ThreadKill not handled by ThreadFiber

[e2]

Celluloid tests are failing consistently on JRuby (jruby-1.7.19 and jruby-head from RVM) with stacks like the following:

Exception in thread "Ruby-0-Fiber-3" org.jruby.exceptions.ThreadKill
        at org.jruby.RubyThread.throwThreadKill(RubyThread.java:517)
        at org.jruby.RubyThread.checkMail(RubyThread.java:216)
        at org.jruby.RubyThread.pollThreadEvents(RubyThread.java:513)
        at org.jruby.RubyThread.pollThreadEvents(RubyThread.java:509)
        at org.jruby.RubyThread.executeBlockingTask(RubyThread.java:1064)
        at org.jruby.RubyThread.wait_timeout(RubyThread.java:1413)
        at org.jruby.ext.thread.Queue.pop(Queue.java:152)
        at org.jruby.ext.thread.Queue.pop(Queue.java:127)
        at org.jruby.ext.thread.SizedQueue.pop(SizedQueue.java:111)
        at org.jruby.ext.fiber.ThreadFiber$1.run(ThreadFiber.java:214)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
        at java.lang.Thread.run(Thread.java:745)

On jruby-head it's:

Exception in thread "Ruby-0-Fiber-1" org.jruby.exceptions.ThreadKill
        at org.jruby.RubyThread.throwThreadKill(RubyThread.java:625)
        at org.jruby.RubyThread.toKill(RubyThread.java:274)
        at org.jruby.RubyThread.executeInterrupts(RubyThread.java:243)
        at org.jruby.RubyThread.pollThreadEvents(RubyThread.java:615)
        at org.jruby.RubyThread.executeTask(RubyThread.java:1363)
        at org.jruby.ext.thread.Queue.pop(Queue.java:195)
        at org.jruby.ext.thread.Queue.pop(Queue.java:164)
        at org.jruby.ext.fiber.ThreadFiber$1.run(ThreadFiber.java:250)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
        at java.lang.Thread.run(Thread.java:745)

Isn't ThreadFiber.run supposed to catch and handle ThreadKill?

Let me know if I can debug further (though coming up with a specific short snippet may be hard).

[headius]

Yes, we probably should just be handling it there, since it needs to propagate out of the fiber into the owning thread.

[headius]

If you can come up with a way to test this, it would be great. Perhaps something that spins up a bunch of threads with fibers and kills them.

[e2]

I'm not sure why/how the fibers are getting killed, actually. Running the Celluloid tests off of master or 0-16-stable fails pretty much every time on JRuby. The "probe_spec.rb" is pretty much what consistently causes this error (though the spec is complex as it is).

It seems to be failing randomly in an "after" block - where celluloid is shut down (tasks are killed, etc.). There's a bit too much stuff there going on for me to create a simple failing example with JRuby. Though the Celluoid tests fail pretty much every time.

I don't get the JRuby side of the threads are interrupted (the flags and stuff) to even being to find where it's failing on the Celluloid side (where the fibers are killed/interrupted).

I thought about just catching the exception in JRuby and throwing any Ruby exception - just to get idea of where the exception is happening.

I'm not even sure how to kill a fiber to begin with...

[e2]

I'll test the fix and let you know (building jruby-head will take a while).

[e2]

Cool! The tests now hang sometimes, but ... no crashes so far!

I'm marking this as closed.

Thanks, you're awesome! :)