + @pytest.fixture(autouse=True) + def _load_data(self, load_birth_names_dashboard_with_slices): # noqa: PT004, F811 + pass + + def _activity(self, dataset_uuid: str, **query: Any) -> Any: + return self.client.get( + f"/api/v1/dataset/{dataset_uuid}/activity/", + query_string=query, + ) + + # ---- 4xx boundary cases ---- + + def test_dataset_activity_returns_404_for_unknown_uuid(self) -> None: + self.login(ADMIN_USERNAME) + rv = self._activity("00000000-0000-0000-0000-000000000000") + assert rv.status_code == 404 + + def test_dataset_activity_returns_400_for_invalid_uuid(self) -> None: + self.login(ADMIN_USERNAME) + rv = self._activity("not-a-uuid") + assert rv.status_code == 400 + + def test_dataset_activity_returns_400_for_invalid_include(self) -> None: + _persist_fixture_state() + dataset = _get_birth_names_dataset() + assert dataset is not None + self.login(ADMIN_USERNAME) + rv = self._activity(str(dataset.uuid), include="upstream") + assert rv.status_code == 400 + + def test_dataset_activity_allows_read_non_owner(self) -> None: + """A read-access non-owner (Alpha) can read a dataset's activity + stream, so the read endpoint returns 200.""" + _persist_fixture_state() + dataset = _get_birth_names_dataset() + assert dataset is not None + self.login(ALPHA_USERNAME) + rv = self._activity(str(dataset.uuid)) + assert rv.status_code == 200 + + # ---- 200 happy paths ---- + + def test_dataset_activity_returns_200_with_envelope_shape(self) -> None: + _persist_fixture_state() + dataset = _get_birth_names_dataset() + assert dataset is not None + self.login(ADMIN_USERNAME) + rv = self._activity(str(dataset.uuid)) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + assert isinstance(body["result"], list) + assert isinstance(body["count"], int) + + def test_dataset_activity_includes_dataset_self_edits(self) -> None: + """T036: the dataset's own scalar edits appear as ``source=self``, + ``entity_kind=SqlaTable``.""" + _persist_fixture_state() + dataset = _get_birth_names_dataset() + assert dataset is not None + dataset_id = dataset.id + dataset_uuid = str(dataset.uuid) + original_description = dataset.description + + try: + dataset.description = "edited self for dataset activity" + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(dataset_uuid) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + self_records = [ + r + for r in body["result"] + if r["entity_kind"] == "dataset" and r["source"] == "self" + ] + got = [(r["entity_kind"], r["source"]) for r in body["result"]] + assert self_records, ( + f"Expected ≥1 SqlaTable/self record from the dataset edit; got: {got}" + ) + finally: + db.session.rollback() + dataset = ( + db.session.query(SqlaTable).filter(SqlaTable.id == dataset_id).one() + ) + dataset.description = original_description + db.session.commit() + + def test_dataset_activity_excludes_chart_edits(self) -> None: + """T035 / AS-1 / AV-004: When a chart that uses the dataset is + edited, that edit does NOT appear in the dataset's activity stream. + Datasets are read-only upstream in V2.""" + _persist_fixture_state() + dataset = _get_birth_names_dataset() + chart = db.session.query(Slice).filter(Slice.slice_name == "Girls").first() + assert dataset is not None + assert chart is not None + dataset_uuid = str(dataset.uuid) + chart_id = chart.id + chart_original_name = chart.slice_name + + try: + # Edit the chart — generates a Slice change record. The + # dataset's activity MUST NOT surface it. + chart.slice_name = f"{chart_original_name} (edited from dataset test)" + db.session.commit() + + self.login(ADMIN_USERNAME) + rv = self._activity(dataset_uuid) + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + for record in body["result"]: + assert record["entity_kind"] == "dataset", ( + "Non-dataset record leaked into dataset's activity " + f"stream: {record}" + ) + assert record["source"] == "self", ( + f"Dataset activity contains a related record: {record}" + ) + finally: + db.session.rollback() + chart = db.session.query(Slice).filter(Slice.id == chart_id).one() + chart.slice_name = chart_original_name + db.session.commit() + + def test_dataset_activity_related_only_returns_empty(self) -> None: + """AV-004: datasets have no transitive layer. ``?include=related`` + returns an empty result list because there are no related entities + to draw from.""" + _persist_fixture_state() + dataset = _get_birth_names_dataset() + assert dataset is not None + self.login(ADMIN_USERNAME) + rv = self._activity(str(dataset.uuid), include="related") + assert rv.status_code == 200 + body = _json.loads(rv.data.decode("utf-8")) + assert body["result"] == [] + assert body["count"] == 0 + + +class TestActivityOpenApiSpec(SupersetTestCase): + """T049 — confirm the three ``/activity/`` endpoints are surfaced by + FAB-generated OpenAPI at ``/api/v1/_openapi``. + + ``base_api_tests.py::TestOpenApiSpec::test_open_api_spec`` already + validates the full spec's YAML correctness on every CI run. This + class adds activity-specific assertions: the paths exist, are + documented with the expected query parameters, and reference an + ``ActivityResponse``-shaped 200 response. + """ + + def _spec(self) -> dict[str, Any]: + self.login(ADMIN_USERNAME) + rv = self.client.get("/api/v1/_openapi") + assert rv.status_code == 200, rv.status_code + return _json.loads(rv.data.decode("utf-8")) + + def test_three_activity_paths_appear_in_openapi(self) -> None: + """One path per endpoint family. Paths are keyed by the URL + template, not the method name, so the FAB-generated keys are + the ``//activity/`` route templates.""" + spec = self._spec() + paths = spec.get("paths", {}) + # FAB templates the path-arg as ``{uuid_str}`` in the OpenAPI dict. + expected = { + "/api/v1/dashboard/{uuid_str}/activity/", + "/api/v1/chart/{uuid_str}/activity/", + "/api/v1/dataset/{uuid_str}/activity/", + } + missing = expected - paths.keys() + assert not missing, f"missing activity paths in OpenAPI: {missing}" + + def test_activity_endpoints_document_query_params(self) -> None: + """Each endpoint declares since / until / include / page / + page_size as query parameters. Spot-check on the dashboard + endpoint — the YAML docstring is the same shape across all + three so this assertion is sufficient.""" + spec = self._spec() + op = spec["paths"]["/api/v1/dashboard/{uuid_str}/activity/"]["get"] + params = {p["name"]: p for p in op.get("parameters", [])} + for expected in ("since", "until", "include", "page", "page_size"): + assert expected in params, ( + f"query param {expected!r} missing from dashboard /activity/" + ) + # include enum is the published contract — verify it's correct. + include_param = params["include"] + assert include_param["in"] == "query" + assert set(include_param["schema"]["enum"]) == {"self", "related", "all"} + + def test_activity_endpoints_declare_200_response(self) -> None: + """Each endpoint declares a 200 response. The exact schema + reference depends on how FAB resolves ``schema: ActivityResponseSchema`` + in the YAML docstring; here we just confirm the 200 + the 4xx + error responses are all present.""" + spec = self._spec() + op = spec["paths"]["/api/v1/dashboard/{uuid_str}/activity/"]["get"] + responses = op.get("responses", {}) + for code in ("200", "400", "401", "403", "404"): + assert code in responses, ( + f"response code {code} missing on dashboard /activity/" + ) diff --git a/tests/integration_tests/versioning/change_records_tests.py b/tests/integration_tests/versioning/change_records_tests.py new file mode 100644 index 000000000000..001001b48dde --- /dev/null +++ b/tests/integration_tests/versioning/change_records_tests.py @@ -0,0 +1,701 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Integration tests for ``version_changes`` capture (T052, partial). + +Covers in this file: + (a) saving a chart with three field changes produces three rows + (f) baseline / INSERT transactions produce zero records *for that entity* + + unchanged-save / dashboard / params-classification cases + +Deferred: + (b) ``GET /versions/`` response includes ``changes`` array — lands with + T050 (API integration). + (c) FK cascade — exercisable in principle (the migration declares + ``ON DELETE CASCADE``) but can't be isolated in a unit-style test + because ``version_transaction`` is referenced by non-cascading FKs + from slices_version / dashboards_version / etc. Covered instead + by (d) below once it lands, and by the structural declaration in + T046's migration. + (d) retention prune drops change records alongside the pruned + version — will land when T049 extends ``VersionDAO.prune_versions`` + to include ``version_changes`` alongside the shadow-row delete. + (e) ``kind`` index query plan on Postgres — deferred to T053 perf + validation. +""" + +from __future__ import annotations + +from datetime import datetime, timedelta +from typing import Any +from uuid import uuid4 + +import pytest +import sqlalchemy as sa +from sqlalchemy_continuum import version_class + +from superset.extensions import db +from superset.models.dashboard import Dashboard +from superset.models.slice import Slice +from superset.utils import json as _json +from tests.integration_tests.base_tests import SupersetTestCase +from tests.integration_tests.fixtures.birth_names_dashboard import ( # noqa: F401 + load_birth_names_dashboard_with_slices, + load_birth_names_data, +) + +_VERSION_CHANGES = sa.table( + "version_changes", + sa.column("id"), + sa.column("transaction_id"), + sa.column("entity_kind"), + sa.column("entity_id"), + sa.column("sequence"), + sa.column("kind"), + sa.column("operation"), + sa.column("path"), + sa.column("from_value"), + sa.column("to_value"), +) + +_VERSION_TRANSACTION = sa.table( + "version_transaction", + sa.column("id"), + sa.column("issued_at"), + sa.column("user_id"), + sa.column("action_kind"), +) + + +def _action_kind_for(tx_id: int) -> str | None: + """Read the ``action_kind`` column from the version_transaction row.""" + return ( + db.session.connection() + .execute( + sa.select(_VERSION_TRANSACTION.c.action_kind).where( + _VERSION_TRANSACTION.c.id == tx_id + ) + ) + .scalar() + ) + + +def _change_rows_for( + tx_id: int, + *, + entity_kind: str | None = None, + entity_id: int | None = None, +) -> list[dict[str, Any]]: + """Raw fetch of ``version_changes`` rows for a tx + optional entity filter.""" + query = sa.select(_VERSION_CHANGES).where( + _VERSION_CHANGES.c.transaction_id == tx_id + ) + if entity_kind is not None: + query = query.where(_VERSION_CHANGES.c.entity_kind == entity_kind) + if entity_id is not None: + query = query.where(_VERSION_CHANGES.c.entity_id == entity_id) + query = query.order_by(_VERSION_CHANGES.c.sequence.asc()) + result = db.session.connection().execute(query) + return [dict(row._mapping) for row in result] + + +def _persist_fixture_state() -> None: + """Commit fixture INSERTs so the baseline row exists before the test edits. + + Without this, the test's first commit batches the fixture's pending + INSERTs with the test's UPDATE into a single Continuum transaction + and no diff records are emitted (no pre-state). + """ + db.session.commit() + + +class TestChartChangeRecords(SupersetTestCase): + """Change-record capture for chart (Slice) saves.""" + + @pytest.fixture(autouse=True) + def _load_data(self, load_birth_names_dashboard_with_slices): # noqa: F811, PT004 + pass + + def test_single_scalar_edit_produces_one_change_record(self) -> None: + """(a) — one field changed, one ``version_changes`` row.""" + _persist_fixture_state() + + chart = db.session.query(Slice).first() + assert chart is not None + chart.slice_name = f"{chart.slice_name[:64]}_renamed" + db.session.commit() + + # The save produces one new version row (the UPDATE). Fetch its tx_id. + ver_cls = version_class(Slice) + update_tx_id = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + .transaction_id + ) + + rows = _change_rows_for(update_tx_id, entity_kind="chart", entity_id=chart.id) + assert len(rows) == 1 + assert rows[0]["kind"] == "field" + path = ( + _json.loads(rows[0]["path"]) + if isinstance(rows[0]["path"], str) + else rows[0]["path"] + ) + assert path == ["slice_name"] + assert rows[0]["sequence"] == 0 + + def test_last_saved_at_is_excluded_as_audit_noise(self) -> None: + """``last_saved_at`` / ``last_saved_by_fk`` are save-side-effect + fields stamped by ``UpdateChartCommand`` and must not produce + change records — same category as ``changed_on``. + + Saving a chart with ONLY a ``last_saved_at`` bump must produce + zero ``version_changes`` rows for that transaction. (Continuum + still records the shadow row; we just don't want to noise up + the per-edit diff log.) + """ + _persist_fixture_state() + + chart = db.session.query(Slice).first() + assert chart is not None + chart.last_saved_at = datetime.now() + timedelta(seconds=1) + db.session.commit() + + ver_cls = version_class(Slice) + latest_tx = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + ) + # If the save produced no version row at all (no actual model + # change beyond the audit field), nothing to assert. If it did, + # there must be no ``last_saved_at`` row in version_changes. + if latest_tx is None: + return + rows = _change_rows_for( + latest_tx.transaction_id, entity_kind="chart", entity_id=chart.id + ) + paths = [ + _json.loads(r["path"]) if isinstance(r["path"], str) else r["path"] + for r in rows + ] + assert ["last_saved_at"] not in paths + assert ["last_saved_by_fk"] not in paths + + def test_three_scalar_edits_produce_three_records_in_sequence(self) -> None: + """(a) — three fields changed, three rows, ``sequence`` 0..2.""" + _persist_fixture_state() + + chart = db.session.query(Slice).first() + assert chart is not None + # Derive from CURRENT values so every run guarantees a real + # change even against a persistent test DB where prior runs + # have already mutated the chart. + chart.slice_name = f"{chart.slice_name[:60]}_x" + chart.description = f"{chart.description or ''}_x" + chart.cache_timeout = (chart.cache_timeout or 0) + 1 + db.session.commit() + + ver_cls = version_class(Slice) + update_tx_id = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + .transaction_id + ) + rows = _change_rows_for(update_tx_id, entity_kind="chart", entity_id=chart.id) + assert len(rows) == 3 + assert [r["sequence"] for r in rows] == [0, 1, 2] + # Sorted by field name (diff engine emits in sorted field order) + paths = [ + _json.loads(r["path"]) if isinstance(r["path"], str) else r["path"] + for r in rows + ] + assert paths == [["cache_timeout"], ["description"], ["slice_name"]] + + def test_params_filter_add_produces_filter_kind_record(self) -> None: + """(a) — params classification still flows through the listener. + + Adds an adhoc_filter with a per-run-unique natural key + (``subject``): the filter differ keys on ``subject``, so a + STABLE subject is only "new" on the first run against a + persistent DB — every later run re-appends an already-present + key and the keyed diff emits nothing. Whatever was in + ``adhoc_filters`` before stays; we only want to confirm at + least one ``kind='filter'`` record is emitted. + """ + _persist_fixture_state() + + chart = db.session.query(Slice).first() + assert chart is not None + unique_subject = f"col_{chart.id}_{uuid4().hex[:8]}" + params = _json.loads(chart.params or "{}") + existing = params.get("adhoc_filters", []) or [] + params["adhoc_filters"] = [ + *existing, + { + "subject": unique_subject, + "operator": "==", + "comparator": "x", + "expressionType": "SIMPLE", + }, + ] + chart.params = _json.dumps(params) + db.session.commit() + + ver_cls = version_class(Slice) + update_tx_id = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + .transaction_id + ) + rows = _change_rows_for(update_tx_id, entity_kind="chart", entity_id=chart.id) + filter_rows = [r for r in rows if r["kind"] == "filter"] + assert len(filter_rows) >= 1, ( + f"expected at least one filter record, got rows: {rows}" + ) + + def test_unchanged_save_produces_zero_change_records(self) -> None: + """An edit that sets fields to identical values emits nothing.""" + _persist_fixture_state() + + chart = db.session.query(Slice).first() + ver_cls = version_class(Slice) + # Capture the latest tx_id BEFORE this test's save so we can + # distinguish "the no-op save produced nothing new" (the intent) + # from "prior tests left tx rows with records on them" (noise). + pre_save_tx_row = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + ) + pre_save_tx_id = pre_save_tx_row.transaction_id if pre_save_tx_row else 0 + + # Touch the object (mark dirty) but assign the same value. + current_name = chart.slice_name + chart.slice_name = current_name + db.session.commit() + + post_save_tx_row = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .filter(ver_cls.transaction_id > pre_save_tx_id) + .order_by(ver_cls.transaction_id.desc()) + .first() + ) + # Either no new tx at all (nothing dirty, best case), or a new + # tx with zero change records for this chart. + if post_save_tx_row is not None: + assert ( + _change_rows_for( + post_save_tx_row.transaction_id, + entity_kind="chart", + entity_id=chart.id, + ) + == [] + ) + + def test_perm_only_rewrite_produces_no_version(self) -> None: + """Bulk permission maintenance rewrites perm / schema_perm / + catalog_perm across many entities; the perm-string class is + derived security state, not user content, and is excluded from + ``__versioned__``. A commit touching ONLY those columns must + produce no shadow row at all — not even an empty transaction. + Regression for the phantom "Chart updated" flood the + version-history UI surfaced (PR #40988: one user save + 10 + perm-rewrite ride-alongs rendered as 10 phantom rows). + """ + _persist_fixture_state() + + chart = db.session.query(Slice).first() + assert chart is not None + ver_cls = version_class(Slice) + original_perms = (chart.perm, chart.schema_perm, chart.catalog_perm) + pre_save_tx_row = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .order_by(ver_cls.transaction_id.desc()) + .first() + ) + pre_save_tx_id = pre_save_tx_row.transaction_id if pre_save_tx_row else 0 + + try: + chart.perm = f"[seed].[perm_rewrite {uuid4().hex[:8]}]" + chart.schema_perm = f"[seed].[schema {uuid4().hex[:8]}]" + chart.catalog_perm = f"[seed].[catalog {uuid4().hex[:8]}]" + db.session.commit() + + post_save_tx_row = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.transaction_id > pre_save_tx_id) + .first() + ) + assert post_save_tx_row is None, ( + "perm-only rewrite created a shadow row " + f"(tx {post_save_tx_row.transaction_id}); the perm-string " + "class must be excluded from versioning" + ) + finally: + # Perm strings are security-routing state consulted by + # datasource-access checks; leaving random probe values on a + # persistent test DB breaks unrelated permission tests. + db.session.rollback() + chart = db.session.query(Slice).filter(Slice.id == chart.id).one() + chart.perm, chart.schema_perm, chart.catalog_perm = original_perms + db.session.commit() + + +class TestDashboardChangeRecords(SupersetTestCase): + """Same flow for dashboards — all scalar fields land in ``kind='field'``.""" + + @pytest.fixture(autouse=True) + def _load_data(self, load_birth_names_dashboard_with_slices): # noqa: F811, PT004 + pass + + def test_dashboard_title_edit_produces_field_record(self) -> None: + _persist_fixture_state() + + dashboard = db.session.query(Dashboard).first() + assert dashboard is not None + dashboard.dashboard_title = f"{dashboard.dashboard_title}_rev" + db.session.commit() + + ver_cls = version_class(Dashboard) + update_tx_id = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == dashboard.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + .transaction_id + ) + rows = _change_rows_for( + update_tx_id, entity_kind="dashboard", entity_id=dashboard.id + ) + assert len(rows) >= 1 + field_rows = [r for r in rows if r["kind"] == "field"] + paths = [ + _json.loads(r["path"]) if isinstance(r["path"], str) else r["path"] + for r in field_rows + ] + assert ["dashboard_title"] in paths + + +class TestDatasetChildChangeRecords(SupersetTestCase): + """T048b — column and metric diff records for dataset saves. + + Two snapshots must exist for any child diff to emit: the prior + save's and the current one. The fixture ``load_birth_names_data`` + has already created the dataset before these tests run; their + first commit produces snapshot #1. The test's edit produces + snapshot #2, and the listener diffs the two. + """ + + @pytest.fixture(autouse=True) + def _load_data(self, load_birth_names_dashboard_with_slices): # noqa: F811, PT004 + pass + + def test_column_description_change_produces_column_record(self) -> None: + # pylint: disable=import-outside-toplevel + from sqlalchemy_continuum import version_class + + from superset.connectors.sqla.models import SqlaTable + + _persist_fixture_state() + + dataset = ( + db.session.query(SqlaTable) + .filter(SqlaTable.table_name == "birth_names") + .first() + ) + assert dataset is not None + assert dataset.columns, "birth_names fixture should produce columns" + # First save establishes snapshot #1 (the pre-edit state). + # Scalar + child diffs won't emit anything yet because there's + # no prior snapshot to diff against. + dataset.description = f"{dataset.description or ''}_v1" + db.session.commit() + # Second save: edit a column AND touch a dataset scalar so + # the parent SqlaTable ends up in session.dirty. In real + # flows DatasetDAO.update_columns() marks the parent via its + # individual session.add / session.delete calls (T011); the + # direct-ORM test here needs an explicit parent touch. + column = dataset.columns[0] + column.description = f"{column.description or ''}_edited" + dataset.description = f"{dataset.description}_v2" + db.session.commit() + + ver_cls = version_class(SqlaTable) + latest_tx_id = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == dataset.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + .transaction_id + ) + rows = _change_rows_for( + latest_tx_id, entity_kind="dataset", entity_id=dataset.id + ) + column_rows = [r for r in rows if r["kind"] == "column"] + assert len(column_rows) >= 1, ( + f"expected at least one kind='column' record, got {rows}" + ) + + +class TestBaselineProducesZeroChangeRecords(SupersetTestCase): + """(f) — operation_type=0 (baseline / INSERT) transactions emit no records.""" + + @pytest.fixture(autouse=True) + def _load_data(self, load_birth_names_dashboard_with_slices): # noqa: F811, PT004 + pass + + def test_baseline_transaction_has_no_change_records_for_this_entity( + self, + ) -> None: + """(f) — baseline tx produces zero records *for that entity*. + + A single transaction can touch multiple entities (fixture loads, + import pipelines). A tx that's a baseline for this chart might + still legitimately carry update records for some *other* entity + that shared the flush. The spec's M4 clarification means: + records filtered to this entity's (tx, entity_kind, entity_id) + are empty for its baseline tx. + """ + _persist_fixture_state() + + chart = db.session.query(Slice).first() + chart.slice_name = f"{chart.slice_name[:64]}_force_baseline" + db.session.commit() + + ver_cls = version_class(Slice) + rows_by_tx = ( + db.session.query(ver_cls.transaction_id, ver_cls.operation_type) + .filter(ver_cls.id == chart.id) + .order_by(ver_cls.transaction_id.asc()) + .all() + ) + baseline_tx_ids = [tx for tx, op in rows_by_tx if op == 0] + assert baseline_tx_ids, "expected at least one baseline version row" + + for tx_id in baseline_tx_ids: + records_for_this_chart = _change_rows_for( + tx_id, entity_kind="chart", entity_id=chart.id + ) + assert records_for_this_chart == [], ( + f"baseline tx {tx_id} unexpectedly has change records for " + f"chart id={chart.id}: {records_for_this_chart}" + ) + + +class TestTransactionActionKindPropagation(SupersetTestCase): + """Confirm ``version_transaction.action_kind`` is stamped when a + command declares one via ``session.info["_versioning_action_kind"]``, + and stays ``NULL`` on ordinary saves.""" + + @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices") + def test_ordinary_save_has_null_action_kind(self) -> None: + """No command sets the key → version_transaction.action_kind + is NULL for a normal user-initiated save.""" + from superset.versioning.changes import ACTION_KIND_KEY + + _persist_fixture_state() + # Sanity: the key shouldn't already be on the session. + assert ACTION_KIND_KEY not in db.session.info + + chart = db.session.query(Slice).first() + assert chart is not None + chart.slice_name = f"{chart.slice_name[:60]}_baseline" + db.session.commit() + + ver_cls = version_class(Slice) + tx_id = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + .transaction_id + ) + assert _action_kind_for(tx_id) is None + + @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices") + def test_session_info_action_kind_propagates_to_transaction(self) -> None: + """The listener reads ``session.info[ACTION_KIND_KEY]`` and + stamps it on the version_transaction row. Exercises the wiring + directly so we don't need a full end-to-end command run for the + propagation test (the per-command tests below cover the + calling side).""" + from superset.versioning.changes import ACTION_KIND_KEY + + _persist_fixture_state() + chart = db.session.query(Slice).first() + assert chart is not None + + db.session.info[ACTION_KIND_KEY] = "restore" + chart.slice_name = f"{chart.slice_name[:60]}_trig" + db.session.commit() + + ver_cls = version_class(Slice) + tx_id = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + .transaction_id + ) + assert _action_kind_for(tx_id) == "restore" + + # And: the key is popped — next save resets to NULL action_kind. + assert ACTION_KIND_KEY not in db.session.info + + @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices") + def test_action_kind_pops_so_next_save_is_clean(self) -> None: + """After the listener stamps the action_kind, subsequent saves + on the same session must not carry it forward.""" + from superset.versioning.changes import ACTION_KIND_KEY + + _persist_fixture_state() + chart = db.session.query(Slice).first() + assert chart is not None + + # First save with action_kind. + db.session.info[ACTION_KIND_KEY] = "import" + chart.slice_name = f"{chart.slice_name[:60]}_a" + db.session.commit() + + # Second save without setting the key. + chart.slice_name = f"{chart.slice_name[:60]}_b" + db.session.commit() + + ver_cls = version_class(Slice) + # Get the two most-recent edit tx_ids. + rows = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .limit(2) + .all() + ) + assert len(rows) == 2 + second_tx, first_tx = rows[0].transaction_id, rows[1].transaction_id + + assert _action_kind_for(first_tx) == "import" + assert _action_kind_for(second_tx) is None + + @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices") + def test_action_kind_dropped_on_rollback(self) -> None: + """When a command sets ACTION_KIND_KEY and then an exception + fires before any flush stamps it (e.g. validation error after + the key is set), the value must not leak into the next save on + the same session. Regression for sqlalchemy-review C3.""" + from superset.versioning.changes import ACTION_KIND_KEY + + _persist_fixture_state() + chart = db.session.query(Slice).first() + assert chart is not None + + # Declare an action_kind, then force a rollback before the + # listener's flush stamps it. + db.session.info[ACTION_KIND_KEY] = "restore" + db.session.rollback() + + # The after_rollback listener must have popped the key. + assert ACTION_KIND_KEY not in db.session.info + + # And: a normal save now records NULL action_kind, not "restore". + chart.slice_name = f"{chart.slice_name[:60]}_postrollback" + db.session.commit() + + ver_cls = version_class(Slice) + tx_id = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + .transaction_id + ) + assert _action_kind_for(tx_id) is None + + @pytest.mark.usefixtures("load_birth_names_dashboard_with_slices") + def test_action_kind_survives_multiple_flushes_in_one_transaction(self) -> None: + """A restore/import can flush more than once before commit (an + explicit ``flush()`` or an autoflush from a mid-commit query). The + listener stamps ``action_kind`` *eagerly* on the first firing and + then dedups later firings of the same Continuum tx via the + ``_PROCESSED_TXS_KEY`` guard. + + This pins the eager-stamp + dedup interaction: across two flushes + in a single transaction the action_kind lands exactly once on the + one tx that carries the change records, the key is popped (so it + can't leak to the next save), and the second flush does not + re-emit records (which would trip the UNIQUE(transaction_id, + entity_kind, entity_id, sequence) constraint). Regression for the + amin-review finding on stamp-before-short-circuit ordering. + """ + from superset.versioning.changes import ACTION_KIND_KEY + + _persist_fixture_state() + chart = db.session.query(Slice).first() + assert chart is not None + + # One transaction, two flushes: declare the action_kind, edit + + # flush (first after_flush firing stamps and persists), then edit + # again + commit (second firing for the same tx must short-circuit). + db.session.info[ACTION_KIND_KEY] = "restore" + chart.slice_name = f"{chart.slice_name[:60]}_f1" + db.session.flush() + chart.slice_name = f"{chart.slice_name[:60]}_f2" + db.session.commit() + + ver_cls = version_class(Slice) + tx_id = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .filter(ver_cls.operation_type == 1) + .order_by(ver_cls.transaction_id.desc()) + .first() + .transaction_id + ) + + # Stamped exactly once on the records-bearing tx, and popped. + assert _action_kind_for(tx_id) == "restore" + assert ACTION_KIND_KEY not in db.session.info + # Records exist for the tx and the dedup guard prevented a + # duplicate-sequence re-emit on the second flush. + rows = _change_rows_for(tx_id, entity_kind="chart", entity_id=chart.id) + assert rows, "expected change records on the multi-flush transaction" + sequences = [r["sequence"] for r in rows] + assert len(sequences) == len(set(sequences)), ( + f"duplicate sequences on tx {tx_id}: {sequences}" + ) diff --git a/tests/integration_tests/versioning/perf_validation_tests.py b/tests/integration_tests/versioning/perf_validation_tests.py new file mode 100644 index 000000000000..5e792db50c05 --- /dev/null +++ b/tests/integration_tests/versioning/perf_validation_tests.py @@ -0,0 +1,447 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""T044 — Performance validation for entity version history. + +Skipped by default. Run on demand: + + SUPERSET_PERF_VALIDATION=1 pytest \ + tests/integration_tests/versioning/perf_validation_tests.py -v -s + +Measures the three success criteria defined in the spec: + + * SC-002: version list endpoint responds in under 1 second + * SC-003: restore endpoint completes in under 3 seconds + * SC-004: save path p95 overhead under 50 ms with Continuum tracking + on vs. off (FR-014) + +The test prints a summary table suitable for pasting into the PR +description. It also asserts each target so regressions fail loudly +when the harness is re-run. +""" + +from __future__ import annotations + +import os +import statistics +import time +from typing import Any + +import pytest +import sqlalchemy as sa +from sqlalchemy_continuum import version_class, versioning_manager + +from superset.extensions import db +from superset.models.slice import Slice +from tests.integration_tests.base_tests import SupersetTestCase +from tests.integration_tests.constants import ADMIN_USERNAME +from tests.integration_tests.fixtures.birth_names_dashboard import ( # noqa: F401 + load_birth_names_dashboard_with_slices, + load_birth_names_data, +) + +SKIP_REASON = "Performance validation is manual. Set SUPERSET_PERF_VALIDATION=1 to run." + +# Thresholds from spec.md §Success Criteria. +LIST_ENDPOINT_MAX_MS = 1000 # SC-002 +RESTORE_ENDPOINT_MAX_MS = 3000 # SC-003 +SAVE_OVERHEAD_P95_MAX_MS = 50 # SC-004 + +# Activity-view thresholds (sc-107283 §Success Criteria). +ACTIVITY_ENDPOINT_P95_MAX_MS = 1500 # SC-AV-001 + + +def _save_chart_once(chart: Slice, suffix: str) -> None: + """One ORM-level save path, mimicking what ChartDAO.update does.""" + chart.slice_name = f"{chart.slice_name[:64]}_{suffix}" + db.session.commit() + + +def _timings_ms(seconds: list[float]) -> dict[str, float]: + ms = sorted(s * 1000.0 for s in seconds) + return { + "p50": statistics.median(ms), + "p95": ms[int(len(ms) * 0.95) - 1] if len(ms) >= 20 else max(ms), + "max": max(ms), + "n": len(ms), + } + + +@pytest.mark.skipif( + not os.environ.get("SUPERSET_PERF_VALIDATION"), + reason=SKIP_REASON, +) +class PerfValidationTests(SupersetTestCase): + """Runs only when SUPERSET_PERF_VALIDATION=1 is set.""" + + @pytest.fixture(autouse=True) + def _load_data(self, load_birth_names_dashboard_with_slices: Any) -> None: # noqa: F811, PT004 + pass + + def _seed_chart_with_n_versions(self, n: int) -> Slice: + """Save a chart N times to produce N version rows.""" + chart = db.session.query(Slice).first() + assert chart is not None, "birth_names fixture should provide charts" + + for i in range(n): + _save_chart_once(chart, f"v{i}") + db.session.commit() + return chart + + def test_sc002_list_endpoint_under_1s(self) -> None: + """SC-002: list endpoint responds in under 1 second.""" + self.login(ADMIN_USERNAME) + + # Generate enough versions to exercise the retention-capped state. + chart = self._seed_chart_with_n_versions(24) + chart_uuid = str(chart.uuid) + url = f"/api/v1/chart/{chart_uuid}/versions/" + + # Warm up the endpoint once (JIT caching, mapper configuration, etc.) + self.client.get(url) + + timings: list[float] = [] + for _ in range(10): + t0 = time.perf_counter() + response = self.client.get(url) + timings.append(time.perf_counter() - t0) + assert response.status_code == 200 + + stats = _timings_ms(timings) + print( + f"\n[SC-002] GET /versions/ (24 versions) " + f"p50={stats['p50']:.1f}ms p95={stats['p95']:.1f}ms " + f"max={stats['max']:.1f}ms n={stats['n']}" + ) + assert stats["p95"] < LIST_ENDPOINT_MAX_MS, ( + f"SC-002 failed: list endpoint p95 {stats['p95']:.1f}ms " + f">= {LIST_ENDPOINT_MAX_MS}ms" + ) + + def test_sc003_restore_endpoint_under_3s(self) -> None: + """SC-003: restore endpoint completes in under 3 seconds.""" + self.login(ADMIN_USERNAME) + + chart = self._seed_chart_with_n_versions(5) + chart_uuid = str(chart.uuid) + + list_response = self.client.get(f"/api/v1/chart/{chart_uuid}/versions/") + assert list_response.status_code == 200 + versions = list_response.get_json()["result"] + assert len(versions) >= 2, "need at least two versions to restore" + target_version_uuid = versions[-1]["version_uuid"] + + restore_url = ( + f"/api/v1/chart/{chart_uuid}/versions/{target_version_uuid}/restore" + ) + + # Warm up once + self.client.post(restore_url) + + timings: list[float] = [] + for _ in range(5): + t0 = time.perf_counter() + response = self.client.post(restore_url) + timings.append(time.perf_counter() - t0) + assert response.status_code == 200 + + stats = _timings_ms(timings) + print( + f"\n[SC-003] POST /restore chart " + f"p50={stats['p50']:.1f}ms max={stats['max']:.1f}ms n={stats['n']}" + ) + assert stats["max"] < RESTORE_ENDPOINT_MAX_MS, ( + f"SC-003 failed: restore max {stats['max']:.1f}ms " + f">= {RESTORE_ENDPOINT_MAX_MS}ms" + ) + + def test_sc004_save_overhead_under_50ms(self) -> None: + """SC-004: save path p95 overhead under 50ms (FR-014). + + Toggling Continuum on and off mid-process corrupts its internal + ``units_of_work`` state and is not a reliable measurement. Instead + this test directly measures the wall-clock time spent inside the + four session-level listeners Continuum attaches to + ``sa.orm.session.Session`` — ``before_flush``, ``after_flush``, + ``after_commit``, ``after_rollback`` — plus Superset's own + baseline / snapshot / retention-prune listeners (attached to + ``db.session``). The cumulative listener time per save is the + marginal overhead version capture adds over a save with + versioning removed entirely, because without these listeners + the ORM would not execute any of that code. + + The approach: + 1. Wrap each known listener with a timing proxy that adds its + wall-clock time to a per-save accumulator. + 2. Save the same chart N times, recording each save's + accumulator value. + 3. Compute p50 / p95 of the per-save overhead. + + This matches the measurement intent of SC-004 (how much does + versioning cost per save) without the fragility of toggling + Continuum mid-test. + """ + self.login(ADMIN_USERNAME) + + chart = db.session.query(Slice).first() + assert chart is not None + + # Per-save accumulator incremented by the wrapped listeners. + acc = [0.0] + + def wrap_listener(original: Any) -> Any: + def wrapper(*args: Any, **kwargs: Any) -> Any: + t0 = time.perf_counter() + try: + return original(*args, **kwargs) + finally: + acc[0] += time.perf_counter() - t0 + + wrapper.__wrapped__ = original # type: ignore[attr-defined] + return wrapper + + # Instrument Continuum's four session listeners by detaching the + # bound method, wrapping, and re-attaching under a single-use + # listener handle we can cleanly remove on teardown. + session_target = sa.orm.session.Session + attached: list[tuple[str, Any]] = [] + for event_name, listener in list(versioning_manager.session_listeners.items()): + sa.event.remove(session_target, event_name, listener) + wrapped = wrap_listener(listener) + sa.event.listen(session_target, event_name, wrapped) + attached.append((event_name, wrapped)) + + iterations = 100 + warmup = 5 + try: + # Warmup (first baseline INSERT, JIT, cache warming). + for i in range(warmup): + _save_chart_once(chart, f"warm_{i}") + acc[0] = 0.0 + + total_timings: list[float] = [] + overhead_timings: list[float] = [] + for i in range(iterations): + acc[0] = 0.0 + t0 = time.perf_counter() + _save_chart_once(chart, f"run_{i}") + total_timings.append(time.perf_counter() - t0) + overhead_timings.append(acc[0]) + finally: + for event_name, wrapped in attached: + sa.event.remove(session_target, event_name, wrapped) + sa.event.listen( + session_target, + event_name, + wrapped.__wrapped__, + ) + + total = _timings_ms(total_timings) + overhead = _timings_ms(overhead_timings) + + ver_cls = version_class(Slice) + produced = db.session.query(ver_cls).filter(ver_cls.id == chart.id).count() + print( + f"\n[SC-004] save iterations={iterations} chart_id={chart.id} " + f"version_rows_produced={produced}" + ) + print( + f"[SC-004] full save: " + f"p50={total['p50']:.2f}ms p95={total['p95']:.2f}ms " + f"max={total['max']:.2f}ms" + ) + print( + f"[SC-004] version-cap overhead: " + f"p50={overhead['p50']:.2f}ms p95={overhead['p95']:.2f}ms " + f"max={overhead['max']:.2f}ms" + ) + + assert overhead["p95"] < SAVE_OVERHEAD_P95_MAX_MS, ( + f"SC-004 failed: version-capture p95 overhead " + f"{overhead['p95']:.2f}ms >= {SAVE_OVERHEAD_P95_MAX_MS}ms" + ) + + # ---- T045: Activity-view perf validation ----------------------------- + + def _seed_activity_history(self) -> str: + """Generate dense history on the birth_names dashboard so the + activity endpoint has something realistic to read. + + T045's spec target is "25 charts × 3 dataset windows each". The + birth_names fixture has ~12 charts on a single dataset (no + multi-dataset support without a bespoke fixture). We approximate + the load by: (a) editing many charts on the dashboard, (b) + editing the dataset's description several times, (c) editing the + dashboard's own title once. That yields ~30+ change records + spanning all three entity kinds — enough to exercise the + decoration, visibility, and impact-batch paths without needing a + multi-dataset fixture builder. Returns the dashboard UUID. + + **Why this commits without rollback** (unlike the test bodies in + ``activity_view_tests.py``): the whole point of a perf seed is + that the rows it produces have actually been persisted, so the + endpoint hit that follows reads a realistic state of the + ``version_changes`` / shadow tables. T053's + ``try/finally``+``rollback`` convention is for tests that + assert on *which records were captured*; here the seed IS the + setup, not the unit under test. The fixture's session-scoped + ``_cleanup`` removes the dashboard / slices at session teardown, + which is when the shadow rows age out too. + """ + # pylint: disable=import-outside-toplevel + from superset.connectors.sqla.models import SqlaTable + from superset.models.dashboard import Dashboard + + dashboard = ( + db.session.query(Dashboard) + .filter(Dashboard.dashboard_title.like("USA Births%")) + .first() + ) + dataset = ( + db.session.query(SqlaTable) + .filter(SqlaTable.table_name == "birth_names") + .first() + ) + assert dashboard is not None + assert dataset is not None + dashboard_uuid = str(dashboard.uuid) + + # Many chart edits — most of the activity volume. + for chart in dashboard.slices[:12]: + chart.slice_name = f"{chart.slice_name[:48]}_perf" + db.session.commit() + + # A handful of dataset edits — exercises the impact-batch path + # (Dashboard path + SqlaTable related). + for i in range(5): + dataset.description = f"perf seed iteration {i}" + db.session.commit() + + # One dashboard self-edit. + dashboard.dashboard_title = f"{dashboard.dashboard_title}_perf" + db.session.commit() + + return dashboard_uuid + + def test_av_sc001_activity_endpoint_p95_under_1500ms(self) -> None: + """SC-AV-001: dashboard activity endpoint p95 < 1500ms across 50 + invocations against a realistic history.""" + self.login(ADMIN_USERNAME) + + dashboard_uuid = self._seed_activity_history() + url = f"/api/v1/dashboard/{dashboard_uuid}/activity/" + + # Warmup — JIT, mapper config, identity-map population. + for _ in range(3): + self.client.get(url) + + timings: list[float] = [] + for _ in range(50): + t0 = time.perf_counter() + response = self.client.get(url) + timings.append(time.perf_counter() - t0) + assert response.status_code == 200 + + stats = _timings_ms(timings) + body = response.get_json() + print( + f"\n[AV-SC-001] GET /dashboard//activity/ " + f"records_returned={len(body['result'])} count={body['count']}" + ) + print( + f"[AV-SC-001] p50={stats['p50']:.1f}ms " + f"p95={stats['p95']:.1f}ms max={stats['max']:.1f}ms " + f"n={stats['n']}" + ) + assert stats["p95"] < ACTIVITY_ENDPOINT_P95_MAX_MS, ( + f"AV-SC-001 failed: activity endpoint p95 {stats['p95']:.1f}ms " + f">= {ACTIVITY_ENDPOINT_P95_MAX_MS}ms — profile the query " + f"plan and consider the T046 index migration (see " + f"specs/sc-107283/data-model.md §Possible additive indexes)" + ) + + def test_av_sc003_save_path_p95_unaffected_by_activity_view(self) -> None: + """AV-SC-003: the activity-view feature is read-only. Save path + p95 must remain within sc-103156's SC-004 budget (50ms version- + capture overhead) even with the activity tables in place. + + We re-measure the same overhead SC-004 measures, with the + activity-view branch's code in scope, to catch any accidental + regression from a save-path coupling. + """ + self.login(ADMIN_USERNAME) + # Seed some history so the M2M shadow + version_changes have + # enough rows that any pathological save-time read against them + # would surface. + self._seed_activity_history() + + chart = db.session.query(Slice).first() + assert chart is not None + + acc = [0.0] + + def wrap_listener(original: Any) -> Any: + def wrapper(*args: Any, **kwargs: Any) -> Any: + t0 = time.perf_counter() + try: + return original(*args, **kwargs) + finally: + acc[0] += time.perf_counter() - t0 + + wrapper.__wrapped__ = original # type: ignore[attr-defined] + return wrapper + + session_target = sa.orm.session.Session + attached: list[tuple[str, Any]] = [] + for event_name, listener in list(versioning_manager.session_listeners.items()): + sa.event.remove(session_target, event_name, listener) + wrapped = wrap_listener(listener) + sa.event.listen(session_target, event_name, wrapped) + attached.append((event_name, wrapped)) + + iterations = 50 + warmup = 3 + try: + for i in range(warmup): + _save_chart_once(chart, f"av_warm_{i}") + acc[0] = 0.0 + + overhead_timings: list[float] = [] + for i in range(iterations): + acc[0] = 0.0 + _save_chart_once(chart, f"av_run_{i}") + overhead_timings.append(acc[0]) + finally: + for event_name, wrapped in attached: + sa.event.remove(session_target, event_name, wrapped) + sa.event.listen( + session_target, + event_name, + wrapped.__wrapped__, + ) + + overhead = _timings_ms(overhead_timings) + print( + f"\n[AV-SC-003] save-path overhead with activity-view in scope: " + f"p50={overhead['p50']:.2f}ms p95={overhead['p95']:.2f}ms " + f"max={overhead['max']:.2f}ms" + ) + assert overhead["p95"] < SAVE_OVERHEAD_P95_MAX_MS, ( + f"AV-SC-003 failed: save-path p95 overhead " + f"{overhead['p95']:.2f}ms >= {SAVE_OVERHEAD_P95_MAX_MS}ms — " + f"the activity-view branch has regressed sc-103156's SC-004 " + f"budget; check for a new save-path read coupling." + ) diff --git a/tests/unit_tests/versioning/test_activity.py b/tests/unit_tests/versioning/test_activity.py new file mode 100644 index 000000000000..c3067370b6a2 --- /dev/null +++ b/tests/unit_tests/versioning/test_activity.py @@ -0,0 +1,678 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""Unit tests for ``superset.versioning.activity`` pure helpers (sc-107283). + +No app context, no DB, no Flask. Covers the helpers that can be exercised +in isolation: window intersection, scope resolution branching, entity- +window merging, AV-012 summary headlines, ``changed_by`` projection, +read-predicate fall-through, and the no-impact paths of +``_compute_impact``. The DB-touching helpers +(``charts_attached_to_dashboard``, ``datasets_used_by_chart``, +``fetch_change_records``, ``apply_entity_name_denormalization``, +``check_entity_tombstones``, ``_lookup_entity_uuids``) are exercised +by the integration suite in +``tests/integration_tests/versioning/activity_view_tests.py``. +""" + +from __future__ import annotations + +from datetime import datetime +from unittest.mock import patch + +import pytest + +from superset.versioning.activity import ( + ActivityParamsError, + EntityWindows, + parse_activity_query_params, + Window, +) +from superset.versioning.activity.impact import ( + collect_impact_pairs, + impact_for_record, +) +from superset.versioning.activity.kinds import API_KIND_TO_TABLE, TABLE_KIND_TO_API +from superset.versioning.activity.orchestrator import ( + _DEFAULT_PAGE_SIZE, + _emit_request_shape_attributes, + _MAX_PAGE_SIZE, +) +from superset.versioning.activity.render import _build_summary, _changed_by_dict +from superset.versioning.activity.scope import resolve_scope +from superset.versioning.activity.windows import ( + intersect_windows, + merge_entity_windows, + row_within_any_window, + union_windows, +) + +# ---- intersect_windows --------------------------------------------------- + + +@pytest.mark.parametrize( + "outer, inner, expected", + [ + # Inner fully inside outer + (Window(10, 20), Window(15, 18), Window(15, 18)), + # Left overlap — clipped on the left + (Window(10, 20), Window(5, 15), Window(10, 15)), + # Right overlap — clipped on the right + (Window(10, 20), Window(15, 25), Window(15, 20)), + # Outer fully inside inner + (Window(10, 20), Window(5, 25), Window(10, 20)), + # Touching at end → half-open semantics yield disjoint + (Window(10, 20), Window(20, 30), None), + # Disjoint to the right + (Window(10, 20), Window(25, 30), None), + # Disjoint to the left + (Window(10, 20), Window(0, 5), None), + # Open-ended outer (end_tx=None means +∞) + (Window(10, None), Window(5, 25), Window(10, 25)), + # Open-ended inner + (Window(10, 20), Window(5, None), Window(10, 20)), + # Both open-ended + (Window(10, None), Window(5, None), Window(10, None)), + # Identical + (Window(10, 20), Window(10, 20), Window(10, 20)), + ], +) +def test_intersect_windows( + outer: Window, inner: Window, expected: Window | None +) -> None: + assert intersect_windows(outer, inner) == expected + + +# ---- resolve_scope ------------------------------------------------------- + + +def test_resolve_scope_self_only_for_dashboard() -> None: + """``include='self'`` yields exactly one tuple covering all transactions.""" + assert resolve_scope("Dashboard", 42, "self") == [ + ("Dashboard", 42, [Window(0, None)]), + ] + + +def test_resolve_scope_self_only_for_chart() -> None: + assert resolve_scope("Slice", 7, "self") == [("Slice", 7, [Window(0, None)])] + + +def test_resolve_scope_self_only_for_dataset() -> None: + assert resolve_scope("SqlaTable", 9, "self") == [ + ("SqlaTable", 9, [Window(0, None)]), + ] + + +def test_dataset_has_no_related_scope() -> None: + """AV-004: datasets are not transitive recipients of activity in V2.""" + assert resolve_scope("SqlaTable", 9, "related") == [] + + +def test_dataset_all_returns_only_self() -> None: + """For datasets, ``include='all'`` == ``include='self'`` (AV-004).""" + assert resolve_scope("SqlaTable", 9, "all") == [ + ("SqlaTable", 9, [Window(0, None)]), + ] + + +# ---- merge_entity_windows ----------------------------------------------- + + +def test_merge_entity_windows_collapses_repeated_keys() -> None: + """Repeated ``(api_kind, entity_id)`` entries union their window lists + so the fetch query's OR-clause stays compact.""" + merged = merge_entity_windows( + [ + ("Slice", 1, [Window(0, 100)]), + ("Slice", 1, [Window(200, 300)]), + ("SqlaTable", 5, [Window(0, None)]), + ] + ) + by_key = {(kind, eid): windows for kind, eid, windows in merged} + assert by_key[("Slice", 1)] == [Window(0, 100), Window(200, 300)] + assert by_key[("SqlaTable", 5)] == [Window(0, None)] + + +def test_merge_entity_windows_preserves_singletons() -> None: + """Non-duplicated entries pass through unchanged.""" + inputs: list[EntityWindows] = [ + ("Slice", 1, [Window(0, 100)]), + ("Dashboard", 2, [Window(10, 20)]), + ] + merged = merge_entity_windows(inputs) + assert sorted(merged) == sorted(inputs) + + +def test_merge_entity_windows_unions_overlapping_windows_for_one_entity() -> None: + """Same entity, many redundant attachment windows → collapsed to one. + + This guards the SQLite expression-tree limit: a fixture that + re-creates a chart-on-dashboard association across many transactions + used to produce N separate OR branches in the fetch query (one per + redundant window). merge_entity_windows must coalesce them. + """ + scope: list[EntityWindows] = [ + ("Slice", 1, [Window(10, 20)]), + ("Slice", 1, [Window(15, 25)]), # overlaps + ("Slice", 1, [Window(25, 30)]), # touches + ("Slice", 1, [Window(40, 50)]), # disjoint + ] + merged = merge_entity_windows(scope) + assert merged == [("Slice", 1, [Window(10, 30), Window(40, 50)])] + + +# ---- union_windows ------------------------------------------------------- + + +@pytest.mark.parametrize( + "windows, expected", + [ + # Disjoint windows pass through + ( + [Window(10, 20), Window(30, 40)], + [Window(10, 20), Window(30, 40)], + ), + # Overlapping windows merge + ([Window(10, 20), Window(15, 25)], [Window(10, 25)]), + # Touching windows merge (half-open: [10,20) + [20,30) = [10,30)) + ([Window(10, 20), Window(20, 30)], [Window(10, 30)]), + # Many overlapping windows collapse to one + ( + [Window(10, 20), Window(15, 25), Window(20, 30), Window(25, 35)], + [Window(10, 35)], + ), + # Input order doesn't matter + ( + [Window(30, 40), Window(10, 20), Window(15, 25)], + [Window(10, 25), Window(30, 40)], + ), + # Open-ended absorbs everything to the right + ([Window(10, None), Window(50, 60)], [Window(10, None)]), + # Open-ended at the right merges into open-ended + ([Window(10, 20), Window(15, None)], [Window(10, None)]), + # Empty input + ([], []), + # Single window pass-through + ([Window(5, 10)], [Window(5, 10)]), + ], +) +def test_union_windows(windows: list[Window], expected: list[Window]) -> None: + assert union_windows(windows) == expected + + +# ---- row_within_any_window (Python post-filter for the fetch query) ------ + + +def test_row_in_window_inside() -> None: + assert row_within_any_window({"transaction_id": 15}, [Window(10, 20)]) + + +def test_row_in_window_at_start_boundary_inclusive() -> None: + """Half-open: ``[10, 20)`` includes 10.""" + assert row_within_any_window({"transaction_id": 10}, [Window(10, 20)]) + + +def test_row_in_window_at_end_boundary_exclusive() -> None: + """Half-open: ``[10, 20)`` excludes 20.""" + assert not row_within_any_window({"transaction_id": 20}, [Window(10, 20)]) + + +def test_row_in_open_ended_window() -> None: + """``end=None`` means +∞.""" + assert row_within_any_window({"transaction_id": 999}, [Window(10, None)]) + + +def test_row_in_any_of_several_windows() -> None: + assert row_within_any_window( + {"transaction_id": 50}, + [Window(10, 20), Window(40, 60), Window(90, 100)], + ) + + +def test_row_in_no_windows_returns_false() -> None: + assert not row_within_any_window({"transaction_id": 50}, []) + assert not row_within_any_window( + {"transaction_id": 25}, [Window(10, 20), Window(30, 40)] + ) + + +# ---- Kind translation round-trip ----------------------------------------- + + +def test_kind_translation_is_bijective_for_supported_kinds() -> None: + """Every API kind maps to a table kind and back to the same value. + Locks in the contract that the two maps don't drift.""" + for api_kind, table_kind in API_KIND_TO_TABLE.items(): + assert TABLE_KIND_TO_API[table_kind] == api_kind + + +# ---- _build_summary (AV-012) --------------------------------------------- + + +def test_summary_for_dataset_column_change() -> None: + rec = {"kind": "column", "entity_name": "Sales Transactions"} + assert _build_summary("SqlaTable", rec) == ( + "Dataset column changed: Sales Transactions" + ) + + +def test_summary_for_chart_filter_change() -> None: + rec = {"kind": "filter", "entity_name": "Top Charts"} + assert _build_summary("Slice", rec) == "Chart filter changed: Top Charts" + + +def test_summary_for_restore_event() -> None: + rec = {"kind": "restore", "entity_name": "Q4 Dashboard"} + assert _build_summary("Dashboard", rec) == "Dashboard restored: Q4 Dashboard" + + +def test_summary_unknown_kind_falls_back_to_updated() -> None: + """Unmapped change kinds collapse to a generic 'updated' verb.""" + rec = {"kind": "mystery_kind", "entity_name": "X"} + assert _build_summary("Dashboard", rec) == "Dashboard updated: X" + + +def test_summary_without_entity_name_drops_colon() -> None: + """Tombstoned entities have no name; the headline reads naturally + without a trailing colon and empty value.""" + rec = {"kind": "filter", "entity_name": ""} + assert _build_summary("Slice", rec) == "Chart filter changed" + + +# ---- _changed_by_dict ---------------------------------------------------- + + +def test_changed_by_returns_none_when_no_user_attached() -> None: + """Saves from CLI/Celery/import have no Flask user (sc-103156 §Session + 2026-05-18 clarification).""" + assert _changed_by_dict({"changed_by_id": None}) is None + + +def test_changed_by_projects_only_display_fields() -> None: + """Per the ActivityChangedBy contract: id + first_name + last_name only. + Username is intentionally omitted (data-model.md).""" + record = { + "changed_by_id": 5, + "first_name": "Mike", + "last_name": "Bridge", + "user_id": 5, # internal column, must not leak + } + result = _changed_by_dict(record) + assert result == {"id": 5, "first_name": "Mike", "last_name": "Bridge"} + assert result is not None + assert "username" not in result + + +# ---- impact_for_record (pure, post-batch) ------------------------------- + + +def test_impact_for_record_dashboard_path_dataset_related_uses_count() -> None: + """The only path/related shape that carries impact: ``Dashboard`` → + ``SqlaTable``. The count comes from the pre-batched lookup.""" + record = {"entity_kind": "dataset", "entity_id": 5, "transaction_id": 100} + counts = {(5, 100): 3} + assert impact_for_record(record, "Dashboard", counts) == {"charts": 3} + + +def test_impact_for_record_missing_count_yields_none() -> None: + """A pair the batch query didn't return (no matching siblings) + collapses to ``None`` rather than ``{"charts": 0}``.""" + record = {"entity_kind": "dataset", "entity_id": 5, "transaction_id": 100} + assert impact_for_record(record, "Dashboard", {}) is None + + +def test_impact_for_record_zero_count_yields_none() -> None: + """Explicit zero in the counts map is treated the same as missing — + no impact field on the wire.""" + record = {"entity_kind": "dataset", "entity_id": 5, "transaction_id": 100} + assert impact_for_record(record, "Dashboard", {(5, 100): 0}) is None + + +def test_impact_for_record_dashboard_path_chart_related_yields_none() -> None: + """Dashboard → chart is a direct dependency; no further sibling + layer to count.""" + record = {"entity_kind": "chart", "entity_id": 5, "transaction_id": 100} + assert impact_for_record(record, "Dashboard", {(5, 100): 999}) is None + + +def test_impact_for_record_chart_path_with_dataset_related_yields_none() -> None: + """Chart → dataset: the chart is itself the only dependent of the + dataset edit.""" + record = {"entity_kind": "dataset", "entity_id": 5, "transaction_id": 100} + assert impact_for_record(record, "Slice", {(5, 100): 999}) is None + + +def test_impact_for_record_dataset_path_yields_none() -> None: + """Datasets have no transitive layer (AV-004).""" + record = {"entity_kind": "dataset", "entity_id": 5, "transaction_id": 100} + assert impact_for_record(record, "SqlaTable", {(5, 100): 999}) is None + + +# ---- collect_impact_pairs ----------------------------------------------- + + +def test_collect_impact_pairs_dashboard_path_collects_only_datasets() -> None: + """The batched pre-query only needs ``(dataset_id, tx)`` pairs. + Chart-related and self records aren't relevant.""" + records = [ + {"entity_kind": "dataset", "entity_id": 5, "transaction_id": 100}, + {"entity_kind": "dataset", "entity_id": 7, "transaction_id": 200}, + {"entity_kind": "chart", "entity_id": 9, "transaction_id": 300}, + {"entity_kind": "dashboard", "entity_id": 1, "transaction_id": 400}, + ] + assert collect_impact_pairs(records, "Dashboard") == {(5, 100), (7, 200)} + + +def test_collect_impact_pairs_dedupes_repeated_pairs() -> None: + """Multiple change records for the same (dataset, tx) collapse to + one pair — the batch query computes the count once.""" + records = [ + {"entity_kind": "dataset", "entity_id": 5, "transaction_id": 100}, + {"entity_kind": "dataset", "entity_id": 5, "transaction_id": 100}, + {"entity_kind": "dataset", "entity_id": 5, "transaction_id": 100}, + ] + pairs = collect_impact_pairs(records, "Dashboard") + assert pairs == {(5, 100)} + + +def test_collect_impact_pairs_chart_path_returns_empty() -> None: + """Chart paths have no dashboard layer to count siblings on, so the + batch never needs to fire.""" + records = [ + {"entity_kind": "dataset", "entity_id": 5, "transaction_id": 100}, + ] + assert collect_impact_pairs(records, "Slice") == set() + + +def test_collect_impact_pairs_dataset_path_returns_empty() -> None: + records = [ + {"entity_kind": "chart", "entity_id": 5, "transaction_id": 100}, + ] + assert collect_impact_pairs(records, "SqlaTable") == set() + + +def test_collect_impact_pairs_empty_records_returns_empty() -> None: + assert collect_impact_pairs([], "Dashboard") == set() + + +# ---- parse_activity_query_params (shared API helper) --------------------- + + +def test_parser_defaults_when_empty() -> None: + """No params → ``include='all'``, ``page=0``, ``page_size=DEFAULT``.""" + assert parse_activity_query_params({}) == { + "include": "all", + "page": 0, + "page_size": _DEFAULT_PAGE_SIZE, + } + + +def test_parser_clamps_page_size_to_max() -> None: + """A request for more than the contract maximum is clamped, not 400'd + (silent clamp matches AV-019's bounded-payload guarantee).""" + params = parse_activity_query_params({"page_size": str(_MAX_PAGE_SIZE * 5)}) + assert params["page_size"] == _MAX_PAGE_SIZE + + +def test_parser_accepts_iso_datetime_with_z_suffix() -> None: + """Python <3.11 fromisoformat rejects 'Z'; the parser tolerates it.""" + params = parse_activity_query_params({"since": "2026-01-01T00:00:00Z"}) + assert params["since"].year == 2026 + + +def test_parser_normalises_z_suffix_to_naive_utc() -> None: + """The 'Z' result must be tz-NAIVE: ``issued_at`` is a naive column, so a + tz-aware bind shifts the comparison by the session offset (or raises) on + PostgreSQL. The 'Z' instant is already UTC, so the value is unchanged.""" + since = parse_activity_query_params({"since": "2026-01-01T00:00:00Z"})["since"] + assert since.tzinfo is None + assert since == datetime(2026, 1, 1, 0, 0, 0) + + +def test_parser_normalises_offset_to_naive_utc() -> None: + """A non-UTC offset is converted to UTC and stripped to naive, so the + comparison against the naive ``issued_at`` column is in the same frame.""" + since = parse_activity_query_params({"since": "2026-01-01T05:00:00+02:00"})["since"] + assert since.tzinfo is None + assert since == datetime(2026, 1, 1, 3, 0, 0) # 05:00 +02:00 -> 03:00 UTC + + +def test_parser_rejects_invalid_include() -> None: + with pytest.raises(ActivityParamsError, match="include"): + parse_activity_query_params({"include": "sibling"}) + + +def test_parser_rejects_malformed_datetime() -> None: + with pytest.raises(ActivityParamsError, match="since"): + parse_activity_query_params({"since": "yesterday"}) + + +def test_parser_rejects_negative_page() -> None: + with pytest.raises(ActivityParamsError, match="page"): + parse_activity_query_params({"page": "-1"}) + + +def test_parser_rejects_zero_page_size() -> None: + with pytest.raises(ActivityParamsError, match="page_size"): + parse_activity_query_params({"page_size": "0"}) + + +def test_parser_error_is_a_value_error() -> None: + """``ActivityParamsError`` subclasses ``ValueError`` so callers that + only know about the standard library exception hierarchy still catch + it correctly.""" + with pytest.raises(ValueError, match="include"): + parse_activity_query_params({"include": "nope"}) + + +# ---- Observability metric-key convention (T050 cross-coupling) ---------- + + +def test_metric_prefix_matches_versioning_namespace_convention() -> None: + """T050: cross-coupling sanity. The activity-view's instrumentation + prefix (``superset.activity_view.*``) must be a sibling of sc-103156's + eventual ``superset.versioning.*`` namespace, not nested under + a different root. Both endpoint families belong to the versioning + feature; their metrics should be discoverable from one Grafana + filter (``superset.activity_view.*`` OR ``superset.versioning.*``). + + Locking the prefix in a test catches accidental drift in a code + review — a future PR renaming the prefix would fail this assertion + and require explicit acknowledgement. + """ + from superset.versioning.activity.orchestrator import _METRIC_PREFIX + + assert _METRIC_PREFIX == "superset.activity_view", ( + f"Activity-view metrics prefix changed from " + f"'superset.activity_view' to {_METRIC_PREFIX!r}. If this was " + "intentional, update sc-103156's FR-027 instrumentation to " + "match the new convention OR document the new naming in plan §D-17." + ) + # Sibling-namespace check: starts with the versioning-feature root. + assert _METRIC_PREFIX.startswith("superset."), ( + "All Superset metrics live under 'superset.*'; activity_view must too." + ) + + +# ---- _emit_request_shape_attributes: related-entity counts --------------- +# +# The ``related_entity_count.*`` gauges report how many *other* entities an +# activity request fanned out to. ``resolve_scope`` always prepends the path +# entity itself (the "self" window) to the scope list, so the metric loop +# must exclude that self entry — otherwise a chart/dataset request reports +# one phantom related entity of its own kind. + + +def _gauge_value(mock_sl: object, suffix: str) -> float: + """Return the value of the single ``gauge`` call whose metric name ends + with *suffix*. Fails loudly if absent so a renamed metric surfaces here.""" + for call in mock_sl.gauge.call_args_list: # type: ignore[attr-defined] + name = call.args[0] + if name.endswith(suffix): + return call.args[1] + emitted = [c.args[0] for c in mock_sl.gauge.call_args_list] # type: ignore[attr-defined] + raise AssertionError(f"no gauge ending {suffix!r}; emitted {emitted}") + + +@patch("superset.extensions.stats_logger_manager") +def test_related_entity_count_excludes_self_for_chart(mock_mgr) -> None: + """A chart request scopes to itself + the datasets it used. The charts + gauge must read 0 (no *related* charts) even though the self Slice is in + the scope list; the datasets gauge counts only the two related datasets.""" + sl = mock_mgr.instance + entity_windows: list[EntityWindows] = [ + ("Slice", 7, [Window(0, None)]), # self — must not be counted + ("SqlaTable", 5, [Window(0, None)]), # related dataset + ("SqlaTable", 9, [Window(0, None)]), # related dataset + ] + + _emit_request_shape_attributes( + "slice", + include="all", + has_since_filter=False, + page_size=25, + record_count=3, + entity_windows=entity_windows, + path_kind="Slice", + path_id=7, + ) + + assert _gauge_value(sl, "related_entity_count.charts") == 0.0 + assert _gauge_value(sl, "related_entity_count.datasets") == 2.0 + + +@patch("superset.extensions.stats_logger_manager") +def test_related_entity_count_excludes_self_for_dataset(mock_mgr) -> None: + """Datasets have no related scope, so an ``include=all`` dataset request + scopes to itself only. The datasets gauge must read 0, not 1.""" + sl = mock_mgr.instance + entity_windows: list[EntityWindows] = [ + ("SqlaTable", 9, [Window(0, None)]), # self only + ] + + _emit_request_shape_attributes( + "sqlatable", + include="all", + has_since_filter=False, + page_size=25, + record_count=1, + entity_windows=entity_windows, + path_kind="SqlaTable", + path_id=9, + ) + + assert _gauge_value(sl, "related_entity_count.datasets") == 0.0 + + +@patch("superset.extensions.stats_logger_manager") +def test_related_entity_count_counts_genuine_related_of_same_kind(mock_mgr) -> None: + """Self-exclusion keys on (kind, id), not kind alone: a dashboard whose + scope happened to include another dashboard would still count it.""" + sl = mock_mgr.instance + entity_windows: list[EntityWindows] = [ + ("Dashboard", 1, [Window(0, None)]), # self + ("Slice", 5, [Window(0, None)]), # related chart + ("Slice", 6, [Window(0, None)]), # related chart + ] + + _emit_request_shape_attributes( + "dashboard", + include="all", + has_since_filter=False, + page_size=25, + record_count=2, + entity_windows=entity_windows, + path_kind="Dashboard", + path_id=1, + ) + + assert _gauge_value(sl, "related_entity_count.charts") == 2.0 + + +def test_parser_passes_q_through_and_drops_blank() -> None: + """``q`` reaches get_activity stripped; blank/missing stays absent.""" + assert parse_activity_query_params({"q": " revenue "})["q"] == "revenue" + assert "q" not in parse_activity_query_params({}) + assert "q" not in parse_activity_query_params({"q": " "}) + + +def test_record_matches_searches_decorated_surfaces() -> None: + """The q filter covers summary, entity_name, kind, path, and values — + case-insensitively (PR #40988: client search only covered loaded + pages; the server filter must cover the same surfaces).""" + from superset.versioning.activity.orchestrator import _record_matches + + record = { + "summary": "Dataset updated: Sales Transactions", + "entity_name": "Sales Transactions", + "kind": "field", + "path": ["params", "adhoc_filters", "country"], + "from_value": None, + "to_value": {"label": "Revenue (EUR)"}, + } + assert _record_matches(record, "sales") # entity_name/summary + assert _record_matches(record, "COUNTRY") # path segment + assert _record_matches(record, "revenue (eur)") # to_value + assert not _record_matches(record, "nonexistent") + + +def test_record_matches_falsy_values_and_json_form() -> None: + """Falsy values must stay searchable (False/0 must not collapse to + ''), and values match in their JSON form — the text the client + renders — not Python repr.""" + from superset.versioning.activity.orchestrator import _record_matches + + record = { + "summary": "", + "entity_name": "", + "kind": "field", + "path": ["params", "show_legend"], + "from_value": True, + "to_value": False, + } + assert _record_matches(record, "false") # JSON 'false', not Python 'False' + assert _record_matches(record, "true") + zero = {**record, "path": ["params", "row_limit"], "from_value": 10, "to_value": 0} + assert _record_matches(zero, "0") + nested = {**record, "to_value": {"label": "Revenue"}} + assert _record_matches(nested, '"label"') # JSON double-quoted key + + +def test_build_summary_meta_headline_branches() -> None: + """The __meta__ headline dispatches on the transaction's action_kind + (path is pure navigation): restore renders 'restored to version N' + (with entity_name when present); unknown meta actions fall back to + 'updated'.""" + restore = { + "kind": "__meta__", + "action_kind": "restore", + "path": ["__meta__"], + "to_value": {"version_uuid": "u", "version_number": 3}, + "entity_name": "Top 10 Girls", + } + assert _build_summary("Slice", restore) == ( + "Chart restored to version 3: Top 10 Girls" + ) + nameless = {**restore, "entity_name": ""} + assert _build_summary("Slice", nameless) == "Chart restored to version 3" + unknown = { + "kind": "__meta__", + "action_kind": "import", + "path": ["__meta__"], + "to_value": {}, + "entity_name": "", + } + assert _build_summary("Dashboard", unknown) == "Dashboard updated" From da6068c412453d6275919ea3ec4c0099bddeb0a7 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 22 Jun 2026 11:24:28 -0600 Subject: [PATCH 33/35] fix(activity-view): address review findings MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Production cleanup: - Remove the throwaway debug view + page + route (superset/views/activity_debug.py, ActivityDebug page, the /activity-debug route, and the initializer registration) — self-described "delete when the feature ships" code. - Remove the three dead VersionHistoryDropdown.tsx demo components (unreferenced, carried "revert before pushing" banners). - Strip internal tracker/spec/task references (sc-*, AV-*, FR-*, T0*, plan §, data-model.md, US#, PR #) from all shipped code — including the OpenAPI endpoint summaries and schema field descriptions that publish to /swagger. Query correctness: - fetch_change_records: push each kind's window tx-id bounds into the SQL WHERE so the per-statement LIMIT selects from in-window rows (previously the newest N out-of-window rows could be fetched and discarded, silently dropping older in-window records). - Make the record sort key total (append entity_kind, entity_id) so pagination is stable across requests instead of depending on set-iteration order. - Route every entity-id IN-clause through a shared chunked_ids helper (visibility, impact, name-resolution, tombstones, batch dataset walk) so a wide dashboard can't exceed SQLite's bind-variable floor. - Bound the q search param length (cheap-DoS guard). Tests: - Add a negative visibility test proving the AV-008 silent-drop contract: a Gamma principal's inaccessible related record is dropped from the result with no placeholder and no count contribution. Co-Authored-By: Claude Opus 4.8 (1M context) --- .../src/pages/ActivityDebug/index.tsx | 420 ------------------ .../ChartList/VersionHistoryDropdown.tsx | 368 --------------- .../DashboardList/VersionHistoryDropdown.tsx | 388 ---------------- .../DatasetList/VersionHistoryDropdown.tsx | 364 --------------- superset-frontend/src/views/routes.tsx | 13 - superset/charts/api.py | 2 +- superset/datasets/api.py | 4 +- superset/initialization/__init__.py | 6 - superset/versioning/activity/__init__.py | 6 +- superset/versioning/activity/impact.py | 49 +- superset/versioning/activity/kinds.py | 26 +- superset/versioning/activity/orchestrator.py | 36 +- superset/versioning/activity/queries.py | 236 ++++++---- superset/versioning/activity/render.py | 20 +- superset/versioning/activity/scope.py | 2 +- superset/versioning/activity/visibility.py | 54 ++- superset/versioning/schemas.py | 22 +- superset/views/activity_debug.py | 53 --- .../versioning/activity_view_tests.py | 63 ++- 19 files changed, 332 insertions(+), 1800 deletions(-) delete mode 100644 superset-frontend/src/pages/ActivityDebug/index.tsx delete mode 100644 superset-frontend/src/pages/ChartList/VersionHistoryDropdown.tsx delete mode 100644 superset-frontend/src/pages/DashboardList/VersionHistoryDropdown.tsx delete mode 100644 superset-frontend/src/pages/DatasetList/VersionHistoryDropdown.tsx delete mode 100644 superset/views/activity_debug.py diff --git a/superset-frontend/src/pages/ActivityDebug/index.tsx b/superset-frontend/src/pages/ActivityDebug/index.tsx deleted file mode 100644 index bce13f0dded0..000000000000 --- a/superset-frontend/src/pages/ActivityDebug/index.tsx +++ /dev/null @@ -1,420 +0,0 @@ -/** - * Licensed to the Apache Software Foundation (ASF) under one - * or more contributor license agreements. See the NOTICE file - * distributed with this work for additional information - * regarding copyright ownership. The ASF licenses this file - * to you under the Apache License, Version 2.0 (the - * "License"); you may not use this file except in compliance - * with the License. You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, - * software distributed under the License is distributed on an - * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY - * KIND, either express or implied. See the License for the - * specific language governing permissions and limitations - * under the License. - */ - -// Throwaway debug UI for sc-107283 activity-view endpoints. Verify the -// JSON responses look right by eye. Delete this directory + the route -// entry in views/routes.tsx when the activity-view feature ships. - -import { useEffect, useMemo, useState } from 'react'; -import { useParams } from 'react-router-dom'; -import { SupersetClient } from '@superset-ui/core'; -import { t } from '@apache-superset/core/translation'; -import { - Card, - Empty, - Input, - Loading, - Radio, - Space, - Tag, - Typography, -} from '@superset-ui/core/components'; - -type ResourceKind = 'dashboard' | 'chart' | 'dataset'; -type IncludeMode = 'self' | 'related' | 'all'; -type EntityKind = 'dashboard' | 'chart' | 'dataset'; - -interface ChangedBy { - id: number; - first_name: string; - last_name: string; -} - -interface ActivityRecord { - version_uuid: string; - entity_kind: EntityKind; - entity_uuid: string | null; - entity_name: string; - entity_deleted: boolean; - entity_deletion_state: string | null; - source: 'self' | 'related'; - transaction_id: number; - // Transaction-level avenue: restore / import / clone / null - // (= ordinary save). Shared by every record in the same tx. - action_kind: string | null; - issued_at: string; - changed_by: ChangedBy | null; - kind: string; - // Per-record verb: add / remove / move / edit. - operation: string; - path: string[]; - from_value: unknown; - to_value: unknown; - summary: string; - impact: { charts?: number; datasets?: number } | null; -} - -interface ActivityResponse { - result: ActivityRecord[]; - count: number; -} - -const PAGE_SIZE_OPTIONS = [10, 25, 50, 100, 200]; - -const KIND_COLOR: Record = { - filter: 'blue', - metric: 'green', - dimension: 'cyan', - column: 'geekblue', - chart: 'purple', - time_range: 'gold', - color_palette: 'magenta', - restore: 'orange', - field: 'default', -}; - -const ENTITY_KIND_COLOR: Record = { - dashboard: 'blue', - chart: 'purple', - dataset: 'green', -}; - -function formatValue(value: unknown): string { - if (value === null || value === undefined) return 'null'; - if (typeof value === 'string') return JSON.stringify(value); - return JSON.stringify(value); -} - -function ChangedByDisplay({ changedBy }: { changedBy: ChangedBy | null }) { - if (changedBy === null) { - return ( - - {t('system / unknown')} - - ); - } - return ( - - {changedBy.first_name} {changedBy.last_name} (id={changedBy.id}) - - ); -} - -function RecordCard({ record }: { record: ActivityRecord }) { - const headerExtra = ( - - - {record.entity_kind} - - - {record.source} - - {record.kind} - {record.operation} - {record.action_kind && {record.action_kind}} - {record.entity_deleted && deleted} - {record.entity_deletion_state === 'soft_deleted' && ( - soft-deleted - )} - - ); - - const headline = record.summary - ? record.summary - : `${record.entity_kind}: ${record.entity_name || '(unnamed)'}`; - - return ( - - {headline} - - {record.issued_at} · tx={record.transaction_id} ·{' '} - - - - } - extra={headerExtra} - style={{ marginBottom: 8 }} - > - -
- - {t('entity_name:')} - {' '} - - {record.entity_name || t('(no name)')} - -
-
- - {t('entity_uuid:')} - {' '} - - {record.entity_uuid || t('null (tombstoned)')} - -
-
- - {t('version_uuid:')} - {' '} - - {record.version_uuid || t('null')} - -
-
- {t('path:')}{' '} - {JSON.stringify(record.path)} -
-
- {t('from →')}{' '} - - {formatValue(record.from_value)} - {' '} - {t('→ to')}{' '} - {formatValue(record.to_value)} -
- {record.impact !== null && ( -
- {t('impact:')}{' '} - - {JSON.stringify(record.impact)} - -
- )} - - - ); -} - -function ActivityDebug() { - const { resource: resourceParam, uuid } = useParams<{ - resource: string; - uuid: string; - }>(); - - const resource = useMemo(() => { - if ( - resourceParam === 'dashboard' || - resourceParam === 'chart' || - resourceParam === 'dataset' - ) { - return resourceParam; - } - return null; - }, [resourceParam]); - - const [include, setInclude] = useState('all'); - const [page, setPage] = useState(0); - const [pageSize, setPageSize] = useState(25); - const [since, setSince] = useState(''); - const [until, setUntil] = useState(''); - const [data, setData] = useState(null); - const [loading, setLoading] = useState(false); - const [error, setError] = useState(null); - const [reloadCounter, setReloadCounter] = useState(0); - - useEffect(() => { - if (!resource || !uuid) return; - const params = new URLSearchParams(); - params.set('include', include); - params.set('page', String(page)); - params.set('page_size', String(pageSize)); - if (since) params.set('since', since); - if (until) params.set('until', until); - setLoading(true); - setError(null); - SupersetClient.get({ - endpoint: `/api/v1/${resource}/${uuid}/activity/?${params.toString()}`, - }) - .then(({ json }) => { - setData(json as ActivityResponse); - }) - .catch(err => { - const msg = err?.message || String(err); - setError(msg); - setData(null); - }) - .finally(() => setLoading(false)); - }, [resource, uuid, include, page, pageSize, since, until, reloadCounter]); - - if (!resource) { - return ( -
- - {t('Activity Debug — invalid URL')} - - - {t( - 'Use /activity-debug/{dashboard|chart|dataset}/{uuid} — e.g. /activity-debug/dashboard/4a8f3c2e-...', - )} - -
- ); - } - - const records = data?.result ?? []; - const totalCount = data?.count ?? 0; - const lastPage = Math.max(0, Math.ceil(totalCount / pageSize) - 1); - - return ( -
- - {t('Activity Debug')} - - - {t('Throwaway tool for verifying sc-107283 activity-view responses.')} - - - - - - {t('endpoint')} - - {`GET /api/v1/${resource}/${uuid}/activity/`} - - - - - - include - { - setInclude(e.target.value as IncludeMode); - setPage(0); - }} - size="small" - > - all - self - related - - - - - page_size - - - - - since - { - setSince(e.target.value); - setPage(0); - }} - style={{ width: 220 }} - /> - - - - until - { - setUntil(e.target.value); - setPage(0); - }} - style={{ width: 220 }} - /> - - - - - - - - - - {t('count')}: {totalCount} - - - {t('page')}: {page} / {lastPage} - - - - - - {loading && } - - {error && ( - - - {t('error')}: - {' '} - {error} - - )} - - {!loading && !error && records.length === 0 && ( - - )} - - {records.map(record => ( - - ))} -
- ); -} - -export default ActivityDebug; diff --git a/superset-frontend/src/pages/ChartList/VersionHistoryDropdown.tsx b/superset-frontend/src/pages/ChartList/VersionHistoryDropdown.tsx deleted file mode 100644 index 2a1bf5c0a6af..000000000000 --- a/superset-frontend/src/pages/ChartList/VersionHistoryDropdown.tsx +++ /dev/null @@ -1,368 +0,0 @@ -/** - * Licensed to the Apache Software Foundation (ASF) under one - * or more contributor license agreements. See the NOTICE file - * distributed with this work for additional information - * regarding copyright ownership. The ASF licenses this file - * to you under the Apache License, Version 2.0 (the - * "License"); you may not use this file except in compliance - * with the License. You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, - * software distributed under the License is distributed on an - * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY - * KIND, either express or implied. See the License for the - * specific language governing permissions and limitations - * under the License. - */ - -// TEMP: Demo aid for sc-103156 entity-versioning. Lets a user open a -// dropdown of recent versions on a chart and restore one. Not part -// of the merged feature scope (ADR-005 limits v1 to backend); revert -// before pushing the versioning branch. - -import { useState, useCallback } from 'react'; -import { SupersetClient } from '@superset-ui/core'; -import { t } from '@apache-superset/core/translation'; -import { Dropdown, Tooltip, Icons } from '@superset-ui/core/components'; - -interface Change { - kind: string; - // Per-record verb: add / remove / move / edit. Explicit instead of - // inferred from from_value / to_value null-tests or path[0]. - operation: string; - path: string[]; - from_value: unknown; - to_value: unknown; -} - -interface ChangedBy { - id: number; - username: string; - first_name: string; - last_name: string; -} - -interface Version { - version_uuid: string; - version_number: number; - transaction_id: number; - operation_type: string; - // Transaction-level avenue: restore / import / clone / null (= save). - action_kind: string | null; - issued_at: string; - changed_by: ChangedBy | null; - changes: Change[]; -} - -interface Props { - chartUuid: string; - onRestored?: () => void; -} - -// Layout element kinds — used to decide whether a record's -// ``operation`` describes a node-level layout action vs a per-leaf -// change inside a layout-edit. Charts don't normally produce layout -// records, but the formatter stays uniform across the three widgets. -const LAYOUT_KINDS = new Set([ - 'chart', - 'row', - 'column', - 'tab', - 'tabs', - 'header', - 'markdown', - 'divider', -]); - -// Localized labels for the kinds emitted by the backend (layout walker -// + dataset child diff). Defined statically so xgettext can extract them. -const KIND_LABELS: Record = { - chart: t('chart'), - row: t('row'), - column: t('column'), - tab: t('tab'), - tabs: t('tabs'), - header: t('header'), - markdown: t('markdown'), - divider: t('divider'), - metric: t('metric'), -}; -const localizedKind = (k: string): string => KIND_LABELS[k] ?? k; - -function summarizeChange(c: Change): string { - // Layout record at the node level: path = [node_id], operation+kind - // in columns. - if (LAYOUT_KINDS.has(c.kind) && c.path.length === 1) { - const kind = localizedKind(c.kind); - const payload = - ((c.to_value ?? c.from_value) as { name?: string } | null) ?? null; - const name = payload?.name; - if (c.operation === 'add') { - return name - ? t('Added %(kind)s "%(name)s"', { kind, name }) - : t('Added %(kind)s', { kind }); - } - if (c.operation === 'remove') { - return name - ? t('Removed %(kind)s "%(name)s"', { kind, name }) - : t('Removed %(kind)s', { kind }); - } - if (c.operation === 'move') { - return name - ? t('Moved %(kind)s "%(name)s"', { kind, name }) - : t('Moved %(kind)s', { kind }); - } - return name - ? t('Edited %(kind)s "%(name)s"', { kind, name }) - : t('Edited %(kind)s', { kind }); - } - - // Layout edit at the leaf level: path = [node_id, ...leaf-path]. - if (LAYOUT_KINDS.has(c.kind) && c.path.length >= 2) { - const kind = localizedKind(c.kind); - const leaf = String(c.path[c.path.length - 1]); - return t('Changed %(kind)s %(leaf)s', { kind, leaf }); - } - - const isAdd = c.operation === 'add'; - const isRemove = c.operation === 'remove'; - - if (c.path.length === 2 && (c.kind === 'column' || c.kind === 'metric')) { - const kind = localizedKind(c.kind); - const name = String(c.path[1]); - if (isAdd) return t('Added %(kind)s "%(name)s"', { kind, name }); - if (isRemove) return t('Removed %(kind)s "%(name)s"', { kind, name }); - return t('Changed %(kind)s "%(name)s"', { kind, name }); - } - - if (c.path[0] === 'slices') { - const id = String(c.path[1] ?? ''); - if (isAdd) return t('Added chart %(id)s', { id }).trim(); - if (isRemove) return t('Removed chart %(id)s', { id }).trim(); - return t('Changed chart %(id)s', { id }).trim(); - } - - if (c.kind === 'field') { - const fieldName = String(c.path[c.path.length - 1]); - const fieldLabel: string = - fieldName === 'dashboard_title' - ? t('title') - : fieldName === 'slice_name' - ? t('chart name') - : fieldName === 'table_name' - ? t('table name') - : fieldName; - const isShortScalar = - c.to_value !== null && - c.to_value !== undefined && - (typeof c.to_value === 'string' || - typeof c.to_value === 'number' || - typeof c.to_value === 'boolean') && - String(c.to_value).length <= 80; - if (!isAdd && !isRemove && isShortScalar) { - return t('Changed %(field)s to "%(value)s"', { - field: fieldLabel, - value: String(c.to_value), - }); - } - if (isRemove) { - return t('Cleared %(field)s', { field: fieldLabel }); - } - if (isAdd && isShortScalar) { - return t('Set %(field)s to "%(value)s"', { - field: fieldLabel, - value: String(c.to_value), - }); - } - if (isAdd) return t('Added %(field)s', { field: fieldLabel }); - if (isRemove) return t('Removed %(field)s', { field: fieldLabel }); - return t('Changed %(field)s', { field: fieldLabel }); - } - - const kind = localizedKind(c.kind); - if (c.path.length) { - const detail = String(c.path[c.path.length - 1]); - if (isAdd) return t('Added %(kind)s %(detail)s', { kind, detail }); - if (isRemove) return t('Removed %(kind)s %(detail)s', { kind, detail }); - return t('Changed %(kind)s %(detail)s', { kind, detail }); - } - if (isAdd) return t('Added %(kind)s', { kind }); - if (isRemove) return t('Removed %(kind)s', { kind }); - return t('Changed %(kind)s', { kind }); -} - -function formatChangeTitle(changes: Change[]): string { - if (!changes.length) return t('Baseline'); - const first = summarizeChange(changes[0]); - if (changes.length === 1) return first; - return t('%(first)s (+%(more)s more)', { - first, - more: changes.length - 1, - }); -} - -function formatUser(by: ChangedBy | null): string { - if (!by) return t('system'); - if (by.first_name || by.last_name) { - return `${by.first_name ?? ''} ${by.last_name ?? ''}`.trim(); - } - return by.username; -} - -function formatDate(iso: string): string { - try { - // Match the Superset locale set in src/views/App.tsx on - // ``document.documentElement.lang`` rather than the browser default. - const lang = document.documentElement.lang || undefined; - return new Date(iso).toLocaleString(lang); - } catch { - return iso; - } -} - -export default function VersionHistoryDropdown({ - chartUuid, - onRestored, -}: Props) { - const [versions, setVersions] = useState(null); - const [loading, setLoading] = useState(false); - - const loadVersions = useCallback(async () => { - setLoading(true); - try { - const { json } = await SupersetClient.get({ - endpoint: `/api/v1/chart/${chartUuid}/versions/`, - }); - const result = (json as { result: Version[] }).result || []; - // Newest first (API returns oldest-first) - setVersions([...result].reverse().slice(0, 20)); - } catch (e) { - console.error('Failed to load versions', e); - setVersions([]); - } finally { - setLoading(false); - } - }, [chartUuid]); - - const handleRestore = useCallback( - async (version: Version) => { - const summary = formatChangeTitle(version.changes); - if ( - // eslint-disable-next-line no-alert - !window.confirm( - t( - 'Restore this chart to version %(num)s (%(summary)s)? This will overwrite the current state.', - { num: version.version_number, summary }, - ), - ) - ) { - return; - } - try { - await SupersetClient.post({ - endpoint: `/api/v1/chart/${chartUuid}/versions/${version.version_uuid}/restore`, - }); - // eslint-disable-next-line no-alert - window.alert(t('Restored. Reload the page to see the change.')); - if (onRestored) onRestored(); - } catch (e) { - console.error('Restore failed', e); - // eslint-disable-next-line no-alert - window.alert(t('Restore failed — see browser console for details.')); - } - }, - [chartUuid, onRestored], - ); - - const items = (() => { - if (loading) { - return [{ key: 'loading', label: t('Loading…'), disabled: true }]; - } - if (!versions) { - return [ - { key: 'empty', label: t('Click to load versions'), disabled: true }, - ]; - } - if (versions.length === 0) { - return [{ key: 'empty', label: t('No versions yet'), disabled: true }]; - } - // versions is already newest-first, so [0] is the live/current version. - return versions.map((v, idx) => { - const isCurrent = idx === 0; - return { - key: String(v.transaction_id), - // antd's `disabled: true` greys the item and blocks default - // click handling; combined with the inner div NOT having an - // onClick when current, the row becomes informational only. - disabled: isCurrent, - label: ( -
handleRestore(v)} - > -
- #{v.version_number} — {formatChangeTitle(v.changes)} - {isCurrent && ( - - {t('(current)')} - - )} -
-
- {formatUser(v.changed_by)} · {formatDate(v.issued_at)} -
- {v.changes.length > 1 && ( -
    - {v.changes.slice(0, 5).map((c, i) => ( -
  • {summarizeChange(c)}
    • - ))} - {v.changes.length > 5 && ( -
  • - {t('+%(n)s more', { n: v.changes.length - 5 })} -
    • - )} -
- )} -
- ), - }; - }); - })(); - - return ( - { - if (open && versions === null && !loading) loadVersions(); - }} - > - - - - - - - ); -} diff --git a/superset-frontend/src/pages/DashboardList/VersionHistoryDropdown.tsx b/superset-frontend/src/pages/DashboardList/VersionHistoryDropdown.tsx deleted file mode 100644 index 0c2c4fb1a885..000000000000 --- a/superset-frontend/src/pages/DashboardList/VersionHistoryDropdown.tsx +++ /dev/null @@ -1,388 +0,0 @@ -/** - * Licensed to the Apache Software Foundation (ASF) under one - * or more contributor license agreements. See the NOTICE file - * distributed with this work for additional information - * regarding copyright ownership. The ASF licenses this file - * to you under the Apache License, Version 2.0 (the - * "License"); you may not use this file except in compliance - * with the License. You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, - * software distributed under the License is distributed on an - * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY - * KIND, either express or implied. See the License for the - * specific language governing permissions and limitations - * under the License. - */ - -// TEMP: Demo aid for sc-103156 entity-versioning. Lets a user open a -// dropdown of recent versions on a dashboard and restore one. Not part -// of the merged feature scope (ADR-005 limits v1 to backend); revert -// before pushing the versioning branch. - -import { useState, useCallback } from 'react'; -import { SupersetClient } from '@superset-ui/core'; -import { t } from '@apache-superset/core/translation'; -import { Dropdown, Tooltip, Icons } from '@superset-ui/core/components'; - -interface Change { - kind: string; - // Per-record verb: add / remove / move / edit. Explicit instead of - // inferred from from_value / to_value null-tests or path[0]. - operation: string; - path: string[]; - from_value: unknown; - to_value: unknown; -} - -interface ChangedBy { - id: number; - username: string; - first_name: string; - last_name: string; -} - -interface Version { - version_uuid: string; - version_number: number; - transaction_id: number; - operation_type: string; - // Transaction-level avenue: restore / import / clone / null (= save). - // All records sharing a transaction share this. Rendered as a save- - // container headline before the per-record changes are listed. - action_kind: string | null; - issued_at: string; - changed_by: ChangedBy | null; - changes: Change[]; -} - -interface Props { - dashboardUuid: string; - onRestored?: () => void; -} - -// Localized labels for the kinds emitted by the backend (layout walker -// + dataset child diff). Defined statically so xgettext can extract them. -const KIND_LABELS: Record = { - chart: t('chart'), - row: t('row'), - column: t('column'), - tab: t('tab'), - tabs: t('tabs'), - header: t('header'), - markdown: t('markdown'), - divider: t('divider'), - metric: t('metric'), -}; -const localizedKind = (k: string): string => KIND_LABELS[k] ?? k; - -// Layout element kinds — used to decide whether a record's -// ``operation`` describes a node-level layout action vs a per-leaf -// change inside a layout-edit. (Layout edits at depth >= 2 carry the -// node_id + leaf path; we render those via the field-record branch.) -const LAYOUT_KINDS = new Set([ - 'chart', - 'row', - 'column', - 'tab', - 'tabs', - 'header', - 'markdown', - 'divider', -]); - -function summarizeChange(c: Change): string { - // Layout record at the node level: path = [node_id]; operation - // and kind both live in columns. Payload carries name / chartId etc. - if (LAYOUT_KINDS.has(c.kind) && c.path.length === 1) { - const kind = localizedKind(c.kind); - const payload = - ((c.to_value ?? c.from_value) as { name?: string } | null) ?? null; - const name = payload?.name; - if (c.operation === 'add') { - return name - ? t('Added %(kind)s "%(name)s"', { kind, name }) - : t('Added %(kind)s', { kind }); - } - if (c.operation === 'remove') { - return name - ? t('Removed %(kind)s "%(name)s"', { kind, name }) - : t('Removed %(kind)s', { kind }); - } - if (c.operation === 'move') { - return name - ? t('Moved %(kind)s "%(name)s"', { kind, name }) - : t('Moved %(kind)s', { kind }); - } - return name - ? t('Edited %(kind)s "%(name)s"', { kind, name }) - : t('Edited %(kind)s', { kind }); - } - - // Layout edit at the leaf level: path = [node_id, ...leaf-path]. - // kind is the layout element kind; the leaf key is path[1+]. - if (LAYOUT_KINDS.has(c.kind) && c.path.length >= 2) { - const kind = localizedKind(c.kind); - const leaf = String(c.path[c.path.length - 1]); - return t('Changed %(kind)s %(leaf)s', { kind, leaf }); - } - - const isAdd = c.operation === 'add'; - const isRemove = c.operation === 'remove'; - - // Dataset child: path = [columns | metrics, ]. ``kind`` is - // ``column`` / ``metric`` so we can rebuild a readable summary. - if (c.path.length === 2 && (c.kind === 'column' || c.kind === 'metric')) { - const kind = localizedKind(c.kind); - const name = String(c.path[1]); - if (isAdd) return t('Added %(kind)s "%(name)s"', { kind, name }); - if (isRemove) return t('Removed %(kind)s "%(name)s"', { kind, name }); - return t('Changed %(kind)s "%(name)s"', { kind, name }); - } - - // Slice membership (mostly folded into layout records server-side, - // but may still appear if the layout walk didn't catch a chart). - if (c.path[0] === 'slices') { - const id = String(c.path[1] ?? ''); - if (isAdd) return t('Added chart %(id)s', { id }).trim(); - if (isRemove) return t('Removed chart %(id)s', { id }).trim(); - return t('Changed chart %(id)s', { id }).trim(); - } - - // Scalar field record: path = [field_name] or [json_field, sub_key]. - if (c.kind === 'field') { - const fieldName = String(c.path[c.path.length - 1]); - // Friendly labels for the most user-visible fields. - const fieldLabel: string = - fieldName === 'dashboard_title' - ? t('title') - : fieldName === 'slice_name' - ? t('chart name') - : fieldName === 'table_name' - ? t('table name') - : fieldName; - // If the new value is a short primitive (string/number/bool), show - // "Changed to " — much more useful than just naming - // the field. Long strings, dicts and arrays fall through to the - // generic verb-only summary. - const isShortScalar = - c.to_value !== null && - c.to_value !== undefined && - (typeof c.to_value === 'string' || - typeof c.to_value === 'number' || - typeof c.to_value === 'boolean') && - String(c.to_value).length <= 80; - if (!isAdd && !isRemove && isShortScalar) { - return t('Changed %(field)s to "%(value)s"', { - field: fieldLabel, - value: String(c.to_value), - }); - } - if (isRemove) { - return t('Cleared %(field)s', { field: fieldLabel }); - } - if (isAdd && isShortScalar) { - return t('Set %(field)s to "%(value)s"', { - field: fieldLabel, - value: String(c.to_value), - }); - } - if (isAdd) return t('Added %(field)s', { field: fieldLabel }); - if (isRemove) return t('Removed %(field)s', { field: fieldLabel }); - return t('Changed %(field)s', { field: fieldLabel }); - } - - // Fallback: kind plus the trailing path segment (if any). - const kind = localizedKind(c.kind); - if (c.path.length) { - const detail = String(c.path[c.path.length - 1]); - if (isAdd) return t('Added %(kind)s %(detail)s', { kind, detail }); - if (isRemove) return t('Removed %(kind)s %(detail)s', { kind, detail }); - return t('Changed %(kind)s %(detail)s', { kind, detail }); - } - if (isAdd) return t('Added %(kind)s', { kind }); - if (isRemove) return t('Removed %(kind)s', { kind }); - return t('Changed %(kind)s', { kind }); -} - -function formatChangeTitle(changes: Change[]): string { - if (!changes.length) return t('Baseline'); - const first = summarizeChange(changes[0]); - if (changes.length === 1) return first; - return t('%(first)s (+%(more)s more)', { - first, - more: changes.length - 1, - }); -} - -function formatUser(by: ChangedBy | null): string { - if (!by) return t('system'); - if (by.first_name || by.last_name) { - return `${by.first_name ?? ''} ${by.last_name ?? ''}`.trim(); - } - return by.username; -} - -function formatDate(iso: string): string { - try { - // Match the Superset locale set in src/views/App.tsx on - // ``document.documentElement.lang`` rather than the browser default. - const lang = document.documentElement.lang || undefined; - return new Date(iso).toLocaleString(lang); - } catch { - return iso; - } -} - -export default function VersionHistoryDropdown({ - dashboardUuid, - onRestored, -}: Props) { - const [versions, setVersions] = useState(null); - const [loading, setLoading] = useState(false); - - const loadVersions = useCallback(async () => { - setLoading(true); - try { - const { json } = await SupersetClient.get({ - endpoint: `/api/v1/dashboard/${dashboardUuid}/versions/`, - }); - const result = (json as { result: Version[] }).result || []; - // Newest first (API returns oldest-first) - setVersions([...result].reverse().slice(0, 20)); - } catch (e) { - console.error('Failed to load versions', e); - setVersions([]); - } finally { - setLoading(false); - } - }, [dashboardUuid]); - - const handleRestore = useCallback( - async (version: Version) => { - const summary = formatChangeTitle(version.changes); - if ( - // eslint-disable-next-line no-alert - !window.confirm( - t( - 'Restore this dashboard to version %(num)s (%(summary)s)? This will overwrite the current state.', - { num: version.version_number, summary }, - ), - ) - ) { - return; - } - try { - await SupersetClient.post({ - endpoint: `/api/v1/dashboard/${dashboardUuid}/versions/${version.version_uuid}/restore`, - }); - onRestored?.(); - // Navigate to the dashboard with no URL params. A previous - // ``?native_filters_key=…`` (or ``permalink_key`` / ``form_data_key``) - // points at a server-cached snapshot from before the restore; - // the next page hydration would merge it on top of the freshly - // restored ``json_metadata`` and effectively mask the rollback - // (e.g. dashboard-level colour scheme changes don't appear). - // A clean URL forces hydration from the restored DB state. - window.location.href = `/superset/dashboard/${dashboardUuid}/`; - } catch (e) { - console.error('Restore failed', e); - // eslint-disable-next-line no-alert - window.alert(t('Restore failed — see browser console for details.')); - } - }, - [dashboardUuid, onRestored], - ); - - const items = (() => { - if (loading) { - return [{ key: 'loading', label: t('Loading…'), disabled: true }]; - } - if (!versions) { - return [ - { key: 'empty', label: t('Click to load versions'), disabled: true }, - ]; - } - if (versions.length === 0) { - return [{ key: 'empty', label: t('No versions yet'), disabled: true }]; - } - // versions is already newest-first, so [0] is the live/current version. - return versions.map((v, idx) => { - const isCurrent = idx === 0; - return { - key: String(v.transaction_id), - // antd's `disabled: true` greys the item and blocks default - // click handling; combined with the inner div NOT having an - // onClick when current, the row becomes informational only. - disabled: isCurrent, - label: ( -
handleRestore(v)} - > -
- #{v.version_number} — {formatChangeTitle(v.changes)} - {isCurrent && ( - - {t('(current)')} - - )} -
-
- {formatUser(v.changed_by)} · {formatDate(v.issued_at)} -
- {v.changes.length > 1 && ( -
    - {v.changes.slice(0, 5).map((c, i) => ( -
  • {summarizeChange(c)}
    • - ))} - {v.changes.length > 5 && ( -
  • - {t('+%(n)s more', { n: v.changes.length - 5 })} -
    • - )} -
- )} -
- ), - }; - }); - })(); - - return ( - { - if (open && versions === null && !loading) loadVersions(); - }} - > - - - - - - - ); -} diff --git a/superset-frontend/src/pages/DatasetList/VersionHistoryDropdown.tsx b/superset-frontend/src/pages/DatasetList/VersionHistoryDropdown.tsx deleted file mode 100644 index dbf1b4cea774..000000000000 --- a/superset-frontend/src/pages/DatasetList/VersionHistoryDropdown.tsx +++ /dev/null @@ -1,364 +0,0 @@ -/** - * Licensed to the Apache Software Foundation (ASF) under one - * or more contributor license agreements. See the NOTICE file - * distributed with this work for additional information - * regarding copyright ownership. The ASF licenses this file - * to you under the Apache License, Version 2.0 (the - * "License"); you may not use this file except in compliance - * with the License. You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, - * software distributed under the License is distributed on an - * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY - * KIND, either express or implied. See the License for the - * specific language governing permissions and limitations - * under the License. - */ - -// TEMP: Demo aid for sc-103156 entity-versioning. Lets a user open a -// dropdown of recent versions on a dataset and restore one. Not part -// of the merged feature scope (ADR-005 limits v1 to backend); revert -// before pushing the versioning branch. - -import { useState, useCallback } from 'react'; -import { SupersetClient } from '@superset-ui/core'; -import { t } from '@apache-superset/core/translation'; -import { Dropdown, Tooltip, Icons } from '@superset-ui/core/components'; - -interface Change { - kind: string; - // Per-record verb: add / remove / move / edit. Explicit instead of - // inferred from from_value / to_value null-tests or path[0]. - operation: string; - path: string[]; - from_value: unknown; - to_value: unknown; -} - -interface ChangedBy { - id: number; - username: string; - first_name: string; - last_name: string; -} - -interface Version { - version_uuid: string; - version_number: number; - transaction_id: number; - operation_type: string; - // Transaction-level avenue: restore / import / clone / null (= save). - action_kind: string | null; - issued_at: string; - changed_by: ChangedBy | null; - changes: Change[]; -} - -interface Props { - datasetUuid: string; - onRestored?: () => void; -} - -// Layout element kinds — used to decide whether a record's -// ``operation`` describes a node-level layout action vs a per-leaf -// change inside a layout-edit. Datasets don't normally produce layout -// records, but the formatter stays uniform across the three widgets. -const LAYOUT_KINDS = new Set([ - 'chart', - 'row', - 'column', - 'tab', - 'tabs', - 'header', - 'markdown', - 'divider', -]); - -// Localized labels for the kinds emitted by the backend (layout walker -// + dataset child diff). Defined statically so xgettext can extract them. -const KIND_LABELS: Record = { - chart: t('chart'), - row: t('row'), - column: t('column'), - tab: t('tab'), - tabs: t('tabs'), - header: t('header'), - markdown: t('markdown'), - divider: t('divider'), - metric: t('metric'), -}; -const localizedKind = (k: string): string => KIND_LABELS[k] ?? k; - -function summarizeChange(c: Change): string { - // Layout record at the node level: path = [node_id], operation+kind - // in columns. - if (LAYOUT_KINDS.has(c.kind) && c.path.length === 1) { - const kind = localizedKind(c.kind); - const payload = - ((c.to_value ?? c.from_value) as { name?: string } | null) ?? null; - const name = payload?.name; - if (c.operation === 'add') { - return name - ? t('Added %(kind)s "%(name)s"', { kind, name }) - : t('Added %(kind)s', { kind }); - } - if (c.operation === 'remove') { - return name - ? t('Removed %(kind)s "%(name)s"', { kind, name }) - : t('Removed %(kind)s', { kind }); - } - if (c.operation === 'move') { - return name - ? t('Moved %(kind)s "%(name)s"', { kind, name }) - : t('Moved %(kind)s', { kind }); - } - return name - ? t('Edited %(kind)s "%(name)s"', { kind, name }) - : t('Edited %(kind)s', { kind }); - } - - // Layout edit at the leaf level: path = [node_id, ...leaf-path]. - if (LAYOUT_KINDS.has(c.kind) && c.path.length >= 2) { - const kind = localizedKind(c.kind); - const leaf = String(c.path[c.path.length - 1]); - return t('Changed %(kind)s %(leaf)s', { kind, leaf }); - } - - const isAdd = c.operation === 'add'; - const isRemove = c.operation === 'remove'; - - if (c.path.length === 2 && (c.kind === 'column' || c.kind === 'metric')) { - const kind = localizedKind(c.kind); - const name = String(c.path[1]); - if (isAdd) return t('Added %(kind)s "%(name)s"', { kind, name }); - if (isRemove) return t('Removed %(kind)s "%(name)s"', { kind, name }); - return t('Changed %(kind)s "%(name)s"', { kind, name }); - } - - if (c.path[0] === 'slices') { - const id = String(c.path[1] ?? ''); - if (isAdd) return t('Added chart %(id)s', { id }).trim(); - if (isRemove) return t('Removed chart %(id)s', { id }).trim(); - return t('Changed chart %(id)s', { id }).trim(); - } - - if (c.kind === 'field') { - const fieldName = String(c.path[c.path.length - 1]); - const fieldLabel: string = - fieldName === 'dashboard_title' - ? t('title') - : fieldName === 'slice_name' - ? t('chart name') - : fieldName === 'table_name' - ? t('table name') - : fieldName; - const isShortScalar = - c.to_value !== null && - c.to_value !== undefined && - (typeof c.to_value === 'string' || - typeof c.to_value === 'number' || - typeof c.to_value === 'boolean') && - String(c.to_value).length <= 80; - if (!isAdd && !isRemove && isShortScalar) { - return t('Changed %(field)s to "%(value)s"', { - field: fieldLabel, - value: String(c.to_value), - }); - } - if (isRemove) { - return t('Cleared %(field)s', { field: fieldLabel }); - } - if (isAdd && isShortScalar) { - return t('Set %(field)s to "%(value)s"', { - field: fieldLabel, - value: String(c.to_value), - }); - } - if (isAdd) return t('Added %(field)s', { field: fieldLabel }); - if (isRemove) return t('Removed %(field)s', { field: fieldLabel }); - return t('Changed %(field)s', { field: fieldLabel }); - } - - const kind = localizedKind(c.kind); - if (c.path.length) { - const detail = String(c.path[c.path.length - 1]); - if (isAdd) return t('Added %(kind)s %(detail)s', { kind, detail }); - if (isRemove) return t('Removed %(kind)s %(detail)s', { kind, detail }); - return t('Changed %(kind)s %(detail)s', { kind, detail }); - } - if (isAdd) return t('Added %(kind)s', { kind }); - if (isRemove) return t('Removed %(kind)s', { kind }); - return t('Changed %(kind)s', { kind }); -} - -function formatChangeTitle(changes: Change[]): string { - if (!changes.length) return t('Baseline'); - const first = summarizeChange(changes[0]); - if (changes.length === 1) return first; - return t('%(first)s (+%(more)s more)', { - first, - more: changes.length - 1, - }); -} - -function formatUser(by: ChangedBy | null): string { - if (!by) return t('system'); - if (by.first_name || by.last_name) { - return `${by.first_name ?? ''} ${by.last_name ?? ''}`.trim(); - } - return by.username; -} - -function formatDate(iso: string): string { - try { - // Match the Superset locale set in src/views/App.tsx on - // ``document.documentElement.lang`` rather than the browser default. - const lang = document.documentElement.lang || undefined; - return new Date(iso).toLocaleString(lang); - } catch { - return iso; - } -} - -export default function VersionHistoryDropdown({ - datasetUuid, - onRestored, -}: Props) { - const [versions, setVersions] = useState(null); - const [loading, setLoading] = useState(false); - - const loadVersions = useCallback(async () => { - setLoading(true); - try { - const { json } = await SupersetClient.get({ - endpoint: `/api/v1/dataset/${datasetUuid}/versions/`, - }); - const result = (json as { result: Version[] }).result || []; - // Newest first (API returns oldest-first) - setVersions([...result].reverse().slice(0, 20)); - } catch (e) { - console.error('Failed to load versions', e); - setVersions([]); - } finally { - setLoading(false); - } - }, [datasetUuid]); - - const handleRestore = useCallback( - async (version: Version) => { - const summary = formatChangeTitle(version.changes); - if ( - // eslint-disable-next-line no-alert - !window.confirm( - t( - 'Restore this dataset to version %(num)s (%(summary)s)? This will overwrite the current state.', - { num: version.version_number, summary }, - ), - ) - ) { - return; - } - try { - await SupersetClient.post({ - endpoint: `/api/v1/dataset/${datasetUuid}/versions/${version.version_uuid}/restore`, - }); - // eslint-disable-next-line no-alert - window.alert(t('Restored. Reload the page to see the change.')); - if (onRestored) onRestored(); - } catch (e) { - console.error('Restore failed', e); - // eslint-disable-next-line no-alert - window.alert(t('Restore failed — see browser console for details.')); - } - }, - [datasetUuid, onRestored], - ); - - const items = (() => { - if (loading) { - return [{ key: 'loading', label: t('Loading…'), disabled: true }]; - } - if (!versions) { - return [ - { key: 'empty', label: t('Click to load versions'), disabled: true }, - ]; - } - if (versions.length === 0) { - return [{ key: 'empty', label: t('No versions yet'), disabled: true }]; - } - return versions.map((v, idx) => { - const isCurrent = idx === 0; - return { - key: String(v.transaction_id), - disabled: isCurrent, - label: ( -
handleRestore(v)} - > -
- #{v.version_number} — {formatChangeTitle(v.changes)} - {isCurrent && ( - - {t('(current)')} - - )} -
-
- {formatUser(v.changed_by)} · {formatDate(v.issued_at)} -
- {v.changes.length > 1 && ( -
    - {v.changes.slice(0, 5).map((c, i) => ( -
  • {summarizeChange(c)}
    • - ))} - {v.changes.length > 5 && ( -
  • - {t('+%(n)s more', { n: v.changes.length - 5 })} -
    • - )} -
- )} -
- ), - }; - }); - })(); - - return ( - { - if (open && versions === null && !loading) loadVersions(); - }} - > - - - - - - - ); -} diff --git a/superset-frontend/src/views/routes.tsx b/superset-frontend/src/views/routes.tsx index ec6f44a54348..4f066e3ec2cb 100644 --- a/superset-frontend/src/views/routes.tsx +++ b/superset-frontend/src/views/routes.tsx @@ -74,14 +74,6 @@ const DashboardList = lazy( import(/* webpackChunkName: "DashboardList" */ 'src/pages/DashboardList'), ); -// Throwaway: sc-107283 activity-view debug UI. Delete this lazy import + -// the route below + the src/pages/ActivityDebug directory when the -// feature ships. -const ActivityDebug = lazy( - () => - import(/* webpackChunkName: "ActivityDebug" */ 'src/pages/ActivityDebug'), -); - const Dashboard = lazy( () => import(/* webpackChunkName: "Dashboard" */ 'src/pages/Dashboard'), ); @@ -235,11 +227,6 @@ export const routes: Routes = [ path: '/dashboard/list/', Component: DashboardList, }, - // Throwaway: sc-107283 activity-view debug UI. - { - path: '/activity-debug/:resource/:uuid', - Component: ActivityDebug, - }, { path: '/superset/dashboard/:idOrSlug/', Component: Dashboard, diff --git a/superset/charts/api.py b/superset/charts/api.py index fd59f779003f..8363fcce156b 100644 --- a/superset/charts/api.py +++ b/superset/charts/api.py @@ -1409,7 +1409,7 @@ def activity(self, uuid_str: str) -> Response: --- get: summary: Activity stream — chart own edits + datasets the - chart pointed at during association (sc-107283 US2) + chart pointed at during association parameters: - in: path schema: diff --git a/superset/datasets/api.py b/superset/datasets/api.py index 9b36dcf1017d..767c2acc8067 100644 --- a/superset/datasets/api.py +++ b/superset/datasets/api.py @@ -1686,8 +1686,8 @@ def activity(self, uuid_str: str) -> Response: """Return the activity stream for a dataset. --- get: - summary: Activity stream — dataset's own edits only (sc-107283 US3). - Per AV-004, datasets have no transitive layer in V2 — chart and + summary: Activity stream — dataset's own edits only. + Datasets have no transitive layer in V2 — chart and dashboard edits that touch this dataset do NOT appear here; ``?include=related`` and ``?include=all`` collapse to the same self-only stream as ``?include=self``. diff --git a/superset/initialization/__init__.py b/superset/initialization/__init__.py index 7a1493ec2a48..a0b0d8ed6562 100644 --- a/superset/initialization/__init__.py +++ b/superset/initialization/__init__.py @@ -464,12 +464,6 @@ def init_views(self) -> None: appbuilder.add_view_no_menu(RedirectView) appbuilder.add_view_no_menu(RoleRestAPI) appbuilder.add_view_no_menu(UserInfoView) - # Throwaway: sc-107283 activity-view debug page. Delete this - # import + add_view_no_menu when the debug UI is no longer - # needed (see superset/views/activity_debug.py). - from superset.views.activity_debug import ActivityDebugView - - appbuilder.add_view_no_menu(ActivityDebugView) # # Add links diff --git a/superset/versioning/activity/__init__.py b/superset/versioning/activity/__init__.py index 9d17483ab00a..0d277bfb8e69 100644 --- a/superset/versioning/activity/__init__.py +++ b/superset/versioning/activity/__init__.py @@ -14,7 +14,7 @@ # KIND, either express or implied. See the License for the # specific language governing permissions and limitations # under the License. -"""Read-side queries for the cross-entity activity-view API (sc-107283). +"""Read-side queries for the cross-entity activity-view API. Companion to :mod:`superset.versioning.queries`. Whereas ``queries.py`` returns transaction-level history for a single entity, the helpers @@ -39,7 +39,7 @@ * :mod:`.orchestrator` — :func:`get_activity` (public), the ``activity_endpoint`` REST helper, the request param parser (:func:`parse_activity_query_params`), and the observability - instrumentation that T037/T038 specify. + instrumentation (request-shape + per-kind metrics). * :mod:`.scope` — scope resolution (DB-touching): :func:`resolve_scope` / :func:`_resolve_dashboard_scope` / :func:`_resolve_chart_scope` / :func:`_resolve_related_scope`. @@ -54,7 +54,7 @@ path-entity resolution, and tombstone-state lookup. * :mod:`.impact` — per-record impact-count computation (the only field that requires its own batched query). -* :mod:`.visibility` — the AV-008 silent visibility filter; uses +* :mod:`.visibility` — the silent visibility filter; uses the same SQL-side access filters FAB applies on list endpoints. * :mod:`.render` — record-decoration helpers that turn raw rows into the ActivityRecord DTO (summary headlines, ``changed_by`` projection, diff --git a/superset/versioning/activity/impact.py b/superset/versioning/activity/impact.py index 651f81f10604..71312a366275 100644 --- a/superset/versioning/activity/impact.py +++ b/superset/versioning/activity/impact.py @@ -29,8 +29,7 @@ the matching charts validity-strategy-style. * :func:`impact_for_record` — pure projection from the pre-fetched counts onto each record (returns ``None`` for non-Dashboard paths - or non-SqlaTable kinds, matching data-model.md §"``impact`` - computation"). + or non-SqlaTable kinds, matching the ``impact`` computation). Splitting the count batching from the pure projection keeps the SQL inside one function (the batched read) and the per-record decoration @@ -44,14 +43,18 @@ import sqlalchemy as sa from superset.extensions import db -from superset.versioning.activity.kinds import TABLE_KIND_TO_API +from superset.versioning.activity.kinds import ( + chunked_ids, + ENTITY_ID_CHUNK_SIZE, + TABLE_KIND_TO_API, +) def collect_impact_pairs( records: list[dict[str, Any]], path_kind: str ) -> set[tuple[int, int]]: """Distinct ``(dataset_id, transaction_id)`` pairs from *records* - that require an impact computation per data-model.md. + that require an impact computation. Only dashboard-path records whose related entity is a ``SqlaTable`` produce a non-null ``impact`` field; for any other shape this set @@ -99,22 +102,26 @@ def batch_chart_counts( return {} dataset_ids = {dataset_id for dataset_id, _ in pairs} - stmt = sa.select( - m2m_tbl.c.slice_id, - slices_tbl.c.datasource_id, - m2m_tbl.c.transaction_id.label("m2m_start"), - m2m_tbl.c.end_transaction_id.label("m2m_end"), - slices_tbl.c.transaction_id.label("slice_start"), - slices_tbl.c.end_transaction_id.label("slice_end"), - ).where( - m2m_tbl.c.dashboard_id == dashboard_id, - m2m_tbl.c.operation_type != 2, - slices_tbl.c.id == m2m_tbl.c.slice_id, - slices_tbl.c.datasource_id.in_(dataset_ids), - slices_tbl.c.datasource_type == "table", - slices_tbl.c.operation_type != 2, - ) - rows = db.session.connection().execute(stmt).mappings().all() + # Chunk the datasource_id IN-clause to stay under SQLite's bind-variable + # floor (a dashboard pointing at very many datasets can exceed it). + rows: list[Any] = [] + for chunk in chunked_ids(dataset_ids, ENTITY_ID_CHUNK_SIZE): + stmt = sa.select( + m2m_tbl.c.slice_id, + slices_tbl.c.datasource_id, + m2m_tbl.c.transaction_id.label("m2m_start"), + m2m_tbl.c.end_transaction_id.label("m2m_end"), + slices_tbl.c.transaction_id.label("slice_start"), + slices_tbl.c.end_transaction_id.label("slice_end"), + ).where( + m2m_tbl.c.dashboard_id == dashboard_id, + m2m_tbl.c.operation_type != 2, + slices_tbl.c.id == m2m_tbl.c.slice_id, + slices_tbl.c.datasource_id.in_(chunk), + slices_tbl.c.datasource_type == "table", + slices_tbl.c.operation_type != 2, + ) + rows.extend(db.session.connection().execute(stmt).mappings().all()) # For each pair, collect the slice_ids whose two validity windows # both straddle target_tx. ``set`` dedupes within a pair. @@ -146,7 +153,7 @@ def impact_for_record( """Synthesize the ``impact`` field for one record using the pre- fetched *counts* mapping. Pure function — no DB. - Per data-model.md §"``impact`` computation": only + For the ``impact`` computation: only ``path=Dashboard`` and ``related=SqlaTable`` shapes carry an impact; everything else returns ``None``. """ diff --git a/superset/versioning/activity/kinds.py b/superset/versioning/activity/kinds.py index 4c11a05ad13c..21e8a12c018a 100644 --- a/superset/versioning/activity/kinds.py +++ b/superset/versioning/activity/kinds.py @@ -33,6 +33,7 @@ from __future__ import annotations +from collections.abc import Iterable, Iterator from dataclasses import dataclass from flask_appbuilder import Model @@ -42,21 +43,40 @@ from superset.commands.dataset.exceptions import DatasetNotFoundError from superset.versioning.changes import ENTITY_KIND_BY_CLASS_NAME +# Max entity ids per ``IN (...)`` clause across the activity-view queries. +# Bounds the bind-parameter count under SQLite's ``SQLITE_MAX_VARIABLE_NUMBER`` +# floor (default 999 in many builds); Postgres/MySQL accept the full list but +# the chunk is dialect-agnostic. 500 leaves headroom for the other bound +# params on each statement. +ENTITY_ID_CHUNK_SIZE = 500 + + +def chunked_ids( + ids: Iterable[int], size: int = ENTITY_ID_CHUNK_SIZE +) -> Iterator[list[int]]: + """Yield *ids* in fixed-size lists (final chunk may be smaller). Shared by + the activity-view query helpers so every ``IN (...)`` stays under the + SQLite bind-variable floor.""" + items = list(ids) + for i in range(0, len(items), size): + yield items[i : i + size] + + # ---- Kind translation ----------------------------------------------------- # ``version_changes.entity_kind`` stores the friendly downstream-tooling -# value (``"chart"``, ``"dashboard"``, ``"dataset"``) per sc-103156's +# value (``"chart"``, ``"dashboard"``, ``"dataset"``) per the # ``ENTITY_KIND_BY_CLASS_NAME``. The activity-view DTO returns the # Python class name instead (``"Slice"``, ``"Dashboard"``, # ``"SqlaTable"``) so the contract aligns with ``__class__.__name__`` -# (data-model.md §"``ActivityRecord`` DTO"). Translate at the boundary. +# (the ``ActivityRecord`` DTO). Translate at the boundary. TABLE_KIND_TO_API: dict[str, str] = { table_kind: class_name for class_name, table_kind in ENTITY_KIND_BY_CLASS_NAME.items() } API_KIND_TO_TABLE: dict[str, str] = dict(ENTITY_KIND_BY_CLASS_NAME) -# Human-readable label for AV-012 summary headlines +# Human-readable label for summary headlines # ("Dataset updated: Sales Transactions"). Keyed by the internal API kind # (Python class name; matches ``model_cls.__name__``). API_KIND_LABEL: dict[str, str] = { diff --git a/superset/versioning/activity/orchestrator.py b/superset/versioning/activity/orchestrator.py index 7cfbd582b0a0..5deb0a3fd507 100644 --- a/superset/versioning/activity/orchestrator.py +++ b/superset/versioning/activity/orchestrator.py @@ -24,7 +24,7 @@ 2. ``resolve_scope`` (scope.py) — build the related-entity window list. 3. ``fetch_change_records`` (queries.py) — pull rows from ``version_changes`` joined with ``version_transaction`` and ``ab_user``. -4. ``filter_records_by_visibility`` (visibility.py) — silent AV-008 +4. ``filter_records_by_visibility`` (visibility.py) — silent drop of records the requester can't read. 5. ``apply_entity_name_denormalization`` (queries.py) — resolve entity names from the shadow row valid at each record's transaction_id. @@ -36,8 +36,8 @@ :func:`parse_activity_query_params` is called by the three ``/activity/`` endpoint handlers before they call ``get_activity``. Same for the observability instrumentation: ``_phase_timer`` and -``_emit_request_shape_attributes`` emit the metrics that T037/T038 -specify, on the same prefix the cross-coupling test pins. +``_emit_request_shape_attributes`` emit the per-phase timing and +request-shape metrics, on the same prefix the cross-coupling test pins. """ from __future__ import annotations @@ -72,6 +72,12 @@ _VALID_INCLUDE_VALUES: frozenset[str] = frozenset({"self", "related", "all"}) +# Upper bound on the ``q`` search string. The search is a substring scan over +# the (already-capped) materialized record set, so this is a cheap-DoS guard, +# not a correctness limit. +_MAX_Q_LENGTH = 1024 + + class ActivityParamsError(ValueError): """Raised by :func:`parse_activity_query_params` when a query param is malformed. The endpoint catches this and maps to ``response_400``; @@ -97,6 +103,8 @@ def parse_activity_query_params(args: Any) -> dict[str, Any]: if (until := _parse_optional_iso(args.get("until"), name="until")) is not None: params["until"] = until if q := (args.get("q") or "").strip(): + if len(q) > _MAX_Q_LENGTH: + raise ActivityParamsError(f"'q' must be at most {_MAX_Q_LENGTH} characters") params["q"] = q return params @@ -215,21 +223,21 @@ def get_activity( Single polymorphic entry point. Dispatches on *model_cls* to assemble the path entity's self records plus the transitive related- entity records (charts attached to a dashboard, datasets a chart - pointed at, etc.) per data-model.md §"Query phases". + pointed at, etc.). Returns ``(records, total_count, truncated)``. ``truncated`` is ``True`` when the per-request fetch ceiling (``queries._MAX_FETCHED_RECORDS``) bit — older records exist beyond what was materialized, so ``count`` is a floor, not the absolute - total. The count is post-visibility - (AV-008), post-include-filter, and — when ``q`` is supplied — post- + total. The count is post-visibility (silent visibility filter), + post-include-filter, and — when ``q`` is supplied — post- search-filter (``count`` reflects the matches, the contract the server-side search exists to provide), not just the size of the returned slice — clients paginate by passing ``page`` forward until ``page * page_size >= count``. Raises ``DashboardNotFoundError`` / ``ChartNotFoundError`` / - ``DatasetNotFoundError`` when the path entity doesn't exist (AV-009). + ``DatasetNotFoundError`` when the path entity doesn't exist. """ _path_entity, path_id = resolve_path_entity(model_cls, entity_uuid) path_kind = model_cls.__name__ @@ -254,7 +262,7 @@ def get_activity( # ``entity_id`` column (which decoration strips), and dropping # invisible records early means we don't pay for name lookup + # tombstone probes + impact counts on records the requester - # can't see (AV-008's silent-filter contract). + # can't see (the silent-filter contract). with _phase_timer(kind_key, "fetch_ms"): records, truncated = fetch_change_records(entity_windows, since, until) with _phase_timer(kind_key, "visibility_filter_ms"): @@ -268,11 +276,11 @@ def get_activity( with _phase_timer(kind_key, "decorate_ms"): apply_record_decoration(records, path_kind, path_id) - # Server-side search (PR #40988 feedback: the panel's client-side + # Server-side search (the panel's client-side # search only covers loaded pages). Applied post-decoration so the # synthesized ``summary`` / ``entity_name`` participate, and pre- # count so pagination paginates the MATCHES — the full record set - # is already materialized in Python (the documented AV-008 design), + # is already materialized in Python (the documented design), # so the filter adds no extra query. if q: records = [r for r in records if _record_matches(r, q)] @@ -330,9 +338,9 @@ def activity_endpoint( return api.response(200, **payload) -# ---- Observability (T037 / T038) ------------------------------------------ +# ---- Observability ------------------------------------------------------- -#: Common prefix for every metric this module emits. Per plan §D-17. +#: Common prefix for every metric this module emits. _METRIC_PREFIX = "superset.activity_view" @@ -367,7 +375,7 @@ def _emit_request_shape_attributes( ) -> None: """Emit non-PII shape counters about the request and its result set. - Per T038: include_mode / has_since_filter / page_size / record_count + Emits: include_mode / has_since_filter / page_size / record_count + per-related-kind entity counts. **No PII**: entity names, diff content, user identifiers — none of those reach the metric layer. The counters use ``incr`` (counters) since they're tags, not @@ -389,7 +397,7 @@ def _emit_request_shape_attributes( sl.gauge(f"{_METRIC_PREFIX}.{kind_key}.page_size", float(page_size)) sl.gauge(f"{_METRIC_PREFIX}.{kind_key}.record_count", float(record_count)) - # Per-related-kind entity counts (T038 explicit fields). The scope + # Per-related-kind entity counts. The scope # list includes the path entity itself (the "self" window); exclude # it so the gauge reflects only the *related* entities the request # fanned out to, not "this request touched itself". diff --git a/superset/versioning/activity/queries.py b/superset/versioning/activity/queries.py index a84b7352dc6f..be7ffc4d81cd 100644 --- a/superset/versioning/activity/queries.py +++ b/superset/versioning/activity/queries.py @@ -36,7 +36,6 @@ from __future__ import annotations -from collections.abc import Iterator from datetime import datetime from typing import Any from uuid import UUID @@ -47,6 +46,7 @@ from superset.extensions import db from superset.versioning.activity.kinds import ( API_KIND_TO_TABLE, + chunked_ids, EntityWindows, load_shadow_model, NAME_COLUMN, @@ -62,9 +62,9 @@ def resolve_path_entity(model_cls: type[Model], entity_uuid: UUID) -> tuple[Any, int]: """Resolve *entity_uuid* to ``(live_entity, entity_id)`` or raise a - typed 404 per AV-009. + typed 404. - Soft-delete handling (sc-103157) is inherited transparently from + Soft-delete handling is inherited transparently from :func:`superset.versioning.queries.find_active_by_uuid` once it learns to filter out ``deleted_at IS NOT NULL`` rows; at that point soft-deleted paths will also raise here. @@ -161,25 +161,29 @@ def batch_datasets_used_by_charts( from superset.models.slice import Slice slices_tbl = version_class(Slice).__table__ - rows = ( - db.session.connection() - .execute( - sa.select( - slices_tbl.c.id, - slices_tbl.c.datasource_id, - slices_tbl.c.transaction_id, - slices_tbl.c.end_transaction_id, - ).where( - slices_tbl.c.id.in_(slice_ids), - slices_tbl.c.datasource_type == "table", - slices_tbl.c.operation_type != 2, - slices_tbl.c.datasource_id.is_not(None), + grouped: dict[int, list[tuple[int, Window]]] = {} + # Chunk the IN-clause under SQLite's bind-variable floor (a dashboard can + # carry more charts than the floor allows in one statement). + rows: list[Any] = [] + for chunk in chunked_ids(slice_ids): + rows.extend( + db.session.connection() + .execute( + sa.select( + slices_tbl.c.id, + slices_tbl.c.datasource_id, + slices_tbl.c.transaction_id, + slices_tbl.c.end_transaction_id, + ).where( + slices_tbl.c.id.in_(chunk), + slices_tbl.c.datasource_type == "table", + slices_tbl.c.operation_type != 2, + slices_tbl.c.datasource_id.is_not(None), + ) ) + .mappings() + .all() ) - .mappings() - .all() - ) - grouped: dict[int, list[tuple[int, Window]]] = {} for row in rows: grouped.setdefault(row["id"], []).append( ( @@ -218,7 +222,7 @@ def fetch_change_records( ``SQLITE_MAX_EXPR_DEPTH`` limit on dashboards with many slices or many historical attachment windows. - Per AV-008 the visibility filter runs after this function (records + The visibility filter runs after this function (records the requester can't read are silently dropped and must not contribute to ``count``), so the orchestrator paginates in Python over the filtered list — there is no DB-level page ``OFFSET`` here. @@ -229,7 +233,7 @@ def fetch_change_records( Returns ``(records, truncated)``. Records are ordered by ``(issued_at DESC, transaction_id DESC, sequence DESC)`` — the - secondary keys break ties for AV-006's stable-ordering contract. + secondary keys break ties for the stable-ordering contract. """ if not entity_window_tuples: return [], False @@ -248,8 +252,26 @@ def fetch_change_records( if not ids_by_kind: return [], False + # Per-kind transaction_id bounds = the union of that kind's windows. + # Pushing these into the SQL WHERE ensures the per-statement LIMIT + # selects from IN-WINDOW rows. Without it, a related entity whose + # association window is far in the past would have the newest ``limit`` + # (out-of-window) rows fetched and discarded, silently dropping its + # in-window records that lie beyond the limit. ``end_tx = None`` + # (open-ended/current) means no upper bound for that kind. + bounds_by_kind: dict[str, tuple[int, int | None]] = {} + for (table_kind, _entity_id), windows in windows_by_entity.items(): + for w in windows: + cur = bounds_by_kind.get(table_kind) + if cur is None: + bounds_by_kind[table_kind] = (w.start_tx, w.end_tx) + continue + lo = min(cur[0], w.start_tx) + hi = None if (cur[1] is None or w.end_tx is None) else max(cur[1], w.end_tx) + bounds_by_kind[table_kind] = (lo, hi) + rows, truncated = _select_change_rows_for_kinds( - ids_by_kind, since, until, _MAX_FETCHED_RECORDS + ids_by_kind, bounds_by_kind, since, until, _MAX_FETCHED_RECORDS ) filtered = [ row @@ -258,8 +280,19 @@ def fetch_change_records( row, windows_by_entity.get((row["entity_kind"], row["entity_id"]), []) ) ] + # Sort key must be TOTAL so pagination is stable across requests: two + # records from different entities can share (issued_at, transaction_id, + # sequence), so append (entity_kind, entity_id) to break remaining ties + # deterministically. Without these the relative order of tied records + # depends on set-iteration order and a record could shift pages. filtered.sort( - key=lambda r: (r["issued_at"], r["transaction_id"], r["sequence"]), + key=lambda r: ( + r["issued_at"], + r["transaction_id"], + r["sequence"], + r["entity_kind"], + r["entity_id"], + ), reverse=True, ) return filtered, truncated @@ -267,6 +300,7 @@ def fetch_change_records( def _select_change_rows_for_kinds( ids_by_kind: dict[str, set[int]], + bounds_by_kind: dict[str, tuple[int, int | None]], since: datetime | None, until: datetime | None, limit: int, @@ -324,7 +358,7 @@ def _select_change_rows_for_kinds( # ``transaction_id`` share the same value. The column is # declared on the Continuum Table by ``VersionTransactionFactory``, # so ``tx_tbl.c.action_kind`` resolves cleanly here. See - # sc-103156 data-model.md §"Three dimensions". + # the three change-record dimensions. tx_tbl.c.action_kind, user_tbl.c.id.label("changed_by_id"), user_tbl.c.first_name, @@ -341,7 +375,7 @@ def _select_change_rows_for_kinds( # dashboard built from a huge chart library can reach the floor. # Postgres + MySQL accept the full list, but the chunk is # dialect-agnostic for simplicity. - for chunk in _chunked_ids(entity_ids, _ENTITY_ID_CHUNK_SIZE): + for chunk in chunked_ids(entity_ids): stmt = ( sa.select(*select_cols) .select_from(join_tree) @@ -350,6 +384,13 @@ def _select_change_rows_for_kinds( vc.c.entity_id.in_(chunk), ) ) + # Bound by the kind's window union so the LIMIT picks in-window + # rows (see fetch_change_records). The per-entity window filter + # still runs in Python afterwards for exact membership. + tx_lo, tx_hi = bounds_by_kind[table_kind] + stmt = stmt.where(vc.c.transaction_id >= tx_lo) + if tx_hi is not None: + stmt = stmt.where(vc.c.transaction_id < tx_hi) if since is not None: stmt = stmt.where(tx_tbl.c.issued_at >= since) if until is not None: @@ -365,6 +406,7 @@ def _select_change_rows_for_kinds( stmt = stmt.order_by( tx_tbl.c.issued_at.desc(), vc.c.transaction_id.desc(), + vc.c.entity_id.desc(), vc.c.sequence.desc(), ).limit(limit) rows = [ @@ -377,11 +419,6 @@ def _select_change_rows_for_kinds( return out, truncated -# Bind-parameter floor: see ``_select_change_rows_for_kinds`` docstring. -# 500 leaves room for the two literal-string filters and the optional -# since/until datetime params. -_ENTITY_ID_CHUNK_SIZE = 500 - # Per-statement safety ceiling on how many change rows a single activity # request will materialize (per kind-chunk). Bounds memory/CPU for a path # entity with very long history or many related entities; when a statement @@ -390,13 +427,6 @@ def _select_change_rows_for_kinds( _MAX_FETCHED_RECORDS = 5000 -def _chunked_ids(ids: set[int], size: int) -> Iterator[list[int]]: - """Yield *ids* in fixed-size lists. Final chunk may be smaller.""" - items = list(ids) - for i in range(0, len(items), size): - yield items[i : i + size] - - def mark_first_tracked_saves(records: list[dict[str, Any]]) -> None: """Set ``first_tracked_save`` on each record in place: ``True`` when the record's transaction is the entity's FIRST UPDATE (op=1) in its @@ -405,7 +435,7 @@ def mark_first_tracked_saves(records: list[dict[str, Any]]) -> None: The first save of an entity that predates versioning replays every params-normalization delta against the retroactive baseline — a legacy chart's first Explore save produced ~74 records in one - transaction (version-history UI feedback, PR #40988). The server + transaction (version-history UI feedback). The server can't distinguish "normalization" from "the user changed 74 things", but it CAN say "this was the entity's first tracked save"; clients use the marker to collapse such transactions. @@ -437,7 +467,7 @@ def mark_first_tracked_saves(records: list[dict[str, Any]]) -> None: live_model = load_shadow_model(model_name) live_tbl = live_model.__table__ shadow_tbl = version_class(live_model).__table__ - for chunk in _chunked_ids(entity_ids, _ENTITY_ID_CHUNK_SIZE): + for chunk in chunked_ids(entity_ids): stmt = ( sa.select( shadow_tbl.c.id, @@ -487,22 +517,25 @@ def _resolve_names_for_kind( model_name, name_col = NAME_COLUMN[api_kind] model_cls = load_shadow_model(model_name) shadow_tbl = version_class(model_cls).__table__ - ids = sorted({eid for eid, _ in pairs}) - rows = ( - db.session.connection() - .execute( - sa.select( - shadow_tbl.c.id, - shadow_tbl.c.transaction_id, - shadow_tbl.c.end_transaction_id, - shadow_tbl.c[name_col], - ).where(shadow_tbl.c.id.in_(ids)) - ) - .all() - ) + ids = {eid for eid, _ in pairs} per_entity: dict[int, list[tuple[int, int | None, Any]]] = {} - for row in rows: - per_entity.setdefault(row[0], []).append((row[1], row[2], row[3])) + # Chunk the IN-clause to stay under SQLite's bind-variable floor (the + # same reason _select_change_rows_for_kinds chunks). + for chunk in chunked_ids(ids): + rows = ( + db.session.connection() + .execute( + sa.select( + shadow_tbl.c.id, + shadow_tbl.c.transaction_id, + shadow_tbl.c.end_transaction_id, + shadow_tbl.c[name_col], + ).where(shadow_tbl.c.id.in_(chunk)) + ) + .all() + ) + for row in rows: + per_entity.setdefault(row[0], []).append((row[1], row[2], row[3])) resolved: dict[tuple[int, int], str] = {} for entity_id, target_tx in pairs: @@ -559,12 +592,12 @@ def check_entity_tombstones( ) -> dict[tuple[str, int], dict[str, Any]]: """For each ``(api_kind, entity_id)``, report ``deleted`` (no live row) and ``deletion_state`` (``"soft_deleted"`` iff the live row has - a non-null ``deleted_at`` per sc-103157, else ``None``). + a non-null ``deleted_at``, else ``None``). - Pre-sc-103157 the model classes don't have a ``deleted_at`` column; - we probe with ``hasattr`` and report ``deletion_state=None`` - universally in that case. Once sc-103157 lands, this helper picks up - the new column automatically. + The ``deleted_at`` column is probed for at runtime: when the model + classes don't have one, entities are reported as never soft-deleted + (``deletion_state=None``); when a ``deleted_at`` column exists, this + helper picks it up automatically. """ result: dict[tuple[str, int], dict[str, Any]] = {} if not distinct_entities: @@ -580,41 +613,50 @@ def check_entity_tombstones( # flush would otherwise trigger autoflush mid-read. with db.session.no_autoflush: for api_kind, entity_ids in by_kind.items(): - if api_kind not in NAME_COLUMN: - for entity_id in entity_ids: - result[(api_kind, entity_id)] = { - "deleted": True, - "deletion_state": None, - } - continue - - model_name, _ = NAME_COLUMN[api_kind] - model_cls = load_shadow_model(model_name) - live_tbl = model_cls.__table__ - has_deleted_at = "deleted_at" in live_tbl.c - - cols = [live_tbl.c.id] - if has_deleted_at: - cols.append(live_tbl.c.deleted_at) - rows = ( - db.session.connection() - .execute(sa.select(*cols).where(live_tbl.c.id.in_(entity_ids))) - .all() - ) - live: dict[int, Any] = {} - for row in rows: - live[row[0]] = row[1] if has_deleted_at else None - - for entity_id in entity_ids: - if entity_id not in live: - result[(api_kind, entity_id)] = { - "deleted": True, - "deletion_state": None, - } - else: - deleted_at = live[entity_id] - result[(api_kind, entity_id)] = { - "deleted": False, - "deletion_state": "soft_deleted" if deleted_at else None, - } + for entity_id, state in _tombstone_states_for_kind( + api_kind, entity_ids + ).items(): + result[(api_kind, entity_id)] = state return result + + +_TOMBSTONE = {"deleted": True, "deletion_state": None} + + +def _tombstone_states_for_kind( + api_kind: str, entity_ids: list[int] +) -> dict[int, dict[str, Any]]: + """Resolve ``{entity_id: {deleted, deletion_state}}`` for one kind. + + Kinds outside the change-record taxonomy report as tombstoned. For a + known kind, an id with no live row is tombstoned; a live row with a + non-null ``deleted_at`` (when the column exists) is ``soft_deleted``. + """ + if api_kind not in NAME_COLUMN: + return {entity_id: dict(_TOMBSTONE) for entity_id in entity_ids} + + model_cls = load_shadow_model(NAME_COLUMN[api_kind][0]) + live_tbl = model_cls.__table__ + has_deleted_at = "deleted_at" in live_tbl.c + cols = [live_tbl.c.id] + ([live_tbl.c.deleted_at] if has_deleted_at else []) + + live: dict[int, Any] = {} + # Chunk the IN-clause to stay under SQLite's bind-variable floor. + for chunk in chunked_ids(entity_ids): + for row in ( + db.session.connection() + .execute(sa.select(*cols).where(live_tbl.c.id.in_(chunk))) + .all() + ): + live[row[0]] = row[1] if has_deleted_at else None + + states: dict[int, dict[str, Any]] = {} + for entity_id in entity_ids: + if entity_id not in live: + states[entity_id] = dict(_TOMBSTONE) + else: + states[entity_id] = { + "deleted": False, + "deletion_state": "soft_deleted" if live[entity_id] else None, + } + return states diff --git a/superset/versioning/activity/render.py b/superset/versioning/activity/render.py index b8606a551b86..7e1358ef9985 100644 --- a/superset/versioning/activity/render.py +++ b/superset/versioning/activity/render.py @@ -20,7 +20,7 @@ the API contract documents — ``entity_kind`` translated to the user- facing form, ``entity_uuid``, ``entity_deleted`` / ``entity_deletion_state``, ``source`` (self vs. related), -``summary`` (the AV-012 headline), ``impact`` (chart-count for +``summary`` (the headline), ``impact`` (chart-count for dashboard→dataset records), ``version_uuid``, ``changed_by``. This module collects all those decorations: @@ -32,7 +32,7 @@ * :func:`_lookup_entity_uuids` — one IN-clause query per kind to resolve live ``uuid`` for non-tombstoned entities. * :func:`_build_summary` — pure projection of (api_kind, record kind, - entity_name) onto the AV-012 headline string. + entity_name) onto the headline string. * :func:`_changed_by_dict` — projects the user columns onto the ``changed_by`` DTO shape. """ @@ -61,8 +61,8 @@ from superset.versioning.queries import derive_version_uuid _SUMMARY_VERBS: dict[str, str] = { - # The kind taxonomy from FR-016 mapped to past-tense verbs for the - # AV-012 " : " headline. "field" is + # The kind taxonomy mapped to past-tense verbs for the + # " : " headline. "field" is # the fallback for scalar changes that don't map to a named verb. "filter": "filter changed", "metric": "metric changed", @@ -137,7 +137,7 @@ def apply_record_decoration( record["changed_by"] = _changed_by_dict(record) if is_self: - # AV-012 leaves self records summary-less (the panel renders + # Self records are left summary-less (the panel renders # them from kind/path/values) — EXCEPT synthetic ``__meta__`` # headlines, whose entire payload IS the summary and whose # primary surface is the entity's own stream ("restored to @@ -158,8 +158,8 @@ def apply_record_decoration( # diff CONTENT — filter values, column names, SQL/adhoc # expressions — so a requester entitled only to the path # entity can't read the internal change values of a deleted - # related entity. The entity_name and the AV-012 headline - # are kept deliberately (D-15: the panel shows "(deleted) + # related entity. The entity_name and the headline + # are kept deliberately (the panel shows "(deleted) # "); only the value payload is stripped. Self-path # tombstones are untouched — the endpoint already gated them # via ``raise_for_access`` on the path entity. @@ -184,7 +184,7 @@ def _lookup_entity_uuids( tombstones: dict[tuple[str, int], dict[str, Any]], ) -> dict[tuple[str, int], UUID | None]: """Batch-fetch live ``uuid`` per ``(api_kind, entity_id)``. Tombstoned - entities are skipped (their ``entity_uuid`` is null per data-model.md). + entities are skipped (their ``entity_uuid`` is null). """ result: dict[tuple[str, int], UUID | None] = {} by_kind: dict[str, list[int]] = {} @@ -220,13 +220,13 @@ def _lookup_entity_uuids( def _build_summary(api_kind: str, record: dict[str, Any]) -> str: - """Build the AV-012 headline for a related record: + """Build the headline for a related record: ``" : "``.""" label = API_KIND_LABEL.get(api_kind, api_kind) # Synthetic ``__meta__`` headlines carry their payload in to_value # and their verb on the transaction's ``action_kind`` (path stays # pure navigation). The restore variant names the version it - # restored to (PR #40988: "Restored to X from [date]" is not + # restored to ("Restored to X from [date]" is not # renderable from field diffs). if record.get("kind") == "__meta__": name = record.get("entity_name") or "" diff --git a/superset/versioning/activity/scope.py b/superset/versioning/activity/scope.py index f24140c68823..9d7c018049e8 100644 --- a/superset/versioning/activity/scope.py +++ b/superset/versioning/activity/scope.py @@ -61,7 +61,7 @@ def resolve_scope(path_kind: str, path_id: int, include: str) -> list[EntityWind def _resolve_related_scope(path_kind: str, path_id: int) -> list[EntityWindows]: """Walk the dependency edges from the path entity to its related - entities. Per AV-004, datasets have no transitive layer in V2.""" + entities. Datasets have no transitive layer in V2.""" if path_kind == "Dashboard": return _resolve_dashboard_scope(path_id) if path_kind == "Slice": diff --git a/superset/versioning/activity/visibility.py b/superset/versioning/activity/visibility.py index dbc278f17fa0..2b2137ed2795 100644 --- a/superset/versioning/activity/visibility.py +++ b/superset/versioning/activity/visibility.py @@ -14,7 +14,7 @@ # KIND, either express or implied. See the License for the # specific language governing permissions and limitations # under the License. -"""Per-AV-008 silent visibility filter for activity-view records. +"""Silent visibility filter for activity-view records. Drops records whose source entity the requester can't read. Silent in the sense that dropped records contribute no count and no placeholder @@ -58,6 +58,7 @@ from superset.extensions import db from superset.versioning.activity.kinds import ( + ENTITY_ID_CHUNK_SIZE, load_shadow_model, NAME_COLUMN, TABLE_KIND_TO_API, @@ -71,7 +72,7 @@ def filter_records_by_visibility( ) -> list[dict[str, Any]]: """Drop records whose source entity the requester can't read. - Per AV-008 the filter is silent: dropped records contribute no count + The filter is silent: dropped records contribute no count and no placeholder. Tombstoned entities (no live row) pass through — the decorator step marks them ``entity_deleted: true`` and the payload exposes no navigable ``entity_uuid``, so there's nothing @@ -169,29 +170,34 @@ def _resolve_visibility( visible[(api_kind, entity_id)] = True continue model_cls = load_shadow_model(NAME_COLUMN[api_kind][0]) - - # Live ids — what exists at all. Used to decide tombstone vs - # not-visible: an id missing from this set is tombstoned and - # passes through (True); an id in this set but absent from the - # access-filtered set is denied (False). - live_ids = { - row[0] - for row in db.session.query(model_cls.id) - .filter(model_cls.id.in_(entity_ids)) - .all() - } - - # Apply the SQL-side access filter to a query restricted to the - # candidate ids. Same predicate FAB uses for list endpoints, so - # results are consistent with the rest of the read surface. access_filter = access_filter_classes[api_kind]("id", SQLAInterface(model_cls)) - visible_ids = { - row[0] - for row in access_filter.apply( - db.session.query(model_cls.id).filter(model_cls.id.in_(entity_ids)), - value=None, - ).all() - } + + # Chunk the candidate ids to stay under SQLite's bind-variable floor + # (a dashboard built from a large chart library can exceed it). + live_ids: set[int] = set() + visible_ids: set[int] = set() + for start in range(0, len(entity_ids), ENTITY_ID_CHUNK_SIZE): + chunk = entity_ids[start : start + ENTITY_ID_CHUNK_SIZE] + # Live ids — what exists at all. Used to decide tombstone vs + # not-visible: an id missing from this set is tombstoned and + # passes through (True); an id in this set but absent from the + # access-filtered set is denied (False). + live_ids.update( + row[0] + for row in db.session.query(model_cls.id) + .filter(model_cls.id.in_(chunk)) + .all() + ) + # Apply the SQL-side access filter to a query restricted to the + # candidate ids. Same predicate FAB uses for list endpoints, so + # results are consistent with the rest of the read surface. + visible_ids.update( + row[0] + for row in access_filter.apply( + db.session.query(model_cls.id).filter(model_cls.id.in_(chunk)), + value=None, + ).all() + ) for entity_id in entity_ids: if entity_id not in live_ids: diff --git a/superset/versioning/schemas.py b/superset/versioning/schemas.py index c39d7b22c5fd..55533aeb1f54 100644 --- a/superset/versioning/schemas.py +++ b/superset/versioning/schemas.py @@ -142,7 +142,7 @@ class VersionListResponseSchema(Schema): count = fields.Integer() -# ---- Cross-entity activity view (sc-107283) ------------------------------- +# ---- Cross-entity activity view ------------------------------------------ #: Allowed values for ``ActivityRecordSchema.entity_kind``. User-facing #: lowercase strings; the activity layer's internal kind dispatch keys off @@ -151,16 +151,16 @@ class VersionListResponseSchema(Schema): #: :func:`superset.versioning.activity.apply_record_decoration`. ACTIVITY_ENTITY_KINDS: tuple[str, ...] = ("dashboard", "chart", "dataset") -#: Allowed values for ``ActivityRecordSchema.source`` (spec AV-013). +#: Allowed values for ``ActivityRecordSchema.source``. ACTIVITY_SOURCES: tuple[str, ...] = ("self", "related") #: Allowed values for ``ActivityRecordSchema.entity_deletion_state``. #: Hard-delete is communicated separately via ``entity_deleted=true``; -#: the remaining state is the soft-delete sentinel (sc-103157). +#: the remaining state is the soft-delete sentinel. ACTIVITY_DELETION_STATES: tuple[str, ...] = ("soft_deleted",) #: Allowed values for ``ActivityRecordSchema.kind`` — mirrors the -#: change-record taxonomy from sc-103156 FR-016. ``"field"`` is the +#: change-record taxonomy. ``"field"`` is the #: fallback for scalar changes without a more specific category. #: #: ``"restore"`` (previously the synthetic kind for restore events) is @@ -185,7 +185,7 @@ class VersionListResponseSchema(Schema): # ACTION_META_KEY (the ``__meta__`` path convention): machine # namespace, clearly non-content. The canonical case is restore, # whose record's ``to_value`` carries the restored-to - # ``version_uuid`` / ``version_number`` (PR #40988 feedback). + # ``version_uuid`` / ``version_number``. "__meta__", ) @@ -218,8 +218,8 @@ class ActivityChangedBySchema(Schema): The activity-view payload exposes only the display fields (``id`` + given/family name); ``username`` is omitted by design (see - data-model.md §"ActivityRecord DTO"). ``null`` when the saving user - has been deleted from ``ab_user`` (sc-103156 §Session 2026-05-18 + the ActivityRecord DTO). ``null`` when the saving user + has been deleted from ``ab_user`` (the saving user row is gone clarification). """ @@ -251,7 +251,7 @@ class ActivityRecordSchema(Schema): """One change record in the activity stream. One record per atomic field-level change. Fields mirror - data-model.md §"``ActivityRecord`` DTO" — see that doc for source + the ``ActivityRecord`` DTO — see the schema for source and required/optional details. """ @@ -308,7 +308,7 @@ class ActivityRecordSchema(Schema): metadata={ "description": ( "Present when the source entity has non-null ``deleted_at`` " - "(sc-103157). Absent or ``null`` otherwise." + "Absent or ``null`` otherwise." ) }, ) @@ -318,7 +318,7 @@ class ActivityRecordSchema(Schema): "description": ( '``"self"`` if ``(entity_kind, entity_id)`` matches the ' 'path entity; else ``"related"``. Drives the frontend\'s ' - "no-group-under-save rendering rule (AV-013)." + "no-group-under-save rendering rule." ) }, ) @@ -399,7 +399,7 @@ class ActivityRecordSchema(Schema): "description": ( 'Synthesized headline for ``source: "related"`` records — ' 'e.g., ``"Dataset updated: Sales Transactions"`` ' - '(AV-012). Absent for ``source: "self"`` records.' + 'Absent for ``source: "self"`` records.' ) }, ) diff --git a/superset/views/activity_debug.py b/superset/views/activity_debug.py deleted file mode 100644 index 61072fd55cfc..000000000000 --- a/superset/views/activity_debug.py +++ /dev/null @@ -1,53 +0,0 @@ -# Licensed to the Apache Software Foundation (ASF) under one -# or more contributor license agreements. See the NOTICE file -# distributed with this work for additional information -# regarding copyright ownership. The ASF licenses this file -# to you under the Apache License, Version 2.0 (the -# "License"); you may not use this file except in compliance -# with the License. You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, -# software distributed under the License is distributed on an -# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -# KIND, either express or implied. See the License for the -# specific language governing permissions and limitations -# under the License. - -# Throwaway: sc-107283 activity-view debug UI. This Flask view exists -# only to serve the React shell on a fresh page-load of -# /activity-debug//. Without it, Flask returns its -# API-style 404 because the SPA doesn't have a true catch-all — each -# React route needs a corresponding render_app_template call. Delete -# this file + the AppBuilder registration in -# ``superset.initialization`` when the activity-view feature ships -# (or when you no longer need the debug UI). - -from flask_appbuilder import expose - -from superset.superset_typing import FlaskResponse -from superset.views.base import BaseSupersetView - - -class ActivityDebugView(BaseSupersetView): - """Serves the React shell for the throwaway activity-view debug page. - - No auth decorator on the shell itself — the shell page exposes no - data of its own. The React component renders inside it and fires - calls to ``/api/v1/{resource}/{uuid}/activity/`` which gate access - via ``security_manager.raise_for_access`` (read access) on the path - entity — ``raise_for_ownership`` is the write/restore path. Anonymous - users - who somehow land here will see the React UI and the API errors - surface inline as "error: 401 ...". That's a fine UX for a debug - tool — and avoids the FAB ``@has_access`` redirect-to-home - behavior that masked real failures earlier. - """ - - route_base = "/activity-debug" - - @expose("///") - @expose("//") - def show(self, resource: str, uuid: str) -> FlaskResponse: # noqa: ARG002 - return super().render_app_template() diff --git a/tests/integration_tests/versioning/activity_view_tests.py b/tests/integration_tests/versioning/activity_view_tests.py index 9bac7e64a8b5..95b6db7dbc3c 100644 --- a/tests/integration_tests/versioning/activity_view_tests.py +++ b/tests/integration_tests/versioning/activity_view_tests.py @@ -41,7 +41,11 @@ from superset.models.slice import Slice from superset.utils import json as _json from tests.integration_tests.base_tests import SupersetTestCase -from tests.integration_tests.constants import ADMIN_USERNAME, ALPHA_USERNAME +from tests.integration_tests.constants import ( + ADMIN_USERNAME, + ALPHA_USERNAME, + GAMMA_USERNAME, +) from tests.integration_tests.fixtures.birth_names_dashboard import ( # noqa: F401 load_birth_names_dashboard_with_slices, load_birth_names_data, @@ -131,6 +135,63 @@ def test_activity_allows_read_non_owner(self) -> None: rv = self._activity(dashboard_uuid) assert rv.status_code == 200 + def test_visibility_filter_silently_drops_inaccessible_related(self) -> None: + """AV-008 security control: a related record whose entity the caller + cannot read is *silently* dropped — absent from the result, no + placeholder, no contribution to count. Exercises the real + enforcement point (``filter_records_by_visibility`` → + ``DashboardAccessFilter``) with a restricted Gamma principal. + + Setup uses two dashboards (no datasource needed): one owned by + Gamma (readable), and an admin-owned one Gamma may not read. + """ + from superset import security_manager + from superset.versioning.activity.visibility import ( + filter_records_by_visibility, + ) + + admin = security_manager.find_user(ADMIN_USERNAME) + gamma = security_manager.find_user(GAMMA_USERNAME) + visible = Dashboard( + dashboard_title=f"vis-probe-owned {uuid4().hex[:8]}", + slug=f"vis-owned-{uuid4().hex[:8]}", + published=False, + owners=[gamma], + ) + hidden = Dashboard( + dashboard_title=f"vis-probe-hidden {uuid4().hex[:8]}", + slug=f"vis-hidden-{uuid4().hex[:8]}", + published=False, + owners=[admin], + ) + db.session.add_all([visible, hidden]) + db.session.commit() + visible_id, hidden_id = visible.id, hidden.id + try: + records = [ + {"entity_kind": "dashboard", "entity_id": visible_id}, + {"entity_kind": "dashboard", "entity_id": hidden_id}, + ] + self.login(GAMMA_USERNAME) + filtered = filter_records_by_visibility(records) + kept_ids = {r["entity_id"] for r in filtered} + + assert visible_id in kept_ids, ( + "published dashboard must stay visible to Gamma" + ) + assert hidden_id not in kept_ids, ( + "unpublished admin-owned dashboard must be dropped for Gamma" + ) + # Silent: the dropped record leaves nothing behind (no placeholder). + assert len(filtered) == 1 + finally: + db.session.rollback() + for did in (visible_id, hidden_id): + obj = db.session.query(Dashboard).filter(Dashboard.id == did).first() + if obj is not None: + db.session.delete(obj) + db.session.commit() + # ---- 200 happy paths ---- def test_activity_returns_200_with_envelope_shape(self) -> None: From a0b1e263a8c7fd353d9e3f52fa68fe122fd8fb2e Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 22 Jun 2026 11:51:45 -0600 Subject: [PATCH 34/35] fix(activity-view): verification follow-ups MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit From the verification review pass: - Chunk the last unchunked entity-id IN-clause (render._lookup_entity_uuids), which runs on the request path — closes the remaining SQLite bind-floor gap. - Route visibility._resolve_visibility through the shared chunked_ids helper instead of a hand-rolled range-slice (no second chunking implementation). - Align the per-statement SQL order_by with fetch_change_records' final Python sort key (issued_at, transaction_id, sequence, entity_id) so a truncating LIMIT keeps exactly the rows the final sort ranks highest. - Fix the negative-visibility test's assert message (the dashboard is visible to Gamma via ownership, not publication). Co-Authored-By: Claude Opus 4.8 (1M context) --- superset/versioning/activity/queries.py | 6 ++++- superset/versioning/activity/render.py | 22 +++++++++++-------- superset/versioning/activity/visibility.py | 5 ++--- .../versioning/activity_view_tests.py | 2 +- 4 files changed, 21 insertions(+), 14 deletions(-) diff --git a/superset/versioning/activity/queries.py b/superset/versioning/activity/queries.py index be7ffc4d81cd..375895809684 100644 --- a/superset/versioning/activity/queries.py +++ b/superset/versioning/activity/queries.py @@ -403,11 +403,15 @@ def _select_change_rows_for_kinds( # the cap take the newest records; if a statement returns a # full ``limit``, older records exist beyond it and the caller # surfaces ``truncated`` on the response. + # Match fetch_change_records' final Python sort key order + # (issued_at, transaction_id, sequence, entity_id) so a truncating + # LIMIT keeps exactly the rows the final sort ranks highest. + # entity_kind is constant within a per-kind statement. stmt = stmt.order_by( tx_tbl.c.issued_at.desc(), vc.c.transaction_id.desc(), - vc.c.entity_id.desc(), vc.c.sequence.desc(), + vc.c.entity_id.desc(), ).limit(limit) rows = [ dict(row) diff --git a/superset/versioning/activity/render.py b/superset/versioning/activity/render.py index 7e1358ef9985..35ba75c7a449 100644 --- a/superset/versioning/activity/render.py +++ b/superset/versioning/activity/render.py @@ -52,6 +52,7 @@ ) from superset.versioning.activity.kinds import ( API_KIND_LABEL, + chunked_ids, load_shadow_model, NAME_COLUMN, TABLE_KIND_TO_API, @@ -205,17 +206,20 @@ def _lookup_entity_uuids( continue model_cls = load_shadow_model(NAME_COLUMN[api_kind][0]) live_tbl = model_cls.__table__ - rows = ( - db.session.connection() - .execute( - sa.select(live_tbl.c.id, live_tbl.c.uuid).where( - live_tbl.c.id.in_(entity_ids) + # Chunk the IN-clause to stay under SQLite's bind-variable floor + # (a wide dashboard can have more related entities than the floor). + for chunk in chunked_ids(entity_ids): + rows = ( + db.session.connection() + .execute( + sa.select(live_tbl.c.id, live_tbl.c.uuid).where( + live_tbl.c.id.in_(chunk) + ) ) + .all() ) - .all() - ) - for row in rows: - result[(api_kind, row[0])] = row[1] + for row in rows: + result[(api_kind, row[0])] = row[1] return result diff --git a/superset/versioning/activity/visibility.py b/superset/versioning/activity/visibility.py index 2b2137ed2795..a06262472740 100644 --- a/superset/versioning/activity/visibility.py +++ b/superset/versioning/activity/visibility.py @@ -58,7 +58,7 @@ from superset.extensions import db from superset.versioning.activity.kinds import ( - ENTITY_ID_CHUNK_SIZE, + chunked_ids, load_shadow_model, NAME_COLUMN, TABLE_KIND_TO_API, @@ -176,8 +176,7 @@ def _resolve_visibility( # (a dashboard built from a large chart library can exceed it). live_ids: set[int] = set() visible_ids: set[int] = set() - for start in range(0, len(entity_ids), ENTITY_ID_CHUNK_SIZE): - chunk = entity_ids[start : start + ENTITY_ID_CHUNK_SIZE] + for chunk in chunked_ids(entity_ids): # Live ids — what exists at all. Used to decide tombstone vs # not-visible: an id missing from this set is tombstoned and # passes through (True); an id in this set but absent from the diff --git a/tests/integration_tests/versioning/activity_view_tests.py b/tests/integration_tests/versioning/activity_view_tests.py index 95b6db7dbc3c..2936702c0540 100644 --- a/tests/integration_tests/versioning/activity_view_tests.py +++ b/tests/integration_tests/versioning/activity_view_tests.py @@ -177,7 +177,7 @@ def test_visibility_filter_silently_drops_inaccessible_related(self) -> None: kept_ids = {r["entity_id"] for r in filtered} assert visible_id in kept_ids, ( - "published dashboard must stay visible to Gamma" + "gamma-owned dashboard must stay visible to Gamma" ) assert hidden_id not in kept_ids, ( "unpublished admin-owned dashboard must be dropped for Gamma" From b0a166e514265aecd53130df034047bc649fc036 Mon Sep 17 00:00:00 2001 From: Mike Bridge Date: Mon, 22 Jun 2026 12:27:41 -0600 Subject: [PATCH 35/35] test(activity-view): seed a baseline in the perm-only-rewrite test MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit test_perm_only_rewrite_produces_no_version assumed the chart it picks already has a baseline shadow row, so a perm-only edit produces nothing. With version tables cleared between tests, the chart could start with zero shadow rows, making the perm-only commit trigger the synthetic baseline — a shadow row that intermittently failed the assertion when the dir ran in one process. Seed a real content edit first so a baseline provably exists, isolating the "perm-only edit creates no version" behaviour from "first-ever version". Co-Authored-By: Claude Opus 4.8 (1M context) --- .../versioning/change_records_tests.py | 28 ++++++++++++++----- 1 file changed, 21 insertions(+), 7 deletions(-) diff --git a/tests/integration_tests/versioning/change_records_tests.py b/tests/integration_tests/versioning/change_records_tests.py index 001001b48dde..5df3aa88cbaf 100644 --- a/tests/integration_tests/versioning/change_records_tests.py +++ b/tests/integration_tests/versioning/change_records_tests.py @@ -336,16 +336,29 @@ def test_perm_only_rewrite_produces_no_version(self) -> None: chart = db.session.query(Slice).first() assert chart is not None ver_cls = version_class(Slice) + original_name = chart.slice_name original_perms = (chart.perm, chart.schema_perm, chart.catalog_perm) - pre_save_tx_row = ( - db.session.query(ver_cls.transaction_id) - .filter(ver_cls.id == chart.id) - .order_by(ver_cls.transaction_id.desc()) - .first() - ) - pre_save_tx_id = pre_save_tx_row.transaction_id if pre_save_tx_row else 0 try: + # Seed a baseline first: a real content edit guarantees the chart + # already has shadow rows, so the perm-only edit below can't be the + # event that triggers the *synthetic baseline* (which would + # otherwise create a shadow row and confound this assertion — the + # test must isolate "perm-only edit" from "first-ever version"). + # This matters because the suite clears version tables between + # tests, so the chart may start with zero shadow rows. + chart.slice_name = f"{original_name} (baseline-seed)" + db.session.commit() + + pre_save_tx_row = ( + db.session.query(ver_cls.transaction_id) + .filter(ver_cls.id == chart.id) + .order_by(ver_cls.transaction_id.desc()) + .first() + ) + assert pre_save_tx_row is not None, "content edit should have versioned" + pre_save_tx_id = pre_save_tx_row.transaction_id + chart.perm = f"[seed].[perm_rewrite {uuid4().hex[:8]}]" chart.schema_perm = f"[seed].[schema {uuid4().hex[:8]}]" chart.catalog_perm = f"[seed].[catalog {uuid4().hex[:8]}]" @@ -368,6 +381,7 @@ def test_perm_only_rewrite_produces_no_version(self) -> None: # persistent test DB breaks unrelated permission tests. db.session.rollback() chart = db.session.query(Slice).filter(Slice.id == chart.id).one() + chart.slice_name = original_name chart.perm, chart.schema_perm, chart.catalog_perm = original_perms db.session.commit()