[runtime] thread leak in DefaultBroadcasterFactory

[nite23]

Might be related to #716 #717

I use atmosphere with shared thread pools and IDLE_DESTROY policy.
If I call a following function
DefaultBroadcasterFactory.getDefault().lookup(DefaultBroadcaster.class, id, true)
concurrently, at higher rate, it spawns thread pools containing a single thread, which, sometimes, remain waiting indefinitely. Thread count grows at a constant rate until the VM runs out of memory.

It might be caused by a race condition, because when I synchronize the broadcaster lookup like this, the issue is not present, and the thread count stabilizes around 1000

[jfarcand]

Which version are you using?

[nite23]

1.1 snapshot

[jfarcand]

Thanks. Can you gist your test case? Thanks!