Error in shutting down server with Starlette app immediately after starting up

[kichanyurd]

When running the following code (pytest):

import asyncio
from contextlib import asynccontextmanager
from random import random
from typing import AsyncIterator
from fastapi import FastAPI
from fastapi.responses import JSONResponse
import uvicorn


@asynccontextmanager
async def run_server(app: FastAPI) -> AsyncIterator[None]:
    config = uvicorn.Config(app=app, port=8089)
    server = uvicorn.Server(config)
    task = asyncio.create_task(server.serve())
    yield
    server.should_exit = True
    await task


def rng_app() -> FastAPI:
    app = FastAPI()

    @app.get("/random_number")
    async def get_random_number(max: int) -> JSONResponse:
        return JSONResponse({"number": int(random() * max)})

    return app


async def test_this() -> None:
    async with run_server(rng_app()):
        pass

I'm getting the following error:

Task was destroyed but it is pending!
task: <Task pending name='Task-3' coro=<LifespanOn.main() running at /workspaces/emcie/server/.venv/lib/python3.12/site-packages/uvicorn/lifespan/on.py:86> wait_for=<Future pending cb=[Task.task_wakeup()]>>
ERROR:    Traceback (most recent call last):
  File "/home/vscode/.pyenv/versions/3.12.4/lib/python3.12/asyncio/queues.py", line 158, in get
    await getter
GeneratorExit

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/workspaces/emcie/server/.venv/lib/python3.12/site-packages/starlette/routing.py", line 743, in lifespan
    await receive()
  File "/workspaces/emcie/server/.venv/lib/python3.12/site-packages/uvicorn/lifespan/on.py", line 137, in receive
    return await self.receive_queue.get()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/vscode/.pyenv/versions/3.12.4/lib/python3.12/asyncio/queues.py", line 160, in get
    getter.cancel()  # Just in case getter is not done yet.
    ^^^^^^^^^^^^^^^
  File "/home/vscode/.pyenv/versions/3.12.4/lib/python3.12/asyncio/base_events.py", line 795, in call_soon
    self._check_closed()
  File "/home/vscode/.pyenv/versions/3.12.4/lib/python3.12/asyncio/base_events.py", line 541, in _check_closed
    raise RuntimeError('Event loop is closed')
RuntimeError: Event loop is closed

Upon inspection I found that the problem lies in server.py at _serve():

    async def _serve(self, sockets: list[socket.socket] | None = None) -> None:
        process_id = os.getpid()

        config = self.config
        if not config.loaded:
            config.load()

        self.lifespan = config.lifespan_class(config)

        message = "Started server process [%d]"
        color_message = "Started server process [" + click.style("%d", fg="cyan") + "]"
        logger.info(message, process_id, extra={"color_message": color_message})

        await self.startup(sockets=sockets)
        if self.should_exit:
            return  # <---- HERE'S THE PROBLEM
        await self.main_loop()
        await self.shutdown(sockets=sockets)

        message = "Finished server process [%d]"
        color_message = "Finished server process [" + click.style("%d", fg="cyan") + "]"
        logger.info(message, process_id, extra={"color_message": color_message})

Specifically, the lifespan class is not shut down properly in this case, which would be triggered in self.shutdown().
Looking at the implementation, I tried a fix, which works, and which I suspect should be there anyhow.

This is how I resolved the issue I was having:

I changed this:

await self.startup(sockets=sockets)
if self.should_exit:
    return  # <---- HERE'S THE PROBLEM
await self.main_loop()
await self.shutdown(sockets=sockets)

to this:

await self.startup(sockets=sockets)
try:
    if self.should_exit:
        return
    await self.main_loop()
finally:
    await self.shutdown(sockets=sockets)

Thoughts?
P.S. I have a PR with this fix at #2433

[kichanyurd]

@Kludex Please let me know your thoughts on this and #2433 (comment) .
It's causing false errors in our test suite so we would really like to have this fixed without a personal fork to maintain.