Upgrade notes

This page is intended to deal with specific, special circumstances you may encounter when upgrading between versions of Charmed Kubernetes. The notes are organised according to the upgrade path below, but also be aware that any upgrade that spans more than one minor version may need to beware of notes in any of the intervening steps.

Upgrading to 1.16

Docker Registry with Containerd

Prior to 1.16, some fixes were required to support using the Docker Registry charm with Containerd.

This charm, if used, is now supported through standard relations. Before upgrading, remove any reference of the registry in the custom_registries containerd charm configuration.

After upgrading, see the docker registry instructions for details of how to configure a registry.

Admission control plugins

In Charmed Kubernetes 1.16, the API server parameter by which additional, non-default admission control plugins is specified has changed. The old parameter was --admission-control; the new parameter is --enable-admission-plugins.

For example, prior to 1.16, The 'PodSecurityPolicy' admission plugin could be applied like this:

juju config kubernetes-master api-extra-args="admission-control=PodSecurityPolicy"

As of 1.16, this changes to:

juju config kubernetes-master api-extra-args="enable-admission-plugins=PodSecurityPolicy"

If using non-default admission plugins, be sure to upgrade your charm config accordingly after upgrading to 1.16.

Upgrading to 1.15

This upgrade switches the container runtime to make use of containerd, rather than Docker. You have the option of keeping Docker as the container runtime, but even in that case you must perform the upgrade steps. To facilitate different container runtimes, the architecture of Charmed Kubernetes has changed slightly, and the container runtime is now part of a separate, subordinate charm rather than being included in the kubernetes-master and kubernetes-worker charms.

To keep Docker as the container runtime

Docker is currently installed on your kubernetes-worker units. The Docker subordinate charm includes clean-up code to manage the transition to the new pluggable architecture. To upgrade whilst retaining Docker as the runtime, you need to additionally deploy the new charm and add relations to the master and worker components:

juju deploy cs:~containers/docker
juju add-relation docker kubernetes-master
juju add-relation docker kubernetes-worker

The upgrade is complete, and your worker nodes will continue to use the Docker runtime. For information on configuring the Docker charm, see the Docker configuration page.

To migrate to containerd

If you intend to switch to containerd, it’s recommended that you first add some temporary extra worker units. While not strictly necessary, skipping this step will result in some cluster down time. Adding temporary additional workers provides a place for keeping pods running while new workers are brought online. The temporary workers can then be removed as the pods are migrated to the new workers. The rest of these instructions assume that you have deployed temporary workers.

Deploy temporary workers

CURRENT_WORKER_REV=$(juju status | grep '^kubernetes-worker\s' | awk '{print $7}')
CURRENT_WORKER_COUNT=$(juju status | grep '^kubernetes-worker/' | wc -l | sed -e 's/^[ \t]*//')

juju deploy cs:~containers/kubernetes-worker-$CURRENT_WORKER_REV  kubernetes-worker-temp -n $CURRENT_WORKER_COUNT

Add necessary relations

juju status --relations | grep worker: | awk '{print $1,$2}' | sed 's/worker:/worker-temp:/g' | xargs -n2 juju relate

Wait for the temporary workers to become active before continuing. Upgrade the master and worker charms:

juju upgrade-charm kubernetes-master
juju upgrade-charm kubernetes-worker

The kubernetes-worker units will enter a blocked state, with status message “Connect a container runtime.”

Deploy and relate the new Docker charm

This step is needed even if you do not intend to use Docker following the upgrade. Docker is already installed on your kubernetes-worker units, and the Docker subordinate includes clean-up code to uninstall Docker when the Docker charm is replaced with the containerd charm.

juju deploy cs:~containers/docker
juju add-relation docker kubernetes-master
juju add-relation docker kubernetes-worker

Switching to Containerd

Now, pause the existing workers, which will move pods to the temporary workers.

juju run-action [unit] pause --wait

For example:

juju run-action kubernetes-worker/0 pause --wait
juju run-action kubernetes-worker/1 pause --wait
juju run-action kubernetes-worker/2 pause --wait

One-liner:

juju status | grep ^kubernetes-worker/ | awk '{print $1}' | tr -d '*' | xargs -n1 -I '{}' juju run-action {} pause --wait

Remove Docker

This will uninstall Docker from the workers.

juju remove-relation docker kubernetes-master
juju remove-relation docker kubernetes-worker
juju remove-application docker

Deploy Containerd

juju deploy cs:~containers/containerd
juju add-relation containerd kubernetes-master
juju add-relation containerd kubernetes-worker

Resume workers

Now we can allow pods to be rescheduled to our original workers.

juju run-action [unit] resume --wait

E.g.

juju run-action kubernetes-worker/0 resume --wait
juju run-action kubernetes-worker/1 resume --wait
juju run-action kubernetes-worker/2 resume --wait

One-liner:

juju status | grep ^kubernetes-worker/ | awk '{print $1}' | tr -d '*' | xargs -n1 -I '{}' juju run-action {} resume --wait

Cleanup

You can now pause the temporary workers to force all pods to migrate back to your “real” workers, then remove the temporary workers.

juju status | grep ^kubernetes-worker-temp/ | awk '{print $1}' | tr -d '*' | xargs -n1 -I '{}' juju run-action {} pause --wait

juju remove-application kubernetes-worker-temp

Mixing Containerd and Docker

Once you have a Containerd backed Charmed Kubernetes running, you can add Docker backed workers like so:

juju deploy cs:~containers/kubernetes-worker kubernetes-worker-docker
juju deploy cs:~containers/docker
juju relate docker kubernetes-worker-docker

Clusters which previously ran etcd 2.3

If your cluster has run etcd 2.3 at any point in the past, then it is strongly recommended not to upgrade the etcd charm to revision 449. Doing so will cause etcd to lose all of its data. For details, see https://bugs.launchpad.net/charm-etcd/+bug/1843497

We recommend upgrading etcd directly to charm revision 460 instead.

Upgrading to 1.14

This upgrade includes support for CoreDNS 1.4.0. All new deployments of Charmed Kubernetes will install CoreDNS by default instead of KubeDNS.

Existing deployments which are upgraded to Charmed Kubernetes 1.14 will continue to use KubeDNS until the operator chooses to upgrade to CoreDNS. To upgrade, set the new dns-provider config:

juju config kubernetes-master dns-provider=core-dns

Please be aware that changing DNS providers will momentarily interrupt DNS availability within the cluster. It is not necessary to recreate pods after the upgrade completes.

The enable-kube-dns option has been removed to avoid confusion. The new dns-provider option allows you to enable or disable KubeDNS as needed.

For more information on the new dns-provider config, see the dns-provider config description.

Upgrading from 1.9.x to 1.10.x

This upgrade includes a transistion between major versions of etcd, from 2.3 to 3.x. Between these releases, etcd changed the way it accesses storage, so it is necessary to reconfigure this. There is more detailed information on the change and the upgrade proceedure in the upstream etcd documentation.

Caution: This upgrade must be completed before attempting to upgrade the running cluster.

To make this upgrade more convenient for users of Charmed Kubernetes, a script has been prepared to manage the transition. The script can be examined here.

To use the script to update etcd, follow these steps:

1. Download the script with the command:

curl -O https://raw.githubusercontent.com/juju-solutions/cdk-etcd-2to3/master/migrate

2. Make the script executable:

chmod +x migrate

3. Execute the script:

./migrate

4. etcd OutputSed

The script should update etcd and you will see output similar to the following:

· Waiting for etcd units to be active and idle...
· Acquiring configured version of etcd...
· Upgrading etcd to version 3.0/stable from 2.3/stable.
· Waiting for etcd upgrade to 3.0/stable............................................................
· Waiting for etcd units to be active and idle....................................................
· Waiting for etcd cluster convergence...
· Stopping all Kubernetes api servers...
· Waiting for etcd cluster convergence...
· Stopping etcd...
· Migrating etcd/0...
· Migrating etcd/1...
· Migrating etcd/2...
· Starting etcd...
· Configuring storage-backend to etcd3...
· Waiting for all Kubernetes api servers to restart.......
· Done.

You can now proceed with the rest of the upgrade.

We appreciate your feedback on the documentation. You can edit this page or file a bug here.