Kubernetes Architecture

Take a moment to think about this

The Control Plane is the 'brain' of the cluster — it includes the API Server (the receptionist that accepts commands, like a hotel front desk), etcd (the database that stores all cluster state), the Scheduler (decides which node a container runs on), and the Controller Manager (keeps watching to make sure desired state matches actual state). Worker Nodes are the servers where containers actually run — each node has a kubelet (which receives instructions from the Control Plane and runs containers) and a container runtime (Docker/containerd).

Let's connect this to a real scenario

Let's continue the restaurant chain analogy — the Control Plane is the head office (deciding how many branches to open, which branch manager makes which calls). The Worker Nodes are each individual branch (where customers actually get served). kubectl (which we'll cover in a later lesson) is the phone line you use to give the head office instructions — say 'open 5 branches' and the Control Plane assigns the work across the worker nodes.

Control Plane = API Server + etcd + Scheduler + Controller Manager
Worker Node = kubelet + container runtime, where containers actually run
Every command goes through the API Server

Let's look at it together

text

Kubernetes Cluster
===================

Control Plane (the "brain")
├── API Server      — receives all kubectl commands
├── etcd             — stores the entire cluster's state
├── Scheduler         — decides which node runs what
└── Controller Manager — keeps actual state == desired state

Worker Nodes (where containers actually run)
├── Node 1: kubelet + container runtime + your pods
├── Node 2: kubelet + container runtime + your pods
└── Node 3: kubelet + container runtime + your pods

You should see

You should be able to explain the difference in responsibilities between the Control Plane and Worker Nodes.

5-minute try-it

In a cluster with three nodes, predict what happens if Node 1 suddenly goes down (what does the Control Plane do?) and write it out.

A quick word of caution

For local learning (minikube/kind), the Control Plane and Worker Node are usually mixed together on a single VM/container — in a production cluster, these are split across separate servers.