Wait for other worker thread on shutdown #2484
Conversation
When worker stop due to max-lifetime, the first thread will now wait for additional thread to complet before terminating the worker process.
|
Is there something missing for this PR ? |
|
Any chance of getting a review on this one? It looks like a straightforward 4-liner bugfix to a very painful issue and I think is impacting my company as well, because the symptoms line up quite well (read: prod outages with the upgrade from Python 3.9.7 to 3.9.8 -- with the problem appearing to happen in uWSGI internal code) @xrmx ? |
| @@ -3718,6 +3718,10 @@ void uwsgi_ignition() { | |||
| } | |||
| } | |||
|
|
|||
| if (uwsgi.threads > 1 && !uwsgi_instance_is_dying) { | |||
There was a problem hiding this comment.
#define uwsgi_instance_is_dying (uwsgi.status.gracefully_destroying || uwsgi.status.brutally_destroying)
uwsgi_instance_is_dying includes graceful shutdown.
In graceful shutdown, we should wait other thread too.
I think simple if (uwsgi.threads > 1) test is OK.
|
This was fixed in #2626 which is now commit 06a2259 The fix was shipped in v2.0.25, though the version's release notes are less explicit about the change than the commits 2.0.24...2.0.25 |
When worker stop due to max-lifetime, the first thread will now wait for additional thread to complet before terminating the worker process.
This fix the issue described in #1894. With this change I never get any request error nor request that don't finish.
I think this will also fix the issue seen on production described in #2480
If you want that I make this wait behind an option, just ask and I'll update this PR.