Forum Discussion

Hywel_Mallett's avatar
14 years ago

Splitting a cluster into two

Hi all,

I look after a four-node cluster (using SFWHA). We now need to move some servers into a separate network segment, so I need to split the cluster into 2 two-node clusters.

Has anyone done a checklist of things that I need to do, and things I need to check?

So far I've got:

  • Modify service groups so that they are either on node 0 + 1 or node 2 + 3
  • Modify llttab on nodes 2 + 3 to change cluster number (to current number + 1) and node number (to 0 + 1)
  • Modify llthosts on all nodes to remove entries about the other cluster
  • Edit gabconfig on each node to change number of nodes required for cluster startup
  • Modify main.cf on nodes 0 + 1 to remove all references to nodes 2 + 3
  • Modify main.cf on nodes 2 + 3 to remove all references to nodes 0 + 1
  • Change Virtual IP addresses on 2 + 3
  • Modify Windows networking and move 2 + 3 to new network segment

At this point I'm not actually splitting the storage - I'll do that later.

What else am I forgetting? I've got a VCS certification, but I took it about 5 years ago!

  • Hello,

       It looks like you have pretty much everything covered. The only item I see missing is changing the cluster name (so they don't match) in the main.cf as well. If you have configure the 'ClusterAddress' attribute, you'll need to change the IP there as well.

    Here is an example:

    cluster FileCluster01
              Usernames = {admin = xxxxxx}
              ClusterAddress = 192.168.0.5

    Here, you would want to perhaps change the 2nd cluster to name FileCluster02, and if the ClusterAddress attribute is configured, the IP here should match the IP configured in your Cluster Service group. This is attribute is generally used for the Global Cluster Option (GCO), so there is a very good chance this attribute has NOT been configured.

    Lastly, as I'm sure you are aware -- After making the changes, you will need to stop had, llt, and gab (net stop had, net stop llt, net stop gab), then restart them to pick up the config changes manually made.

    Hope this helps,
    rhanley

  • Hello,

       It looks like you have pretty much everything covered. The only item I see missing is changing the cluster name (so they don't match) in the main.cf as well. If you have configure the 'ClusterAddress' attribute, you'll need to change the IP there as well.

    Here is an example:

    cluster FileCluster01
              Usernames = {admin = xxxxxx}
              ClusterAddress = 192.168.0.5

    Here, you would want to perhaps change the 2nd cluster to name FileCluster02, and if the ClusterAddress attribute is configured, the IP here should match the IP configured in your Cluster Service group. This is attribute is generally used for the Global Cluster Option (GCO), so there is a very good chance this attribute has NOT been configured.

    Lastly, as I'm sure you are aware -- After making the changes, you will need to stop had, llt, and gab (net stop had, net stop llt, net stop gab), then restart them to pick up the config changes manually made.

    Hope this helps,
    rhanley

  • I am more comfortable splitting and joining clusters on Unix clusters, since all config info is held in text files. Seems Windows is a mix of Registry and text files.

    This TN on how to rebuild a cluster node : http://www.symantec.com/docs/TECH49451

    lists these Registry keys:

    HKEY_LOCAL_MACHINE\SOFTWARE\VERITAS\VCS\Base\ClusterName
    HKEY_LOCAL_MACHINE\SOFTWARE\VERITAS\VCS\Base\ClusterID
    HKEY_LOCAL_MACHINE\SOFTWARE\VERITAS\VCS\Base\NodeID

     

    Depending on your VCS version, you can also use this TN as guideline: http://www.symantec.com/docs/TECH33036

    Symantec also recommends to use lltconfig to modify llt links.  It was removed in 4.3, but can be downloaded from: http://www.symantec.com/docs/TECH43662

     

    If at all possible to keep heartbeats connected, I'd rather stick with Riaan's recommendation.

  • In the end I decided to implement Riaan's solution, as it makes things much simpler.

    I'll give the points to rhanley though, as he/she answered the question and thought of a few things I'd forgotten about.

    Thanks all!