Comment 6 for bug 1560201

Eric, I was able to reproduce this with beta3 on an lxd deployment. In my original report, I failed to notice that the hang occurs when the AllWatcherId is consumed by an AllWatcher API call. I'll attach a full trace.

I have pushed some code and a short test script to demonstrate this to lp:~fginther/python-jujuclient/lp-1560201-test.

To reproduce:
$ bzr branch lp:~fginther/python-jujuclient/lp-1560201-test
$ cd lp-1560201-test
$ virtualenv .venv
$ . .venv/bin/activate
$ python setup.py develop
$ ./test-case.py local.lxd-ctl:default

This assumes that you have bootstrapped an lxd controller named 'lxd-ctl' with an empty model named 'default'. The test script will dump out the juju API calls as they are being made. If the targeted model has no machines, it should hang. Adding a machine should break the hang.