Kubernetes from Scratch: The Core Concepts

The biggest barrier to learning Kubernetes is the density of terminology. But all of it answers a single question: 'How do we reliably run and scale a container application?' Let's view the concepts around that question.

Pod: the smallest unit

Kubernetes doesn't run containers directly; it wraps them in Pods. A Pod holds one or more containers that share networking and storage. Most of the time, one Pod = one app container.

Deployment: manages Pods

A Pod alone is fragile — if the node dies, the Pod is gone. A Deployment is where you say 'always run 3 copies of this app.' If a Pod dies, the Deployment creates a new one; when updating versions it does a rolling update.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: web
spec:
  replicas: 3
  selector:
    matchLabels: { app: web }
  template:
    metadata:
      labels: { app: web }
    spec:
      containers:
        - name: web
          image: nginx:1.27

Service: a stable address

Pods come and go and their IPs keep changing. A Service puts a stable name/IP in front of a group of Pods and load-balances traffic among them. So you talk to the 'web' service and don't care that the Pods behind it change.

Putting it together

• You put your container in a Pod.
• With a Deployment you specify how many copies and how they update.
• With a Service you give it a stable access point.
• With Ingress you route external HTTP traffic inward.

Your first Deployment hit CrashLoopBackOff — fix it

You deployed your first app but the Pod keeps restarting. This is the #1 place beginners get stuck. The diagnosis flow:

1. 1kubectl get pods → look at STATUS and RESTARTS
2. 2kubectl logs <pod> → why does the app die? (and the previous attempt: --previous)
3. 3kubectl describe pod <pod> → Events + probe failures
4. 4Fix the root cause (config, image, port), verify the rollout

$ kubectl get pods
NAME                   READY   STATUS             RESTARTS   AGE
web-6d9f-abc            0/1     CrashLoopBackOff   4          2m
$ kubectl logs web-6d9f-abc --previous
Error: connect ECONNREFUSED postgres:5432   # the app can't reach the DB
$ kubectl describe pod web-6d9f-abc | grep -A2 Liveness
  Liveness: http-get :3000/health delay=0s timeout=1s   # /health hits the DB → keeps failing
# Fix: correct the DB env + raise the liveness delay, then:
$ kubectl rollout status deploy/web
deployment "web" successfully rolled out   # ✓

These four concepts (Pod, Deployment, Service, Ingress) make up 80% of Kubernetes. You learn the rest far faster by trying it on a cluster — which is exactly what you do in Cloudpuz labs.