resetting console-access-protocol does not propagate along relations to compute nodes

[Nick Moffitt]

We have an openstack stein deployment on Ubuntu Bionic which has compute nodes for four architectures: amd64, arm64, ppc64el and s390x. When we set console-access-protocol to 'spice' we started getting failures on the ppc64el and s390x compute nodes when spinning up new instances:

Failed to start libvirt guest: libvirt.libvirtError: unsupported configuration: spice graphics are not supported with this QEMU

We rolled back the console setting by running `juju config nova-cloud-controller --reset=console-access-protocol` but the problem persisted.

It turns out that doing the config rollback did not cause the relations to remove the spice configurations on the compute node nova.conf files. We had to manually remove the [spice] stanzas and restart services on the compute nodes.

I believe the correct behaviour would be for the config-changed back to '' or 'None' or whatever the charm has as its default state would be to communicate down the relation to the compute nodes that the setting has changed, and the compute nodes' hooks should remove the [spice] configuration and restart daemons.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/684547

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.opendev.org/684547
Committed: https://git.openstack.org/cgit/openstack/charm-nova-cloud-controller/commit/?id=dee96d620e80d035a5a2dd3bda47d5d9e12c549d
Submitter: Zuul
Branch: master

commit dee96d620e80d035a5a2dd3bda47d5d9e12c549d
Author: Liam Young <email address hidden>
Date: Wed Sep 25 07:41:41 2019 +0000

    Explicitly unset console_access_protocol

    If console_access_protocol was switched to none then
    console_access_protocol was not set at all by the relation set
    acting on the relation with the compute nodes. This meant the
    operator had no mechanism to unset console_access_protocol. This
    change explicitly unsets it so the change is passed to the
    compute units.

    Change-Id: Ifdca0ec6626732ee0ea30a6847130ea97478df6f
    Closes-Bug: #1844356

[Paul Collins (pjdc)]

Workaround (not tested in this precisely this form, but it should work):

for relid in $(juju run --unit nova-cloud-controller/0 -- relation-ids cloud-compute) ; do juju run --application nova-cloud-controller -- relation-set -r ${relid?} console_access_protocol= ; done