Hystrix.reset() now shuts down the HystrixTimer

[benjchristensen]

• the HystrixTimer can be shutdown via an interrupt (which Hystrix.reset uses)
• the HystrixTimer will correctly restart the thread if it is used again after a reset

#109

[daveray]

Looks reasonable to me. The listeners map and intervals queue seem like they're coordinated so you could have race conditions between the reset code and listener registration. I could also imagine pathological TimerListener.tick() code that swallows/suppresses interrupts which would cause shutdown not to occur. But, honestly, this code should only be executed during a controlled shutdown so it's probably fine.

[benjchristensen]

Yes, if someone tries to reset while still using Hystrix then it won't be thread-safe and race conditions can occur. I comment on that in the javadoc:

NOTE: This will result in race conditions if {@link #addTimerListener(com.netflix.hystrix.util.HystrixTimer.TimerListener)} is being concurrently called.

And yes, if the tick() implementation was evil it could swallow the interrupt - good point, I hadn't considered that. We are the only ones who create tick() implementations so that should be fine - but I could instead use a separate variable to ask for it to stop rather than using interrupt.