There are two mechanisms through which WebLogic can detect server failures in a cluster:
Monitoring IP socket connections to a peer server
Sending and receiving heartbeat messages
When a server connects to one of its peers in a cluster through an IP socket, WebLogic Server monitors errors on that socket. Should there be an unexpected closure, the peer server is marked as failed and all of its associated services are removed from the cluster-wide JNDI naming tree.
If the clustered server instances are not using open sockets for peer-to-peer communication, failed servers can still be detected via the heartbeat. All server members in a cluster can use multicast to broadcast and receive heartbeat messages between other servers in the cluster. The default regular intervals are set at 10 seconds. If there are three consecutive missed heartbeats (30 seconds), the peer server is marked as failed and all of its associated services are removed from the clusterwide JNDI naming tree.