destroy-model fails to remove all machines, invalid controller credential

[Thomas Cuthbert]

Upon attempting to remove a model against a 2.0.3 controller on GCE the process is unable to destroy/remove the last remaining machine.

See https://pastebin.canonical.com/194981/ for some debug info.

I have left the environment as is so we can do some debugging if required.

[Paul Collins (pjdc)]

What seems to be happening here is that the credential for this controller and/or model is no longer valid, so it can't remove the VM. We have created a new credential and updated it with "juju add-credential --replace" but the controller seems to have cached the old one. I was able to load a new credential under a new name, but obviously this doesn't help us here, and in fact the cached old credential also prevents us from upgrading to a newer version of Juju.

[Paul Collins (pjdc)]

That was rather unclear. We have created a new *key* in GCE and updated the *existing* Juju credential with "juju add-credential --replace", and confirmed it's pointing at the new file.

[Anastasia (anastasia-macmood)]

It seems like the best course of action would be to have functionality to update credential and use that to remove remaining machine.

We have an 'update credential' request, bug # 1713582. I think that this should be marked as its duplicate.