Kubernetes in One Page
Kubernetes on one page: control plane and worker node architecture, Pods, Deployments, StatefulSets, Services and Ingress, a minimal Deployment manifest, the daily kubectl commands, why a pod will not start, rolling updates and a production checklist.
Cheat sheet · Offered as a template
In this diagram
- Kubernetes in One Page
- The control plane decides, the nodes run. The objects you declare, and the commands that show what actually happened.
- ① The idea
- You declare the state you want, in YAML
- Controllers keep nudging reality to match it
- Pods are disposable; identity lives in labels and Services
- Everything is an object behind one API
- ② Cluster architecture
- Control plane
- Worker node
- kubectl
- API server
- etcd
- Scheduler
- Controllers
- kubelet
- kube-proxy
- Pod
- ③ A minimal Deployment
- deploy.yaml
- Then expose it
- The label app=web is the only thing that ties the Service to its pods.
- ④ Objects you will meet
- ⑤ A request reaches a pod
- Client
- Load balancer
- Cloud
- Ingress
- Host and path
- Service
- Stable virtual IP
- Any ready replica
- ⑥ kubectl, daily
- shell
- ⑦ When a pod will not start
- ⑧ Requests and limits
- resources
- Requests are what the scheduler reserves. Limits are where the kernel stops you.
- ⑨ Rolling update
- Change the image in the Deployment
- A new ReplicaSet starts one pod
- Wait until it passes its readiness probe
- Remove one old pod
- Repeat until all are new; undo rolls it back
- ⑩ Production checklist
- Readiness and liveness probes
- Requests and limits set
- Two or more replicas, plus a PodDisruptionBudget