Scheduler driver re-registration backoff

[tarnfeld]

MESOS-1903 was implemented and is due to come out in 0.22.0 so when Pesos supposed re-registration we should probably do something similar.

[wickman]

Implemented at a835b12. Please take a look.

[tarnfeld]

I was doing some testing locally and there seems to be a bug relating to when a master disconnection happens. It appears the re-registration loop is never started.

For example, after the first registration, the next tick on the loop effectively disables the backoff and from what I can tell in the event of a master disconnection, it isn't re-enabled.

Perhaps I'm missing something, or maybe I didn't wait long enough for the framework to pick up the error. I doing a simple test with the standalone master detector and stopping/starting the same master a few times.