yarn.resourcemanager.cluster-id is set to a constant. This value is visible and correct on each RM's /conf page; however, the "Cluster ID" on /cluster/cluster page is a timestamp. Each RM has a different "Cluster ID" on /cluster/cluster. Similarly, the cluster id that is embedded in application id's is the one that matches /cluster/cluster.
The ZK fence for HA seems to be working as expected. Are these two cluster ids the same thing, or is the term cluster id overloaded?