allowed_units sometimes out of sync
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| OpenStack Percona Cluster Charm |
New
|
Undecided
|
Unassigned | ||
Bug Description
This is mostly a sighting; I don't have deep details into why this happened, but this is present on a customer cloud right now.
Seen on: cs:percona-
While debugging a login issue for a gnocchi peer via the shared-db relation, I saw that the correct password was being shared by all MySQL units, but the allowed_units field was not up to date on one of the MySQL units.
Concrete example: 2 units had "gnocchi/2 gnocchi/3 gnocchi/7" for allowed units, while one still had "gnocchi/1 gnocchi/2 gnocchi/3".
I see no record of hooks being forcibly resolved in the juju audit logs for the percona-cluster app, and while I see one "juju resolved" event for gnocchi, it was for a different unit, not the one which was missing.
Possibly related (but perhaps a different bug): for the newest Gnocchi unit on this cloud (gnocchi/7), no user record was created; I had to manually create it for that unit to be able to log in.
