Correct way to sentry_sdk.init() in async code that makes use of a synchronous Django app?

[jklaiho]

I've got a synchronous, wsgi.py based Django project. sentry_sdk.init() is called in settings.py as instructed by DjangoIntegration docs.

I've now added a separate, partially async, long-running worker script. It needs to use some Django facilities, so it calls django.setup() at the top level of the module. This, of course, ends up calling init() in a sync context as the settings are loaded.

The docs for the asyncio integration state the following:

If you call sentry_sdk.init() outside of an async function in an async application, the SDK will not behave as expected.

I'm not (yet) using asyncio integration, but that warning seems to apply regardless. And indeed, some of the Sentry events from the async script are a bit odd (traces being listed as Unknown in particular), though I'm not sure if this is the reason—it could be some lack of configuration, missing spans etc.

Do I need to move sentry_sdk.init() out of Django settings and e.g. produce helper functions (one def, one async def) for initializing the SDK? For example, I'd call the sync version in wsgi.py for my Django app and call the async version in some suitable async one-time initialization function inside the async worker script.

[jklaiho]

Well, I ended up doing the sync/async init() split, and everything still works. The strange traces remain, so that's caused by something else.

Gotta say, the placement of the SDK async warning exclusively inside the AsyncioIntegration docs is a bit of an odd choice, since that's not enabled by default (or even by default in an async context).