none
HPC 2008 Headnode/cluster name change question

    Question

  • I inherited a preconfigured cluster that was setup as a standalone domain.  Now I'd like to join it to the corporate domain and rename the headnode (and all of the compute nodes).  I already removed the AD services, renamed the headnode and joined the domain, but it would appear that the HPC CLuster Manager isn't completely happy about this.  Some operations fail in the configuration screens of Cluster Manager.

    I'm guessing that at this point, I need to uninstall and reinstall the HPC Pack on the headnode.  In order to minimize configuration losses, other than exporting the Job and Node Templates, are there any other pieces that can be preserved through the reinstall?

    Is there some other procedure I should have used up to this point and could continue with?
    Friday, October 31, 2008 8:55 PM

Answers

  • I am going to do exactly the same thing as you describe sometime this week, i.e. decomission existing standalone active dir, rename existing head and compute nodes then add to corp domain.
    My plan is to remove the cluster pack from all nodes, removing the SQL components also. Stop all SQL related services first otherwise it'll complain. The SQL services will still look for the original headnode hostname.
    I suppose you could just choose to create a new instance of the SQL server during the cluster pack reinstall as opposed to completely uninstalling  and deleting the \program files\ (x86)\Microsoft SQL Server folder, but we've had probs in the past reinstalling SQL with existing directory structures in place, so my preferred method is to completely wipe MS SQL and reinstall clean.
    Other than that once you've renamed your nodes, and added them to the existing corp domain, plus applied your node templates, ensure that %system root%\system32\drivers\etc\hosts have the new IP addresses that correspond with your corp network.
    If your old standalone domain used a different IP Address /subnet mask setup then there might be some remnants of those address in the host file. You may have to manually tidy that up.
    The cluster pack should add the new hostnames / IP addresses when you configure your network but double check that to make sure that this has happened.
    Cheers.
    • Marked as answer by msmoritz Thursday, April 16, 2009 1:17 PM
    Monday, November 3, 2008 11:13 AM

All replies

  • I am going to do exactly the same thing as you describe sometime this week, i.e. decomission existing standalone active dir, rename existing head and compute nodes then add to corp domain.
    My plan is to remove the cluster pack from all nodes, removing the SQL components also. Stop all SQL related services first otherwise it'll complain. The SQL services will still look for the original headnode hostname.
    I suppose you could just choose to create a new instance of the SQL server during the cluster pack reinstall as opposed to completely uninstalling  and deleting the \program files\ (x86)\Microsoft SQL Server folder, but we've had probs in the past reinstalling SQL with existing directory structures in place, so my preferred method is to completely wipe MS SQL and reinstall clean.
    Other than that once you've renamed your nodes, and added them to the existing corp domain, plus applied your node templates, ensure that %system root%\system32\drivers\etc\hosts have the new IP addresses that correspond with your corp network.
    If your old standalone domain used a different IP Address /subnet mask setup then there might be some remnants of those address in the host file. You may have to manually tidy that up.
    The cluster pack should add the new hostnames / IP addresses when you configure your network but double check that to make sure that this has happened.
    Cheers.
    • Marked as answer by msmoritz Thursday, April 16, 2009 1:17 PM
    Monday, November 3, 2008 11:13 AM
  • I combined what you suggested with what I had already done.  Unfortunately, had issues with the SQL Server uninstallation that took a bit to resolve.

    Now, when I try to configure the cluster, getting several Error 5005's when running the network configuration and can't continue the configuration.

    The full errors are:

    Error 5005: Failed to deserialize InstanceName using the definition UniqueStringConstraint. The attribute NodeNamesMustBeUnique is invalid.
    Error 5005: Failed to deserialize InstanceName using the definition UniqueStringConstraint. The attribute NetworkNamesMustBeUnique is invalid.
    • Edited by msmoritz Monday, November 3, 2008 6:08 PM
    Monday, November 3, 2008 5:59 PM
  • The resolution to the Error 5005 seems to be to remove the DHCP scopes for the Application and Private networks before running the cluster network configuration step.
    Monday, November 3, 2008 7:54 PM
  • I'm getting this after reinstalling HPC Pack and recreating the cluster/head node (at the network config step):

    Error 5005: Failed to deserialize InstanceName using the definition UniqueStringConstraint. The attribute NetworkNamesMustBeUnique is invalid.

    This machine was never a DHCP server, though, so what does it mean to "remove the DHCP scopes for the Application and Private networks"?

    Thanks,

     

    Thursday, July 7, 2011 7:18 PM
  • Hmm. In my case it seems a reboot solved the problem...
    Thursday, July 7, 2011 8:50 PM