uWSGI worker throwing Runtime Exception when run with `py-call-uwsgi-fork-hooks` flag

[zaheerabbas-prodigal]

uWSGI worker throwing Runtime Exception when run with `py-call-uwsgi-fork-hooks` flag

[sentrivana]

Is there any other way we could make the sentry integration work?

Not that I know of. We're usually happy to either patch or work around things in libraries and frameworks that don't work with the SDK but the missing fork hooks is something so fundamental that there's not much we can do about it. The only solution as long as the upstream PR is not merged is to skip the broken forking altogether with --lazy-apps or, unfortunately, to switch away from uWSGI.

Has no one else in the community faced this problem with uWSGI <> Sentry post Python 3.7+ changes where this was introduced?

We've only started recommending using --py-call-uwsgi-fork-hooks fairly recently which might explain why there haven't been other reports about this so far. That being said, I didn't encounter this myself while working on the original issue. Granted, I was using a very bare-bones Flask app to repro and wasn't doing any logging stuff in the app itself, but things were still being logged indirectly via e.g. the SDK's debug mode. So the RuntimeError might be triggered by something specific in your logging setup -- it's a long shot but playing around with where and when you log things that might also be a workaround.

[sentrivana]

Hey @zaheerabbas-prodigal, thank you for reporting and for providing all the details!

As far as I can tell this is caused by uWSGI not calling all the necessary fork hooks. Basically, in pre-fork mode, everything is initialized in one process that is then forked into additional workers. When forking a process, C extensions are expected to call PyOS_BeforeFork(), PyOS_AfterFork_Parent() in the parent and PyOS_AfterFork_Child() in the child to do stuff like clean up threads in the child processes.

The problem is that by default, uWSGI calls none of them. If you supply --py-call-uwsgi-fork-hooks as we recommend, PyOS_AfterFork_Child() will be called (and take care of some of the necessary thread cleanup), but the other two hooks still won't, and I assume PyOS_BeforeFork() acquires the logging lock in question that PyOS_AfterFork_Child() is then trying to release.

There is a long standing uWSGI pull request to utilize all three hooks.

As for what to do in your case, you can disable pre-forking mode completely with --lazy-apps, in which case every worker sets up the app itself and there is no broken forking involved.

[zaheerabbas-prodigal]

@sentrivana - thank you for explaining the issue. I appreciate the quick response.

I have asked an update on the PR you linked. The lazy-app might cause some memory issues for us. Is there any other way we could make the sentry integration work? Has no one else in the community faced this problem with uWSGI <> Sentry post Python 3.7+ changes where this was introduced?