Comment 3 for bug 1459274

This was instrumental in spotting the difference...

Extract from email from Curtis (brilliant)...

The first bug reported about failures caused by a tool chain is
    https://bugs.launchpad.net/juju-core/+bug/1533792

This issue is very specific and is not the unreliability we are seeing in 1.25

The fix was merge into both master
  https://github.com/juju/juju/commit/36b26048cd8f5fb5665e6e9c95deb466c55750e3
and 1.25
 https://github.com/juju/juju/commit/0377d2ae02060149e6e3aa85cb57b79a53af80d1

We can see before the tool chain change, run-unit-tests-trusty-ppc64el passed
first try
    http://reports.vapour.ws/releases/3490
and after the merge (master was not directly tested after the tool chain
change, but feature branches were). We can see that master did pass (with
reties) after the fix was merged and tested on Timestamp:
Fri Jan 15 22:03:10 2016
    http://reports.vapour.ws/releases/3513/job/run-unit-tests-trusty-ppc64el/attempt/4414
While the test set to non-voting as we tried to get a tool chain fix, we can
see master continued to pass
    http://reports.vapour.ws/releases/3516

1.25 was not so lucky. IT has not passed since Tue Jan 12 22:37:22 2016
    http://reports.vapour.ws/releases#1.25
in addition to the first bug reported, 1.25 had linker failures seen between
Thu Jan 14 01:29:08 2016 and Fri Jan 15 18:41:16 2016
    http://reports.vapour.ws/releases/3505/job/run-unit-tests-trusty-ppc64el/attempt/4391
    http://reports.vapour.ws/releases/3512/job/run-unit-tests-trusty-ppc64el/attempt/4411

After a change to either the host or source code we see the current error.
    http://reports.vapour.ws/releases/3515/job/run-unit-tests-trusty-ppc64el/attempt/4417
This error looks too similar to a previous error. We can see the date change
on
    http://reports.vapour.ws/releases/issue/5565d588749a56738657898c
to know the issue appeared on Jan 15 with
    http://reports.vapour.ws/releases/3511