kubezero/docs/v1.19.md

2.4 KiB

Upgrade to KubeZero V2.19.0

CloudBender / Kubernetes v1.19

Changes

  • worker nodes names are now the default AWS private hostnames, rather than the CloudBender provided unique static hostnames :-(
    This change was required to enable node restrictions via the upstream aws-iam-authenticator as well as prepare for support of the Horizontal Autoscaler and Spot Instances in the next releases
  • cluster-admin kubectl config now stored on SSM Parameter store, contains no more secrets leveraging IAM roles
  • fully encrypted cluster backup on S3 ( automated migration and cleanup of previous files )
  • backup password and cluster version stored on SSM Parameter store
  • worker nodes authenticate via IAM roles rather than tokens
  • improved resource reservations on all nodes
  • various security / reliability improvements and bug fixes

Upgrade

  • Set Kubernetes version in the controller config to eg. 1.19

  • Update controller and worker stacks with latest CFN code

  • terminate controller00 first, afterwards remaining controllers

  • replace worker nodes in a rolling fashion via. drain / terminate / rinse-repeat

KubeZero

Changes

  • Version bump to latest releases of EVERY component
  • optional support for fuse-device-plugin
  • KubeZero now supports bare-metal, all AWS components optional
  • resource definitions for most admin pods, incl. apiserver, etcd, etc.
  • Logging:
    • ES resources are now defined using standard config vs. custom settings like jvm_heap and cpu_request
    • Optional ability to add nodeAffinity rules for ES/Kibana and Fluentd
    • Fluentd replicaCount default from 2 to 1

Upgrade - Without ArgoCD

  1. Update CRDs of all enabled components:
    ./bootstrap.sh crds all clusters/$CLUSTER

  2. Prepare upgrade

  • delete old fluentd deployement because labels are immutable and they changed due to the migration to new upstream helm chart
    kubectl delete deployment logging-fluentd -n logging
  1. Upgrade all components
    ./bootstrap.sh deploy all clusters/$CLUSTER

Upgrade - ArgoCD

  • ArgoCD needs to be upgraded first to support latest Helm chart requirements: ./bootstrap.sh deploy argocd clusters/$CLUSTER
  • push latest cluster config to your git repo
  • verify correct branch etc. ( eg. argoless branch has been retired ! )
  • trigger sync in ArgoCD starting with the KubeZero root app
    ( only if auto-sync is not enabled )

Verification / Tests

  • check if all pods are RUNNING
  • check any Ingress services
  • ...