Kubernetes Basics

Docker Compose is great for one machine. Kubernetes (often "k8s") solves the much harder problem: running containers reliably across a whole cluster of machines, automatically replacing them when they fail.

Pods

The smallest unit Kubernetes schedules is a Pod - one or more containers that always run together, on the same machine, sharing the same network address. Almost always, that's exactly one container per Pod.

Deployments

You don't create Pods directly in practice - you create a Deployment, and tell it how many copies (replicas) of a Pod you want running:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: web
spec:
  replicas: 3
  template:
    spec:
      containers:
        - name: web
          image: nginx:1.25

If a Pod crashes, or the machine it's on dies, the Deployment notices and starts a replacement - without anyone paging a human in the middle of the night to do it by hand. Change replicas: 3 to replicas: 10 and apply it again, and Kubernetes starts seven more, on whichever machines in the cluster have room.

Services

Pods come and go, and each one gets a new IP address every time it's replaced - so nothing else in the cluster can reliably point directly at one. A Service gives a stable name and address that always routes to whichever Pods are currently healthy, no matter how many times they've been replaced underneath it.

Rolling updates

Pushing nginx:1.26 over the top of nginx:1.25, a Deployment replaces Pods a few at a time by default - never all at once - so there's no moment where zero healthy Pods are serving traffic.

Try it

Everything above ultimately runs as ordinary containers on an ordinary Linux node:

docker ps

When you're ready, hit Mark complete and move to the next lesson.