1. Concepts for Release 2
  2. Cluster Administration
  3. Cluster Updates

Cluster Updates

Learn how to update a Kubernetes cluster by updating the Oracle Container Host for Kubernetes (OCK) image on each Kubernetes node.

This section shows you how to update nodes to the latest Kubernetes patch release, or to update them to the next Kubernetes minor release.

Patch releases include errata updates and might include Common Vulnerabilities and Exposures (CVE) fixes, Kubernetes updates, OS updates, and so on. An update to the next Kubernetes minor version is performed in the same way as patch updates, with one extra step to set the Kubernetes version number.

Oracle CNE delivers all updates through updated Oracle Container Host for Kubernetes (OCK) images. Updates are delivered through an OCK image that's specific to the Kubernetes minor version, for example for Kubernetes Release 1.31.

Each node periodically polls the container registry to check for updates to the OCK image it's running, or for an image for the target Kubernetes version if you're upgrading Kubernetes. When you set the Kubernetes version for an upgrade, the image for that version is pulled and staged on the nodes in the cluster. Patch updates are downloaded to each node automatically and don't need to be staged before a node update.

When an update is available, use the ocne node update command to reboot a node to use the new image. Running the ocne node update command for a node completes the following actions:

  1. The node is drained (using the kubectl drain command) from the cluster. This evicts the pods from the node.

  2. The host OCK image is installed on the node, and the node is restarted.

  3. The node is returned to the cluster (using the kubectl uncordon command) and is made available to run pods.

Update nodes sequentially, starting with the control plane nodes.

Tip:

To save time, you can start the update process as soon as one of the control plane nodes has been annotated as having an update available.

You can update a Highly Available cluster without bringing the cluster down. As one control plane node is taken offline, another control plane node takes control of the cluster. In a cluster with a single control plane node, the control plane node is offline for a short time while the update is performed.

If applications are running on more than one worker node, they remain up, and available, during an update.