7 Replies Latest reply on Oct 21, 2003 10:59 AM by Claus Kolb

    Clustering and Non-multicast JavaGroups config

    Claus Kolb Newbie

      Hello from Germany,

      I am desperately trying to get a jboss 3.2.1 cluster working on windows 2000. I hvae 2 jboss server residing on two machines having a stateless session bean service deployed. I want this service clustered with load-balancing and failover. I do NOT want to use multicast since I want to avoid having test and production server being in the same cluster accidentally (I want to have control over cluster members).

      I am using the JavaGroups config recommended in the JBoss Clustering documentation (2nd Ed):

      Server 1
      <TCP start_port="7800" />
      <TCPPING initial_hosts="FSDEHAMWS032[7800]"
      port_range="5" timeout="3000" num_initial_members="2"
      up_thread="true" down_thread="true" />
      <VERIFY_SUSPECT timeout="1500"
      up_thread="false" down_thread="false" />
      <pbcast.STABLE desired_avg_gossip="20000"
      up_thread="false" down_thread="false" />
      <pbcast.NAKACK gc_lag="100" retransmit_timeout="3000"
      up_thread="true" down_thread="true" />
      <pbcast.GMS join_timeout="5000" join_retry_timeout="2000"
      shun="false" print_local_addr="false" up_thread="true" down_thread="true" />

      Server 2

      <TCP start_port="7800" />
      <TCPPING initial_hosts="FSDEHAMWS007[7800]"
      port_range="5" timeout="3000" num_initial_members="2"
      up_thread="true" down_thread="true" />
      <VERIFY_SUSPECT timeout="1500"
      up_thread="false" down_thread="false" />
      <pbcast.STABLE desired_avg_gossip="20000"
      up_thread="false" down_thread="false" />
      <pbcast.NAKACK gc_lag="100" retransmit_timeout="3000"
      up_thread="true" down_thread="true" />
      <pbcast.GMS join_timeout="5000" join_retry_timeout="2000"
      shun="false" print_local_addr="false" up_thread="true" down_thread="true" />



      The problem:
      It is not working at all: The two jboss server do not know each other and are far from building a cluster. When I use the default UDP-multicast-configuration everything (load-balancing, failover) is working fine.

      Thanks for your help,
      ckmod

        • 1. Re: Clustering and Non-multicast JavaGroups config
          Bela Ban Master

          In initial_hosts, you need to have *both* hosts listed
          Bela

          • 2. Re: Clustering and Non-multicast JavaGroups config
            Claus Kolb Newbie

            Hi Bela,

            this does not work either. The servers still do not see each other.

            • 3. Re: Clustering and Non-multicast JavaGroups config
              Bela Ban Master

              Show your prot spec here. Also, does the member address show 'localhost' ?
              Bela

              • 4. Re: Clustering and Non-multicast JavaGroups config
                Claus Kolb Newbie

                I am sorry for the delay but other things kept me busy...

                I am not sure what you mean. In the meanwhile I have two servers working on the same machine.

                Protocol Stack 1


                <TCP start_port="7900" />
                <TCPPING initial_hosts="FSDEHAMWS007[7800]"
                port_range="5"
                timeout="3000"
                num_initial_members="2"
                up_thread="true"
                down_thread="true" />
                <VERIFY_SUSPECT timeout="1500"
                up_thread="false"
                down_thread="false" />
                <pbcast.STABLE desired_avg_gossip="20000"
                up_thread="false"
                down_thread="false" />
                <pbcast.NAKACK gc_lag="100"
                retransmit_timeout="3000"
                up_thread="true"
                down_thread="true" />
                <pbcast.GMS join_timeout="5000"
                join_retry_timeout="2000"
                shun="false"
                print_local_addr="false"
                up_thread="true"
                down_thread="true" />



                Protocol Stack 2


                <TCP start_port="7800" />
                <TCPPING initial_hosts="FSDEHAMWS007[7900]"
                port_range="5"
                timeout="3000"
                num_initial_members="2"
                up_thread="true"
                down_thread="true" />
                <VERIFY_SUSPECT timeout="1500"
                up_thread="false"
                down_thread="false" />
                <pbcast.STABLE desired_avg_gossip="20000"
                up_thread="false"
                down_thread="false" />
                <pbcast.NAKACK gc_lag="100"
                retransmit_timeout="3000"
                up_thread="true"
                down_thread="true" />
                <pbcast.GMS join_timeout="5000"
                join_retry_timeout="2000"
                shun="false"
                print_local_addr="false"
                up_thread="true"
                down_thread="true" />

                The server see each other, failover is working.

                BUT: The calls from the client application do not load balance

                Here is the clients jndi.properties
                java.naming.factory.initial=org.jnp.interfaces.NamingContextFactory
                java.naming.provider.url=10.0.144.77:1100,10.0.144.77:1200
                java.naming.factory.url.pkgs=org.jboss.naming:org.jnp.interfaces
                jnp.timeout=3000
                jnp.sotimeout=3000

                • 5. Re: Clustering and Non-multicast JavaGroups config
                  Claus Kolb Newbie

                  Here are my cluster logfiles of both servers.

                  • 6. Re: Clustering and Non-multicast JavaGroups config
                    Bela Ban Master

                    Regarding client calls not load balancing: you said client calls do failover (when server1 crashes, they get redirected to server2 right). Load balancing uses afaik a sticky LB policy, calls go to server1 as long as server1 is up. You can change the LB policy to use for example random robin, or round robin.
                    Bela

                    • 7. Re: Clustering and Non-multicast JavaGroups config
                      Claus Kolb Newbie

                      The failover might be a result of the client knowing both servers in his jndi.properties (?).

                      I have specified the load balancing in jboss.xml


                      <ejb-name>MyService</ejb-name>
                      <jndi-name>ejb/MyService</jndi-name>
                      <local-jndi-name>MyServiceLocal</local-jndi-name>
                      true
                      <cluster-config>
                      <partition-name>MyServicePartition</partition-name>
                      <home-load-balance-policy>org.jboss.ha.framework.interfaces.RoundRobin</home-load-balance-policy>
                      <bean-load-balance-policy>org.jboss.ha.framework.interfaces.RoundRobin</bean-load-balance-policy>
                      </cluster-config>



                      I will use different names for the different partitions: DevPartition, TestPartition and ProdPartition. This should avoid that servers from development, test and production env accidentally see each other.

                      But it is still weird why we can't bring this protocol stack to work as we want to...

                      Thank you very much for your help, Bela !