Kubernetes Fixes Itself. Here’s How.

One container is easy. Three copies on three servers, and you become the person restarting them at 3 a.m. That is the job Kubernetes exists to take, and it does it in a way nobody expects: it is not a launcher, it is a room full of stubborn thermostats. We build it ourselves from a 4-item checklist: the API server as the shared notebook, controllers that compare what you want with what you have, the scheduler, the kubelet, and Services. And we finally find out why a pod can sit on Pending with no error. Key terms: cluster, control plane, desired state, control loop, controller, kubectl apply, API server, etcd, pod, ReplicaSet, Deployment, scheduler, node, kubelet, Service, kube-proxy. Chapters 0:00 The Problem 1:16 The Mystery 1:55 What We Need 2:39 Attempt 1: A Script 3:29 The Better Question 4:02 The Loop 4:32 The Notebook 5:10 The Replica Loop 6:00 The Scheduler 6:42 The Kubelet 7:18 The Notebook 7:54 The Control Plane 8:43 Healing 9:22 Services 10:04 The Bill 10:44 The Big Idea 11:06 The Problem, Revisited 12:00 Proof 12:38 To Summarise 13:12 End