failover transport and rebalanceClusterClients=true should fall back to original broker list if no broker on updated cluster information is available.

A workaround is to not use rebalanceClusterClients and updateClusterClients . However this can lead to an unfair distribution of clients across brokers in a cluster in particular cluster environments. E.g. in a cluster of two master/slave broker pairs where master brokerA is shut down, all clients will reconnect to the second master brokerB even before the slaveA broker has become the new masterA broker. If clients use long-lived connections - as recommended - they will all remain connected to master brokerB and no further load balancing will happen.