luminocean / k8s-setup

k8s-setup

Scripts to setup an kubernetes environment based on kubeadm.

1. prepare hosts file

# adjust as needed after copy
cp hosts.example hosts

2. add public keys to known_hosts

ssh-keyscan '192.168.0.1' >> ~/.ssh/known_hosts

3. run playbook

ansible-playbook -i hosts playbook.yml

4. run kubeadm

kubeadm init

5. prepare kubectl config

mkdir -p $HOME/.kube
cp -i /etc/kubernetes/admin.conf $HOME/.kube/config

6. install weave-net

sysctl net.bridge.bridge-nf-call-iptables=1

kubectl apply -f "https://cloud.weave.works/k8s/net?k8s-version=$(kubectl version | base64 | tr -d '\n')"

# run the following to check
kubectl get pods --all-namespaces

When the installation is done, you should be able to see nodes ready by running:

kubectl get nodes

7. (optional) install metrics-server

git clone https://github.com/kubernetes-incubator/metrics-server.git

cd metrics-server

By default, the metrics server queries the k8s API server to get node hostnames and try to access them. However in a simple multi-VM environment, we might have a functional DNS service ready. Also the metrics server connects to each node through https, so there's a certificate issue here.

To make things work, we can modify the command line used to deploy the metrics server by adding the following lines to deploy/1.8+/metrics-server-deployment.yaml in the container spec section:

command:
- /metrics-server
- --kubelet-preferred-address-types=InternalIP
- --kubelet-insecure-tls

And then apply files:

kubectl create -f deploy/1.8+/

We can forward k8s API server's endpoint to local for later use:

kubectl proxy --port=8080 &

Test the deployment with:

curl http://localhost:8080/apis/metrics.k8s.io/v1beta1/nodes

and you shall see the node info in JSON format.

Now you can use the following command to see the node resource usages：

kubectl top node

8. (optional) install Prometheus-operator and a prometheus instance

kubectl apply -f prometheus-demo.yml