race in TestDeepStressStaysSane

[Ian Booth]

As seen in

http://reports.vapour.ws/releases/5334/job/run-unit-tests-race/attempt/2834

[John A Meinel (jameinel)]

Wow. The test is supposed to run in about 3s, using 3 slots for pinger. In this case it actually runs in 50s, and I have the feeling its the sort of "hung not doing anything for 48s, and then goes into a mad dash and finishes everything in 2s which means our counts are off.)

The test thought it was being generous by saying "pingers shouldn't trigger more than 4 active pingers in a given window", but when your load is off by a factor of 10x, the measurements can be a bit off.
And we don't want to give ourselves too large of a window, because then we wouldn't actually be testing that pruning is doing anything.

In this particular case, we were only off by about 5 pings out of 2000, but given we should really be more like 500 pings total our 'generous window of 1500 extra' wasn't enough.

I'll think about whether I can redesign it to pay explicit attention to how many pingers fired in the last window that shouldn't be pruned.

[John A Meinel (jameinel)]

https://github.com/juju/juju/pull/7578

[Anastasia (anastasia-macmood)]

I suspect that the fix for this has been already ported into 2.3 (develop) as part of one of the larger merge we have undertaken recently.