Messages sometimes not sent when nova fails to accept "extend volume" events

[Eric Fried]

This was noticed by inspection.

When a volume is extended, cinder sends events to nova through the os-server-external-events API [1]. Nova is supposed to react to those events by doing whatever's necessary to reflect the new size to the instance(s).

When something goes wrong with the os-server-external-events API, cinder sends a user message [2]:

    NOTIFY_COMPUTE_SERVICE_FAILED = (
        '009',
        _("Compute service failed to extend volume."))

Prior to the fix for bug 1855752 in nova [3], this message would *not* be sent under the following conditions:

- The volume is attached to multiple instances; and
- The event was rejected for a *subset* of those instances.

After the fix, the conditions will extend to include:

- The event was rejected for *all* instances, even if that's just one instance.

[1] https://opendev.org/openstack/cinder/src/commit/6088d115e27407a756851456587c2bccb3725464/cinder/compute/nova.py#L145
[2] https://opendev.org/openstack/cinder/src/commit/6088d115e27407a756851456587c2bccb3725464/cinder/compute/nova.py#L221-L226
[3] https://review.opendev.org/#/c/698037/

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/698340