Gossip-48 shutdown message FTW

[Edward Capriolo <ed...@gmail.com>]

Hey all,

I was poking around a bit and realized the exponential weighted average
makes convergence slow in some cases. I realized it made sense as an
optimization  to send a message before shutting down so that hosts will
mark down the node faster regardless of what the FD thinks the node state
is.

https://github.com/apache/incubator-gossip/pull/31

https://issues.apache.org/jira/browse/GOSSIP-48

During the process I found an executor service that was not shut down and
also realized shutting down was the majority of time for some of the tests.

With all that in place the tests now run in 40 seconds down from 2:00- 2:30.

If someone wants to give it a look over I am going to leave it 2-3 days
before merge.

This would be a good one to get some eyes on to poke logical holes in my
theory (which I believe is right) :)

Thanks,
Edward