sidekiq scope potentially leaks when there's an exception

[sl0thentr0py]

we're manually clearing scope in the middleware but if there's an exception in the job this might not run. We should just use push_scope instead to ensure cleanup.

Try this out with a proper test case and fix it.

[st0012]

Yeah you're right, the scope may not be cleared if there's an error. I'll take a look at this.

[st0012]

Ok I should've read my own comment first. I intentionally didn't use ensure there because at that moment, the exception is not reported yet. So we need to NOT clear the scope until we reach ErrorHandler later.

Was there an incident that made you started looking into this?

[sl0thentr0py]

I think @untitaker reported this at some point because he was using the SDK on his mastodon instance

[untitaker]

So we need to NOT clear the scope until we reach ErrorHandler later.

The problem is that the ErrorHandler also does not clear the scope.

[st0012]

@untitaker Yes you're right, good catch 👍 I've opened #2059 to address it.