Updating hardware on nodes running on vSphere

You must ensure that your nodes running in vSphere are running on the hardware version supported by OKD. Currently, hardware version 13 or later is supported for vSphere virtual machines in a cluster.

You can update your virtual hardware immediately or schedule an update in vCenter.

Using hardware version 13 for your cluster nodes running on vSphere is now deprecated. This version is still fully supported, but support will be removed in a future version of OKD. Hardware version 15 is now the default for vSphere virtual machines in OKD.

Updating virtual hardware on vSphere

To update the hardware of your virtual machines (VMs) on VMware vSphere, update your virtual machines separately to reduce the risk of downtime for your cluster.

Updating the virtual hardware for control plane nodes on vSphere

To reduce the risk of downtime, it is recommended that control plane nodes be upgraded serially. This ensures that the Kubernetes API remains available and etcd retains quorum.

Prerequisites

  • You have cluster administrator permissions to execute the required permissions in the vCenter instance hosting your OKD cluster.

  • Your vSphere ESXi hosts are version 6.7U3 or later.

Procedure

  1. List the control plane nodes in your cluster.

    1. $ oc get nodes -l node-role.kubernetes.io/master

    Example output

    1. NAME STATUS ROLES AGE VERSION
    2. control-plane-node-0 Ready master 75m v1.23.0
    3. control-plane-node-1 Ready master 75m v1.23.0
    4. control-plane-node-2 Ready master 75m v1.23.0

    Note the names of your control plane nodes.

  2. Mark the control plane node as unschedulable.

    1. $ oc adm cordon <control_plane_node>
  3. Shut down the virtual machine (VM) associated with the control plane node. Do this in the vSphere client by right-clicking the VM and selecting PowerShut Down Guest OS. Do not shut down the VM using Power Off because it might not shut down safely.

  4. Upgrade the VM in the vSphere client. Follow Upgrading a virtual machine to the latest hardware version in the VMware documentation for more information.

  5. Power on the VM associated with the control plane node. Do this in the vSphere client by right-clicking the VM and selecting Power On.

  6. Wait for the node to report as Ready:

    1. $ oc wait --for=condition=Ready node/<control_plane_node>
  7. Mark the control plane node as schedulable again:

    1. $ oc adm uncordon <control_plane_node>
  8. Repeat this procedure for each control plane node in your cluster.

Updating the virtual hardware for compute nodes on vSphere

To reduce the risk of downtime, it is recommended that compute nodes be upgraded serially.

Multiple compute nodes can be upgraded in parallel given workloads are tolerant of having multiple nodes in a NotReady state. It is the responsibility of the administrator to ensure that the required compute nodes are available.

Prerequisites

  • You have cluster administrator permissions to execute the required permissions in the vCenter instance hosting your OKD cluster.

  • Your vSphere ESXi hosts are version 6.7U3 or later.

Procedure

  1. List the compute nodes in your cluster.

    1. $ oc get nodes -l node-role.kubernetes.io/worker

    Example output

    1. NAME STATUS ROLES AGE VERSION
    2. compute-node-0 Ready worker 30m v1.23.0
    3. compute-node-1 Ready worker 30m v1.23.0
    4. compute-node-2 Ready worker 30m v1.23.0

    Note the names of your compute nodes.

  2. Mark the compute node as unschedulable:

    1. $ oc adm cordon <compute_node>
  3. Evacuate the pods from the compute node. There are several ways to do this. For example, you can evacuate all or selected pods on a node:

    1. $ oc adm drain <compute_node> [--pod-selector=<pod_selector>]

    See the “Understanding how to evacuate pods on nodes” section for other options to evacuate pods from a node.

  4. Shut down the virtual machine (VM) associated with the compute node. Do this in the vSphere client by right-clicking the VM and selecting PowerShut Down Guest OS. Do not shut down the VM using Power Off because it might not shut down safely.

  5. Upgrade the VM in the vSphere client. Follow Upgrading a virtual machine to the latest hardware version in the VMware documentation for more information.

  6. Power on the VM associated with the compute node. Do this in the vSphere client by right-clicking the VM and selecting Power On.

  7. Wait for the node to report as Ready:

    1. $ oc wait --for=condition=Ready node/<compute_node>
  8. Mark the compute node as schedulable again:

    1. $ oc adm uncordon <compute_node>
  9. Repeat this procedure for each compute node in your cluster.

Additional resources

Scheduling an update for virtual hardware on vSphere

Virtual hardware updates can be scheduled to occur when a virtual machine is powered on or rebooted. You can schedule your virtual hardware updates exclusively in vCenter by following Schedule a Compatibility Upgrade for a Virtual Machine in the VMware documentation.

When scheduling an upgrade prior to performing an upgrade of OKD, the virtual hardware update occurs when the nodes are rebooted during the course of the OKD upgrade.