1.x: add optional tracking of worker creation sites + report it on error

[akarnokd]

This is an alternative to #3937 to capture worker creation sites for computation, io and newThread schedulers.

[artem-zinnatullin]

Move to public rx.schedulers and mark as @Experimental?

[akarnokd]

No need to expose this, there is the Schedulers.setWorkerTracking. Plus, this way it can be an enum.

[artem-zinnatullin]

Ok, got it.

[artem-zinnatullin]

I see potential problem with this solution: if I jump between threads in the Rx chain it looks like I'll loose stacktrace of subscribe() call.

Observable
  .just(value)
  .observeOn(scheduler1)
  .map(funcA())
  .observeOn(scheduler2)
  .map(funcB())
  .subscribe()

If exception happens in map(funcB()), we will have stacktrace pointing to observeOn(scheduler2), not to observeOn(scheduler1).

Looks like we need to have ThreadLocal<Throwable> and use it's value if it exists in createWorker() to keep stacktrace while jumping between threads.

[akarnokd]

What you are describing is value-origin tracking, not what this and #3937: if a scheduled task crashes, show where the worker has been created, possibly pointing to the assembly site of the sequence.

[artem-zinnatullin]

Then this solution won't cover major percentage of cases when chain jumps between threads and then unexpected exception happens.

For now I vote for keeping initial stacktrace through threads, so 👎, sorry.

[akarnokd]

If exception happens in map(funcB()), we will have stacktrace pointing to observeOn(scheduler2), not to observeOn(scheduler1).

The scheduler2 site is only reported if funcB crashes fatally, which is detected only on the scheduler2's thread. Why would it have to report scheduler1?