claude-memory/graph/solutions/pytest-asyncio-sqlalchemy-async-db-test-fixture-pattern-fecad3.md

id	type	title	tags	importance	confidence	created	updated
fecad34a-4cc1-4f00-878c-393a9ecb5df0	solution	pytest-asyncio SQLAlchemy async DB test fixture pattern	mantimon-tcg pytest pytest-asyncio sqlalchemy async testing pattern	0.8	0.8	2026-01-27T15:50:36.216118+00:00	2026-01-27T15:50:36.216118+00:00

When testing SQLAlchemy async with pytest-asyncio, the fixture teardown runs in a DIFFERENT event loop than the test body, causing 'Future attached to different loop' errors on async cleanup.

SOLUTION:

1. Use sync psycopg2 for fixture setup/teardown operations (migrations, truncate)
2. Create fresh NullPool engine per test (no connection reuse)
3. Use TRUNCATE via sync psycopg2 in teardown instead of async rollback
4. Suppress warnings for connection cleanup (harmless GC noise)

KEY CODE (conftest.py):

def truncate_all_tables():
    conn = psycopg2.connect(**DB_PARAMS)
    try:
        conn.autocommit = True
        with conn.cursor() as cur:
            for table in TABLES_TO_TRUNCATE:
                cur.execute(f'TRUNCATE TABLE {table} CASCADE')
    finally:
        conn.close()

@pytest_asyncio.fixture
async def db_session():
    engine = create_async_engine(URL, poolclass=pool.NullPool)
    session_factory = async_sessionmaker(engine, expire_on_commit=False)
    session = session_factory()
    try:
        yield session
    finally:
        try:
            await session.close()
        except RuntimeError:
            pass  # Ignore event loop errors
        truncate_all_tables()  # SYNC cleanup always works

ADDITIONAL NOTES:

• When testing relationship loading (selectinload), must expire() the parent object first if children were added after initial load
• ON DELETE SET NULL requires passive_deletes=True on relationship to let DB handle it
• Suppress warnings in pyproject.toml: 'ignore:The garbage collector is trying to clean up:sqlalchemy.exc.SAWarning'