fix issue #40: make internal message queues unlimited

[niamster]

No description provided.

[tarcieri]

Seems ok, but keep this in mind:

https://twitter.com/mrb_bk/status/395301262238838784

[digitalextremist]

I see this is brought up by #40, but could we rather set a configurable buffer size?

[niamster]

@tarcieri I totally agree though I don't see a better way to fix this. Maybe before sending an IPC message check the queue status and reschedule. I don't know whether this will work as the reader and writer are in the same celluloid context if I understand correctly.

@digitalextremist we can though we need a mechanism to fix the deadlock w/o throwing exception "stop producing too many work".

[niamster]

In fact the root cause of the deadlock is that celluloid holds a mutex while calling reactor.wakeup.
Does it make sense to call wake under the lock? I'll create another issue on celluloid side to raise discussion.

[tarcieri]

@niamster the general idea of this sort of wakeup system is it should be possible to wake up the loop even though another thread is holding the lock. If not, it's a bug

[niamster]

The real issue is celluloid/celluloid#500