How can one rename or remove a node from a GPFS cluster?
Occasionally administrators need to rename or remove a node from a GPFS cluster. Typically the operation can be done without causing an outage across the cluster. The "Changing IP addresses and host names" procedures linked in the related URL sections details the appropriate method to use for changing IPs or host names. The "Deleting nodes from a GPFS cluster" procedure also linked in the related URL section details the method to remove a node.
Sometimes the node is renamed or removed before performing the proper procedures and attempts afterwards to follow the procedure fail. Circumvention will be required to get proper cluster configuration again.
While there is no supported method for performing an operation incorrectly, there are some general guidelines you might try for circumventing a configuration error such as moving a node and renaming one outside of GPFS.
- Removing a node that no longer is in the cluster
1. Remove the secondary configuration if node is a secondary NSD server
mmchcluster -s ""
2. unmount filesystems from the primary node
3. Change the NSDs to be directly attached using the mmchnsd command.
4. Use the mmdelnode command to remove the old secondary configuration
- Adding a renamed node back into the cluster
5. Verify that the remote shell (ssh or rsh) is still working using the new node name
6. Add the node back to cluster using mmaddnode command.
7. Re-specify a secondary configuration node for cluster
8. Re-specify NSD servers.
9. Start gpfs on new node
10. mount both file system on each node if they are not mounted.
The best practice is to plan for changing names or removing nodes prior to making the changes and follow the procedures listed in the GPFS Administration and Programming Reference when making the changes to avoid having to perform the general actions listed above. You can find more information about proper procedures within the IBM Cluster Information Center.