Leadership appears broken in 1.25

[Charles Butler]

When working with charm leadership i've managed to enter a situation where juju never re-elects a leader in 1.25+

Reproduction instructions:

juju deploy some service -n 3

Identify the leader of the service.

juju destroy-machine # --force (of the leader)

Juju doesn't appear to recognize that there is no longer a leader in the service pool, thus none of the units seek to obtain leadership and reconfigure the cluster with a leader. This has implications on service that make use of juju's is-leader method call to configure and launch a leadership service, such as the implementation of docker-swarm which runs a scheduler only on the leader node.

[Adam Collard (adam-collard)]

Smells like this could be https://bugs.launchpad.net/juju-core/+bug/1488166 - there's an open request for debug information in this situation in that bug

[Charles Butler (lazypower)]

I'll add some logs from an environment later today to make this bug more useful.

[Stuart Bishop (stub)]

This would be Bug #1511659, which already contains complete logs.