Summary - Internals

Combine etcd, scheduler and controller to explain the architecture.

Summary: internals

In this group:

etcd stores all state consistently with Raft; an odd number of members (3, 5) is used for quorum; keys are /registry/....
kube-scheduler filters and scores; affinity, anti-affinity and topology spread steer placement.
kube-controller-manager runs dozens of controllers; in HA the leader is chosen with a Lease.
kube-apiserver: authentication → RBAC → admission (mutating/validating) → validation → etcd.
All controllers are level-triggered reconcile loops: read the spec, observe the world, close the gap, write the status.

kubectl get --raw /readyz?verbose
kubectl get lease -n kube-system

Exercise Spread across zones

Write a pod template's topologySpreadConstraints section (starting with topologySpreadConstraints:): maxSkew: 1, topologyKey: topology.kubernetes.io/zone, whenUnsatisfiable: DoNotSchedule, labelSelector.matchLabels app: api.

Show example solution

topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: topology.kubernetes.io/zone
    whenUnsatisfiable: DoNotSchedule
    labelSelector:
      matchLabels:
        app: api

Test yourself

What is the order of a pod creation request?
- kubelet → scheduler → etcd
- scheduler → apiserver → kubelet → etcd
- etcd → kubelet → apiserver
- apiserver → etcd → the scheduler picks a node → the kubelet starts it
Show answer

apiserver → etcd → the scheduler picks a node → the kubelet starts it — Each component only talks to the API server.

Query the API server's readiness endpoint.

kubectl get --raw '/____?verbose'

Show answer

kubectl get --raw '/readyz?verbose'

Match.

etcd

State
scheduler

Placement
controller-manager

Reconciliation
apiserver

Entry gate

Next lesson: What is a CRD?

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies