Following an upgrade of a ~15.04 environment to 17.11, cinder (originally r91, revision-id: <email address hidden>) stopped working. I could no longer create, delete, detach, or attach volumes. Although cinder is related to ceph and block-devices is set to "None", following the upgrade, the service was configured to use a non-existent "LVM" backend definition:
root@juju-pjdc-ps45-1711-test-machine-7:~# grep enabled_backends /etc/cinder/cinder.conf
enabled_backends = LVM
root@juju-pjdc-ps45-1711-test-machine-7:~# grep '^\[' /etc/cinder/cinder.conf
[DEFAULT]
[keystone_authtoken]
[database]
root@juju-pjdc-ps45-1711-test-machine-7:~# _
The Ceph configuration was still present in the "DEFAULT" section, and simply commenting out "enabled_backends = LVM" and restarting the cinder services made it work again, but of course Juju may revert this at any time.
At first I thought I was required to deploy cinder-ceph following the upgrade, and although this restored cinder to life, as far as I can tell from reading the charm release notes and consulting https://jujucharms.com/cinder/ the simple case of a single Ceph cluster directly related to cinder seems like it is intended to remain supported.
If attempting to reproduce, run the following (adjusted as needed to match the cinder config) before upgrading cinder to 17.11. The version of the charmhelpers code as used in 15.04 did not set request-id as expected and required by the later versions, and so the 17.11 version of the ceph charm will not process the request.
juju run --unit cinder/0 'relation-set -r $(relation-ids ceph) broker_req="{\"api-version\": 1, \"request-id\": \"'$(uuid -v1)'\", \"ops\": [{\"replicas\": 3, \"name\": \"cinder\", \"op\": \"create-pool\"}]}"'
I also tried breaking and remaking the cinder<->ceph relation, to no effect.
This also affects a fresh icehouse deploy with the 17.11 charms.
Here's a fix (I forgot, yet again, that changes go through Gerrit. I'll look into that): https:/