4 Replies Latest reply: Aug 19, 2013 5:21 PM by Dude! RSS

    OCFS and fencing

    user10177484

      Hello,

       

      I have two nodes OEL 5.8, and i configured ocfs cluster for application server. Shared volume is located on SAN and configured via multipath. The OCFS2 volume is available on both nodes.if

       

      Problem:

       

      If I stop the network service on node1.

        node2 get fence and rebooted, after this OCFS volume is not mounted and i cann't mount also.  (even though it is configured in Fstab).

       

       

      if i stop the network service on node2.

        Nothing happens,  Node1 don't fence.

       

       

      Can any one help on this issue.

       

      thank

        • 1. Re: OCFS and fencing
          Billy~Verreynne

          What is the contents of /etc/ocfs2/cluster.conf - and do the 2 nodes have identical files?

           

          How do you stop the network service with your testing? What is the output of ifconfig after stopping the network service?

          • 2. Re: OCFS and fencing
            user10177484

            Hello

             

            To test cluster, I am stopping the service,

            root#service network stop


            and here is the configuration on both nodes


            node:
                    ip_port = 7777
                    ip_address = 192.168.0.32
                    number = 0
                    name = prodapps1.mydomain.local
                    cluster = ocfs2

            node:
                    ip_port = 7777
                    ip_address = 192.168.0.33
                    number = 1
                    name = prodapps2.mydomain.local
                    cluster = ocfs2

            cluster:
                    node_count = 2
                    name = ocfs2

             

             

            thanks,

            • 3. Re: OCFS and fencing
              Billy~Verreynne

              I cannot see any obvious issues from the config and testing side.

               

              Next step in my view is to stop the network on node 1, and monitor the kernel log for ocfs2 messages.

               

              Reset both nodes and then repeat for node 2, where the network is stopped and both kernel logs are monitored.

              • 4. Re: OCFS and fencing
                Dude!

                Fencing and not mounting the OCFS volume are not necessarily due to the same reason or cause. You may have a stability, performance or networking issue that can lead to a cluster fencing situation. If a node does not mount the OCFS volume after a reboot, perhaps you have not configured persistent device naming using multipath or udev and therefore the order of devices and device names can change after a reboot of the system and /etc/fstab can mount the wrong volumes. You can rely on /etc/fstab only, but then you should use UUID or label for mounting and not device names.