To Reproduce Steps to reproduce the behavior:
Customer is running RHOSP 17.1.4 (upgraded from RHOSP 16.2) with cinder-backup resource managed by pacemaker. When cinder-backup resource is migrated (and original cinder-backup instance is marked as "down" in "openstack volume service list --service cinder-backup --long" output), cinder-scheduler processes continue to schedule backup creation to original cinder-backup instance (so some backups are scheduled on new host and some backups are scheduled on old ones).

This problem can be solved by cinder-scheduler restart (at this point problem is triggered by overcloud control plane backup, so it is possible to predict when it will happen and apply workaround). But this is time consuming + it is not going to address situations when resource is migrated for other reasons and migration wasn't tracked by cloud operator.

Expected behavior
Cinder-scheduler doesn't schedule backup creation on cinder-backup instances in down state.

Bug impact
Extra effort is needed to run RHOSP control plane with this problem.

Known workaround
Manual workaround exists, but it is time consuming + it is not going to address situations when resource is migrated for other reasons and migration wasn't tracked by cloud operator.

Additional context
Outputs and information about data will be provided privately.