Code after yield in the async dependecy executed before response has been sent to the client

First Check

• I added a very descriptive title to this issue.
• I used the GitHub search to find a similar issue and didn’t find it.
• I searched the FastAPI documentation, with the integrated search.
• I already searched in Google “How to X in FastAPI” and didn’t find any information.
• I already read and followed all the tutorial in the docs and didn’t find an answer.
• I already checked if it is not related to FastAPI but to Pydantic.
• I already checked if it is not related to FastAPI but to Swagger UI.
• I already checked if it is not related to FastAPI but to ReDoc.

Commit to Help

• I commit to help with one of those options 👆

Example Code

import asyncio
from typing import AsyncGenerator

import structlog
import uvicorn
from fastapi import FastAPI, Depends
from starlette.middleware.base import BaseHTTPMiddleware, RequestResponseEndpoint
from starlette.requests import Request
from starlette.responses import Response

logger = structlog.getLogger(__name__)


app = FastAPI()


class BarMiddleware(BaseHTTPMiddleware):
    async def dispatch(self, request: Request, call_next: RequestResponseEndpoint) -> Response:
        logger.info("BAR: Before call_next()")
        response = await call_next(request)
        logger.info("BAR: After call_next()")
        logger.info(f"BAR: Status code {response.status_code}")

        logger.info("BAR: Before sleep()")
        await asyncio.sleep(2)
        logger.info("BAR: After sleep()")

        return response


class ZxcMiddleware(BaseHTTPMiddleware):
    async def dispatch(self, request: Request, call_next: RequestResponseEndpoint) -> Response:
        logger.info("ZXC: Before call_next()")
        response = await call_next(request)
        logger.info("ZXC: After call_next()")
        logger.info(f"ZXC: Status code {response.status_code}")

        logger.info("ZXC: Before sleep()")
        await asyncio.sleep(2)
        logger.info("ZXC: After sleep()")

        return response


async def get_foo() -> AsyncGenerator[str, None]:
    logger.info("BEFORE YIELD FOO")
    yield "foo"
    logger.info("AFTER YIELD FOO")


async def get_qwerty(foo: str = Depends(get_foo)) -> AsyncGenerator[str, None]:
    logger.info("BEFORE YIELD QWERTY")
    yield "qwerty"
    logger.info("AFTER YIELD QWERTY")


@app.get("/")
async def root(qwerty: str = Depends(get_qwerty)):
    return {"message": qwerty}


app.add_middleware(BarMiddleware)
app.add_middleware(ZxcMiddleware)


if __name__ == "__main__":
    uvicorn.run(
        "test:app",
        host="0.0.0.0",
        port=8080,
        reload=True,
    )

Description

Hi.

From the documentation:

The code following the yield statement is executed after the response has been delivered:

But this is not the case:

2022-06-23 09:21:29,932 uvicorn.error INFO: Finished server process [30197]
2022-06-23 09:21:30,309 uvicorn.error INFO: Started server process [30216]
2022-06-23 09:21:30,309 uvicorn.error INFO: Waiting for application startup.
2022-06-23 09:21:30,309 uvicorn.error INFO: Application startup complete.
2022-06-23 09:22.12 [info     ] ZXC: Before call_next()
2022-06-23 09:22.12 [info     ] BAR: Before call_next()
2022-06-23 09:22.12 [info     ] BEFORE YIELD FOO
2022-06-23 09:22.12 [info     ] BEFORE YIELD QWERTY
2022-06-23 09:22.12 [info     ] BAR: After call_next()
2022-06-23 09:22.12 [info     ] BAR: Status code 200
2022-06-23 09:22.12 [info     ] BAR: Before sleep()
2022-06-23 09:22.14 [info     ] BAR: After sleep()
2022-06-23 09:22.14 [info     ] ZXC: After call_next()
2022-06-23 09:22.14 [info     ] ZXC: Status code 200
2022-06-23 09:22.14 [info     ] ZXC: Before sleep()
2022-06-23 09:22.14 [info     ] AFTER YIELD QWERTY
2022-06-23 09:22.14 [info     ] AFTER YIELD FOO
2022-06-23 09:22.16 [info     ] ZXC: After sleep()
2022-06-23 09:22:16,494 uvicorn.access INFO: 127.0.0.1:55382 - "GET / HTTP/1.1" 200

Response has not been sent to the client yet, but we’re already in the “after yield” code. I can not reproduce it with a single middleware though.

Using native ASGI middlewares (instead of inheriting from BaseHTTPMiddleware) does not solve the issue.

It’s a problem for us, because we can not rely on the fact that transaction has been commited already (in the middleware), when we’re in the “after yield” code.

Either docs are misleading, or something works incorrectly.

Operating System

Linux

Operating System Details

No response

FastAPI Version

0.78.0

Python Version

3.8.10

Additional Context

No response

[JarroVGIT]

BaseHTTPMiddleware will be deprecated (see this open issue on the Starlette project: Issue). So having the starting point where you see this behaviour with native ASGI middleware would help me reproduce this (and hopefully, help you fix it 😉)

That’s not for sure yet… There has been a lot of discussion about BaseHTTPMiddleware. Also, the limitations of BaseHTTPMiddleware are usually not a problem for most of the users.

That being said… They are still limitations, and if you are going to have them, for the time being, pure ASGI middlewares are recommended.

Fair point, I corrected my comment above 😇

[Kludex]

BaseHTTPMiddleware will be deprecated (see this open issue on the Starlette project: Issue). So having the starting point where you see this behaviour with native ASGI middleware would help me reproduce this (and hopefully, help you fix it 😉)

That’s not for sure yet… There has been a lot of discussion about BaseHTTPMiddleware. Also, the limitations of BaseHTTPMiddleware are usually not a problem for most of the users.

That being said… They are still limitations, and if you are going to have them, for the time being, pure ASGI middlewares are recommended.