Kubernetes The Hard Way (Vagrant)

Vagrant configuration and scripts for a Kubernetes setup, the hard way.

The setup follows https://github.com/kelseyhightower/kubernetes-the-hard-way with the following exceptions:

cri-o is used as a container runtime, not cri-containerd
The pod-cidr is 10.2${i}.0.0/16, routes are provisioned from scripts/vagrant-setup-routes.bash automatically
For crio, an explicit --stream-address must be set, as the address of the default interface isn't routable (see e.g. config/worker-0-crio.service)
192.168.199.40 is the IP of the loadbalancer (haproxy) for HA controllers

Requirements Host

Vagrant (with VirtualBox)
Minimum of 7x 512MB of free RAM
cfssl, cfssljson and kubectl (on Linux, scripts/install-tools can be used to download and install the binaries to /usr/local/bin)

Setup

Manually

To learn Kubernetes from the bottom up, it's recommended to go through KTHW manually. vagrant up gives you three controller and three worker nodes to do that.

The pod-cidr is 10.2${i}.0.0/16, for which the Vagrant nodes have configured routes (see route -n).

The following KTHW parts can/should be skipped:

Everything in regard to the frontend loadbalancer
Pod network rules are automatically setup via Vagrant

The scripts in scripts/ loosely match the setup steps in KTHW by Hightower and can be used as reference and/or to save typing. See scripts/setup also.

Single script

vagrant destroy -f   # remove previous setup
./scripts/setup      # takes about 5 minutes or more
[...]

If everything looks good, continue with "Using the cluster"

Multiple scripts

Remove previously created certificates, tools kubeconfig files:

./scripts/distclean

Download required tools and files:

./scripts/download-tools

Start the virtual machines (optionally, go drink a coffee or tee):

vagrant up
[...]
vagrant status

Current machine states:

controller-0              running (virtualbox)
controller-1              running (virtualbox)
controller-2              running (virtualbox)
worker-0                  running (virtualbox)
worker-1                  running (virtualbox)
worker-2                  running (virtualbox)

Generate the required certificates:

./scripts/generate-certs

Generate the kubeconfig files (as those include copies of the previously generated certificates):

./scripts/generate-kubeconfig-kube-proxy
./scripts/generate-kubeconfig-worker

Setup etcd on the controller nodes and verify it has started:

./scripts/setup-etcd
[...]
vagrant ssh controller-0
ETCDCTL_API=3 etcdctl member list

6c500a9f4f9113de, started, controller-0, https://192.168.199.10:2380, https://192.168.199.10:2379
e206d150eae73959, started, controller-2, https://192.168.199.12:2380, https://192.168.199.12:2379
e7e775a3da74a469, started, controller-1, https://192.168.199.11:2380, https://192.168.199.11:2379

Setup the controller services and verify they are up and running:

./scripts/setup-controller-services
[...]
for c in controller-0 controller-1 controller-2; do vagrant ssh $c -- kubectl get componentstatuses; done

NAME                 STATUS    MESSAGE              ERROR
controller-manager   Healthy   ok
scheduler            Healthy   ok
etcd-1               Healthy   {"health": "true"}
etcd-2               Healthy   {"health": "true"}
etcd-0               Healthy   {"health": "true"}
[...]

Create ClusterRole's for kubelet API auth:

./scripts/setup-kubelet-api-cluster-role

Setup the worker binaries, services and configuration:

./scripts/setup-worker-services
[...]
vagrant ssh controller-0
kubectl get nodes

NAME       STATUS    AGE       VERSION
worker-0   Ready     1m        v1.9.2
worker-1   Ready     55s       v1.9.2
worker-2   Ready     12s       v1.9.2

Configure a kubernetes-the-hard-way context on your host, set it as default and verify everything is ok:

./scripts/configure-kubectl-on-host

kubectl get componentstatuses
[...]
kubectl get nodes
[...]

Using the cluster

Setup DNS add-on

Deploy the DNS add-on and verify it's working:

kubectl create -f ./manifests/kube-dns.yaml
[...]
kubectl get pods -l k8s-app=kube-dns -n kube-system
[...]
kubectl run busybox --image=busybox --command -- sleep 3600
[...]
POD_NAME=$(kubectl get pods -l run=busybox -o jsonpath="{.items[0].metadata.name}")
kubectl exec -ti $POD_NAME -- nslookup kubernetes

Smoke tests

kubectl create -f ./manifests/nginx.yaml
deployment "nginx" created
service "nginx" created

NODE_PORT=$(kubectl get svc nginx --output=jsonpath='{range .spec.ports[0]}{.nodePort}')
for i in {0..2}; do curl -sS 192.168.199.2${i}:${NODE_PORT} | awk '/<h1>/{gsub("<[/]*h1>", ""); print $0}'; done
Welcome to nginx!
Welcome to nginx!
Welcome to nginx!

Connect to services from host

10.32.0.0/24 is the IP range for services. In order to connect to a service from the host, one of the worker nodes (with kube-proxy) must be used as a gateway. Example:

# On Linux
sudo route add -net 10.32.0.0/24 gw 192.168.199.22

# On macOS
sudo route -n add -net 10.32.0.0/24 192.168.199.22

Use Traefik loadbalancer

./scripts/setup-traefik
[...]
curl 192.168.199.30
404 page not found

To test traefik is actually doing its job, you can create an ingress rule for the nginx service that you created above:

kubectl apply -f ./manifests/nginx-ingress.yaml
echo "192.168.199.30 nginx.kthw" | sudo tee -a /etc/hosts
curl nginx.kthw
<!DOCTYPE html>
[...]

Pitfalls

Error loading config file "/var/log": read /var/log: is a directory

On OSX, KUBECONFIG apparently needs to be set explicitly. ~/.kube/config is a good place and the default on Linux.

ramnes/kubernetes-the-hard-way-vagrant