Native ASGI worker leaks active connection count and delays SIGTERM shutdown[Triage]

[frankxzw]

Type

Bug Report

Description

The native ASGI worker does not decrement worker.nr_conns when the server closes a transport.

After at least one HTTP request has been handled, sending SIGTERM to the Gunicorn master causes the worker to wait for the entire graceful_timeout, even though the client connection has already been closed and there are no in-flight requests.

Steps to Reproduce (for bugs)

• Gunicorn: 26.0.0
• Python: 3.12.13
• Platform: Ubuntu 24.04.4 LTS
• Event loop: uvloop 0.22.1 (--asgi-loop auto)

Minimal application

async def app(scope, receive, send):
    if scope["type"] == "lifespan":
        while True:
            message = await receive()
            if message["type"] == "lifespan.startup":
                await send({"type": "lifespan.startup.complete"})
            elif message["type"] == "lifespan.shutdown":
                await send({"type": "lifespan.shutdown.complete"})
                return

    if scope["type"] == "http":
        await send(
            {
                "type": "http.response.start",
                "status": 200,
                "headers": [(b"content-length", b"2")],
            }
        )
        await send({"type": "http.response.body", "body": b"ok"})

Configuration

Start Gunicorn:


gunicorn app:app \
    --worker-class asgi \
    --workers 1 \
    --bind 127.0.0.1:8000 \
    --keep-alive 0 \
    --graceful-timeout 3 \
    --no-control-socket \
    --pid /tmp/gunicorn-asgi.pid \
    --access-logfile -


Send one request and explicitly close the connection:

```bash
curl --http1.1 -H 'Connection: close' http://127.0.0.1:8000/
kill -TERM "$(cat /tmp/gunicorn-asgi.pid)"

### Logs / Error Output

```text
### Actual behavior

The worker waits for the full graceful timeout:


[INFO] Handling signal: term
[INFO] Worker shutting down...
[INFO] Waiting for 1 connections to finish...
[INFO] Shutting down: Master


With `graceful_timeout=3`, shutdown takes approximately 3 seconds. Without handling any request, the same worker exits in approximately 1 second.

The issue also reproduces with `keep-alive=0`, so this is not an active keep-alive connection.

Gunicorn Version

gunicorn 26.0.0

Python Version

Python 3.12.13

Worker Class

asgi (beta)

Operating System

Ubuntu 24.04.4 LTS

Additional Context

Expected behavior

Once the transport has closed and there are no in-flight requests, worker.nr_conns should return to zero and SIGTERM should not consume the entire graceful timeout.

Suspected root cause

ASGIProtocol.connection_made() increments the connection count:

https://github.com/benoitc/gunicorn/blob/26.0.0/gunicorn/asgi/protocol.py#L389-L392

_close_transport() closes the transport and immediately sets _closed = True:

https://github.com/benoitc/gunicorn/blob/26.0.0/gunicorn/asgi/protocol.py#L1450-L1464

When asyncio subsequently invokes connection_lost(), it returns early because _closed is already true. Therefore, worker.nr_conns -= 1 is never executed:

https://github.com/benoitc/gunicorn/blob/26.0.0/gunicorn/asgi/protocol.py#L672-L687

The ASGI worker then waits on the leaked count during graceful shutdown:

https://github.com/benoitc/gunicorn/blob/26.0.0/gunicorn/workers/gasgi.py#L239-L251

A direct state-level reproduction gives:

after _close_transport: _closed=True, nr_conns=1
after connection_lost:  _closed=True, nr_conns=1

Possible fix

Connection accounting should use a state flag separate from _closed, for example _connection_lost_handled, so connection_lost() decrements nr_conns exactly once even if _close_transport() has already marked the transport closed.

A regression test could verify that:

1. connection_made() increments nr_conns.
2. _close_transport() followed by connection_lost() decrements it to zero.
3. SIGTERM after a completed, closed request does not wait for graceful_timeout.

Checklist

• I have searched existing discussions and issues for duplicates
• I have checked the documentation and FAQ

[jeffsawatzky]

Related to #3661?