Unexpected api shutdown followed by panic

[Tim Van Steenburgh]

Looking through the logs of the recent test run failures (http://reports.vapour.ws/charm-test-details/charm-bundle-test-parent-245), I noticed that all failures seemed to stem from the python-jujuclient losing the websocket connection to the juju api server. Looking into the all-machines log for each provider, I found something similar in each. Here's the log from the HP cloud (start reading on line 721):

http://pastebin.ubuntu.com/11098434/

Sure looks like something bad is happening, but I can't diagnose beyond that...

[Menno Finlay-Smits (menno.smits)]

The panic is caused by the "not a valid unit name" error. This looks like a duplicate of bug 1437266.