TestProvisionerObservesMachineJobs fails intermittently

[John A Meinel]

I ran into this today. I spent some time debugging it, and it looks like sometimes it takes long enough for the NewProvisionerTask to start, that by the time it starts up, the configuration has already changed. So the task starts is HarvestAny mode and never sends the signal (there is never a change).
So we need the test to wait until NewProvisionerTask has started before it changes the configuration.

[John A Meinel (jameinel)]

I've been able to reproduce this on multiple machines by just doing it in parallel:

for i in `seq 4`; do go test -v -check.v -check.f ProvisionerObserves & done ; time wait

[John A Meinel (jameinel)]

I wanted to make the test more deterministic, but there is a race condition before you get the observer set. So you can't wait for a first config observation, because you can't create the provisioner with an observer.

[John A Meinel (jameinel)]

This might be a useful fix for 1.25, but I don't think it is mandatory.

[Cheryl Jennings (cherylj)]

John's PR for master: https://github.com/juju/juju/pull/4614

[Anastasia (anastasia-macmood)]

Backport for 1.25: https://github.com/juju/juju/pull/4724