https://sourceforge.net/tracker/?func=detail&aid=3078911&group_id=152762&atid=785194

From [http://www.zkoss.org/forum/listComment/11057]

Currently in a clustering environment, ZK does not "detects" when a cluster node fails or is not longer available. The application "stops" working and the only way to let zk know is to hit the reload button for the failover to work properly.

Zk could have a way to detect this and reload the page automatically, soo everything keeps on working "normally" assuming that the cluster, session replication etc. its correctly configured.