Greetings and salutations. We’re investigating whether our Keycloak nodes are leaving the cluster cleanly during deploy shutdown.
Setup: 3 individual VM hosts, each running one Keycloak 26.4.2 Docker container. The nodes discover each other through DB/JDBC_PING, but runtime cluster communication is over Infinispan/JGroups between hosts.
I confirmed the stopping node logs:
[com.arjuna.ats.jbossatx] (main) ARJUNA032014: Stopping transaction recovery manager
So it appears to receive the shutdown signal and enter normal shutdown. However, another node logs this when one node is brought down:
host-01-2913: failed sending message to host-02-18951: java.net.ConnectException: Connection refused
Question: is that transient Connection refused error level message expected during a normal JGroups/Infinispan view change, or does it suggest the node is closing its transport before cleanly leaving the cluster?