evaluator.py

"""
# SnapshotEvaluator

A snapshot evaluator is responsible for evaluating a snapshot given some runtime arguments, e.g. start
and end timestamps.

# Evaluation

Snapshot evaluation involves determining the queries necessary to evaluate a snapshot and using
`sqlmesh.core.engine_adapter` to execute the queries. Schemas, tables, and views are created if
they don't exist and data is inserted when applicable.

A snapshot evaluator also promotes and demotes snapshots to a given environment.

# Audits

A snapshot evaluator can also run the audits for a snapshot's node. This is often done after a snapshot
has been evaluated to check for data quality issues.

For more information about audits, see `sqlmesh.core.audit`.
"""

from __future__ import annotations

import abc
import logging
import typing as t
import sys
from collections import defaultdict
from contextlib import contextmanager
from functools import reduce

from sqlglot import exp, select
from sqlglot.executor import execute

from sqlmesh.core import constants as c
from sqlmesh.core import dialect as d
from sqlmesh.core.audit import Audit, StandaloneAudit
from sqlmesh.core.dialect import schema_
from sqlmesh.core.engine_adapter import EngineAdapter
from sqlmesh.core.engine_adapter.shared import InsertOverwriteStrategy, DataObjectType
from sqlmesh.core.macros import RuntimeStage
from sqlmesh.core.model import (
    AuditResult,
    IncrementalUnmanagedKind,
    Model,
    SeedModel,
    SCDType2ByColumnKind,
    SCDType2ByTimeKind,
    ViewKind,
    CustomKind,
)
from sqlmesh.utils import CompletionStatus
from sqlmesh.core.schema_diff import has_drop_alteration, get_dropped_column_names
from sqlmesh.core.snapshot import (
    DeployabilityIndex,
    Intervals,
    Snapshot,
    SnapshotId,
    SnapshotInfoLike,
    SnapshotTableCleanupTask,
)
from sqlmesh.utils import random_id, CorrelationId
from sqlmesh.utils.concurrency import (
    concurrent_apply_to_snapshots,
    concurrent_apply_to_values,
    NodeExecutionFailedError,
)
from sqlmesh.utils.date import TimeLike, now, time_like_to_str
from sqlmesh.utils.errors import (
    ConfigError,
    DestructiveChangeError,
    SQLMeshError,
    format_destructive_change_msg,
)

if sys.version_info >= (3, 12):
    from importlib import metadata
else:
    import importlib_metadata as metadata  # type: ignore

if t.TYPE_CHECKING:
    from sqlmesh.core.engine_adapter._typing import DF, QueryOrDF
    from sqlmesh.core.environment import EnvironmentNamingInfo

logger = logging.getLogger(__name__)


class SnapshotCreationFailedError(SQLMeshError):
    def __init__(
        self, errors: t.List[NodeExecutionFailedError[SnapshotId]], skipped: t.List[SnapshotId]
    ):
        messages = "\n\n".join(f"{error}\n  {error.__cause__}" for error in errors)
        super().__init__(f"Physical table creation failed:\n\n{messages}")
        self.errors = errors
        self.skipped = skipped


class SnapshotEvaluator:
    """Evaluates a snapshot given runtime arguments through an arbitrary EngineAdapter.

    The SnapshotEvaluator contains the business logic to generically evaluate a snapshot.
    It is responsible for delegating queries to the EngineAdapter. The SnapshotEvaluator
    does not directly communicate with the underlying execution engine.

    Args:
        adapters: A single EngineAdapter or a dictionary of EngineAdapters where
            the key is the gateway name. When a dictionary is provided, and not an
            explicit default gateway its first item is treated as the default
            adapter and used for the virtual layer.
        ddl_concurrent_tasks: The number of concurrent tasks used for DDL
            operations (table / view creation, deletion, etc). Default: 1.
    """

    def __init__(
        self,
        adapters: EngineAdapter | t.Dict[str, EngineAdapter],
        ddl_concurrent_tasks: int = 1,
        selected_gateway: t.Optional[str] = None,
    ):
        self.adapters = (
            adapters if isinstance(adapters, t.Dict) else {selected_gateway or "": adapters}
        )
        self.adapter = (
            next(iter(self.adapters.values()))
            if not selected_gateway
            else self.adapters[selected_gateway]
        )
        self.selected_gateway = selected_gateway
        self.ddl_concurrent_tasks = ddl_concurrent_tasks

    def evaluate(
        self,
        snapshot: Snapshot,
        *,
        start: TimeLike,
        end: TimeLike,
        execution_time: TimeLike,
        snapshots: t.Dict[str, Snapshot],
        allow_destructive_snapshots: t.Optional[t.Set[str]] = None,
        deployability_index: t.Optional[DeployabilityIndex] = None,
        batch_index: int = 0,
        target_table_exists: t.Optional[bool] = None,
        **kwargs: t.Any,
    ) -> t.Optional[str]:
        """Renders the snapshot's model, executes it and stores the result in the snapshot's physical table.

        Args:
            snapshot: Snapshot to evaluate.
            start: The start datetime to render.
            end: The end datetime to render.
            execution_time: The date/time time reference to use for execution time.
            snapshots: All upstream snapshots (by name) to use for expansion and mapping of physical locations.
            allow_destructive_snapshots: Snapshots for which destructive schema changes are allowed.
            deployability_index: Determines snapshots that are deployable in the context of this evaluation.
            batch_index: If the snapshot is part of a batch of related snapshots; which index in the batch is it
            target_table_exists: Whether the target table exists. If None, the table will be checked for existence.
            kwargs: Additional kwargs to pass to the renderer.

        Returns:
            The WAP ID of this evaluation if supported, None otherwise.
        """
        result = self._evaluate_snapshot(
            start=start,
            end=end,
            execution_time=execution_time,
            snapshot=snapshot,
            snapshots=snapshots,
            allow_destructive_snapshots=allow_destructive_snapshots or set(),
            deployability_index=deployability_index,
            batch_index=batch_index,
            target_table_exists=target_table_exists,
            **kwargs,
        )
        if result is None or isinstance(result, str):
            return result
        raise SQLMeshError(
            f"Unexpected result {result} when evaluating snapshot {snapshot.snapshot_id}."
        )

    def evaluate_and_fetch(
        self,
        snapshot: Snapshot,
        *,
        start: TimeLike,
        end: TimeLike,
        execution_time: TimeLike,
        snapshots: t.Dict[str, Snapshot],
        limit: int,
        deployability_index: t.Optional[DeployabilityIndex] = None,
        **kwargs: t.Any,
    ) -> DF:
        """Renders the snapshot's model, executes it and returns a dataframe with the result.

        Args:
            snapshot: Snapshot to evaluate.
            start: The start datetime to render.
            end: The end datetime to render.
            execution_time: The date/time time reference to use for execution time.
            snapshots: All upstream snapshots (by name) to use for expansion and mapping of physical locations.
            limit: The maximum number of rows to fetch.
            deployability_index: Determines snapshots that are deployable in the context of this evaluation.
            kwargs: Additional kwargs to pass to the renderer.

        Returns:
            The result of the evaluation as a dataframe.
        """
        import pandas as pd

        adapter = self.get_adapter(snapshot.model.gateway)
        render_kwargs = dict(
            start=start,
            end=end,
            execution_time=execution_time,
            snapshot=snapshot,
            runtime_stage=RuntimeStage.EVALUATING,
            **kwargs,
        )
        queries_or_dfs = self._render_snapshot_for_evaluation(
            snapshot,
            snapshots,
            deployability_index or DeployabilityIndex.all_deployable(),
            render_kwargs,
        )
        query_or_df = next(queries_or_dfs)
        if isinstance(query_or_df, pd.DataFrame):
            return query_or_df.head(limit)
        if not isinstance(query_or_df, exp.Expression):
            # We assume that if this branch is reached, `query_or_df` is a pyspark / snowpark / bigframe dataframe,
            # so we use `limit` instead of `head` to get back a dataframe instead of List[Row]
            # https://spark.apache.org/docs/3.1.1/api/python/reference/api/pyspark.sql.DataFrame.head.html#pyspark.sql.DataFrame.head
            return query_or_df.limit(limit)

        assert isinstance(query_or_df, exp.Query)

        existing_limit = query_or_df.args.get("limit")
        if existing_limit:
            limit = min(limit, execute(exp.select(existing_limit.expression)).rows[0][0])
            assert limit is not None

        return adapter._fetch_native_df(query_or_df.limit(limit))

    def promote(
        self,
        target_snapshots: t.Iterable[Snapshot],
        environment_naming_info: EnvironmentNamingInfo,
        deployability_index: t.Optional[DeployabilityIndex] = None,
        start: t.Optional[TimeLike] = None,
        end: t.Optional[TimeLike] = None,
        execution_time: t.Optional[TimeLike] = None,
        snapshots: t.Optional[t.Dict[SnapshotId, Snapshot]] = None,
        table_mapping: t.Optional[t.Dict[str, str]] = None,
        on_complete: t.Optional[t.Callable[[SnapshotInfoLike], None]] = None,
    ) -> None:
        """Promotes the given collection of snapshots in the target environment by replacing a corresponding
        view with a physical table associated with the given snapshot.

        Args:
            target_snapshots: Snapshots to promote.
            environment_naming_info: Naming information for the target environment.
            deployability_index: Determines snapshots that are deployable in the context of this promotion.
            on_complete: A callback to call on each successfully promoted snapshot.
        """

        tables_by_gateway: t.Dict[t.Union[str, None], t.List[exp.Table]] = defaultdict(list)
        for snapshot in target_snapshots:
            if snapshot.is_model and not snapshot.is_symbolic:
                gateway = (
                    snapshot.model_gateway if environment_naming_info.gateway_managed else None
                )
                adapter = self.get_adapter(gateway)
                table = snapshot.qualified_view_name.table_for_environment(
                    environment_naming_info, dialect=adapter.dialect
                )
                tables_by_gateway[gateway].append(table)

        # A schema can be shared across multiple engines, so we need to group by gateway
        for gateway, tables in tables_by_gateway.items():
            if environment_naming_info.suffix_target.is_catalog:
                self._create_catalogs(tables=tables, gateway=gateway)

        gateway_table_pairs = [
            (gateway, table) for gateway, tables in tables_by_gateway.items() for table in tables
        ]
        self._create_schemas(gateway_table_pairs=gateway_table_pairs)

        deployability_index = deployability_index or DeployabilityIndex.all_deployable()
        with self.concurrent_context():
            concurrent_apply_to_snapshots(
                target_snapshots,
                lambda s: self._promote_snapshot(
                    s,
                    start=start,
                    end=end,
                    execution_time=execution_time,
                    snapshots=snapshots,
                    table_mapping=table_mapping,
                    environment_naming_info=environment_naming_info,
                    deployability_index=deployability_index,  # type: ignore
                    on_complete=on_complete,
                ),
                self.ddl_concurrent_tasks,
            )

    def demote(
        self,
        target_snapshots: t.Iterable[Snapshot],
        environment_naming_info: EnvironmentNamingInfo,
        table_mapping: t.Optional[t.Dict[str, str]] = None,
        deployability_index: t.Optional[DeployabilityIndex] = None,
        on_complete: t.Optional[t.Callable[[SnapshotInfoLike], None]] = None,
    ) -> None:
        """Demotes the given collection of snapshots in the target environment by removing its view.

        Args:
            target_snapshots: Snapshots to demote.
            environment_naming_info: Naming info for the target environment.
            on_complete: A callback to call on each successfully demoted snapshot.
        """
        with self.concurrent_context():
            concurrent_apply_to_snapshots(
                target_snapshots,
                lambda s: self._demote_snapshot(
                    s,
                    environment_naming_info,
                    deployability_index=deployability_index,
                    on_complete=on_complete,
                    table_mapping=table_mapping,
                ),
                self.ddl_concurrent_tasks,
            )

    def create(
        self,
        target_snapshots: t.Iterable[Snapshot],
        snapshots: t.Dict[SnapshotId, Snapshot],
        deployability_index: t.Optional[DeployabilityIndex] = None,
        on_start: t.Optional[t.Callable] = None,
        on_complete: t.Optional[t.Callable[[SnapshotInfoLike], None]] = None,
        allow_destructive_snapshots: t.Optional[t.Set[str]] = None,
    ) -> CompletionStatus:
        """Creates a physical snapshot schema and table for the given collection of snapshots.

        Args:
            target_snapshots: Target snapshots.
            snapshots: Mapping of snapshot ID to snapshot.
            deployability_index: Determines snapshots that are deployable in the context of this creation.
            on_start: A callback to initialize the snapshot creation progress bar.
            on_complete: A callback to call on each successfully created snapshot.
            allow_destructive_snapshots: Set of snapshots that are allowed to have destructive schema changes.

        Returns:
            CompletionStatus: The status of the creation operation (success, failure, nothing to do).
        """
        deployability_index = deployability_index or DeployabilityIndex.all_deployable()

        snapshots_to_create = self.get_snapshots_to_create(target_snapshots, deployability_index)
        if not snapshots_to_create:
            return CompletionStatus.NOTHING_TO_DO
        if on_start:
            on_start(snapshots_to_create)

        self._create_snapshots(
            snapshots_to_create=snapshots_to_create,
            snapshots={s.name: s for s in snapshots.values()},
            deployability_index=deployability_index,
            on_complete=on_complete,
            allow_destructive_snapshots=allow_destructive_snapshots or set(),
        )
        return CompletionStatus.SUCCESS

    def create_physical_schemas(
        self, snapshots: t.Iterable[Snapshot], deployability_index: DeployabilityIndex
    ) -> None:
        """Creates the physical schemas for the given snapshots.

        Args:
            snapshots: Snapshots to create physical schemas for.
            deployability_index: Determines snapshots that are deployable in the context of this creation.
        """
        tables_by_gateway: t.Dict[t.Optional[str], t.List[str]] = defaultdict(list)
        for snapshot in snapshots:
            if snapshot.is_model and not snapshot.is_symbolic:
                tables_by_gateway[snapshot.model_gateway].append(
                    snapshot.table_name(is_deployable=deployability_index.is_deployable(snapshot))
                )

        gateway_table_pairs = [
            (gateway, table) for gateway, tables in tables_by_gateway.items() for table in tables
        ]
        self._create_schemas(gateway_table_pairs=gateway_table_pairs)

    def get_snapshots_to_create(
        self, target_snapshots: t.Iterable[Snapshot], deployability_index: DeployabilityIndex
    ) -> t.List[Snapshot]:
        """Returns a list of snapshots that need to have their physical tables created.

        Args:
            target_snapshots: Target snapshots.
            deployability_index: Determines snapshots that are deployable / representative in the context of this creation.
        """
        snapshots_with_table_names = defaultdict(set)
        tables_by_gateway_and_schema: t.Dict[t.Union[str, None], t.Dict[exp.Table, set[str]]] = (
            defaultdict(lambda: defaultdict(set))
        )

        for snapshot in target_snapshots:
            if not snapshot.is_model or snapshot.is_symbolic:
                continue
            is_deployable = deployability_index.is_deployable(snapshot)
            table = exp.to_table(snapshot.table_name(is_deployable), dialect=snapshot.model.dialect)
            snapshots_with_table_names[snapshot].add(table.name)
            table_schema = d.schema_(table.db, catalog=table.catalog)
            tables_by_gateway_and_schema[snapshot.model_gateway][table_schema].add(table.name)

        def _get_data_objects(
            schema: exp.Table,
            object_names: t.Optional[t.Set[str]] = None,
            gateway: t.Optional[str] = None,
        ) -> t.Set[str]:
            logger.info("Listing data objects in schema %s", schema.sql())
            objs = self.get_adapter(gateway).get_data_objects(schema, object_names)
            return {obj.name for obj in objs}

        with self.concurrent_context():
            existing_objects: t.Set[str] = set()
            # A schema can be shared across multiple engines, so we need to group tables by both gateway and schema
            for gateway, tables_by_schema in tables_by_gateway_and_schema.items():
                objs_for_gateway = {
                    obj
                    for objs in concurrent_apply_to_values(
                        list(tables_by_schema),
                        lambda s: _get_data_objects(
                            schema=s, object_names=tables_by_schema.get(s), gateway=gateway
                        ),
                        self.ddl_concurrent_tasks,
                    )
                    for obj in objs
                }
                existing_objects.update(objs_for_gateway)

        snapshots_to_create = []
        for snapshot, table_names in snapshots_with_table_names.items():
            missing_tables = table_names - existing_objects
            if missing_tables or (snapshot.is_seed and not snapshot.intervals):
                snapshots_to_create.append(snapshot)

        return snapshots_to_create

    def _create_snapshots(
        self,
        snapshots_to_create: t.Iterable[Snapshot],
        snapshots: t.Dict[str, Snapshot],
        deployability_index: DeployabilityIndex,
        on_complete: t.Optional[t.Callable[[SnapshotInfoLike], None]],
        allow_destructive_snapshots: t.Set[str],
    ) -> None:
        """Internal method to create tables in parallel."""
        with self.concurrent_context():
            errors, skipped = concurrent_apply_to_snapshots(
                snapshots_to_create,
                lambda s: self.create_snapshot(
                    s,
                    snapshots=snapshots,
                    deployability_index=deployability_index,
                    allow_destructive_snapshots=allow_destructive_snapshots,
                    on_complete=on_complete,
                ),
                self.ddl_concurrent_tasks,
                raise_on_error=False,
            )
            if errors:
                raise SnapshotCreationFailedError(errors, skipped)

    def migrate(
        self,
        target_snapshots: t.Iterable[Snapshot],
        snapshots: t.Dict[SnapshotId, Snapshot],
        allow_destructive_snapshots: t.Optional[t.Set[str]] = None,
        deployability_index: t.Optional[DeployabilityIndex] = None,
    ) -> None:
        """Alters a physical snapshot table to match its snapshot's schema for the given collection of snapshots.

        Args:
            target_snapshots: Target snapshots.
            snapshots: Mapping of snapshot ID to snapshot.
            allow_destructive_snapshots: Set of snapshots that are allowed to have destructive schema changes.
            deployability_index: Determines snapshots that are deployable in the context of this evaluation.
        """
        allow_destructive_snapshots = allow_destructive_snapshots or set()
        deployability_index = deployability_index or DeployabilityIndex.all_deployable()
        snapshots_by_name = {s.name: s for s in snapshots.values()}
        with self.concurrent_context():
            concurrent_apply_to_snapshots(
                target_snapshots,
                lambda s: self._migrate_snapshot(
                    s,
                    snapshots_by_name,
                    allow_destructive_snapshots,
                    self.get_adapter(s.model_gateway),
                    deployability_index,
                ),
                self.ddl_concurrent_tasks,
            )

    def cleanup(
        self,
        target_snapshots: t.Iterable[SnapshotTableCleanupTask],
        on_complete: t.Optional[t.Callable[[str], None]] = None,
    ) -> None:
        """Cleans up the given snapshots by removing its table

        Args:
            target_snapshots: Snapshots to cleanup.
            on_complete: A callback to call on each successfully deleted database object.
        """
        snapshots_to_dev_table_only = {
            t.snapshot.snapshot_id: t.dev_table_only for t in target_snapshots
        }

        with self.concurrent_context():
            concurrent_apply_to_snapshots(
                [t.snapshot for t in target_snapshots],
                lambda s: self._cleanup_snapshot(
                    s,
                    snapshots_to_dev_table_only[s.snapshot_id],
                    self.get_adapter(s.model_gateway),
                    on_complete,
                ),
                self.ddl_concurrent_tasks,
                reverse_order=True,
            )

    def audit(
        self,
        snapshot: Snapshot,
        *,
        snapshots: t.Dict[str, Snapshot],
        start: t.Optional[TimeLike] = None,
        end: t.Optional[TimeLike] = None,
        execution_time: t.Optional[TimeLike] = None,
        deployability_index: t.Optional[DeployabilityIndex] = None,
        wap_id: t.Optional[str] = None,
        **kwargs: t.Any,
    ) -> t.List[AuditResult]:
        """Execute a snapshot's node's audit queries.

        Args:
            snapshot: Snapshot to evaluate.
            snapshots: All upstream snapshots (by name) to use for expansion and mapping of physical locations.
            start: The start datetime to audit. Defaults to epoch start.
            end: The end datetime to audit. Defaults to epoch start.
            execution_time: The date/time time reference to use for execution time.
            deployability_index: Determines snapshots that are deployable in the context of this evaluation.
            wap_id: The WAP ID if applicable, None otherwise.
            kwargs: Additional kwargs to pass to the renderer.
        """
        deployability_index = deployability_index or DeployabilityIndex.all_deployable()
        adapter = self.get_adapter(snapshot.model_gateway)

        if not snapshot.version:
            raise ConfigError(
                f"Cannot audit '{snapshot.name}' because it has not been versioned yet. Apply a plan first."
            )

        if wap_id is not None:
            deployability_index = deployability_index or DeployabilityIndex.all_deployable()
            original_table_name = snapshot.table_name(
                is_deployable=deployability_index.is_deployable(snapshot)
            )
            wap_table_name = adapter.wap_table_name(original_table_name, wap_id)
            logger.info(
                "Auditing WAP table '%s', snapshot %s",
                wap_table_name,
                snapshot.snapshot_id,
            )

            table_mapping = kwargs.get("table_mapping") or {}
            table_mapping[snapshot.name] = wap_table_name
            kwargs["table_mapping"] = table_mapping
            kwargs["this_model"] = exp.to_table(wap_table_name, dialect=adapter.dialect)

        results = []

        audits_with_args = snapshot.node.audits_with_args

        force_non_blocking = False

        if audits_with_args:
            logger.info("Auditing snapshot %s", snapshot.snapshot_id)

            if not deployability_index.is_deployable(snapshot) and not adapter.SUPPORTS_CLONING:
                # For dev preview tables that aren't based on clones of the production table, only a subset of the data is typically available
                # However, users still expect audits to run anwyay. Some audits (such as row count) are practically guaranteed to fail
                # when run on only a subset of data, so we switch all audits to non blocking and the user can decide if they still want to proceed
                force_non_blocking = True

        for audit, audit_args in audits_with_args:
            if force_non_blocking:
                # remove any blocking indicator on the model itself
                audit_args.pop("blocking", None)
                # so that we can fall back to the audit's setting, which we override to blocking: False
                audit = audit.model_copy(update={"blocking": False})

            results.append(
                self._audit(
                    audit=audit,
                    audit_args=audit_args,
                    snapshot=snapshot,
                    snapshots=snapshots,
                    start=start,
                    end=end,
                    execution_time=execution_time,
                    deployability_index=deployability_index,
                    **kwargs,
                )
            )

        if wap_id is not None:
            logger.info(
                "Publishing evaluation results for snapshot %s, WAP ID '%s'",
                snapshot.snapshot_id,
                wap_id,
            )
            self._wap_publish_snapshot(snapshot, wap_id, deployability_index)

        return results

    @contextmanager
    def concurrent_context(self) -> t.Iterator[None]:
        try:
            yield
        finally:
            self.recycle()

    def recycle(self) -> None:
        """Closes all open connections and releases all allocated resources associated with any thread
        except the calling one."""
        try:
            for adapter in self.adapters.values():
                adapter.recycle()

        except Exception:
            logger.exception("Failed to recycle Snapshot Evaluator")

    def close(self) -> None:
        """Closes all open connections and releases all allocated resources."""
        try:
            for adapter in self.adapters.values():
                adapter.close()
        except Exception:
            logger.exception("Failed to close Snapshot Evaluator")

    def set_correlation_id(self, correlation_id: CorrelationId) -> SnapshotEvaluator:
        return SnapshotEvaluator(
            {
                gateway: adapter.with_settings(correlation_id=correlation_id)
                for gateway, adapter in self.adapters.items()
            },
            self.ddl_concurrent_tasks,
            self.selected_gateway,
        )

    def _evaluate_snapshot(
        self,
        start: TimeLike,
        end: TimeLike,
        execution_time: TimeLike,
        snapshot: Snapshot,
        snapshots: t.Dict[str, Snapshot],
        allow_destructive_snapshots: t.Set[str],
        deployability_index: t.Optional[DeployabilityIndex],
        batch_index: int,
        target_table_exists: t.Optional[bool],
        **kwargs: t.Any,
    ) -> t.Optional[str]:
        """Renders the snapshot's model and executes it. The return value depends on whether the limit was specified.

        Args:
            snapshot: Snapshot to evaluate.
            start: The start datetime to render.
            end: The end datetime to render.
            execution_time: The date/time time reference to use for execution time.
            snapshots: All upstream snapshots to use for expansion and mapping of physical locations.
            allow_destructive_snapshots: Snapshots for which destructive schema changes are allowed.
            deployability_index: Determines snapshots that are deployable in the context of this evaluation.
            batch_index: If the snapshot is part of a batch of related snapshots; which index in the batch is it
            target_table_exists: Whether the target table exists. If None, the table will be checked for existence.
            kwargs: Additional kwargs to pass to the renderer.
        """
        if not snapshot.is_model:
            return None

        model = snapshot.model

        logger.info("Evaluating snapshot %s", snapshot.snapshot_id)

        adapter = self.get_adapter(model.gateway)
        deployability_index = deployability_index or DeployabilityIndex.all_deployable()
        is_snapshot_deployable = deployability_index.is_deployable(snapshot)
        target_table_name = snapshot.table_name(is_deployable=is_snapshot_deployable)
        # https://github.com/TobikoData/sqlmesh/issues/2609
        # If there are no existing intervals yet; only consider this a first insert for the first snapshot in the batch
        if target_table_exists is None:
            target_table_exists = adapter.table_exists(target_table_name)
        is_first_insert = (
            not _intervals(snapshot, deployability_index) or not target_table_exists
        ) and batch_index == 0

        # Use the 'creating' stage if the table doesn't exist yet to preserve backwards compatibility with existing projects
        # that depend on a separate physical table creation stage.
        runtime_stage = RuntimeStage.EVALUATING if target_table_exists else RuntimeStage.CREATING
        common_render_kwargs = dict(
            start=start,
            end=end,
            execution_time=execution_time,
            snapshot=snapshot,
            runtime_stage=runtime_stage,
            **kwargs,
        )
        create_render_kwargs = dict(
            engine_adapter=adapter,
            snapshots=snapshots,
            deployability_index=deployability_index,
            **common_render_kwargs,
        )
        create_render_kwargs["runtime_stage"] = RuntimeStage.CREATING
        render_statements_kwargs = dict(
            engine_adapter=adapter,
            snapshots=snapshots,
            deployability_index=deployability_index,
            **common_render_kwargs,
        )
        rendered_physical_properties = snapshot.model.render_physical_properties(
            **render_statements_kwargs
        )

        with (
            adapter.transaction(),
            adapter.session(snapshot.model.render_session_properties(**render_statements_kwargs)),
        ):
            adapter.execute(model.render_pre_statements(**render_statements_kwargs))

            if not target_table_exists or (model.is_seed and not snapshot.intervals):
                if self._can_clone(snapshot, deployability_index):
                    self._clone_snapshot_in_dev(
                        snapshot=snapshot,
                        snapshots=snapshots,
                        deployability_index=deployability_index,
                        render_kwargs=create_render_kwargs,
                        rendered_physical_properties=rendered_physical_properties,
                        allow_destructive_snapshots=allow_destructive_snapshots,
                    )
                    common_render_kwargs["runtime_stage"] = RuntimeStage.EVALUATING
                elif model.annotated or model.is_seed or model.kind.is_scd_type_2:
                    self._execute_create(
                        snapshot=snapshot,
                        table_name=target_table_name,
                        is_table_deployable=is_snapshot_deployable,
                        deployability_index=deployability_index,
                        create_render_kwargs=create_render_kwargs,
                        rendered_physical_properties=rendered_physical_properties,
                        dry_run=False,
                        run_pre_post_statements=False,
                    )
                    common_render_kwargs["runtime_stage"] = RuntimeStage.EVALUATING

            wap_id: t.Optional[str] = None
            if snapshot.is_materialized and (
                model.wap_supported or adapter.wap_supported(target_table_name)
            ):
                wap_id = random_id()[0:8]
                logger.info("Using WAP ID '%s' for snapshot %s", wap_id, snapshot.snapshot_id)
                target_table_name = adapter.wap_prepare(target_table_name, wap_id)

            self._render_and_insert_snapshot(
                start=start,
                end=end,
                execution_time=execution_time,
                snapshot=snapshot,
                snapshots=snapshots,
                render_kwargs=common_render_kwargs,
                rendered_physical_properties=rendered_physical_properties,
                deployability_index=deployability_index,
                is_first_insert=is_first_insert,
                batch_index=batch_index,
            )

            adapter.execute(model.render_post_statements(**render_statements_kwargs))

            return wap_id

    def create_snapshot(
        self,
        snapshot: Snapshot,
        snapshots: t.Dict[str, Snapshot],
        deployability_index: DeployabilityIndex,
        allow_destructive_snapshots: t.Set[str],
        on_complete: t.Optional[t.Callable[[SnapshotInfoLike], None]] = None,
    ) -> None:
        """Creates a physical table for the given snapshot.

        Args:
            snapshot: Snapshot to create.
            snapshots: All upstream snapshots to use for expansion and mapping of physical locations.
            deployability_index: Determines snapshots that are deployable in the context of this creation.
            on_complete: A callback to call on each successfully created database object.
            allow_destructive_snapshots: Snapshots for which destructive schema changes are allowed.
        """
        if not snapshot.is_model:
            return

        logger.info("Creating a physical table for snapshot %s", snapshot.snapshot_id)

        adapter = self.get_adapter(snapshot.model.gateway)
        create_render_kwargs: t.Dict[str, t.Any] = dict(
            engine_adapter=adapter,
            snapshots=snapshots,
            runtime_stage=RuntimeStage.CREATING,
            deployability_index=deployability_index,
        )

        with (
            adapter.transaction(),
            adapter.session(snapshot.model.render_session_properties(**create_render_kwargs)),
        ):
            rendered_physical_properties = snapshot.model.render_physical_properties(
                **create_render_kwargs
            )

            if self._can_clone(snapshot, deployability_index):
                self._clone_snapshot_in_dev(
                    snapshot=snapshot,
                    snapshots=snapshots,
                    deployability_index=deployability_index,
                    render_kwargs=create_render_kwargs,
                    rendered_physical_properties=rendered_physical_properties,
                    allow_destructive_snapshots=allow_destructive_snapshots,
                )
            else:
                is_table_deployable = deployability_index.is_deployable(snapshot)
                self._execute_create(
                    snapshot=snapshot,
                    table_name=snapshot.table_name(is_deployable=is_table_deployable),
                    is_table_deployable=is_table_deployable,
                    deployability_index=deployability_index,
                    create_render_kwargs=create_render_kwargs,
                    rendered_physical_properties=rendered_physical_properties,
                    dry_run=True,
                )

        if on_complete is not None:
            on_complete(snapshot)

    def _render_and_insert_snapshot(
        self,
        start: TimeLike,
        end: TimeLike,
        execution_time: TimeLike,
        snapshot: Snapshot,
        snapshots: t.Dict[str, Snapshot],
        render_kwargs: t.Dict[str, t.Any],
        rendered_physical_properties: t.Dict[str, exp.Expression],
        deployability_index: DeployabilityIndex,
        is_first_insert: bool,
        batch_index: int,
    ) -> None:
        if not snapshot.is_model or snapshot.is_seed:
            return

        logger.info("Inserting data for snapshot %s", snapshot.snapshot_id)

        model = snapshot.model
        table_name = snapshot.table_name(is_deployable=deployability_index.is_deployable(snapshot))
        adapter = self.get_adapter(model.gateway)
        evaluation_strategy = _evaluation_strategy(snapshot, adapter)

        queries_or_dfs = self._render_snapshot_for_evaluation(
            snapshot,
            snapshots,
            deployability_index,
            render_kwargs,
        )

        def apply(query_or_df: QueryOrDF, index: int = 0) -> None:
            if index > 0:
                evaluation_strategy.append(
                    table_name=table_name,
                    query_or_df=query_or_df,
                    model=snapshot.model,
                    snapshot=snapshot,
                    snapshots=snapshots,
                    deployability_index=deployability_index,
                    batch_index=batch_index,
                    start=start,
                    end=end,
                    execution_time=execution_time,
                    physical_properties=rendered_physical_properties,
                    render_kwargs=render_kwargs,
                )
            else:
                logger.info(
                    "Inserting batch (%s, %s) into %s'",
                    time_like_to_str(start),
                    time_like_to_str(end),
                    table_name,
                )
                evaluation_strategy.insert(
                    table_name=table_name,
                    query_or_df=query_or_df,
                    is_first_insert=is_first_insert,
                    model=snapshot.model,
                    snapshot=snapshot,
                    snapshots=snapshots,
                    deployability_index=deployability_index,
                    batch_index=batch_index,
                    start=start,
                    end=end,
                    execution_time=execution_time,
                    physical_properties=rendered_physical_properties,
                    render_kwargs=render_kwargs,
                )

        # DataFrames, unlike SQL expressions, can provide partial results by yielding dataframes. As a result,
        # if the engine supports INSERT OVERWRITE or REPLACE WHERE and the snapshot is incremental by time range, we risk
        # having a partial result since each dataframe write can re-truncate partitions. To avoid this, we
        # union all the dataframes together before writing. For pandas this could result in OOM and a potential
        # workaround for that would be to serialize pandas to disk and then read it back with Spark.
        # Note: We assume that if multiple things are yielded from `queries_or_dfs` that they are dataframes
        # and not SQL expressions.
        if (
            adapter.INSERT_OVERWRITE_STRATEGY
            in (
                InsertOverwriteStrategy.INSERT_OVERWRITE,
                InsertOverwriteStrategy.REPLACE_WHERE,
            )
            and snapshot.is_incremental_by_time_range
        ):
            import pandas as pd

            query_or_df = reduce(
                lambda a, b: (
                    pd.concat([a, b], ignore_index=True)  # type: ignore
                    if isinstance(a, pd.DataFrame)
                    else a.union_all(b)  # type: ignore
                ),  # type: ignore
                queries_or_dfs,
            )
            apply(query_or_df, index=0)
        else:
            for index, query_or_df in enumerate(queries_or_dfs):
                apply(query_or_df, index)

    def _render_snapshot_for_evaluation(
        self,
        snapshot: Snapshot,
        snapshots: t.Dict[str, Snapshot],
        deployability_index: DeployabilityIndex,
        render_kwargs: t.Dict[str, t.Any],
    ) -> t.Iterator[QueryOrDF]:
        from sqlmesh.core.context import ExecutionContext

        model = snapshot.model
        adapter = self.get_adapter(model.gateway)

        return model.render(
            context=ExecutionContext(
                adapter,
                snapshots,
                deployability_index,
                default_dialect=model.dialect,
                default_catalog=model.default_catalog,
            ),
            **render_kwargs,
        )

    def _clone_snapshot_in_dev(
        self,
        snapshot: Snapshot,
        snapshots: t.Dict[str, Snapshot],
        deployability_index: DeployabilityIndex,
        render_kwargs: t.Dict[str, t.Any],
        rendered_physical_properties: t.Dict[str, exp.Expression],
        allow_destructive_snapshots: t.Set[str],
    ) -> None:
        adapter = self.get_adapter(snapshot.model.gateway)

        target_table_name = snapshot.table_name(is_deployable=False)
        source_table_name = snapshot.table_name()

        try:
            logger.info(f"Cloning table '{source_table_name}' into '{target_table_name}'")
            adapter.clone_table(
                target_table_name,
                snapshot.table_name(),
                replace=True,
                rendered_physical_properties=rendered_physical_properties,
            )
            self._migrate_target_table(
                target_table_name=target_table_name,
                snapshot=snapshot,
                snapshots=snapshots,
                deployability_index=deployability_index,
                render_kwargs=render_kwargs,
                rendered_physical_properties=rendered_physical_properties,
                allow_destructive_snapshots=allow_destructive_snapshots,
            )
        except Exception:
            adapter.drop_table(target_table_name)
            raise

    def _migrate_snapshot(
        self,
        snapshot: Snapshot,
        snapshots: t.Dict[str, Snapshot],
        allow_destructive_snapshots: t.Set[str],
        adapter: EngineAdapter,
        deployability_index: DeployabilityIndex,
    ) -> None:
        if not snapshot.requires_schema_migration_in_prod:
            return

        deployability_index = DeployabilityIndex.all_deployable()
        render_kwargs: t.Dict[str, t.Any] = dict(
            engine_adapter=adapter,
            snapshots=snapshots,
            runtime_stage=RuntimeStage.CREATING,
            deployability_index=deployability_index,
        )
        target_table_name = snapshot.table_name()

        with (
            adapter.transaction(),
            adapter.session(snapshot.model.render_session_properties(**render_kwargs)),
        ):
            target_data_object = adapter.get_data_object(target_table_name)
            table_exists = target_data_object is not None
            if adapter.drop_data_object_on_type_mismatch(
                target_data_object, _snapshot_to_data_object_type(snapshot)
            ):
                table_exists = False

            if table_exists:
                self._migrate_target_table(
                    target_table_name=target_table_name,
                    snapshot=snapshot,
                    snapshots=snapshots,
                    deployability_index=deployability_index,
                    render_kwargs=render_kwargs,
                    rendered_physical_properties=snapshot.model.render_physical_properties(
                        **render_kwargs
                    ),
                    allow_destructive_snapshots=allow_destructive_snapshots,
                    run_pre_post_statements=True,
                )

    def _migrate_target_table(
        self,
        target_table_name: str,
        snapshot: Snapshot,
        snapshots: t.Dict[str, Snapshot],
        deployability_index: DeployabilityIndex,
        render_kwargs: t.Dict[str, t.Any],
        rendered_physical_properties: t.Dict[str, exp.Expression],
        allow_destructive_snapshots: t.Set[str],
        run_pre_post_statements: bool = False,
    ) -> None:
        adapter = self.get_adapter(snapshot.model.gateway)

        tmp_table_name = f"{target_table_name}_schema_tmp"
        if snapshot.is_materialized:
            self._execute_create(
                snapshot=snapshot,
                table_name=tmp_table_name,
                is_table_deployable=False,
                deployability_index=deployability_index,
                create_render_kwargs=render_kwargs,
                rendered_physical_properties=rendered_physical_properties,
                dry_run=False,
                run_pre_post_statements=run_pre_post_statements,
            )
        try:
            evaluation_strategy = _evaluation_strategy(snapshot, adapter)
            logger.info(
                "Migrating table schema from '%s' to '%s'",
                tmp_table_name,
                target_table_name,
            )
            evaluation_strategy.migrate(
                target_table_name=target_table_name,
                source_table_name=tmp_table_name,
                snapshot=snapshot,
                snapshots=snapshots,
                allow_destructive_snapshots=allow_destructive_snapshots,
                ignore_destructive=snapshot.model.on_destructive_change.is_ignore,
            )
        finally:
            if snapshot.is_materialized:
                adapter.drop_table(tmp_table_name)

    def _promote_snapshot(
        self,
        snapshot: Snapshot,
        environment_naming_info: EnvironmentNamingInfo,
        deployability_index: DeployabilityIndex,
        on_complete: t.Optional[t.Callable[[SnapshotInfoLike], None]],
        start: t.Optional[TimeLike] = None,
        end: t.Optional[TimeLike] = None,
        execution_time: t.Optional[TimeLike] = None,
        snapshots: t.Optional[t.Dict[SnapshotId, Snapshot]] = None,
        table_mapping: t.Optional[t.Dict[str, str]] = None,
    ) -> None:
        if not snapshot.is_model:
            return

        adapter = (
            self.get_adapter(snapshot.model_gateway)
            if environment_naming_info.gateway_managed
            else self.adapter
        )
        table_name = snapshot.table_name(deployability_index.is_representative(snapshot))
        view_name = snapshot.qualified_view_name.for_environment(
            environment_naming_info, dialect=adapter.dialect
        )
        render_kwargs: t.Dict[str, t.Any] = dict(
            start=start,
            end=end,
            execution_time=execution_time,
            engine_adapter=adapter,
            deployability_index=deployability_index,
            table_mapping=table_mapping,
            runtime_stage=RuntimeStage.PROMOTING,
        )

        with (
            adapter.transaction(),
            adapter.session(snapshot.model.render_session_properties(**render_kwargs)),
        ):
            _evaluation_strategy(snapshot, adapter).promote(
                table_name=table_name,
                view_name=view_name,
                model=snapshot.model,
                environment=environment_naming_info.name,
                snapshots=snapshots,
                **render_kwargs,
            )

            snapshot_by_name = {s.name: s for s in (snapshots or {}).values()}
            render_kwargs["snapshots"] = snapshot_by_name
            adapter.execute(snapshot.model.render_on_virtual_update(**render_kwargs))

        if on_complete is not None:
            on_complete(snapshot)

    def _demote_snapshot(
        self,
        snapshot: Snapshot,
        environment_naming_info: EnvironmentNamingInfo,
        deployability_index: t.Optional[DeployabilityIndex],
        on_complete: t.Optional[t.Callable[[SnapshotInfoLike], None]],
        table_mapping: t.Optional[t.Dict[str, str]] = None,
    ) -> None:
        if not snapshot.is_model:
            return

        adapter = (
            self.get_adapter(snapshot.model_gateway)
            if environment_naming_info.gateway_managed
            else self.adapter
        )
        view_name = snapshot.qualified_view_name.for_environment(
            environment_naming_info, dialect=adapter.dialect
        )
        with (
            adapter.transaction(),
            adapter.session(
                snapshot.model.render_session_properties(
                    engine_adapter=adapter,
                    deployability_index=deployability_index,
                    table_mapping=table_mapping,
                    runtime_stage=RuntimeStage.DEMOTING,
                )
            ),
        ):
            _evaluation_strategy(snapshot, adapter).demote(view_name)

        if on_complete is not None:
            on_complete(snapshot)

    def _cleanup_snapshot(
        self,
        snapshot: SnapshotInfoLike,
        dev_table_only: bool,
        adapter: EngineAdapter,
        on_complete: t.Optional[t.Callable[[str], None]],
    ) -> None:
        snapshot = snapshot.table_info

        table_names = [(False, snapshot.table_name(is_deployable=False))]
        if not dev_table_only:
            table_names.append((True, snapshot.table_name(is_deployable=True)))

        evaluation_strategy = _evaluation_strategy(snapshot, adapter)
        for is_table_deployable, table_name in table_names:
            try:
                evaluation_strategy.delete(
                    table_name,
                    is_table_deployable=is_table_deployable,
                    physical_schema=snapshot.physical_schema,
                    # we need to set cascade=true or we will get a 'cant drop because other objects depend on it'-style
                    # error on engines that enforce referential integrity, such as Postgres
                    # this situation can happen when a snapshot expires but downstream view snapshots that reference it have not yet expired
                    cascade=True,
                )
            except Exception:
                # Use `get_data_object` to check if the table exists instead of `table_exists` since the former
                # is based on `INFORMATION_SCHEMA` and avoids touching the table directly.
                # This is important when the table name is malformed for some reason and running any statement
                # that touches the table would result in an error.
                if adapter.get_data_object(table_name) is not None:
                    raise
                logger.warning(
                    "Skipping cleanup of table '%s' because it does not exist", table_name
                )

            if on_complete is not None:
                on_complete(table_name)

    def _wap_publish_snapshot(
        self,
        snapshot: Snapshot,
        wap_id: str,
        deployability_index: t.Optional[DeployabilityIndex],
    ) -> None:
        deployability_index = deployability_index or DeployabilityIndex.all_deployable()
        table_name = snapshot.table_name(is_deployable=deployability_index.is_deployable(snapshot))
        adapter = self.get_adapter(snapshot.model_gateway)
        adapter.wap_publish(table_name, wap_id)

    def _audit(
        self,
        audit: Audit,
        audit_args: t.Dict[t.Any, t.Any],
        snapshot: Snapshot,
        snapshots: t.Dict[str, Snapshot],
        start: t.Optional[TimeLike],
        end: t.Optional[TimeLike],
        execution_time: t.Optional[TimeLike],
        deployability_index: t.Optional[DeployabilityIndex],
        **kwargs: t.Any,
    ) -> AuditResult:
        if audit.skip:
            return AuditResult(
                audit=audit,
                audit_args=audit_args,
                model=snapshot.model_or_none,
                skipped=True,
            )

        # Model's "blocking" argument takes precedence over the audit's default setting
        blocking = audit_args.pop("blocking", None)
        blocking = blocking == exp.true() if blocking else audit.blocking

        adapter = self.get_adapter(snapshot.model_gateway)

        kwargs = {
            "start": start,
            "end": end,
            "execution_time": execution_time,
            "snapshots": snapshots,
            "deployability_index": deployability_index,
            "engine_adapter": adapter,
            "runtime_stage": RuntimeStage.AUDITING,
            **audit_args,
            **kwargs,
        }

        if snapshot.is_model:
            query = snapshot.model.render_audit_query(audit, **kwargs)
        elif isinstance(audit, StandaloneAudit):
            query = audit.render_audit_query(**kwargs)
        else:
            raise SQLMeshError("Expected model or standalone audit. {snapshot}: {audit}")

        count, *_ = adapter.fetchone(
            select("COUNT(*)").from_(query.subquery("audit")),
            quote_identifiers=True,
        )  # type: ignore

        return AuditResult(
            audit=audit,
            audit_args=audit_args,
            model=snapshot.model_or_none,
            count=count,
            query=query,
            blocking=blocking,
        )

    def _create_catalogs(
        self,
        tables: t.Iterable[t.Union[exp.Table, str]],
        gateway: t.Optional[str] = None,
    ) -> None:
        # attempt to create catalogs for the virtual layer if possible
        adapter = self.get_adapter(gateway)
        if adapter.SUPPORTS_CREATE_DROP_CATALOG:
            unique_catalogs = {t.catalog for t in [exp.to_table(maybe_t) for maybe_t in tables]}
            for catalog_name in unique_catalogs:
                adapter.create_catalog(catalog_name)

    def _create_schemas(
        self,
        gateway_table_pairs: t.Iterable[t.Tuple[t.Optional[str], t.Union[exp.Table, str]]],
    ) -> None:
        table_exprs = [(gateway, exp.to_table(t)) for gateway, t in gateway_table_pairs]
        unique_schemas = {
            (gateway, t.args["db"], t.args.get("catalog"))
            for gateway, t in table_exprs
            if t and t.db
        }

        def _create_schema(
            gateway: t.Optional[str], schema_name: str, catalog: t.Optional[str]
        ) -> None:
            schema = schema_(schema_name, catalog)
            logger.info("Creating schema '%s'", schema)
            adapter = self.get_adapter(gateway)
            adapter.create_schema(schema)

        with self.concurrent_context():
            concurrent_apply_to_values(
                list(unique_schemas),
                lambda item: _create_schema(item[0], item[1], item[2]),
                self.ddl_concurrent_tasks,
            )

    def get_adapter(self, gateway: t.Optional[str] = None) -> EngineAdapter:
        """Returns the adapter for the specified gateway or the default adapter if none is provided."""
        if gateway:
            if adapter := self.adapters.get(gateway):
                return adapter
            raise SQLMeshError(f"Gateway '{gateway}' not found in the available engine adapters.")
        return self.adapter

    def _execute_create(
        self,
        snapshot: Snapshot,
        table_name: str,
        is_table_deployable: bool,
        deployability_index: DeployabilityIndex,
        create_render_kwargs: t.Dict[str, t.Any],
        rendered_physical_properties: t.Dict[str, exp.Expression],
        dry_run: bool,
        run_pre_post_statements: bool = True,
    ) -> None:
        adapter = self.get_adapter(snapshot.model.gateway)
        evaluation_strategy = _evaluation_strategy(snapshot, adapter)

        # It can still be useful for some strategies to know if the snapshot was actually deployable
        is_snapshot_deployable = deployability_index.is_deployable(snapshot)
        is_snapshot_representative = deployability_index.is_representative(snapshot)

        create_render_kwargs = {
            **create_render_kwargs,
            "table_mapping": {snapshot.name: table_name},
        }
        if run_pre_post_statements:
            adapter.execute(snapshot.model.render_pre_statements(**create_render_kwargs))
        evaluation_strategy.create(
            table_name=table_name,
            model=snapshot.model,
            is_table_deployable=is_table_deployable,
            render_kwargs=create_render_kwargs,
            is_snapshot_deployable=is_snapshot_deployable,
            is_snapshot_representative=is_snapshot_representative,
            dry_run=dry_run,
            physical_properties=rendered_physical_properties,
        )
        if run_pre_post_statements:
            adapter.execute(snapshot.model.render_post_statements(**create_render_kwargs))

    def _can_clone(self, snapshot: Snapshot, deployability_index: DeployabilityIndex) -> bool:
        adapter = self.get_adapter(snapshot.model.gateway)
        return (
            snapshot.is_forward_only
            and snapshot.is_materialized
            and bool(snapshot.previous_versions)
            and adapter.SUPPORTS_CLONING
            # managed models cannot have their schema mutated because theyre based on queries, so clone + alter wont work
            and not snapshot.is_managed
            # If the deployable table is missing we can't clone it
            and not deployability_index.is_deployable(snapshot)
        )


def _evaluation_strategy(snapshot: SnapshotInfoLike, adapter: EngineAdapter) -> EvaluationStrategy:
    klass: t.Type
    if snapshot.is_embedded:
        klass = EmbeddedStrategy
    elif snapshot.is_symbolic or snapshot.is_audit:
        klass = SymbolicStrategy
    elif snapshot.is_full:
        klass = FullRefreshStrategy
    elif snapshot.is_seed:
        klass = SeedStrategy
    elif snapshot.is_incremental_by_time_range:
        klass = IncrementalByTimeRangeStrategy
    elif snapshot.is_incremental_by_unique_key:
        klass = IncrementalByUniqueKeyStrategy
    elif snapshot.is_incremental_by_partition:
        klass = IncrementalByPartitionStrategy
    elif snapshot.is_incremental_unmanaged:
        klass = IncrementalUnmanagedStrategy
    elif snapshot.is_view:
        klass = ViewStrategy
    elif snapshot.is_scd_type_2:
        klass = SCDType2Strategy
    elif snapshot.is_custom:
        if snapshot.custom_materialization is None:
            raise SQLMeshError(
                f"Missing the name of a custom evaluation strategy in model '{snapshot.name}'."
            )
        _, klass = get_custom_materialization_type_or_raise(snapshot.custom_materialization)
        return klass(adapter)
    elif snapshot.is_managed:
        klass = EngineManagedStrategy
    else:
        raise SQLMeshError(f"Unexpected snapshot: {snapshot}")

    return klass(adapter)


class EvaluationStrategy(abc.ABC):
    def __init__(self, adapter: EngineAdapter):
        self.adapter = adapter

    @abc.abstractmethod
    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        """Inserts the given query or a DataFrame into the target table or a view.

        Args:
            table_name: The name of the target table or view.
            query_or_df: A query or a DataFrame to insert.
            model: The target model.
            is_first_insert: Whether this is the first insert for this version of a model. This value is set to True
                if no data has been previously inserted into the target table, or when the entire history of the target model has
                been restated. Note that in the latter case, the table might contain data from previous executions, and it is the
                responsibility of a specific evaluation strategy to handle the truncation of the table if necessary.
            render_kwargs: Additional key-value arguments to pass when rendering the model's query.
        """

    @abc.abstractmethod
    def append(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        """Appends the given query or a DataFrame to the existing table.

        Args:
            table_name: The target table name.
            query_or_df: A query or a DataFrame to insert.
            model: The target model.
            render_kwargs: Additional key-value arguments to pass when rendering the model's query.
        """

    @abc.abstractmethod
    def create(
        self,
        table_name: str,
        model: Model,
        is_table_deployable: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        """Creates the target table or view.

        Note that the intention here is to just create the table structure, data is loaded in insert() and append()

        Args:
            table_name: The name of a table or a view.
            model: The target model.
            is_table_deployable: True if this creation request is for the "main" table that *might* be deployed to a production environment.
                False if this creation request is for the "dev preview" table. Note that this flag is not related to the DeployabilityIndex
                which determines if the snapshot is deployable to production or not
            render_kwargs: Additional key-value arguments to pass when rendering the model's query.
        """

    @abc.abstractmethod
    def migrate(
        self,
        target_table_name: str,
        source_table_name: str,
        snapshot: Snapshot,
        *,
        ignore_destructive: bool,
        **kwargs: t.Any,
    ) -> None:
        """Migrates the target table schema so that it corresponds to the source table schema.

        Args:
            target_table_name: The target table name.
            source_table_name: The source table name.
            snapshot: The target snapshot.
            ignore_destructive: If True, destructive changes are not created when migrating.
                This is used for forward-only models that are being migrated to a new version.
        """

    @abc.abstractmethod
    def delete(self, name: str, **kwargs: t.Any) -> None:
        """Deletes a target table or a view.

        Args:
            name: The name of a table or a view.
        """

    @abc.abstractmethod
    def promote(
        self,
        table_name: str,
        view_name: str,
        model: Model,
        environment: str,
        **kwargs: t.Any,
    ) -> None:
        """Updates the target view to point to the target table.

        Args:
            table_name: The name of a table in the physical layer that is being promoted.
            view_name: The name of the target view in the virtual layer.
            model: The model that is being promoted.
            environment: The name of the target environment.
        """

    @abc.abstractmethod
    def demote(self, view_name: str, **kwargs: t.Any) -> None:
        """Deletes the target view in the virtual layer.

        Args:
            view_name: The name of the target view in the virtual layer.
        """


class SymbolicStrategy(EvaluationStrategy):
    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        pass

    def append(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        pass

    def create(
        self,
        table_name: str,
        model: Model,
        is_table_deployable: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        pass

    def migrate(
        self,
        target_table_name: str,
        source_table_name: str,
        snapshot: Snapshot,
        *,
        ignore_destructive: bool,
        **kwarg: t.Any,
    ) -> None:
        pass

    def delete(self, name: str, **kwargs: t.Any) -> None:
        pass

    def promote(
        self,
        table_name: str,
        view_name: str,
        model: Model,
        environment: str,
        **kwargs: t.Any,
    ) -> None:
        pass

    def demote(self, view_name: str, **kwargs: t.Any) -> None:
        pass


class EmbeddedStrategy(SymbolicStrategy):
    def promote(
        self,
        table_name: str,
        view_name: str,
        model: Model,
        environment: str,
        **kwargs: t.Any,
    ) -> None:
        logger.info("Dropping view '%s' for non-materialized table", view_name)
        self.adapter.drop_view(view_name, cascade=False)


class PromotableStrategy(EvaluationStrategy, abc.ABC):
    def promote(
        self,
        table_name: str,
        view_name: str,
        model: Model,
        environment: str,
        **kwargs: t.Any,
    ) -> None:
        is_prod = environment == c.PROD
        logger.info("Updating view '%s' to point at table '%s'", view_name, table_name)
        render_kwargs: t.Dict[str, t.Any] = dict(
            start=kwargs.get("start"),
            end=kwargs.get("end"),
            execution_time=kwargs.get("execution_time"),
            engine_adapter=kwargs.get("engine_adapter"),
            snapshots=kwargs.get("snapshots"),
            deployability_index=kwargs.get("deployability_index"),
            table_mapping=kwargs.get("table_mapping"),
            runtime_stage=kwargs.get("runtime_stage"),
        )
        self.adapter.create_view(
            view_name,
            exp.select("*").from_(table_name, dialect=self.adapter.dialect),
            table_description=model.description if is_prod else None,
            column_descriptions=model.column_descriptions if is_prod else None,
            view_properties=model.render_virtual_properties(**render_kwargs),
        )

    def demote(self, view_name: str, **kwargs: t.Any) -> None:
        logger.info("Dropping view '%s'", view_name)
        self.adapter.drop_view(view_name, cascade=False)


class MaterializableStrategy(PromotableStrategy, abc.ABC):
    def create(
        self,
        table_name: str,
        model: Model,
        is_table_deployable: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        ctas_query = model.ctas_query(**render_kwargs)
        physical_properties = kwargs.get("physical_properties", model.physical_properties)

        logger.info("Creating table '%s'", table_name)
        if model.annotated:
            self.adapter.create_table(
                table_name,
                target_columns_to_types=model.columns_to_types_or_raise,
                table_format=model.table_format,
                storage_format=model.storage_format,
                partitioned_by=model.partitioned_by,
                partition_interval_unit=model.partition_interval_unit,
                clustered_by=model.clustered_by,
                table_properties=physical_properties,
                table_description=model.description if is_table_deployable else None,
                column_descriptions=model.column_descriptions if is_table_deployable else None,
            )

            # If we create both temp and prod tables, we need to make sure that we dry run once.
            dry_run = kwargs.get("dry_run", True) or not is_table_deployable

            # Only sql models have queries that can be tested.
            # We also need to make sure that we don't dry run on Redshift because its planner / optimizer sometimes
            # breaks on our CTAS queries due to us relying on the WHERE FALSE LIMIT 0 combo.
            if model.is_sql and dry_run and self.adapter.dialect != "redshift":
                logger.info("Dry running model '%s'", model.name)
                self.adapter.fetchall(ctas_query)
        else:
            self.adapter.ctas(
                table_name,
                ctas_query,
                model.columns_to_types,
                table_format=model.table_format,
                storage_format=model.storage_format,
                partitioned_by=model.partitioned_by,
                partition_interval_unit=model.partition_interval_unit,
                clustered_by=model.clustered_by,
                table_properties=physical_properties,
                table_description=model.description if is_table_deployable else None,
                column_descriptions=model.column_descriptions if is_table_deployable else None,
            )

    def migrate(
        self,
        target_table_name: str,
        source_table_name: str,
        snapshot: Snapshot,
        *,
        ignore_destructive: bool,
        **kwargs: t.Any,
    ) -> None:
        logger.info(f"Altering table '{target_table_name}'")
        alter_expressions = self.adapter.get_alter_expressions(
            target_table_name, source_table_name, ignore_destructive=ignore_destructive
        )
        _check_destructive_schema_change(
            snapshot, alter_expressions, kwargs["allow_destructive_snapshots"]
        )
        self.adapter.alter_table(alter_expressions)

    def delete(self, name: str, **kwargs: t.Any) -> None:
        _check_table_db_is_physical_schema(name, kwargs["physical_schema"])
        self.adapter.drop_table(name, cascade=kwargs.pop("cascade", False))
        logger.info("Dropped table '%s'", name)

    def _replace_query_for_model(
        self,
        model: Model,
        name: str,
        query_or_df: QueryOrDF,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        """Replaces the table for the given model.

        Args:
            model: The target model.
            name: The name of the target table.
            query_or_df: The query or DataFrame to replace the target table with.
        """
        if (model.is_seed or model.kind.is_full) and model.annotated:
            columns_to_types = model.columns_to_types_or_raise
            source_columns: t.Optional[t.List[str]] = list(columns_to_types)
        else:
            try:
                # Source columns from the underlying table to prevent unintentional table schema changes during restatement of incremental models.
                columns_to_types, source_columns = self._get_target_and_source_columns(
                    model, name, render_kwargs, force_get_columns_from_target=True
                )
            except Exception:
                columns_to_types, source_columns = None, None

        self.adapter.replace_query(
            name,
            query_or_df,
            table_format=model.table_format,
            storage_format=model.storage_format,
            partitioned_by=model.partitioned_by,
            partition_interval_unit=model.partition_interval_unit,
            clustered_by=model.clustered_by,
            table_properties=kwargs.get("physical_properties", model.physical_properties),
            table_description=model.description,
            column_descriptions=model.column_descriptions,
            target_columns_to_types=columns_to_types,
            source_columns=source_columns,
        )

    def _get_target_and_source_columns(
        self,
        model: Model,
        table_name: str,
        render_kwargs: t.Dict[str, t.Any],
        force_get_columns_from_target: bool = False,
    ) -> t.Tuple[t.Dict[str, exp.DataType], t.Optional[t.List[str]]]:
        if force_get_columns_from_target:
            target_column_to_types = self.adapter.columns(table_name)
        else:
            target_column_to_types = (
                model.columns_to_types  # type: ignore
                if model.annotated and not model.on_destructive_change.is_ignore
                else self.adapter.columns(table_name)
            )
        assert target_column_to_types is not None
        if model.on_destructive_change.is_ignore:
            # We need to identify the columns that are only in the source so we create an empty table with
            # the user query to determine that
            with self.adapter.temp_table(model.ctas_query(**render_kwargs)) as temp_table:
                source_columns = list(self.adapter.columns(temp_table))
        else:
            source_columns = None
        return target_column_to_types, source_columns


class IncrementalStrategy(MaterializableStrategy, abc.ABC):
    def append(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        columns_to_types, source_columns = self._get_target_and_source_columns(
            model, table_name, render_kwargs=render_kwargs
        )
        self.adapter.insert_append(
            table_name,
            query_or_df,
            target_columns_to_types=columns_to_types,
            source_columns=source_columns,
        )


class IncrementalByPartitionStrategy(IncrementalStrategy):
    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        if is_first_insert:
            self._replace_query_for_model(model, table_name, query_or_df, render_kwargs, **kwargs)
        else:
            columns_to_types, source_columns = self._get_target_and_source_columns(
                model, table_name, render_kwargs=render_kwargs
            )
            self.adapter.insert_overwrite_by_partition(
                table_name,
                query_or_df,
                partitioned_by=model.partitioned_by,
                target_columns_to_types=columns_to_types,
                source_columns=source_columns,
            )


class IncrementalByTimeRangeStrategy(IncrementalStrategy):
    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        assert model.time_column
        columns_to_types, source_columns = self._get_target_and_source_columns(
            model, table_name, render_kwargs=render_kwargs
        )
        self.adapter.insert_overwrite_by_time_partition(
            table_name,
            query_or_df,
            time_formatter=model.convert_to_time_column,
            time_column=model.time_column,
            target_columns_to_types=columns_to_types,
            source_columns=source_columns,
            **kwargs,
        )


class IncrementalByUniqueKeyStrategy(IncrementalStrategy):
    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        if is_first_insert:
            self._replace_query_for_model(model, table_name, query_or_df, render_kwargs, **kwargs)
        else:
            columns_to_types, source_columns = self._get_target_and_source_columns(
                model,
                table_name,
                render_kwargs=render_kwargs,
            )
            self.adapter.merge(
                table_name,
                query_or_df,
                target_columns_to_types=columns_to_types,
                unique_key=model.unique_key,
                when_matched=model.when_matched,
                merge_filter=model.render_merge_filter(
                    start=kwargs.get("start"),
                    end=kwargs.get("end"),
                    execution_time=kwargs.get("execution_time"),
                ),
                physical_properties=kwargs.get("physical_properties", model.physical_properties),
                source_columns=source_columns,
            )

    def append(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        columns_to_types, source_columns = self._get_target_and_source_columns(
            model, table_name, render_kwargs=render_kwargs
        )
        self.adapter.merge(
            table_name,
            query_or_df,
            target_columns_to_types=columns_to_types,
            unique_key=model.unique_key,
            when_matched=model.when_matched,
            merge_filter=model.render_merge_filter(
                start=kwargs.get("start"),
                end=kwargs.get("end"),
                execution_time=kwargs.get("execution_time"),
            ),
            physical_properties=kwargs.get("physical_properties", model.physical_properties),
            source_columns=source_columns,
        )


class IncrementalUnmanagedStrategy(IncrementalStrategy):
    def append(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        columns_to_types, source_columns = self._get_target_and_source_columns(
            model, table_name, render_kwargs=render_kwargs
        )
        self.adapter.insert_append(
            table_name,
            query_or_df,
            target_columns_to_types=columns_to_types,
            source_columns=source_columns,
        )

    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        if is_first_insert:
            return self._replace_query_for_model(
                model, table_name, query_or_df, render_kwargs, **kwargs
            )
        if isinstance(model.kind, IncrementalUnmanagedKind) and model.kind.insert_overwrite:
            columns_to_types, source_columns = self._get_target_and_source_columns(
                model,
                table_name,
                render_kwargs=render_kwargs,
            )

            return self.adapter.insert_overwrite_by_partition(
                table_name,
                query_or_df,
                model.partitioned_by,
                target_columns_to_types=columns_to_types,
                source_columns=source_columns,
            )
        return self.append(
            table_name,
            query_or_df,
            model,
            render_kwargs=render_kwargs,
            **kwargs,
        )


class FullRefreshStrategy(MaterializableStrategy):
    def append(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        self.adapter.insert_append(
            table_name,
            query_or_df,
            target_columns_to_types=model.columns_to_types,
        )

    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        self._replace_query_for_model(model, table_name, query_or_df, render_kwargs, **kwargs)


class SeedStrategy(MaterializableStrategy):
    def create(
        self,
        table_name: str,
        model: Model,
        is_table_deployable: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        model = t.cast(SeedModel, model)
        if not model.is_hydrated and self.adapter.table_exists(table_name):
            # This likely means that the table was created and populated previously, but the evaluation stage
            # failed before the interval could be added for this model.
            logger.warning(
                "Seed model '%s' is not hydrated, but the table '%s' exists. Skipping creation",
                model.name,
                table_name,
            )
            return

        super().create(table_name, model, is_table_deployable, render_kwargs, **kwargs)
        if is_table_deployable:
            # For seeds we insert data at the time of table creation.
            try:
                for index, df in enumerate(model.render_seed()):
                    if index == 0:
                        self._replace_query_for_model(
                            model, table_name, df, render_kwargs, **kwargs
                        )
                    else:
                        self.adapter.insert_append(
                            table_name, df, target_columns_to_types=model.columns_to_types
                        )
            except Exception:
                self.adapter.drop_table(table_name)
                raise

    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        # Data has already been inserted at the time of table creation.
        pass

    def append(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        # Data has already been inserted at the time of table creation.
        pass


class SCDType2Strategy(IncrementalStrategy):
    def create(
        self,
        table_name: str,
        model: Model,
        is_table_deployable: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        assert isinstance(model.kind, (SCDType2ByTimeKind, SCDType2ByColumnKind))
        if model.annotated:
            logger.info("Creating table '%s'", table_name)
            columns_to_types = model.columns_to_types_or_raise
            if isinstance(model.kind, SCDType2ByTimeKind):
                columns_to_types[model.kind.updated_at_name.name] = model.kind.time_data_type
            self.adapter.create_table(
                table_name,
                target_columns_to_types=columns_to_types,
                table_format=model.table_format,
                storage_format=model.storage_format,
                partitioned_by=model.partitioned_by,
                partition_interval_unit=model.partition_interval_unit,
                clustered_by=model.clustered_by,
                table_properties=kwargs.get("physical_properties", model.physical_properties),
                table_description=model.description if is_table_deployable else None,
                column_descriptions=model.column_descriptions if is_table_deployable else None,
            )
        else:
            # We assume that the data type for `updated_at_name` matches the data type that is defined for
            # `time_data_type`. If that isn't the case, then the user might get an error about not being able
            # to do comparisons across different data types
            super().create(
                table_name,
                model,
                is_table_deployable,
                render_kwargs,
                **kwargs,
            )

    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        # Source columns from the underlying table to prevent unintentional table schema changes during restatement of incremental models.
        columns_to_types, source_columns = self._get_target_and_source_columns(
            model,
            table_name,
            render_kwargs=render_kwargs,
            force_get_columns_from_target=True,
        )
        if isinstance(model.kind, SCDType2ByTimeKind):
            self.adapter.scd_type_2_by_time(
                target_table=table_name,
                source_table=query_or_df,
                unique_key=model.unique_key,
                valid_from_col=model.kind.valid_from_name,
                valid_to_col=model.kind.valid_to_name,
                execution_time=kwargs["execution_time"],
                updated_at_col=model.kind.updated_at_name,
                invalidate_hard_deletes=model.kind.invalidate_hard_deletes,
                updated_at_as_valid_from=model.kind.updated_at_as_valid_from,
                target_columns_to_types=columns_to_types,
                table_format=model.table_format,
                table_description=model.description,
                column_descriptions=model.column_descriptions,
                truncate=is_first_insert,
                source_columns=source_columns,
            )
        elif isinstance(model.kind, SCDType2ByColumnKind):
            self.adapter.scd_type_2_by_column(
                target_table=table_name,
                source_table=query_or_df,
                unique_key=model.unique_key,
                valid_from_col=model.kind.valid_from_name,
                valid_to_col=model.kind.valid_to_name,
                execution_time=model.kind.updated_at_name or kwargs["execution_time"],
                check_columns=model.kind.columns,
                invalidate_hard_deletes=model.kind.invalidate_hard_deletes,
                execution_time_as_valid_from=model.kind.execution_time_as_valid_from,
                target_columns_to_types=columns_to_types,
                table_format=model.table_format,
                table_description=model.description,
                column_descriptions=model.column_descriptions,
                truncate=is_first_insert,
                source_columns=source_columns,
            )
        else:
            raise SQLMeshError(
                f"Unexpected SCD Type 2 kind: {model.kind}. This is not expected and please report this as a bug."
            )

    def append(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        # Source columns from the underlying table to prevent unintentional table schema changes during restatement of incremental models.
        columns_to_types, source_columns = self._get_target_and_source_columns(
            model,
            table_name,
            render_kwargs=render_kwargs,
            force_get_columns_from_target=True,
        )
        if isinstance(model.kind, SCDType2ByTimeKind):
            self.adapter.scd_type_2_by_time(
                target_table=table_name,
                source_table=query_or_df,
                unique_key=model.unique_key,
                valid_from_col=model.kind.valid_from_name,
                valid_to_col=model.kind.valid_to_name,
                updated_at_col=model.kind.updated_at_name,
                invalidate_hard_deletes=model.kind.invalidate_hard_deletes,
                updated_at_as_valid_from=model.kind.updated_at_as_valid_from,
                target_columns_to_types=columns_to_types,
                table_format=model.table_format,
                table_description=model.description,
                column_descriptions=model.column_descriptions,
                source_columns=source_columns,
                **kwargs,
            )
        elif isinstance(model.kind, SCDType2ByColumnKind):
            self.adapter.scd_type_2_by_column(
                target_table=table_name,
                source_table=query_or_df,
                unique_key=model.unique_key,
                valid_from_col=model.kind.valid_from_name,
                valid_to_col=model.kind.valid_to_name,
                check_columns=model.kind.columns,
                target_columns_to_types=columns_to_types,
                table_format=model.table_format,
                invalidate_hard_deletes=model.kind.invalidate_hard_deletes,
                execution_time_as_valid_from=model.kind.execution_time_as_valid_from,
                table_description=model.description,
                column_descriptions=model.column_descriptions,
                source_columns=source_columns,
                **kwargs,
            )
        else:
            raise SQLMeshError(
                f"Unexpected SCD Type 2 kind: {model.kind}. This is not expected and please report this as a bug."
            )


class ViewStrategy(PromotableStrategy):
    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        snapshot = kwargs["snapshot"]

        if (
            not snapshot.is_materialized_view
            and self.adapter.HAS_VIEW_BINDING
            and self.adapter.table_exists(table_name)
        ):
            logger.info("Skipping creation of the view '%s'", table_name)
            return

        logger.info("Replacing view '%s'", table_name)
        self.adapter.create_view(
            table_name,
            query_or_df,
            model.columns_to_types,
            replace=not self.adapter.HAS_VIEW_BINDING,
            materialized=self._is_materialized_view(model),
            view_properties=kwargs.get("physical_properties", model.physical_properties),
            table_description=model.description,
            column_descriptions=model.column_descriptions,
        )

    def append(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        raise ConfigError(f"Cannot append to a view '{table_name}'.")

    def create(
        self,
        table_name: str,
        model: Model,
        is_table_deployable: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        if self.adapter.table_exists(table_name):
            # Make sure we don't recreate the view to prevent deletion of downstream views in engines with no late
            # binding support (because of DROP CASCADE).
            logger.info("View '%s' already exists", table_name)
            return

        logger.info("Creating view '%s'", table_name)
        materialized = self._is_materialized_view(model)
        materialized_properties = None
        if materialized:
            materialized_properties = {
                "partitioned_by": model.partitioned_by,
                "clustered_by": model.clustered_by,
                "partition_interval_unit": model.partition_interval_unit,
            }
        self.adapter.create_view(
            table_name,
            model.render_query_or_raise(**render_kwargs),
            # Make sure we never replace the view during creation to avoid race conditions in engines with no late binding support.
            replace=False,
            materialized=self._is_materialized_view(model),
            materialized_properties=materialized_properties,
            view_properties=kwargs.get("physical_properties", model.physical_properties),
            table_description=model.description if is_table_deployable else None,
            column_descriptions=model.column_descriptions if is_table_deployable else None,
        )

    def migrate(
        self,
        target_table_name: str,
        source_table_name: str,
        snapshot: Snapshot,
        *,
        ignore_destructive: bool,
        **kwargs: t.Any,
    ) -> None:
        logger.info("Migrating view '%s'", target_table_name)
        model = snapshot.model
        render_kwargs = dict(
            execution_time=now(), snapshots=kwargs["snapshots"], engine_adapter=self.adapter
        )

        self.adapter.create_view(
            target_table_name,
            model.render_query_or_raise(**render_kwargs),
            model.columns_to_types,
            materialized=self._is_materialized_view(model),
            view_properties=model.render_physical_properties(**render_kwargs),
            table_description=model.description,
            column_descriptions=model.column_descriptions,
        )

    def delete(self, name: str, **kwargs: t.Any) -> None:
        cascade = kwargs.pop("cascade", False)
        try:
            self.adapter.drop_view(name, cascade=cascade)
        except Exception:
            logger.debug(
                "Failed to drop view '%s'. Trying to drop the materialized view instead",
                name,
                exc_info=True,
            )
            self.adapter.drop_view(name, materialized=True, cascade=cascade)
        logger.info("Dropped view '%s'", name)

    def _is_materialized_view(self, model: Model) -> bool:
        return isinstance(model.kind, ViewKind) and model.kind.materialized


C = t.TypeVar("C", bound=CustomKind)


class CustomMaterialization(IncrementalStrategy, t.Generic[C]):
    """Base class for custom materializations."""

    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        """Inserts the given query or a DataFrame into the target table or a view.

        Args:
            table_name: The name of the target table or view.
            query_or_df: A query or a DataFrame to insert.
            model: The target model.
            is_first_insert: Whether this is the first insert for this version of a model. This value is set to True
                if no data has been previously inserted into the target table, or when the entire history of the target model has
                been restated. Note that in the latter case, the table might contain data from previous executions, and it is the
                responsibility of a specific evaluation strategy to handle the truncation of the table if necessary.
            render_kwargs: Additional key-value arguments to pass when rendering the model's query.
        """
        raise NotImplementedError(
            "Custom materialization strategies must implement the 'insert' method."
        )


_custom_materialization_type_cache: t.Optional[
    t.Dict[str, t.Tuple[t.Type[CustomKind], t.Type[CustomMaterialization]]]
] = None


def get_custom_materialization_kind_type(st: t.Type[CustomMaterialization]) -> t.Type[CustomKind]:
    # try to read if there is a custom 'kind' type in use by inspecting the type signature
    # eg try to read 'MyCustomKind' from:
    # >>>> class MyCustomMaterialization(CustomMaterialization[MyCustomKind])
    # and fall back to base CustomKind if there is no generic type declared
    if hasattr(st, "__orig_bases__"):
        for base in st.__orig_bases__:
            if hasattr(base, "__origin__") and base.__origin__ == CustomMaterialization:
                for generic_arg in t.get_args(base):
                    if not issubclass(generic_arg, CustomKind):
                        raise SQLMeshError(
                            f"Custom materialization kind '{generic_arg.__name__}' must be a subclass of CustomKind"
                        )

                    return generic_arg

    return CustomKind


def get_custom_materialization_type(
    name: str, raise_errors: bool = True
) -> t.Optional[t.Tuple[t.Type[CustomKind], t.Type[CustomMaterialization]]]:
    global _custom_materialization_type_cache

    strategy_key = name.lower()

    try:
        if (
            _custom_materialization_type_cache is None
            or strategy_key not in _custom_materialization_type_cache
        ):
            strategy_types = list(CustomMaterialization.__subclasses__())

            entry_points = metadata.entry_points(group="sqlmesh.materializations")
            for entry_point in entry_points:
                strategy_type = entry_point.load()
                if not issubclass(strategy_type, CustomMaterialization):
                    raise SQLMeshError(
                        f"Custom materialization entry point '{entry_point.name}' must be a subclass of CustomMaterialization."
                    )
                strategy_types.append(strategy_type)

            _custom_materialization_type_cache = {
                getattr(strategy_type, "NAME", strategy_type.__name__).lower(): (
                    get_custom_materialization_kind_type(strategy_type),
                    strategy_type,
                )
                for strategy_type in strategy_types
            }

        if strategy_key not in _custom_materialization_type_cache:
            raise ConfigError(f"Materialization strategy with name '{name}' was not found.")
    except (SQLMeshError, ConfigError) as e:
        if raise_errors:
            raise e

        from sqlmesh.core.console import get_console

        get_console().log_warning(str(e))
        return None

    strategy_kind_type, strategy_type = _custom_materialization_type_cache[strategy_key]
    logger.debug(
        "Resolved custom materialization '%s' to '%s' (%s)", name, strategy_type, strategy_kind_type
    )

    return strategy_kind_type, strategy_type


def get_custom_materialization_type_or_raise(
    name: str,
) -> t.Tuple[t.Type[CustomKind], t.Type[CustomMaterialization]]:
    types = get_custom_materialization_type(name, raise_errors=True)
    if types is not None:
        return types[0], types[1]

    # Shouldnt get here as get_custom_materialization_type() has raise_errors=True, but just in case...
    raise SQLMeshError(f"Custom materialization '{name}' not present in the Python environment")


class EngineManagedStrategy(MaterializableStrategy):
    def create(
        self,
        table_name: str,
        model: Model,
        is_table_deployable: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        is_snapshot_deployable: bool = kwargs["is_snapshot_deployable"]

        if is_table_deployable and is_snapshot_deployable:
            # We could deploy this to prod; create a proper managed table
            logger.info("Creating managed table: %s", table_name)
            self.adapter.create_managed_table(
                table_name=table_name,
                query=model.render_query_or_raise(**render_kwargs),
                target_columns_to_types=model.columns_to_types,
                partitioned_by=model.partitioned_by,
                clustered_by=model.clustered_by,
                table_properties=kwargs.get("physical_properties", model.physical_properties),
                table_description=model.description,
                column_descriptions=model.column_descriptions,
                table_format=model.table_format,
            )
        elif not is_table_deployable:
            # Only create the dev preview table as a normal table.
            # For the main table, if the snapshot is cant be deployed to prod (eg upstream is forward-only) do nothing.
            # Any downstream models that reference it will be updated to point to the dev preview table.
            # If the user eventually tries to deploy it, the logic in insert() will see it doesnt exist and create it
            super().create(
                table_name=table_name,
                model=model,
                is_table_deployable=is_table_deployable,
                render_kwargs=render_kwargs,
                **kwargs,
            )

    def insert(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        is_first_insert: bool,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        deployability_index: DeployabilityIndex = kwargs["deployability_index"]
        snapshot: Snapshot = kwargs["snapshot"]
        is_snapshot_deployable = deployability_index.is_deployable(snapshot)

        if is_first_insert and is_snapshot_deployable and not self.adapter.table_exists(table_name):
            self.adapter.create_managed_table(
                table_name=table_name,
                query=query_or_df,  # type: ignore
                target_columns_to_types=model.columns_to_types,
                partitioned_by=model.partitioned_by,
                clustered_by=model.clustered_by,
                table_properties=kwargs.get("physical_properties", model.physical_properties),
                table_description=model.description,
                column_descriptions=model.column_descriptions,
                table_format=model.table_format,
            )
        elif not is_snapshot_deployable:
            # Snapshot isnt deployable; update the preview table instead
            # If the snapshot was deployable, then data would have already been loaded in create() because a managed table would have been created
            logger.info(
                "Updating preview table: %s (for managed model: %s)",
                table_name,
                model.name,
            )
            self._replace_query_for_model(
                model=model,
                name=table_name,
                query_or_df=query_or_df,
                render_kwargs=render_kwargs,
                **kwargs,
            )

    def append(
        self,
        table_name: str,
        query_or_df: QueryOrDF,
        model: Model,
        render_kwargs: t.Dict[str, t.Any],
        **kwargs: t.Any,
    ) -> None:
        raise ConfigError(f"Cannot append to a managed table '{table_name}'.")

    def migrate(
        self,
        target_table_name: str,
        source_table_name: str,
        snapshot: Snapshot,
        *,
        ignore_destructive: bool,
        **kwargs: t.Any,
    ) -> None:
        potential_alter_expressions = self.adapter.get_alter_expressions(
            target_table_name, source_table_name, ignore_destructive=ignore_destructive
        )
        if len(potential_alter_expressions) > 0:
            # this can happen if a user changes a managed model and deliberately overrides a plan to be forward only, eg `sqlmesh plan --forward-only`
            raise SQLMeshError(
                f"The schema of the managed model '{target_table_name}' cannot be updated in a forward-only fashion."
            )

    def delete(self, name: str, **kwargs: t.Any) -> None:
        # a dev preview table is created as a normal table, so it needs to be dropped as a normal table
        _check_table_db_is_physical_schema(name, kwargs["physical_schema"])
        if kwargs["is_table_deployable"]:
            self.adapter.drop_managed_table(name)
            logger.info("Dropped managed table '%s'", name)
        else:
            self.adapter.drop_table(name)
            logger.info("Dropped dev preview for managed table '%s'", name)


def _intervals(snapshot: Snapshot, deployability_index: DeployabilityIndex) -> Intervals:
    return (
        snapshot.intervals
        if deployability_index.is_deployable(snapshot)
        else snapshot.dev_intervals
    )


def _check_destructive_schema_change(
    snapshot: Snapshot,
    alter_expressions: t.List[exp.Alter],
    allow_destructive_snapshots: t.Set[str],
) -> None:
    if (
        snapshot.is_no_rebuild
        and snapshot.needs_destructive_check(allow_destructive_snapshots)
        and has_drop_alteration(alter_expressions)
    ):
        snapshot_name = snapshot.name
        dropped_column_names = get_dropped_column_names(alter_expressions)
        model_dialect = snapshot.model.dialect

        if snapshot.model.on_destructive_change.is_warn:
            logger.warning(
                format_destructive_change_msg(
                    snapshot_name,
                    dropped_column_names,
                    alter_expressions,
                    model_dialect,
                    error=False,
                )
            )
            return
        raise DestructiveChangeError(
            format_destructive_change_msg(
                snapshot_name, dropped_column_names, alter_expressions, model_dialect
            )
        )


def _check_table_db_is_physical_schema(table_name: str, physical_schema: str) -> None:
    table = exp.to_table(table_name)
    if table.db != physical_schema:
        raise SQLMeshError(
            f"Table '{table_name}' is not a part of the physical schema '{physical_schema}' and so can't be dropped."
        )


def _snapshot_to_data_object_type(snapshot: Snapshot) -> DataObjectType:
    if snapshot.is_managed:
        return DataObjectType.MANAGED_TABLE
    if snapshot.is_materialized_view:
        return DataObjectType.MATERIALIZED_VIEW
    if snapshot.is_view:
        return DataObjectType.VIEW
    if snapshot.is_materialized:
        return DataObjectType.TABLE
    return DataObjectType.UNKNOWN