In a Cloud upgraded from trusty/mitaka to xenial/queens with stepped upgrades and finally to bionic/queens, there are some older volumes that have wrong host attribute equal to cinder@cinder-ceph#cinder.volume.drivers.rbd.RBDDriver.
When trying to create a snapshot of that volumes or creating a new volume from a snapshot, this fails with:
2022-01-04 12:35:40.750 69884 ERROR cinder.scheduler.manager [req-4f20b00c-37a8-4e2d-b689-a92c75d2dabc 4cc22e877fad4ff580abd78567acab2c 500cec76ab6c49c198a4fe0138ba7dc6 - b71e87f93ff54f4c9c66a8db1a2cab5f b71e87f93ff54f4c9c66a8db1a2cab5f] Failed to schedule_create_snapshot: Cannot place snapshot e157421c-3455-47c9-8dd8-933111309d94 on cinder@cinder-ceph#cinder.volume.drivers.rbd.RBDDriver.: NoValidBackend: Cannot place snapshot e157421c-3455-47c9-8dd8-933111309d94 on cinder@cinder-ceph#cinder.volume.drivers.rbd.RBDDriver.
If I update manually the host on the DB to cinder@cinder-ceph#cinder-ceph everything works fine.
To fix that I tried to use the rename-volume-host action:
$ juju run-action cinder/7 rename-volume-host currenthost='cinder@cinder-ceph#cinder.volume.drivers.rbd.RBDDriver' newhost='cinder@cinder-ceph#cinder-ceph' --wait
unit-cinder-7:
UnitId: cinder/7
id: "11352"
message: Cannot update host attribute from cinder@cinder-ceph#cinder.volume.drivers.rbd.RBDDriver,
cinder@cinder-ceph#cinder.volume.drivers.rbd.RBDDriver not found
results: {}
status: failed
timing:
completed: 2022-01-05 12:34:59 +0000 UTC
enqueued: 2022-01-05 12:34:56 +0000 UTC
started: 2022-01-05 12:34:56 +0000 UTC
This is happening because the charm checks if currenthost exists in hosts from "cinder-manage service list":
# cinder-manage service list
Binary Host Zone Status State Updated At RPC Version Object Version Cluster
cinder-scheduler cinder nova enabled :-) 2022-01-05 12:41:50 3.10 1.35
cinder-backup cinder nova enabled :-) 2022-01-05 12:41:49 2.1 1.35
cinder-volume cinder-ssd@ssd-storage nova enabled :-) 2022-01-05 12:41:41 3.15 1.35
cinder-volume cinder@cinder-ceph nova enabled :-) 2022-01-05 12:41:48 3.15 1.35
and the host is not there but actually from documentation [0] I understand that action is intended to fix that kind of issues when a backend is removed or renamed but that fails and then charm should not check if currenthost is there or not. Also currenthost is passed in the format:
host_uuid@backend-name#pool [1]
And can not be == to host in cinder-manage service list where you have only host_uuid@backend-name. That makes the action unusable.
0. https://github.com/openstack/cinder/blob/master/cinder/cmd/manage.py#L573
1. https://support.platform9.com/hc/en-us/articles/360010968913-Changing-Management-of-Volumes-in-a-Shared-Storage-Backend
Another question will be given that I have the following available pools:
$ cinder get-pools
+------
| Property | Value |
+------
| name | cinder@
+------
+------
| Property | Value |
+------
| name | cinder-
+------
can I workaround the snapshot creation issue using a query directly on the DB like the following?
update volumes set host="cinder@