OAK-3672 describes the fact that the (cluster) id in tarMk case is not persisted and thus doesn't survive server restarts. This breaks the discovery API contract since discovery.oak uses this id directly as the ClusterView.getId - and that must by definition be stable (across restarts that is).
The solution suggested on oak-dev list is to have tarMk return null for this id (while mongoMk still manages this id).
This implies that discovery.oak must now handle the fact that the discovery-lite id is null. In which case it should readOrDefine a clusterId at /var/discovery/oak.
This will auto-support cold standby/failover as well as not harm the repo-clone case (as this solution is equivalent to how discovery.impl is working).