Task was destroyed but it is pending error

[jeffthompsonsd]

Hello, using the following converted fastapi app, serving the resulting wsgi app with gunicorn and then hitting the route with the vegeta load testing tool im seeing the error after about a minute of requests.

import logging

from fastapi import FastAPI, APIRouter
from a2wsgi import ASGIMiddleware

logging.basicConfig(
    filename="app.log",
    level=logging.INFO,
    format="%(asctime)s | [%(thread)d](%(process)d) >>> %(message)s",
    datefmt="%H:%M:%S",
)
logger = logging.getLogger(__name__)

hello = APIRouter()

@hello.get("/hello")
def get_hello():
    return "world"

def create_app():
    app = FastAPI()
    app.include_router(hello)
    return app

def create_wsgi_app():
    return ASGIMiddleware(create_app())

Error starts dumping in the app.log about a minute into the requests

Task was destroyed but it is pending!
task: <Task pending name='Task-4' coro=<FastAPI.__call__() running at ../venv/lib64/python3.11/site-packages/fastapi/applications.py:1055> wait_for=<Future pending cb=[Task.task_wakeup()]> cb=[ASGIResponder.asgi_done_callback()]>

Launching the wsgi app with 10 gunicorn workers

.venv/bin/gunicorn \
        --bind 0.0.0.0:8000 \
        --access-logformat '%(t)s | (%(p)s) %(s)s >>> %(r)s' \
        --log-level 'debug' \
        --error-logfile 'error.log' \
        --access-logfile 'access.log' \
        --max-requests 3 \
        --max-requests-jitter 2 \
        --workers 10 \
        'app:create_wsgi_app()' \
        > gunicorn.log 2> gunicorn_error.log

Hitting the route with the vegeta testing tool

echo "GET http://localhost:8000/hello" | vegeta attack --duration=120s --rate=50 | tee results.bin | vegeta report

The error doesnt happen with an asgi app (uvicorn alone or gunicorn as a process manager with uvicorn workers). Any help is appreciated.

[abersheeran]

This seems to be a race condition in Task state. Seems to be related to https://github.com/abersheeran/a2wsgi/issues/53

[jeffthompsonsd]

It did seem related to that issue, any ideas? We are seeing it outside of tests.