Create three VMs (droplets on digital ocean, ec2 on aws, compute engine on google cloud, virtual machine on azure). Make sure they all have 2 CPUs or more (kubernetes requires this).
ssh root@$YOUR_MASTER_NODE_IP and run node.sh, then master.sh on that machine.
After running master.sh, copy the JOIN comamnd printed out to the console. Paste it into join.sh.
On your local linux machine, run local install. If you're on mac, see this instead:
https://kubernetes.io/docs/tasks/tools/install-kubectl/
Then run talk_to_cluster.sh for linux or mac. Make sure the $KUBE_MASTER variable resolves to the IP address of your master node.
For each worker node, ssh into the box (ssh root@$WORKER_NODE_IP), run node.sh, then run run the JOIN command you copied at the end of step two (if you copied it into join.sh, then just run bash join.sh).
kubectl apply -f deployment.yaml. Then kubectl apply -f service.yaml. Boom, you have an application running on kubernetes!
run kubectl get services to see the port exposed for your service.
To get Hello Kubernetes back, run curl http://$DEMO_WORKER_1:$EXPOSED_PORT and you should get a response from the deployed pod.