mgf-sqlalchemy 0.1.0

Async SQLAlchemy helpers for mgf-common consumers — typed engine factory, sessionmaker, Postgres RLS tenant-scoping. Sibling of mgf-common under the mgf.* namespace.

mgf-sqlalchemy — async SQLAlchemy helpers for mgf-common consumers

Sibling of mgf-common under the mgf.* namespace. Houses the async-SQLAlchemy helpers that previously lived under mgf.common.db.* — extracted at mgf-common v0.30 / mgf-sqlalchemy v0.1 per the federation split plan.

What this provides

Submodule

What

mgf.sqlalchemy

create_engine — async SQLAlchemy engine factory with production-leaning pool defaults (pool_pre_ping, pool_recycle, sane pool_size/max_overflow). SQLite-aware (skips pool kwargs that StaticPool/NullPool reject). create_sessionmaker — async sessionmaker factory with SQLAlchemy 2's recommended expire_on_commit=False default. tenant_session — context manager for Postgres RLS multi-tenancy (SET LOCAL app.current_tenant = '<uuid>'); UUID-validated to prevent SQL injection.

FastAPI helpers live elsewhere. The get_session FastAPI-Depends generator and the setup_db lifespan helper that previously co-located with these in mgf-common moved to mgf.fastapi.db in mgf-fastapi v0.2.0 (which depends on mgf-sqlalchemy via its [sqlalchemy] extra). This sibling stays framework-agnostic — no Starlette / no FastAPI in the dependency graph.

Install

pip install mgf-sqlalchemy
# Or with the test extra (aiosqlite for in-memory SQLite tests):
pip install 'mgf-sqlalchemy[test]'

Pulls in mgf-common + sqlalchemy[asyncio] automatically. Production consumers also need an async driver of their own (asyncpg / aiomysql / asyncmy); we don't pin one — the consumer picks based on their database.

Quick start

import asyncio
from mgf.sqlalchemy import create_engine, create_sessionmaker

async def main() -> None:
    engine = create_engine(
        "postgresql+asyncpg://user:pass@localhost/myapp",
        echo=False,
    )
    sessionmaker = create_sessionmaker(engine)
    try:
        async with sessionmaker() as session:
            from sqlalchemy import text
            row = (await session.execute(text("SELECT 1"))).scalar_one()
            print(row)
    finally:
        await engine.dispose()

asyncio.run(main())

Postgres RLS multi-tenancy

from uuid import UUID
from mgf.sqlalchemy import tenant_session

tenant_id = UUID("550e8400-e29b-41d4-a716-446655440000")

async with sessionmaker() as session:
    async with tenant_session(session, tenant_id) as scoped:
        # Every query on `scoped` (same session, just tenant-scoped)
        # gets `app.current_tenant` set in the current transaction.
        # RLS policies in your schema can read from
        # `current_setting('app.current_tenant')`.
        rows = await scoped.execute(text("SELECT ..."))

FastAPI integration

The FastAPI-shaped helpers (request-scoped session injection + lifespan) live in mgf-fastapi:

# pyproject.toml
dependencies = [
    "mgf-common>=0.30,<0.31",
    "mgf-sqlalchemy>=0.1,<0.2",
    "mgf-fastapi[sqlalchemy]>=0.2,<0.3",
]

from typing import Annotated
from contextlib import asynccontextmanager
from fastapi import FastAPI, Depends
from sqlalchemy.ext.asyncio import AsyncSession
from mgf.fastapi.db import get_session, setup_db

@asynccontextmanager
async def lifespan(app: FastAPI):
    async with setup_db(app, database_url="postgresql+asyncpg://..."):
        yield

app = FastAPI(lifespan=lifespan)

@app.get("/users")
async def list_users(
    session: Annotated[AsyncSession, Depends(get_session)],
) -> list[dict]:
    ...

Documentation

• docs/recipes/sqlalchemy.md — full async-SQLAlchemy walkthrough.
• docs/cutover/v0.1.0.md — maiden voyage migration story (the v0.30 split + the get_session/setup_db relocation to mgf-fastapi).
• PUBLIC_API.md — full public surface contract.
• CHANGELOG.md — release history.

For the federation-wide engineering standards (DESIGN_PRINCIPLES, ERROR_HANDLING, SECURITY, etc.) see mgf-common/docs/standards/. This sibling inherits them by reference; the standards source-of-truth lives in mgf-common.

Status

🚧 Experimental — every public name is experimental per AP-09. Promotion to stable happens release-by-release as consumer feedback in mgf-common/FEEDBACK.md converges. The 0.x window applies. Pin tightly: mgf-sqlalchemy = ">=0.X.0,<0.Y".

Cross-references

• Filing process for sharp edges: open an entry on mgf-common/FEEDBACK.md with [mgf-sqlalchemy] prefix, OR file directly on this repo's Issues → maintainer mirrors into the canonical FEEDBACK.md.
• Federation pattern: mgf-common/docs/design/federation.md.
• The split that created this sibling: mgf-common/docs/release/federation_roadmap.md.
• Companion sibling: mgf-alembic — async-aware alembic env.py helper (paired ship at v0.30; depends on this sibling).