Kubernetes Architecture

Control Plane and Worker Nodes

A Kubernetes cluster has two logical parts. The control plane makes cluster-wide decisions; the worker nodes run your application workloads.

Control Plane Components

• API server (kube-apiserver) - the front door; every kubectl and controller request goes through it.
• etcd - the cluster database; stores all cluster state.
• Scheduler (kube-scheduler) - picks which node should run a new pod.
• Controller manager - runs control loops (replicas, endpoints, nodes) that reconcile desired state.

Worker Node Components

• kubelet - the node agent; talks to the API server and manages pods on the node.
• kube-proxy - maintains network rules for Services.
• Container runtime - the software that actually runs containers (containerd, CRI-O).

The Reconciliation Loop

When you run kubectl apply, the API server stores your desired state in etcd. Controllers watch that state, compare it with reality and act until they match. This loop runs forever - that is why Kubernetes is self-healing.

kubectl get controller-manager -n kube-system
kubectl get pods -n kube-system

Key Points

• The API server is the only component you talk to directly.
• etcd holds the source of truth for the entire cluster.
• kubelet + container runtime run pods on each worker node.
• Controllers continuously reconcile desired versus actual state.