Cluster Architecture

Who does what

The control plane holds the cluster state and makes decisions: the API server (the front door every tool talks to), etcd (the key-value store holding all objects), the scheduler (chooses nodes for new pods) and controller managers (loops that reconcile Deployments, ReplicaSets, Jobs and more). Each node runs the kubelet (starts and monitors containers via a container runtime such as containerd) and a network proxy component. Everything is driven through the API: kubectl, CI tools and controllers all read and write objects there.

Components at a glance

Control plane on top, nodes below.

control plane
  kube-apiserver           REST API; authentication, authorisation, admission
  etcd                     stores every object (desired + observed state)
  kube-scheduler           assigns pending pods to nodes
  kube-controller-manager  reconciliation loops (Deployments, ReplicaSets, Jobs, ...)

every node
  kubelet                  runs pods via the container runtime, reports status, runs probes
  container runtime        containerd / CRI-O
  kube-proxy (or CNI)      implements Service networking

Back up etcd (or trust your provider)

etcd is the cluster's memory; with a managed service, know the provider's backup and recovery guarantees.

Quick check: Which component decides which node a new pod runs on?

• The scheduler
• etcd
• The kubelet
• The container image