When accepting connections in a multi-process environment, it is highly likely that the coroutine will be blocked.

[godtang]

import asyncio
import datetime
from fastapi import FastAPI
import os

app = FastAPI(title="FastAPI Multi-Process Demo", version="1.0.0")


async def print_loop():
    index = 1
    while True:
        print(
            f"[process {os.getpid()}] print_loop {index} {datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')} is running..."
        )
        index += 1
        await asyncio.sleep(1)


@app.on_event("startup")
async def start_print_loop():
    asyncio.create_task(print_loop())

use command uvicorn main:app --host 0.0.0.0 --port 22222 --workers 2 run
After connecting to port 22222 multiple times, the following situation will occur as shown in the figure. If you reconnect, the coroutine printing can be resumed.
On my computer, it can be replicated within 5 connections. However, my colleague's computer requires more than 20 attempts.

python -m uvicorn --version
Running uvicorn 0.30.6 with CPython 3.12.9 on Windows

[Kludex]

I can't reproduce this.

[godtang]

20250716-144335.mp4

@Kludex here is video, i'm certain that this problem has occurred on at least three machines in the company.

[Kludex]

I see. Can you give me a script to reproduce with a tool I can use on the terminal? That would help me debug it.

[godtang]

NetAssist.zip

[vvanglro]

I couldn't reproduce it either, whether using h11 or httptools.

[godtang]

@vvanglro 我看到你的日志里只有一个进程67465，复现需要是多进程运行，你可以用uvicorn main:app --host 0.0.0.0 --port 22222 --workers 2 执行
I noticed that there is only one process 67465 in your log file. To reproduce the issue, multiple processes need to be run. You can execute uvicorn main:app --host 0.0.0.0 --port 22222 --workers 2 to do this.

[vvanglro]

Multi process is the same, cannot be reproduce.

[nggit]

Can't reproduce it.

Before digging further make sure you keep a strong reference to the task because in this case: the startup event handler exits immediately. And the created task may not survive or may be GCed at some point.

app.state.bg_task = set()

@app.on_event("startup")
async def start_print_loop():
    app.state.bg_task.add(asyncio.create_task(print_loop()))
    print('exited')