Beginner — Kubernetes

README · Next: Intermediate

A Pod is one lunch tray. A Deployment keeps N trays out. A Service is the name that stays put. First cluster is kind (Kubernetes in Docker), not kubeadm.

Beginner: core concepts

Desired state and the API server

• What it is: You declare what should exist. kubectl sends that declaration to the API server. etcd is the cluster’s memory of those objects.
• Why it exists: You cannot SSH to every machine and start containers by hand. Controllers watch the API and move reality toward the declaration (reconciliation).
• How it looks: kubectl apply -f examples/nginx-deployment.yaml creates or updates a Deployment object. kubectl get deploy nginx reads it back.
• Common confusion: kubectl is not the cluster. It is a client. The cluster is the API server, etcd, controllers, and kubelets.

Pod

• What it is: The smallest deployable unit: one or more containers that share a network namespace (they see each other as localhost) and can share volumes.
• Why it exists: Some processes need a helper next to them (a proxy, a log shipper). Those helpers are extra containers in the same Pod.
• How it looks: kubectl get pods lists them. Each Pod gets its own IP, which changes if the Pod is recreated.
• Common confusion: A Pod is not a virtual machine, and it is ephemeral. Do not create naked Pods for apps you care about. Use a Deployment.

Deployment / ReplicaSet

• What it is: A Deployment is the replica count plus a Pod template. Kubernetes creates a ReplicaSet; the ReplicaSet creates the Pods.
• Why it exists: You want N copies, a rolling replace when the image changes, and a new Pod when one dies.
• How it looks: spec.replicas: 3 and spec.template in examples/nginx-deployment.yaml.
• Common confusion: You almost never create a ReplicaSet yourself. Edit the Deployment; it owns the ReplicaSet.

Labels and selectors

• What it is: A label is a key-value tag on an object. A selector is a query over those tags.
• Why it exists: Names of Pods are generated (nginx-7d4c8b9f5-xk2m1). Services and ReplicaSets find Pods by label, not by name.
• How it looks: The Pod template has labels: {app: nginx}. The Deployment selector and the Service selector both say app: nginx.
• Common confusion: Matching the Deployment’s own metadata.labels is not enough. The Service must match the Pod template labels.

Service (ClusterIP first)

• What it is: A stable virtual IP and DNS name in front of Pods. ClusterIP is reachable only inside the cluster. That is the default type.
• Why it exists: Pod IPs die with the Pod. Callers need a name that outlives any one tray.
• How it looks: examples/nginx-service.yaml — type: ClusterIP, selector.app: nginx, port: 80.
• Common confusion: ClusterIP is not on your laptop. curl from your shell will not hit it until you kubectl port-forward (or install an Ingress controller later).

kubectl as the client

• What it is: A CLI that talks HTTP to the API server using a kubeconfig file (context, cluster, user).
• Why it exists: One tool to apply YAML, list objects, and debug.
• How it looks: kubectl apply, kubectl get, kubectl describe, kubectl logs.
• Common confusion: Installing kubectl does not create a cluster. kind (this module) or a cloud control plane is the cluster. kubectl config current-context tells you which one you are talking to.

Beginner: install

You need Docker running (module 03). Then install the two binaries from their official pages — do not invent package names from a blog:

1. Install kubectl (match it to the kind node version you will run, 1.31+).
2. Install kind.

Create the learning cluster:

kind create cluster --name learn
kubectl cluster-info
kubectl config current-context

Expected output: kubectl cluster-info prints a control-plane URL (often https://127.0.0.1:…). current-context is kind-learn.

If it failed: kind: command not found → install is not on PATH. Cannot connect to the Docker daemon → start Docker; kind runs nodes as containers. kind create cluster says the name already exists → skip create, or kind delete cluster --name learn and recreate.

Do not run kubeadm init on your laptop. That path is in Production.

kind create cluster --name learn   # skip if it already exists
kubectl apply -f examples/nginx-deployment.yaml
kubectl apply -f examples/nginx-service.yaml
kubectl get deploy,po,svc
kubectl wait --for=condition=available deploy/nginx --timeout=90s
kubectl port-forward svc/nginx 8080:80

curl -s -o /dev/null -w "%{http_code}\n" http://127.0.0.1:8080

kubectl scale deploy/nginx --replicas=1
kubectl get pods
kubectl scale deploy/nginx --replicas=3
kubectl get pods