KubeZero 1.24

TODO

What's new - Major themes

Cilium is now the default CNI, calico gets removed
cluster-autoscaler is enabled by default on AWS
worker nodes are now automatically update to latest AMI and config in a rolling fashion
integrated Bitnami Sealed Secrets controller
reduced avg. CPU load on controller nodes well below the 20% threshold to prevent extra costs from CPU credits

Version upgrades

cilium
metallb
nvidia-device-plugin
aws-node-termination-handler
aws-ebs-csi-driver
aws-efs-csi-driver
istio 1.16
argocd 2.5.5 + tweaks
all things prometheus incl. automated muting of certain alarms, eg. CPUOverCommit when cluster-autoscaler is available

FeatureGates

PodAndContainerStatsFromCRI
DelegateFSGroupToCSIDriver

Upgrade

(No, really, you MUST read this before you upgrade)

Ensure your Kube context points to the correct cluster !

Review CFN config for controller and workers, no mandatory changes during this release though
Upgrade CFN stacks for the control plane ONLY ! Updating the workers CFN stacks would trigger rolling updates right away !
Trigger cluster upgrade:
./admin/upgrade_cluster.sh <path to the argocd app kubezero yaml for THIS cluster>
Review the kubezero-config and if all looks good commit the ArgoApp resouce for Kubezero via regular git
git add / commit / push <cluster/env/kubezero/application.yaml>

DO NOT yet re-enable ArgoCD before all pre v1.24 workers have been replaced !!! *

Reboot controller(s) one by one
Wait each time for controller to join and all pods running. Might take a while ...
Upgrade CFN stacks for the workers.
This in turn will trigger automated worker updates by evicting pods and launching new workers in a rolling fashion. Grab a coffee and keep an eye on the cluster to be safe ... Depending on your cluster size it might take a while to roll over all workers!
Re-enable ArgoCD by hitting on the still waiting upgrade script
Quickly head over to ArgoCD and sync the KubeZero main module as soon as possible to reduce potential back and forth in case ArgoCD has legacy state

Known issues

existing EFS volumes

If pods are getting stuck in Pending during the worker upgrade, check the status of any EFS PVC. In case any PVC is in status Lost, edit the PVC and remove the following annotation: pv.kubernetes.io/bind-completed: "yes" This will instantly rebind the PVC to its PV and allow the pods to migrate.
Going to be fixed during the v1.25 cycle by a planned rework of the EFS storage module.

2.6 KiB Raw Blame History