Re-provision Node OS

Re-provisioning a cluster node via PXE.

Use this procedure for OS updates, disaster recovery, or adding a new node (Control, Compute, Storage, or GPU) to the cluster.

Impact: Full node downtime. The node must be rebooted and re-imaged.

Prerequisites

• IPMI/BMC access to the target node
• OpenStack admin credentials (for compute nodes with running instances)

注記

For OpenStack commands, run platform-setup.sh --shell to enter the management container.

Procedure

1. Evacuate the Node (Compute nodes only)

If the node is a compute node with running instances, migrate them before proceeding.

1. Disable the compute service to prevent new instances from being scheduled:

   openstack compute service set --disable --reason "Re-provisioning" <host-name>

2. List and migrate all instances:

   openstack server list --host <host-name> --all-projects
   
   # Live migrate each instance (repeat for all)
   openstack server migrate --wait --live-migration <instance-id>

   Verify migration success with openstack server show <instance-id>.

2. Update PXE Configuration

Apply the PXE configuration with the desired OS image:

platform-setup.sh --bootstrap

3. Re-provision the Node

1. Boot the node into PXE mode using IPMI:

   ipmitool -I lanplus -H <ipmi-ip> -U <user> -P <pass> chassis bootdev pxe
   ipmitool -I lanplus -H <ipmi-ip> -U <user> -P <pass> power cycle

2. Monitor the installation via IPMI Serial-over-LAN:

   ipmitool -I lanplus -H <ipmi-ip> -U <user> -P <pass> sol activate

4. Verify

1. SSH into the node and verify the OS version:

   cat /etc/os-release
   uname -r

   Check the output against the expected version in the target release notes.

2. For compute nodes, re-enable the service and migrate instances back:

   openstack compute service set --enable <host-name>
   
   openstack server migrate --wait --live-migration <instance-id> --host <host-name>

Rollback

If the node fails to boot or install:

1. Revert the changes in inventory.yml to the previous known-good configuration.
2. Run platform-setup.sh --bootstrap to restore the old boot files.
3. Reboot the node via IPMI again.

References

• Adding and removing hosts