ThreadLocalActiveSpan.deactivate()

ThreadLocalActiveSpan.deactivate() does not deactivate/finish span when it is not currently active span: https://github.com/opentracing/opentracing-java/blob/master/opentracing-util/src/main/java/io/opentracing/util/ThreadLocalActiveSpan.java#L50

Problems:

• response interceptors sometimes run in a different thread, therefore it makes it impossible to deactivate the span (e.g. async jax-rs)
• if a user accidentally does not deactivate local span, span created by instrumentation is not finished

I would like to see what others think and what issues they faced.

[rbtcollins]

I came here tracking a span leak in test code - using MockTracer w/ThreadLocalActiveSpanSource - even without threads the current code has some fragility.

What I have going on is clearly a misuse of the API - and I’m going to be tracking that down now - but the silent leak was much less helpful than we could be.

I’d like to see an exception raised, or a diagnostic stacktrace logged or some such thing: leaking is pretty much the most difficult thing to deal with: its silent, it will cause traces to have bad parents and possibly even join distinct traces together.

[carlosalberto]

Trying to prevent this Issue to become stale, a pair of things:

1. Once we implement some kind of Observer API, this kind of things could at least be reported to the user.
2. With the old 0.30 API it seemed to be much more of a problem - as we haven’t been getting reports about this behavior with the new 0.31 one.

Unless somebody thinks this is still an issue, I will close it by the end of the next week. Let us know 😉

@pavolloffay @rbtcollins