These days it’s everybody’s dream to build their own bare metal K8s cluster. Whether it’s for learning purposes, running Blender, or even building out production infrastructure, people love running their own clusters.
The Raspberry Pi 4 (RPi) with it’s relatively fast CPU cores, up to 8 GB of RAM, and tiny physical footprint presents a great option to run a cluster on. Provisioning all those RPis can be a pain however, and people have wanted to use tools like MAAS.
For years the community has wondered how to use MAAS with RPis, but hasn’t quite cracked it - until now, that is. Two problems needed to be overcome to use MAAS with RPis:
- Booting and installing an operating system with MAAS
- Power control
Thanks to community members Giles, lory696, someo_new and others, these problems have been overcome.
In this tutorial, we will put all their efforts together and show you how to build your own small, fully managed bare metal Raspberry Pi cluster. When you’re finished with this tutorial, you’ll be at a point where you can install microK8s so that you end up with a fully functioning Kubernetes cluster that is ready to be managed by any tool you like, such as Juju.
This tutorial is not meant to be exhaustive for all the possible ways to build a cluster, but should teach you all the basics needed in order for you to try other techniques. For example, perhaps you’d like to use a PoE switch instead of the UUGear Mega4 USB hub used in this tutorial. At the end, you might end up with something that looks a bit like this:
DISCLAIMER
This tutorial is provided “as is” and with no warranty. It makes use of the RPi UEFI project and uses the generic ARM64 kernel and image from Canonical. Canonical does not officially support the RPi UEFI project nor does it support using the generic ARM64 kernel and image on Raspberry Pis.
With the above disclaimer out of the way, let’s get started!