5 Replies Latest reply on Dec 8, 2008 3:48 PM by Bill Robinson

    Clustered DataStore Servers



      We have a clustered datastore. Two servers both accessing the same area of shared SAN storage (not at the same time), with cluster IP's configured that failover to whichever node is active. The following services are configured to run also on the active nodes... (DHCP, PXE and TFTP, and NFS, HTTP and SMB). I've discovered however, that if the clustered IP is brought online the PXE server stops working. For some reason when it binds to the clustered IP address bare metal clients fail to boot with the message "PXE-E55 ProxyDHCP service did not reply to request on port 4011". If I down the cluster IP and restart the PXE / DHCP services it then works.


      Our cluster consists of two RedHat servers running Veritas Storage Foundation HA 4.1. I need to make the datastore services highly available, but obviously this is going to break the PXE server for whatever reason. Does anyone know why this might prevent PXE from working properly? I've attached a basic diagram showing the topology of the datastore cluster...

        • 1. Re: Clustered DataStore Servers
          Bill Robinson

          if you start the pxe server after the clustered ip is online, does it work? and when does this actually fail, during failover, and how does failover occur - ip gets bound by the 2ndary nic, then the pxe service starts up?


          and pxe is config'd right to co-exist w/ a dhcp server ? (the option 60 and a setting in the pxe server start)?

          • 2. Re: Clustered DataStore Servers

            The setup is that each machines in the cluster has a NIC connected to our build LAN. Whichever node in the cluster is active and running PXE brings up a virtual IP on the same interface.


            So for example, our primary node's eth0 interface is, and the virtual IP which is clustered and moves between machines depending upon which one is active is and gets brought up as eth0:0 on the active node.


            If PXE is brought online after this eth0:0 interface is brought up it doesn't work and we get the PXE-E55. However, I found that if I manually brought up PXE first and then the clustered interface it does work, but of course PXE is only bound to the IP of eth0 and not eth0:0.


            I tried changing the pxe.conf so that the interface_to_bind is set to eth0:0 but it doesn't seem to make a difference.


            However since posting this message I did find that if I set the interface in pxe.conf to "all" it does work, however the problem then is that PXE answers requests on the public LAN too which we don't want.


            Although I'm looking at implementing some iptables rules to prevent PXE packets being sent out on all but the build LAN interfaces.

            • 3. Re: Clustered DataStore Servers
              Bill Robinson

              file a defect w/ support on the eth0:0 binding thing if you can.


              does the 'all' work in the PM gui config also?

              • 4. Re: Clustered DataStore Servers

                I believe "all" is valid when input into the PM Config options GUI menu yes.

                • 5. Re: Clustered DataStore Servers
                  Bill Robinson

                  i tried this on a virtual machine - rhel4 - what i get is that it always binds to any active interfaces i have, regardless of what's set in the db or the pxe.conf file. so i think limiting access to the port via iptables is the best best...