SIGINT handler is annoying during socket bind

[pajod]

Type

Bug Report

Description

Because of the 5-second delay in retrying socket bind (which I do not intend to meaningfully change in #3454) hitting Ctrl+C during startup is ignored. I do believe retrying for most OSError is useful, so things do not break simply from (common) incorrect service startup ordering during system/container boot.

Simply moving the signal handler installation up in the arbiter startup code may not be a perfect solution either, because Gunicorn should not ignore other signals during startup.

Looking for ideas to keep startup interruptible without reorganizing current signal installation too much.

Steps to Reproduce (for bugs)

Ask to bind to an address not associated with any accessible interface, e.g.:

python3 -m gunicorn --log-level=info --bind="[fe80::]:8000" myapp:app

Logs / Error Output

[2026-01-26 03:50:05 +0100] [59356] [INFO] Starting gunicorn 24.1.0
[2026-01-26 03:50:05 +0100] [59356] [ERROR] connection to ('fe80::', 8000) failed: [Errno 22] Invalid argument
[2026-01-26 03:50:06 +0100] [59356] [ERROR] connection to ('fe80::', 8000) failed: [Errno 22] Invalid argument
[2026-01-26 03:50:07 +0100] [59356] [ERROR] connection to ('fe80::', 8000) failed: [Errno 22] Invalid argument
[2026-01-26 03:50:08 +0100] [59356] [ERROR] connection to ('fe80::', 8000) failed: [Errno 22] Invalid argument
[2026-01-26 03:50:09 +0100] [59356] [ERROR] connection to ('fe80::', 8000) failed: [Errno 22] Invalid argument
[2026-01-26 03:50:10 +0100] [59356] [ERROR] Can't connect to ('fe80::', 8000)

Gunicorn Version

gunicorn 24.1.0

Python Version

Python 3.11.2

Worker Class

N/A (feature request)

Operating System

Debian

Additional Context

No response

Checklist

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

[benoitc]

why not adding a handler on SIGINT that set a variable. On each retry you check if stop (the variable) is set and then stop ?

in arbiter:

def __init__(..):

    _stop = threading.Event()

def handle_int(self):
   ...
   _stop.set()

in socks:

def create_sockets(conf, log, _stop, fds=None):

    ....

     if i < 5 and not _stop.is_set():
         log.debug("Retrying in 1 second.")
         time.sleep(1)

Or something like it?

[benoitc]

could also use _stop.wait(1) instead of time.sleep(1) ...

[benoitc]

@pajod does it answer to your question?

[pajod]

Sounds much better than anything I had in mind - will probably add this to one of my (numerous..) PRs once I got around actually testing.