This sets up a one node Kubernetes cluster on a RaspberryPI. This is based on the Hypriot Docker image for RPi.
Why ? Because it is the easiest way to start experimenting with containers on ARM architectures which can be extremely useful for the IoT and Embedded business. It can also be a great platform to investigate scalability issues of both Docker and Kubernetes.
You will need a RaspberryPi and you will need to have installed the Hypriot OS image on it
Clone this repository on your Pi
$ git clone email@example.com:Project31/kubernetes-installer-rpi.git
$ cd ./kubernetes-installer-rpi
Install a master
On a different RPi install a node
We use the Kubernetes
kubelet running as a systemd unit to monitor a few Docker containers that make up the Kubernetes cluster.
The Kubelet binary for ARM is downloaded and installed in
/usr/bin/kubelet, a manifest is copied to
/etc/kubernetes/manifests/kubernetes.yaml which represents a Kubernetes Pod that make up all the required containers.
In this pod we have:
etcdcontainer. Which we run by building an etcd image from scratch on the PI.
Several containers based on the Hyperkube image. Hyperkube is a single binary that can start all the Kubernetes components: API server, controller, scheduler. We build a local Hyperkube image.
Then a little trick. Kubernetes does assume that you will run the nodes on x86_64 and automatically pulls an image called the pause container. This container is used to get an IP and share that IP with all the containers in the pod. For the PI, we need to run this pause container on ARM, we trick Kubernetes by building the image
gcr.io/google_containers/pause:0.8.0locally after having compiled the pause Golang code for ARM.
Finally, we download the Kubernetes client
kubectlfor ARM from the official release:
$ curl -fsSL -o kubectl https://storage.googleapis.com/kubernetes-release/release/v1.0.3/bin/linux/arm/kubectl
The ARM binaries for the
hyperkube are downloaded from https://github.com/Project31/kubernetes-arm