The Kudu/HMS integration currently assumes that any Kudu table entry in the HMS (as identified by the Kudu storage handler) belongs to the Kudu cluster instance. It seems increasingly likely that we will need to support deployments where a single HMS is shared among multiple Kudu cluster instances. In order to disambiguate the HMS entries, we could introduce a unique per-cluster ID and add it to the HMS entries. The cluster ID would be written durably to the master's sys catalog during startup. Eventually we would want tools and client APIs to inspect the cluster ID.