Changing IP addresses and host names

GPFS assumes that IP addresses and host names remain constant. In the rare event that such a change becomes necessary or is inadvertently introduced by reinstalling a node with a disk image from a different node for example, follow the steps in this topic.

If you use a federation type configuration, and the affected node is a collector,

If all of the nodes in the cluster are affected:
  1. Issue the mmshutdown -a command to stop GPFS on all nodes.
  2. Using the documented procedures for the operating system, add the new host names or IP addresses but do not remove the old ones yet. This can be achieved, for example, by creating temporary alias entries in /etc/hosts. Do not restart the nodes until the mmchnode command in Step 3 is executed successfully. If any of these conditions cannot be met, use the alternate procedure described in this section.
  3. Update the nodes that the cluster uses as the administration interface node and the daemon interface node, if necessary. To update these values, issue the mmchnode command with the --admin-interface and the --daemon-interface options.
    Note: You cannot specify the --daemon-interface option for a quorum node if CCR is enabled. Temporarily change the node to a nonquorum node. Then issue the mmchnode command with the --daemon-interface option against the nonquorum node. Finally, change the node back into a quorum node.
  4. If the IP addresses over which the subnet attribute is defined are changed, you must update your configuration by issuing the mmchconfig command with the subnets attribute.
  5. Start GPFS on all nodes with mmstartup -a.
  6. Remove the unneeded old host names and IP addresses.
If only a subset of the nodes are affected, it may be easier to make the changes using these steps:
  1. Before any of the host names or IP addresses are changed:
    • Use the mmshutdown command to stop GPFS on all affected nodes.
    • If the host names or IP addresses of the primary or secondary GPFS cluster configuration server nodes must change, use the mmchcluster command to specify another node to serve as the primary or secondary GPFS cluster configuration server.
    • If the host names or IP addresses of an NSD server node must change, temporarily remove the node from being a server with the mmchnsd command. Then, after the node has been added back to the cluster, use the mmchnsd command to change the NSDs to their original configuration. Use the mmlsnsd command to obtain the NSD server node names.
    • If the affected node is a CES node, CES must be disabled from the node using the mmchnode -N <node> --ces-disable command.
    • Unless all nodes in the cluster are being deleted, ensure that the mmdelnode command is run from a node that remains in the cluster.
  2. Change the node names and IP addresses using the documented procedures for the operating system.
  3. If the IP addresses over which the subnet attribute is defined are changed, you need to update your configuration by using the mmchconfig command with the subnets attribute.
  4. Issue the mmaddnode command to restore the nodes to the GPFS cluster.
  5. If necessary, use the mmchcluster, mmchlicense, and mmchnsd commands to restore the original configuration and the NSD servers.
    Note: You can use the mmchnode command if you need to re-enable CES on the node.
Note: When you change your cluster node names and IP addresses, ensure that the performance monitoring configuration file is also changed accordingly. To change the performance monitoring configuration file, follow these steps:
  1. Save the current configuration file in a temporary file using the following command:
    mmperfmon config show --config-file <tmp_file_name>
  2. Change all occurrences of the old node name or IP address to the new one, using an editor.
    Important: If no changes are needed, then you do not need to run the update.
  3. Update the performance monitoring configuration information using the following command:
    mmperfmon config update --config-file <tmp_file_name>
If you use a federation type configuration, and the affected node is a collector, you need to change the names and IP addresses of the peers in the /opt/IBM/zimon/ZIMonCollector.cfg file on all collector nodes as well. Starting with IBM Spectrum Scale version 5.0.2, the peers section will be managed automatically.
Important: If the long admin node names (FQDN) of any call home group members were changed, the customer must delete the affected call home groups, and then create new ones if needed. Run the mmcallhome group list command to verify whether nodes that are members of a call home group are deleted, or their long admin node names (including domain) are changed. In such cases, the mmcallhome group list command displays ------ instead of the names of such nodes. For more information, see mmcallhome command.