Upgrades
This page will outline how to manage and execute a Kubernetes upgrade.
- Before you begin
- Assumptions
- Preparing for an Upgrade
- Upgrade etcd
- Upgrade Kubernetes
- Verify upgrade
- Upgrade Flannel
- Upgrade easyrsa
Before you begin
This page assumes you have a working deployed cluster.
Assumptions
You should always back up all your data before attempting an upgrade. Don’t forget to include the workload inside your cluster! Refer to the backup documentation.
Preparing for an Upgrade
See if upgrades are available. The Kubernetes charms are updated bi-monthly and mentioned in the Kubernetes release notes. Important operational considerations and change in behaviour will always be documented in the release notes.
You can use juju status to see if an upgrade is available. There will either be an upgrade to kubernetes or etcd, or both.
Upgrade etcd
Backing up etcd requires an export and snapshot, refer to the backup documentation to create a snapshot. After the snapshot upgrade the etcd service with:
juju upgrade-charm etcd
This will handle upgrades between minor versions of etcd. Major upgrades from etcd 2.x to 3.x are currently unsupported. Instead, data will be run in etcdv2 stores over the etcdv3 api.
Upgrade Kubernetes
The Kubernetes Charms use snap channels to drive payloads. The channels are defined by X.Y/channel where X.Y is the major.minor release of Kubernetes (e.g. 1.6) and channel is one of the four following channels:
| Channel name | Description |
|---|---|
| stable | The latest stable released patch version of Kubernetes |
| candidate | Release candidate releases of Kubernetes |
| beta | Latest alpha or beta of Kubernetes for that minor release |
| edge | Nightly builds of that minor release of Kubernetes |
If a release isn’t available, the next highest channel is used. For example, 1.6/beta will load /candidate or /stable depending on availablility of release. Development versions of Kubernetes are available in that minor releases edge channel. There is no guarantee that edge or master will work with the current charms.
Master Upgrades
First you need to upgrade the masters:
juju upgrade-charm kubernetes-master
NOTE: Always upgrade the masters before the workers.
Once the latest charm is deployed, the channel for Kubernetes can be selected by issuing the following:
juju config kubernetes-master channel=1.x/stable
Where x is the minor version of Kubernetes. For example, 1.6/stable. See above for Channel definitions
Worker Upgrades
Two methods of upgrading workers are supported. Blue/Green Deployment and upgrade-in-place. Both methods are provided for operational flexibility and both are supported and tested. Blue/Green will require more hardware up front than inplace, but is a safer upgrade route.
Blue/Green Upgrade
Given the following deployment, where the workers are named kubernetes-alpha.
Deploy new worker(s):
juju deploy kubernetes-beta
Pause the old workers so your workload migrates:
juju run-action kubernetes-alpha/# pause
Verify old workloads have migrated with:
kubectl get pod -o wide
Tear down old workers with:
juju remove-application kubernetes-alpha
In place worker upgrade
juju upgrade-charm kubernetes-worker
juju config kubernetes-worker channel=1.x/stable
Where x is the minor version of Kubernetes. For example, 1.6/stable. See above for Channel definitions. Once you’ve configured kubernetes-worker with the appropriate channel, run the upgrade action on each worker:
juju run-action kubernetes-worker/0 upgrade
juju run-action kubernetes-worker/1 upgrade
...
Verify upgrade
kubectl version should return the newer version.
It is recommended to rerun a cluster validation to ensure that the cluster upgrade has successfully completed.
Upgrade Flannel
Upgrading flannel can be done at any time, it is independent of Kubernetes upgrades. Be advised that networking is interrupted during the upgrade. You can initiate a flannel upgrade:
juju upgrade-charm flannel
Upgrade easyrsa
Upgrading easyrsa can be done at any time, it is independent of Kubernetes upgrades. Upgrading easyrsa should result in zero downtime as it is not a running service:
juju upgrade-charm easyrsa
Rolling back etcd
At this time rolling back etcd is unsupported.
Rolling back Kubernetes
At this time rolling back Kubernetes is unsupported.