Becoming an Expert of Diagnosing a Kubernetes Control Plane Problem
summary
Shows stepwise diagnostics that find root causes and fixes: use kubectl get/components/nodes/pods, describe, watch, logs and output yaml/json to see missing fields or finalizers; a stuck terminating pod is often a lingering finalizer that can be removed manually or fixed by removing finalizer or upgrading controller; etcd defaults to 4GB limit and compaction every 5 minutes so a cluster resized from ~1,000 to ~5,000 nodes can hit space exceeded; analyze offline etcd files with augur and reduce preload image lists or increase storage.
AI-written summary · about
chapters (28)
- 0:00Introduction
- 12:19Containers
- 12:53Kubernetes Pod
- 13:18Master VM
- 13:51Master Note
- 14:08Master Components
- 14:46sanity check tools
- 16:05three different examples
- 16:50check controller manager log
- 17:29check online
- 18:19API server
- 19:02Open API
- 20:49Monitoring
- 21:20API Server Log
- 21:38Scaling V1
- 23:39Note Selector
- 24:11Note Affinity Note Ting
- 26:48Healthy Control Plane
- 27:12Puzzle Pieces
- 28:08Resource
- 28:46Upgrade
- 29:31Question Time
- 30:26Shelf Life
- 31:17Best Practice
- 31:27Monitoring First Line of Defense
- 31:42Feedback
- 32:12Use Cases
- 32:58Question
related talks
34mPanel: Improving and Managing Kubernetes at Scale
35mIntro to Control Plane Engineering
01h 32mTutorial: Debug Your Kubernetes Apps
36mKubernetes Confessions: Real Mistakes We’ve All Made (and What They Teach Us)
14mArt of decluttering your Kubernetes cluster
37mKubernetesception: How to run a 1000 control planes
35mKubernetes, Lessons Learnt
36mUnderstanding Scalability and Performance in the Kubernetes Master