Does graphene_sqlalchemy keep sessions open?

まず、使い捨ての Postgres コンテナに対して reset_testdata()、close_all_sessions()、metadata.drop_all()/create_all() を使い、pytest のハングを再現します。示されている FastAPI ミドルウェアありの場合となしの場合で、graphene_sqlalchemy を使ったリクエストを比較し、その後セッションのライフサイクルを追跡します。セッションが残り続ける原因を特定し、期待されるクリーンアップ動作を文書化するか、リグレッションテストでカバーできれば完了です。

tl;dr

Does graphene_sqlalchemy keep sessions open? I'm having the problem that Postgres doesnt run drop_all() in my test suite.

Explanation

Hi, I have a fastAPI app and I'm using graphene_sqlalchemy to generate an API from database models defined on the declarative base of SQLAlchemy. The database is postgres. For testing I am running some pytests against the API while it's running. So, I start a throw-away postgres container, then I start the API app, then I run pytest.

The tests include a reset_testdata() function which also uses the SQLAlchemy ORM. It looks like:

def reset_testdata():
    close_all_sessions()  # from sqlalchemy.orm.session
    models.Base.metadata.drop_all(bind=engine)
    models.Base.metadata.create_all(bind=engine)

    db.add_all([
        # ... add stuff
    ])

    db.commit()
    db.close()

I noticed, that once reset_testdata() is used, the pytest process hangs. No errors, not able to Ctrl+C, it just waits. I had this issue before and it usually stems from Postgres not allowing stuff like drop_all() if there are still active sessions. After some trying out I found that queries from my app (which are implemented by graphene_sqlalchemy) seem to keep a session open.
(I used this example in my app: https://docs.graphene-python.org/projects/sqlalchemy/en/latest/tutorial/#defining-our-models)

I tested this, and basically I can circumvent this problem by adding a middleware that closes all sessions after every request.

@app.middleware('http')
async def close_sessions(request: Request, call_next):
    try:
        response = await call_next(request)
    finally:
        close_all_sessions()
    return response

I wonder whether I am missing something here? Any experience with this issue (if it is even an issue)?