Zero-Downtime restart

[yangbo1024]

Initial Checks

• I confirm this was discussed, and the maintainers suggest I open an issue.
• I'm aware that if I created this issue without a discussion, it may be closed without a response.

Discussion Link

No link.

Description

Hello. I’ve encountered two requirements for which I can’t seem to find a proper solution in Uvicorn for now:

1. For a singleton service that must run with only 1 worker but requires a graceful restart. When sending a SIGHUP to Uvicorn, the process exits.
2. For another service with 2 workers, when sending a SIGHUP to Uvicorn, the restart happens one by one—specifically, shutting down the old worker first before starting a new one.

I’m using Uvicorn directly without Gunicorn. Based on the information I’ve researched and my actual tests, Gunicorn’s connection handling mechanism, even with the reuse_port option, still has issues with uneven load distribution across multiple workers.

What I expect from Uvicorn is:
For the single-worker scenario, when the master process receives a SIGHUP, it should shut down the old worker and start a new one instead of exiting.
Regarding the worker startup order, I believe the correct approach should be to start the new worker first. This ensures the service doesn’t interrupt and prevents a sudden increase in load on the original worker.

I’m wondering if I’m using something incorrectly, or can uvicorn add the features? Thanks!

Example Code

Python, Uvicorn & OS Version

Python 3.12.3, Uvicorn 0.38.0

[theseriff]

maybe my PR fix it
#2752

[yangbo1024]

Thanks for your response, but it seems not the same problem.
My main requirement is to achieve zero-downtime restarts, even when using only one worker.

[nggit]

For the single-worker scenario, when the master process receives a SIGHUP,

In Uvicorn there is no such "master process" for single mode, single-worker itself is the main/master process... So it can't resurrect itself after exit. You need to bring your own PM/supervisor.