Kubernetes the Easy Way – DevOps Series, Part 14

If you have been following this DevOps series you might remember that cool banana-sized kubernetes cluster we built in Part 4 with some Raspberry Pi boards. It was a great way to build a fully functional setup from scratch and also learn a lot in the process.

However, since then things have evolved. And now, instead of just k8s, we also have k3s (which, just judging by the name, must include at least 5 things less). k3s is an easy-to-install, lightweight but fully-compliant, kubernetes distribution (40MB single binary and 512MB RAM) optimized for ARM architectures… like our RPi setup. It does not include several heavy components that might not be really necessary in a common setup, like legacy features, embedded plugins, and other things like… Docker. Yes, you read well. It does not include Docker. What!?! Well, it includes a better option: a low-level component called containerd, much lighter than Docker.

Sounds like a great option for our small cluster, right? Time to get our hands dirty!

Ketchup for kubernetes

To make installation as simple and quick as possible we will use a tool called k3sup (read ketchup). So, let’s get started by running the following steps from your workstation.

First you need to install k3sup:

curl -sLS https://get.k3sup.dev | sh

Then, from your workstation, you can install k3s in your master RPi node (ie. the one with IP 192.168.1.100):

k3sup install --ip 192.168.1.100 --user pi

Save the kubeconfig file in your local directory, and start using it (please make sure you specify the complete path to the file):

export KUBECONFIG=~/Downloads/kubeconfig

In (literally) less than a minute you should be able to see your kubernetes master node up and ready:

kubectl get nodes

Wow, that was quick, huh?

Let’s now configure the rest of RPi boards as worker nodes, by specifying their IP addresses (192.168.1.101-103) and the master node IP address (192.168.1.100):

k3sup join --ip 192.168.1.101 --server-ip 192.168.1.100 --user pi
k3sup join --ip 192.168.1.102 --server-ip 192.168.1.100 --user pi
k3sup join --ip 192.168.1.103 --server-ip 192.168.1.100 --user pi

Again, in less than a minute you should see all of them up and running:

kubectl get nodes

That’s all! If you are a fast typist you can go from zero to a fully-configured and ready-to-use kubernetes cluster in just a few minutes.

THIS has to be the definition of “automagical”… so cool!!!

On top of it, k3s also includes traefik installed by default, so you don’t need to install a bare-metal load-balancer, nor an ingress controller. Everything is included and ready for you to use!

From Heapster to Metrics Server

Something else that changed during the last year is that the solution we used for monitoring cluster resource usage (Heapster) has been deprecated. A suitable replacement is Kubernetes Metrics server, a cluster-wide aggregator of resource usage data. It provides access to CPU & RAM usage per node and per pod, via CLI and API.

To install it please clone the required repo into your workstation:

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

Then edit the deployment file and replace the default image name with the appropriate ARM one (k8s.gcr.io/metrics-server-arm:v0.3.2):

vi metrics-server/deploy/1.8+/metrics-server-deployment.yaml

You are now ready to apply the required Metrics Server manifests:

kubectl apply -f metrics-server/deploy/1.8+ -n kube-system

Once the pod is ready you will be able access resource usage info via CLI:

kubectl top node
kubectl top pod

Or you can also browse its API, as you would with any other kubernetes API:

kubectl get --raw /apis/metrics.k8s.io/v1beta1/nodes | jq .
kubectl get --raw /apis/metrics.k8s.io/v1beta1/pods | jq .

(note: to get nicely formatted output you will need jq installed in your system, ie. brew install jq in your Mac)

Alternatively, if you would rather use HTTP to browse the API (ie. with curl or wget), you can always use kubectl proxy (reverse proxy to help with locating the API server and authentication):

kubectl proxy --port=8080 &
curl http://localhost:8080/apis/metrics.k8s.io/v1beta1/nodes
curl http://localhost:8080/apis/metrics.k8s.io/v1beta1/pods

Ready to rock!

With k3sup you have been able to easily install k3s and get your kubernetes cluster ready in a matter of minutes! It is now ready to get some applications deployed in it, so please feel free to try it out with our classic example microservices-based application: myhero. You may find how to do it in my previous DevOps Part 4 blog post and associated Learning Lab.

Our kubernetes cluster still looks as cool as ever, but now with k3s it has much better performance and can be fully configured from scratch in just minutes!