Operating kube-apiserver Without Hiccups
summary
Describes a concrete solution that keeps kube-apiserver available during upgrades and cert rotation by running static pods managed by an operator which creates immutable revisions and per-node installer pods that write manifests into kubelet directories. Shows a real failure from rolling a DaemonSet over three controller managers that lost leader lease and crashed, then explains use of minimum shutdown duration, separate health vs readiness endpoints, and an operator-driven, ordered cert rotation so no client requests are dropped.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 1:20Static Pods
- 2:51Static Pod CheckPointer
- 8:05No Hiccups
- 15:46Search Rotation
- 21:41Questions
related talks
21mSelf-Healing Kubernetes
44mKubeCeption! A Story of Self-Hosted Kubernetes
16mUnconference Session: Reliable K8s Operator: How to Survive Eventual Consistency
38mCaulking the deployment gap: absolutely zero downtime deployments in Kubernetes
27mWhen SysAdmins Quit: Protecting Kubernetes Clusters When the Owner of Multiple...
35mHandling Risky Business: Cluster Upgrades
34mSelf-Hosted Kubernetes: How and Why [I]
53mAutomating Stateful Applications with Kubernetes Operators