none
Adding New LUNs to the Cluster

    Question

  • Hi,

    I have a W2K3 2-node a/p cluster attached to a SAN.  I need to add new LUNs to the cluster, and would like to do this without bringing down the cluster.  Here's how I'm thinking of doing it --

    1.  shut down passive-node
    2.  map LUNs to passive-node
    3.  bring up passive-node
    4.  on passive-node:  format the new disks, assign drive letters
                                   add new physical disk resources, configure resource dependencies
                                   add the new disk resources to a test resource group
    5.  failover all resources (to the node that now has the new disks)
    6.  shut down the now passive-node
    7  map LUNs to passive-node
    8.  bring up passive-node
    9.  on passive-node:  verify that the new disks can be seen, assign drive letters to match the other node
    10. failover the test resource group to make sure the new disks failover successfully, then failback
    11. move the new disk resources into the appropriate resource group, then delete the test group

    A question about this procedure -- Will I be able to add in the new disk resources into the cluster from the passive node before the other node has even seen the new disks?  I'm asking because everything I've read seem to indicate that access to the new disks should be verified from both nodes, then the disk resources added into the cluster.  Does the cluster service require that both nodes be aware of the new disks before it allows them to be added in as resources?

    Thanks in advance.

    Friday, May 29, 2009 1:45 PM

Answers

  • Q: Will I be able to add in the new disk resources into the cluster from the passive node before the other node has even seen the new disks?

    A: Probably. 2003 cluster does not require that all nodes see the disk before adding it to the cluster. I say "probably" because I have seen many occasions where cluster will not recognize the new signature until you reboot again (and again...). To work around this, you can review http://support.microsoft.com/kb/555312

    Q: Does the cluster service require that both nodes be aware of the new disks before it allows them to be added in as resources?

    A: No, it technically doesn't but it is a REALLY GOOD IDEA to do so. What I have also seen on many occasions is that when you attempt to add a disk as you've described, the other node of the cluster does not update its cluster disk registry hive so when you attempt to failover to the newly added disk, it fails to come online on the other node.


    FYI, in Windows 2008, this process is a bit easier. Newly presented disks come into the host in an "offline" state. This way, you can present the LUN to all nodes at the same time, and then just online/offline on each node one at a time to verify the disk before adding it into cluster.

    Also FYI, in Windows 2008 cluster, it would NOT let you add a disk into the cluster if it is not presented to all nodes...without forcing the disk into the cluster using the command line. One step forward, one step back ;-)

    Hope this helps.

    Visit my blog about multi-site clustering - http://msmvps.com/blogs/jtoner
    Tuesday, June 02, 2009 9:58 PM
    Moderator

All replies

  • Q: Will I be able to add in the new disk resources into the cluster from the passive node before the other node has even seen the new disks?

    A: Probably. 2003 cluster does not require that all nodes see the disk before adding it to the cluster. I say "probably" because I have seen many occasions where cluster will not recognize the new signature until you reboot again (and again...). To work around this, you can review http://support.microsoft.com/kb/555312

    Q: Does the cluster service require that both nodes be aware of the new disks before it allows them to be added in as resources?

    A: No, it technically doesn't but it is a REALLY GOOD IDEA to do so. What I have also seen on many occasions is that when you attempt to add a disk as you've described, the other node of the cluster does not update its cluster disk registry hive so when you attempt to failover to the newly added disk, it fails to come online on the other node.


    FYI, in Windows 2008, this process is a bit easier. Newly presented disks come into the host in an "offline" state. This way, you can present the LUN to all nodes at the same time, and then just online/offline on each node one at a time to verify the disk before adding it into cluster.

    Also FYI, in Windows 2008 cluster, it would NOT let you add a disk into the cluster if it is not presented to all nodes...without forcing the disk into the cluster using the command line. One step forward, one step back ;-)

    Hope this helps.

    Visit my blog about multi-site clustering - http://msmvps.com/blogs/jtoner
    Tuesday, June 02, 2009 9:58 PM
    Moderator
  • Thanks John, appreciate it!
    Monday, June 08, 2009 3:12 PM
  • These steps might be better? Assuming node1 is the Active and Windows could see the new SAN without reboot.

    1. Map the new LUN to Node 1.
    2. Rescan for the new disk, format and assign a drive letter.
    3. Failover the groups to Node2. Don't add the new disk to cluster yet.
    4. Shutdown Node1. This is to ensure no simultaneous access to the disk from both nodes.
    5. Map the new LUN to Node 2.
    6. Rescan for the new disk, assign the same drive letter.
    7. Add the new drive to the cluster. May need use the command line. http://support.microsoft.com/kb/555312
    8. Start up Node 1.
    9. Fail back all resources to Node1.

    Microsoft's KB for adding a new SCSI disk looks too complex. http://support.microsoft.com/kb/175278. It requires disabling the cluster service and cluster disk driver before shutting down the servers. And I never understand why the OS needs to be paused instead of fully shutdown. I suppose all these steps are to ensure they are compatible with old NT4 and non-hot swap disks. But should MS publish a new KB for Win2k3 and SAN?

    • Proposed as answer by satboy Monday, July 13, 2009 10:00 PM
    Monday, July 13, 2009 9:58 PM
  • How would be go about a 4 node cluster

    Thursday, February 24, 2011 10:06 PM