There's a few errors happening in the uniter. This one was mentioned in IRC:
2015-09-16 02:40:53 ERROR juju.worker runner.go:218 exited "uniter": ModeAbide: cannot set invalid status "started"
but I think this is a red herring. This will not prevent the uniter from upgrading, so it'll error, restart, upgrade, then start setting status in the correct way.
I *think* this is the issue:
2015-09-16 05:23:45 DEBUG juju.worker.metrics.collect manifold.go:159 [{github.com/juju/juju/worker/charmdir/manifold.go:102: charmdir not available}]
There's a few errors happening in the uniter. This one was mentioned in IRC:
2015-09-16 02:40:53 ERROR juju.worker runner.go:218 exited "uniter": ModeAbide: cannot set invalid status "started"
but I think this is a red herring. This will not prevent the uniter from upgrading, so it'll error, restart, upgrade, then start setting status in the correct way.
I *think* this is the issue: metrics. collect manifold.go:159 [{github. com/juju/ juju/worker/ charmdir/ manifold. go:102: charmdir not available}]
2015-09-16 05:23:45 DEBUG juju.worker.
I'm getting this in my repro of the bug, and it's in the logs at http:// data.vapour. ws/juju- ci/products/ version- 3059/aws- upgrade- 20-trusty- amd64/build- 29/unit- dummy-sink- 0.log. It's interesting that, in those logs, the token *is* eventually set to the expected value... I think it's just taking longer than it should, due to this bug causing the uniter to bounce up and down.