After hearing Dashaun Carter talk about his Kubernetes cluster on Raspberry Pi on Spring Office Hours , I was inspired to build my own. This project will document the steps I took to set up a Raspberry Pi cluster running Kubernetes.

I wish I had gone to his blog to use his cluster as inspiration first–it looks very cool and was cheaper than mine!

Why

I have been a Kubernetes fan-boy for a few years, but I don’t have much in the way of hardware to run it at home. I wanted something that I could experiment with. Plus I am planning to go for a Kubernetes certification in the short run and figured this would provide a useful playground environment.

Inspiration

I inspired my cluster afterAnthony Simon’s. I especially liked the notion of powering it with Power over Ethernet (PoE) and using a network switch to simplify the setup. I don’t yet have a nifty case like Anthony’s, but it’s on my list of things to print on my Creality Ender 3.

Bill of Materials

Hardware

I was able to find almost everything at the local Micro Center. They were out of PoE hats so I had to resort to Amazon for those.

Item	Quantity	Unit Price (USD)	Total (USD)
Raspberry Pi 5 8GB	3	$80	$240
MicroSD Card 32GB	3	$7	$21
TP-Link PoE Switch	1	$60	$60
TP-Link Travel Router	1	$40	$40
Uctronics PoE Hat	3	$22	$66
		Total	$427

So yeah, I could have gone for a lower-end Pi to save some dough, but I wanted to actually run some stuff on it, so I opted for the most memory I could get.

Software

I imaged the SD cards with Raspberry Pi Imager. Since I wanted the thinnest slice of host OS possible, I chose the smaller “lite” version of Raspberry Pi OS.

To install Kubernetes, I went with K3s, which is a lightweight version of Kubernetes. It is designed to be easy to install and run on resource-constrained devices like the Raspberry Pi.

Since I’d be tearing this down and setting it back up multiple times, I put together some ansible playbooks to automate the process of preparing each node with pre requisites, installing a shared token secret, installing a control plane “server” and 2 “worker” nodes.

Result

I really need to print that case to make this look better, for now, Heinen’s rubber bands will have to do.

Here is the cluster in all its glory: