Hi all,

We are facing a mysterious issue where EJB method invocations are throwing exceptions about failing to connect to some other server even when we did not have a cluster formed between them.

We are pretty sure that although clustering was enabled on each of the servers, the configurations were such that no two jboss servers can group together and form a cluster. We have around  a dozen JBoss servers deployed in the same way in our lab.

Here are the details:

Server-1: pdev21:

     JGoups multicast address :  230. + "last three fragments of the ip address"

ClusterPartition Name:

     fully qualified hostname of pdev21

Server-2 : pdev33:

     JGoups multicast address :  230. + "last three fragments of the ip address"

ClusterPartition Name:

     fully qualified hostname of pdev33

Both the servers are started with -b 0.0.0.0 option; however, in the boot log I'm seeing that the jgroups.bind_address=fully qulified hostname of the respective server.

We have confirmed that cluster is not formed between any of the machines by verifying the ClusterView attribute of the MBean named: org.jboss.ha.framework.server.ClusterPartition.

When an EJB is looked up with Context.PROVIDER_URL="localhost:1399" or with out any provider url, things are working fine and the EJB method invocations are happening properly.

But if the provider url is using any other port (we used 1099, by mistake, and noticed this issue) apart from 1399, the EJB method calls are getting routed to some fixed server, lets say pdev29.

We noticed the same behaviour in all of our servers in the subnet; even when the "MAGNET" server (pdev29) is shutdown, EJB method calls to any of the servers are getting directed to the same one all-attracting server. We checked that there is no hrd coded references to this server in any of the configuration files or in the code. In fact, occasionally we noticed that some method incations have attempted to connect to one other server, but the percentage is very low.

When the same depoloyment is made in another subnet, we noticed that method calls to all the JBoss servers in that subnet are getting routed to one single server , lets say demo1, with in that subnet.

It is very puzzling why all the traffic get routed to some other server, when there is no clustering formed between these two.

It is true that the EJB calls on correct port (1399) are working fine but we are wondering why this strong affinity to one single server, from among many other deployments and whether this would cause any unforeseen isuses in thhe future.

Your help is highly appreciated.

Thanks and Regards,
Samba