Now if ALL agents participating in the backup goes down, PBM observes stale lock only during the next backup/restore run. An only during the next backup/restore that operation would be marked as failed.

The other problem is that any action needed after the backup finished won't be ever run. For example, PBM turns the balancer off before backup and it should be turned on after. But if ALL agents are gone - no one would bring balancer back.

So maybe we need some vacuum process (GC) which would periodically check locks and apply necessary actions (restart balancer, mark backup/restore etc.) if any stale locks found.