Description of problem: Cinder-volume and cinder-backup are both controlled by Pacemaker and both services are always running on single controller node (there is single instance of each service). For cinder-volume backends we generally set "backend_host=hostgroup" and we get consistent picture of "openstack volume service list": despite migrations there is a single instance of cinder-volume service per backend, which is up. Situation is different for cinder-backup: due to the way cinder works or the way TripleO provisions it (I haven't isolated the root cause), a dedicated cinder-backup service is created for every controller node. As a result, there is cinder-backup service in down state when cinder-backup service is migrated from one controller node to another. I am reporting this bug to ask engineering to check if this cosmetic issue could be resolved, or there is no way to change current behavior.
Sorry, at least for now "that's the way it works." cinder-volume supports the notion of a backend_host, but the cinder-backup service doesn't have backends (only drivers). But see bug #1666804, which is intended to address this situation, and bug #1665191 for more background. *** This bug has been marked as a duplicate of bug 1665191 ***