Clustering JMS

It is possible to configure connection factories to run in a clustered environment, but it is sort of a headache and the value is limited. Producers and consumers are clusterable resources. If you have deployed a producer or consumer on each server in the cluster, the producer or consumer will be available as long as at least one server is up.

To distribute a consumer or producer, use the Targets tab to associate a connection factory with two or more servers.

If, however, the server on which the destination (queue or topic) is housed goes down, then the producers and consumers will stop working. The destination is tied to a server by either RAM (in the case of nonpersistent messaging) or disk space (in the case of persistent messaging). Either way, when you lose the server you lose the destination.

Less debilitating but still occasionally a problem is the propagation delay involved in replicating JNDI entries across multiple servers. If a method on one server references a remote destination, there is the potential for a javax.naming.NamingException. This may occur in situations such as server crashes or normal rebooting in which the JNDI on the remote server housing the resource is not yet synchronized with the local machine. This problem was fixed in WLS 6.1 and above by having connections be retried until they become available, so it will be seen less frequently as time goes by.