Catch crashing threads about sentry-python

[mmerickel]

I noticed some crashing threads that weren't getting reported to sentry and was surprised. I've updated my code to catch them myself but I agree that sentry could try harder here.

Sentry should be able to monkeypatch threading.Thread._bootstrap_inner. Unfortunately there is not a good spot to monkeypatch Thread.run (which a public api overridable by users) which Thread._bootstrap_inner calls directly. Is there a standard approach sentry uses to monkeypatching without vendoring the current _bootstrap_inner code into sentry? So far that seems to me to be the only option.

[untitaker]

Two ideas:

• Patch Thread._exc_info to capture the exception (it gets called in _bootstrap_inner)
• Patch Thread._bootstrap_inner, but inside the patched function patch self.run and just call the old _bootstrap_inner.

There is no standard approach to this. An established rule is to patch public/stable APIs rather than private/unstable ones but it seems like we have no choice here.

[untitaker]

Actually we can do something that uses only public API: Patch Thread.start, patched functions patches self.run and calls old Thread.start.