I have a cluster of 6 machines. Each of those has 2 network interfaces, one "uplink" to the outside world and one on which the cluster should do its messaging.
As far as cluster and ha-jndi message go it's easy to configure the bind address so that all communications is over the second "cluster" network interface.
However RMI traffic is always routed over the uplink interface even for rmi calls between 2 cluster nodes. The scenario is as follows: I have a client application outside the cluster that calls a bean that is located on one cluster node, which in turn calls another bean that is on another cluster node. The problem now is that the communication between the 2 beans is over the uplink network interface while I would like it to go over the cluster network interface.
Does any one have a possible solution?
I think the only way around this is to rewrite the HA-JNDI service so that it checks if a call comes from a cluster node or not and returns the correct interface. I haven't really looked in to the code if this is difficult or not, but I assume it will be rather timeconsuming. So I'm hoping there is another way.