A Complete Visual Guide to Kubernetes Architecture

Kubernetes (K8s) is a core component of large‑scale architectures. Its architecture is divided into a control plane and worker nodes.

Control Plane – the “brain” of the cluster, consisting of five main components:

API Server : the single entry point for all requests; it handles authentication, validation, data processing and communication with other components.

etcd : a distributed key‑value store that persists the entire cluster state, effectively the database of Kubernetes.

Scheduler : decides on which node a pending Pod should run based on resource availability and constraints.

Controller Manager : runs various controllers (e.g., ReplicaSet, Node controller) that continuously reconcile the actual state with the desired state, acting as Kubernetes’ “autopilot”.

Other components (e.g., cloud‑controller‑manager) are omitted for brevity.

All components communicate through the API Server. A typical request flow is illustrated below:

kubectl
↓
API Server
↓
… (authentication, etc.)
↓
HTTPS
↓
… (response)

Example command: kubectl get pods etcd stores objects such as Pods, Nodes, Deployments, ConfigMaps, Secrets and Namespaces. Example of stored objects:

Pod

Node

Deployment

ConfigMap

Secret

Namespace

The Scheduler evaluates node metrics; for instance, it may choose Node1 with CPU 20 % and Memory 30 % over Node2 with higher utilization.

Node1
CPU: 20%
Memory: 30%

When a Pod fails, the Controller Manager recreates it to maintain the desired replica count, ensuring that the specified number of Pods (e.g., three) is always running.

Replicas = 3
Pod1
Pod2
Pod3

Worker Nodes – the machines that actually run containers. Each node includes:

kubelet : an agent that receives instructions from the API Server and ensures Pods are running correctly.

kube‑proxy : maintains network rules and load‑balances service traffic.

Container runtime (e.g., containerd, CRI‑O): pulls images and starts containers.

This modular design gives Kubernetes strong scalability and maintainability: control logic is centralized in the control plane, while execution logic is distributed across the nodes.