Comment 15 for bug 1654116

Looking at the rabbitmq logs, it appears to me (I could be misinterpreting):

At 21:39:23 -> unit 2 becomes the leader (the leader-elected hook is run)
At 21:47:25 -> unit 0 becomes the leader (the leader-elected hook is run)
At 22:01:27 -> unit 1 becomes the leader (the leader-elected hook is run)
At 22:13:54 -> unit 1 thinks it is the leader and updates leader settings but Juju thinks it is not
At 22:17:06 -> unit 0 becomes the leader (the leader-elected hook is run)
At 22:22:57 -> unit 0 thinks it is the leader and updates leader settings but Juju thinks it is not

So it appears that Juju ran the leader elected hook on both unit 0 and unit 1 and these units tried to do stuff as leader. But Juju thinks unit 2 is leader.

Curiously, at 22:17:06 and 22:17:58, unit 2 successfully updates leader settings, even though it should not have been the leader anymore.

The charm doesn't appear to use leader elected or deposed hooks - it just symlinks to a common Python file. Juju still thinks that there are hooks though, so it runs the leader-elected hook and logs that. But I can't see any logs of where Juju runs the leader-deposed hook. Perhaps that's a clue? Perhaps that indicates that Juju is getting confused as to who the leader is? It would be useful to have a mongo dump to see what state the model is in.