Traffic loss on different subnets when ClusterXL interface fails back after a failure, or after a cable is reconnected
In some cases, Cluster Control Protocol (CCP) packets might arrive few seconds after interface changed its state to 'UP'. During that period, all cluster members have interface with link up, but they are not able to hear each other (receive each other's CCP packets).
By design, when a cluster member does not receive CCP packets from a peer member (or not able to send its own CCP packets), cluster member uses probing mechanism in order to determine the problematic interface and the problematic member. All cluster members send series of ARP Requests and series of ICMP Requests to all hosts on the subnet.
Since interfaces on all cluster members are OK, these series of ARP Requests and ICMP Requests might confuse ARP tables on Layer 2 / Layer 3 devices.