End-Users Won't Even Notice, Painless Kubernetes Upgrades
summary
Claim: moving to side-by-side clusters lets upgrades happen with near zero downtime by creating a 1.19 green alongside a 1.18 blue and switching traffic after verification. Demonstrated tooling includes flux for GitOps add-ons, cluster-api or kops/kubespray to provision clusters, and pluto to find deprecated APIs (extensions/v1beta1 removed in 1.16, AWS in-tree EBS replaced by CSI in 1.21). Shows handling control planes (3/5/9 nodes), data-plane node swaps, backups and replica promotion for stateful data.
AI-written summary · about
chapters (19)
- 0:00Start
- 1:00Outline
- 3:45K8s Control Plane Upgrades
- 4:33K8s API Versions...
- 5:49K8s 1.17... K8s 1.x Example
- 8:39K8s Data Plane Upgrades
- 9:37In-Place Upgrade
- 13:05BLUE/GREEN CLUSTERS - Nodes
- 13:46Other Blue/Green Approaches
- 16:38Terraform/EKSCTL
- 17:15Flux
- 20:57Other Tools...
- 22:05Pluto
- 23:45Production...
- 25:08Production Stateful Apps?...
- 27:30Future Tools? - Gaps
- 30:55Cloud providers Example: EKS
- 31:45Takeaways
- 34:04References
related talks
19mUpgrading Your K8s Clusters: Key Considerations for a Smooth Transition
25mAutomating Kubernetes Upgrades for Zero Downtime
35mClusters as Cattle: How to Seamlessly Migrate Apps Across Kubernetes Clusters
50mOps issues
35mSIG Cluster Lifecycle (kops)
16mZero-downtime upgrades of Flink applications
06mNever ever upgrade
29m11 5 years of kubernetes and looking to the future