Description
With a situation like HBASE-3258, it's easy to have the region servers stuck on waiting for either the master or the cluster state znode since it has no timeout. You have to kill -9 them to have them shutting down. This is very bad for usability.