Description
From Slack:
It seems that setting the
sentry-propagate-traces
header to allow scheduled celery tasks to be their own trace (in Sentry, Python) doesn't actually work. Here's theapply_async
where the header is provided; the actual tasks clearly share a trace with their whole task tree; here's an error from one of the tasks.. you can see in the stack trace vars we've got trace id headers in there. I can provide additional examples of a trace id being shared between tasks scheduled this way, if that's useful. There's a lot of overriding and runtime patching happening, so it's easy to imagine that being subtly broken at some point, but I haven't seen anything obvious yet. The impact here is that relevant tasks for my team are too large to view in the Tracing UI, which really makes diagnosing our active performance problems harder.