Kubernetes Fundamentals

Kubernetes is a container orchestration platform that automates deploying, scaling, and operating containerized applications across clusters of machines. It schedules workloads, manages networking and storage, and restarts or replaces containers that fail.

itCloud native tools and technologies

Don't Panic — Kubernetes Fundamentals

Kubernetes has a reputation for complexity that is about seventy percent deserved and thirty percent people describing all of it at once. The system itself rests on one idea, and it is a small one.

You write down the state you want. Something else then spends the rest of its life trying to make reality match. That is a thermostat. Kubernetes is a very large, very well-funded thermostat with a Greek name meaning "helmsman", abbreviated K8s because there are eight letters between the K and the s.

Everything follows from that. Each object created carries a spec — what was asked for — and a status — what is actually going on. Control loops watch the gap between the two and close it.

A crashed container, a dead machine, a new version to roll out on a Tuesday afternoon: same shape of problem, same machinery. Once that is visible, the two hundred nouns stop being two hundred nouns.

Three of them come first. A Pod is one or more containers glued together, sharing an address, scheduled onto one machine, and — this part matters — designed to die. Do not grow attached to a Pod.

You will not create them by hand either. Declare a Deployment saying "three of these, please" and its controller handles the creating, the replacing and the rolling forward.

A Service then gives that constantly changing set of Pods one stable name, selecting them by their labels, which is how anything in the cluster finds anything else at all.

The architecture underneath is less alarming than its diagram. The control plane holds an API server every request passes through, a database called etcd keeping the agreed truth, a scheduler deciding where things land, and controllers running the loops. Each worker machine runs a kubelet making sure the right containers are alive, a runtime actually running them, and a proxy pointing traffic at them. That is the entire cast.

Two things nobody says early enough. Kubernetes is not a platform-as-a-service — it builds nothing for you, brings no database, and has cheerfully no opinion about monitoring. Every gap in that sentence is a decision somebody now owns. And by default a cluster is not a security boundary between the things running inside it. Anyone who assumed otherwise should stop assuming it.

For one small application all of this is comically excessive, and a container on a single virtual machine will serve better. With many services, many machines and somebody carrying a pager, it stops being excessive and becomes the reason the pager stays quiet.

Read the Intro next for the whole picture, the Cheatsheet when you have forgotten a noun, and the Slides if you think in shapes. This was only the map.

Don't Panic — Kubernetes Fundamentals

Relevant careers