Comment 6 for bug 1914415

Yeah, that analysis is correct -- this is definitely the same issue as https://github.com/canonical/operator/issues/175. I was able to repro this locally quite easily (using a minikube cluster).

Note that it's the relation.data access that's raising the KeyError, which is due to Relation.app being None (I don't think it's related to RelationEvent, as the event.app access has already executed by this time).

Note that I was also able to work around this using "juju resolved":

juju resolved --no-retry postgresql/0

After that the setup continued and Mattermost and the db came up with an issue.

I'll look at a proper fix tomorrow (spent too much time today getting Docker and minikube going again on my refreshed laptop...).