Kubernetes

Step by step

The example above, written out — the same steps the animation plays.

1. 1kubectl apply -f deployment.yaml. You declare the desired state: "run 3 copies of web:v1". Kubernetes keeps making reality match it.
2. 2Schedule web-1 → node-1. The scheduler puts each pod on the node with the most room.
3. 3Schedule web-2 → node-2. The scheduler puts each pod on the node with the most room.
4. 4Schedule web-3 → node-3. The scheduler puts each pod on the node with the most room.
5. 5Service: one address for all three. A Service gives the pods one stable name and spreads traffic across the healthy ones.
6. 6web-2 crashes. The app inside hit a fatal error. Desired 3, running 2.
7. 7Replace it: web-4. The ReplicaSet notices the difference and creates a new pod. No pager needed.
8. 8node-3 goes down. The whole machine is gone, with 1 pod on it.
9. 9Reschedule → node-1. web-5 starts on a healthy node.
10. 10kubectl scale --replicas=5. Traffic is up: change the desired count.
11. 11Scale up: web-6 → node-2. New pods join the Service as soon as they are ready.
12. 12Scale up: web-7 → node-1. New pods join the Service as soon as they are ready.
13. 13Deploy web:v2 (rolling update). Replace pods one at a time, so the app never goes down.
14. 14Start web-8 (v2). A v2 pod starts and passes its readiness check…

…and 10 more steps — press Play above to watch them all.

What's happening?

1. A Deployment declares "run 3 copies of web:v1"; the scheduler places the pods on nodes.
2. When a pod crashes or a node dies, controllers see fewer pods than desired and start new ones elsewhere.
3. Scaling changes the desired count; a rolling update swaps pods to the new version one at a time, with no downtime.

Where you'll meet it

Running microservices at most mid-to-large companies, on EKS, GKE or AKS — and the platform underneath many PaaS products.

FAQ