2 Replies Latest reply on Mar 11, 2011 4:36 PM by Paul Ferraro

    JBOSS Clustering failure detection

    Luis Conceicao Newbie

      Hi,

       

      We are planning to use JBOSS cluster with 2 nodes. Suppose we have the following scenario in the network:

      Node A fails communication with Node B but all the cluster clients can communicate with both nodes. How will  the cluster behave? Will both nodes detect communication and become master? Will JBOSS handle any conflict when both (Node A and Node B) communicate again with each other?

       

      Thanks for you help.

        • 1. JBOSS Clustering failure detection
          Wolf-Dieter Fink Master

          It depends

           

          Ok, a bit more detailed.

          If the cluster disintegrate, nevertheless how many instances / parts you have, every 'new' cluster will have a master and find correct together again after the network environment is correct.

          But,

          - for EJB it depends on the client view to which part a client stay connected, mean from which server the client will get the 'cluster view'.

          - HTTP will not affected because there is no automatic load balancing and failover, here it depends on the load balancer in front of

          - difficult thing is your application if you use internal sharing of states, here you have to think about

           

          another risk is that the network is wrong configured, I've had this , the instances will see something (most of multicast packages are lost) but build not a cluster and the internal cluster state change sometimes (extremly was a cluster without or with multiple coordinators).

          Fortunately our application share no state, except the database and backend systems, and the clients are still working but the load is not balanced.

          • 2. JBOSS Clustering failure detection
            Paul Ferraro Master

            Wolf-Dieter Fink wrote:

             

            - HTTP will not affected because there is no automatic load balancing and failover, here it depends on the load balancer in front of

            Actually, mod_cluster will attempt to remove nodes from the load balancer upon jgroups membership change.  However, the coordinator will first validate that each node detected to have left is no longer reachable via the load balancer prior to issuing a removal.  This was done specifically to handle the cluster split scenario.