feat(feature-parity) LF <> LFX Services Layer Tiering & Separation

.gitignore

@@ -192,6 +192,10 @@ instance/
 # Scrapy stuff:
 .scrapy
 
+# AI Scratch work
+/scratch
+scratch/
+
 # Sphinx documentation
 docs/_build/
 

AGENTS.md

@@ -89,7 +89,7 @@ src/
 
 ### ORM Model Ownership
 
-All `table=True` SQLModel classes live in `lfx.services.database.models` (flat modules per domain: `flow`, `user`, `folder`, `api_key`, `variable`, `file`, `jobs`, `traces`, `message`, `transactions`, `vertex_builds`, etc.; `auth/` and `deployment_provider_account/` are subpackages). The old `langflow.services.database.models.*` paths remain as re-export shims (class identity preserved — both import paths yield the same class objects, so `SQLModel.metadata` registers each table once). CRUD modules, alembic migrations, and the DB service stay in `langflow-base`; `lfx` owns only the model definitions. Shared model helpers live in `lfx.services.database.utils` (string validators) and `lfx.services.database.models.variable` (the variable type constants); the langflow twins re-export them. <!-- pragma: allowlist secret --> `lfx` depends on `sqlmodel` for these models; import them lazily on lfx hot paths. Known seam wart: `lfx.services.database.models.memory_base` lazily imports `langflow.services.memory_base.preprocessing` inside one default-factory path (embedded-only).
+All `table=True` SQLModel classes live in `lfx.services.database.models` (flat modules per domain: `flow`, `user`, `folder`, `api_key`, `variable`, `file`, `jobs`, `traces`, `message`, `transactions`, `vertex_builds`, etc.; `auth/` and `deployment_provider_account/` are subpackages). The old `langflow.services.database.models.*` paths remain as re-export shims (class identity preserved — both import paths yield the same class objects, so `SQLModel.metadata` registers each table once). The Tier 1 **DatabaseService** (engine/session/migration mechanism) now lives in `lfx.services.database.service`; `langflow.services.database.service.DatabaseService` is a thin subclass that overrides only its migration-stream identity (version table `alembic_version`, `script_location` → `langflow/alembic`) and adds domain bootstrap (superuser assignment, schema-health). `lfx` also owns its **own** alembic lineage (`lfx.services.database.migrations`, version table `lfx_alembic_version`, execution tables only) so bare `lfx serve` can provision its schema without langflow; drive it with `lfx db upgrade/check/...`. langflow keeps its 82-migration alembic lineage and the CRUD modules. The model classes are the single source of truth, so the two streams can only differ in staleness — caught by `alembic check` per stream. See `src/lfx/PLUGGABLE_SERVICES.md` (two-stream migration model) and `src/lfx/deploy/` (hydration). `lfx` owns the model definitions. Shared model helpers live in `lfx.services.database.utils` (string validators) and `lfx.services.database.models.variable` (the variable type constants); the langflow twins re-export them. <!-- pragma: allowlist secret --> `lfx` depends on `sqlmodel` for these models; import them lazily on lfx hot paths. Known seam wart: `lfx.services.database.models.memory_base` lazily imports `langflow.services.memory_base.preprocessing` inside one default-factory path (embedded-only).
 
 ### Service Layer
 Backend services in `src/backend/base/langflow/services/`:

pyproject.toml

@@ -301,6 +301,7 @@ directory = "coverage"
 target-version = "py310"
 exclude = [
     "src/backend/base/langflow/alembic/*",
+    "src/lfx/src/lfx/services/database/migrations/*",
     "src/frontend/tests/assets/*",
     "src/lfx/src/lfx/_assets/component_index.json",
     "docs/**/API-Reference/python-examples/**",

src/backend/base/langflow/helpers/windows_postgres_helper.py

@@ -1,45 +1,18 @@
-"""Helper for Windows + PostgreSQL event loop configuration."""
-
-import asyncio
-import os
-import platform
-
-from lfx.log.logger import logger
-
-LANGFLOW_DATABASE_URL = "LANGFLOW_DATABASE_URL"
-POSTGRESQL_PREFIXES = ("postgresql", "postgres")
-
-
-def configure_windows_postgres_event_loop(source: str | None = None) -> bool:
-    """Configure event loop for Windows + PostgreSQL compatibility.
-
-    Args:
-        source: Optional identifier for logging context
-
-    Returns:
-        True if configuration was applied, False otherwise
-    """
-    if platform.system() != "Windows":
-        return False
-
-    db_url = os.environ.get(LANGFLOW_DATABASE_URL, "")
-    if not db_url or not any(db_url.startswith(prefix) for prefix in POSTGRESQL_PREFIXES):
-        return False
-
-    # Use getattr to safely access the Windows-only class on all platforms
-    selector_policy = getattr(asyncio, "WindowsSelectorEventLoopPolicy", None)
-    if selector_policy is None:
-        return False
-
-    current_policy = asyncio.get_event_loop_policy()
-    if isinstance(current_policy, selector_policy):
-        return False
-
-    asyncio.set_event_loop_policy(selector_policy())
-
-    log_context = {"event_loop": "WindowsSelectorEventLoop", "reason": "psycopg_compatibility"}
-    if source:
-        log_context["source"] = source
-
-    logger.debug("Windows PostgreSQL event loop configured", extra=log_context)
-    return True
+"""Helper for Windows + PostgreSQL event loop configuration.
+
+The implementation moved to ``lfx.utils.windows_postgres_helper`` when the Tier 1
+DatabaseService was extracted into lfx. This shim preserves the historical
+``langflow.helpers.windows_postgres_helper`` import path.
+"""
+
+from lfx.utils.windows_postgres_helper import (
+    LANGFLOW_DATABASE_URL,
+    POSTGRESQL_PREFIXES,
+    configure_windows_postgres_event_loop,
+)
+
+__all__ = [
+    "LANGFLOW_DATABASE_URL",
+    "POSTGRESQL_PREFIXES",
+    "configure_windows_postgres_event_loop",
+]