Kubernetes on Raspberry Pi

If you have a few raspberry pis lying around and you’re looking for an interesting way to get started with kubernetes, this is the tutorial for you! Below, I’ve outlined a fairly straight forward method for getting a stripped down install of kubernetes (k3s) installed across a pair of raspberry pis.

While I was working on this project, I realized that one of my pis was corrupting sd cards. It took me a while to realize that the pi itself was the culprit, so I ended up spending quite a bit of time flashing new raspbian images and getting fresh installs set up with some basic tools. I wrote a script to do most of the work, so this will be a relatively quick tutorial with a minimum of non-kubernetes related setup tasks.

Things you’ll need:

  • Two raspberry pis (with sd cards)
  • Two ethernet cables
  • A network switch
  • A stable internet connection
  • About an hour

Flash your raspberry pi

Start by flashing raspbian onto the raspberry pi sd cards. I used the April 2019 release of raspbian from the official downloads page and etcher to flash the cards.

Log into your first pi (we’ll refer this as node1). The default user:pass is pi:raspberry. Next run sudo raspi-config nonint do_ssh 0, which enables remote ssh access.

If you do not have your pi connected to a monitor and keyboard or simply don’t have physical access to it, add an empty file at /boot/ssh to the sd card before inserting it into the pi. This will enable ssh on first boot.

The easiest way to do this is by attaching a monitor and a keyboard to your pi

A fun note about this command —

The raspi-config binary doesn’t have an official non-interactive API, so this command may not work in the future. You can either use the interactive mode by omitting nonint do_ssh 0 and using the GUI or by looking at the source and updating the command. In case the link to source above is broken, you can also check an archive of the page.

Get the setup script

Run wget https://raw.githubusercontent.com/subtlepseudonym/pi/master/setup/pi_raspbian.sh, which will download a setup script that will set your hostname, create a non-default user, and install some useful packages.

This script is fairly opinionated and will set up some basic directories and change your default shell. I recommend reading through the script before running it and tailoring it to your use case.

Run the setup

Add executable permissions to the setup script with chmod +x setup.sh and run it with sudo ./setup.sh HOSTNAME USERNAME with a hostname and username of your choosing. For this example, we’ll use node1 as the hostname and user as the username.

You will be prompted for a password and then you can go get a coffee, because it takes a few minutes to set things up. When the script is finished, your pi will reboot.

This gif as been truncated to maintain readability

ssh into the pi and run the cleanup script via sudo ./cleanup.sh. This removes the default pi user and its home directory.

Notice the user@host change from the previous gif

Install k3s

We’ll be using k3s, a stripped-down version of kubernetes. Download the install script with curl -sfL https://get.k3s.io | sh -. This will install k3s and get it started as a system service.

Congratulations! Your first pi is all set.

If you run into trouble, you can check the status of the k3s service with
systemctl status k3s. You can also uninstall k3s using
/usr/local/bin/k3s-uninstall.sh.

Set up your second node

Repeat these steps on the second pi (node2), stopping before the k3s installation. We don’t need the whole server, so we’re just going to install the k3s agent.

First, get the node token from node1 with the following:
sudo cat /var/lib/rancher/k3s/server/node-token.

Next, run the following command, replacing $NODE1_IP and $NODE_TOKEN with the IP address of node1 and the output of the cat command above.

This will get the k3s agent started on node2 and register it with your k3s server on node1.

If you run into trouble, try checking the status of the k3s agent service with systemctl status k3s-agent. In the worst case, you can uninstall and reinstall the k3s agent using /usr/local/bin/k3s-agent-uninstall.sh and the command you used above to install the agent. If you do uninstall and reinstall, be sure to remove the password entry corresponding to node2 on node1 in /var/lib/rancher/k3s/server/cred/passwd.

Test your cluster setup

So long as node1 is discoverable by node2, your kubernetes cluster should now be set up! You can confirm this by running k3s kubectl get nodes on node1 and ensuring that both node1 and node2 are listed.

Now that you have a working kubernetes cluster, you’re free to write up a manifest and get your software running resiliently! If you’re looking for a more specific suggestion, check out my article on running pi-hole on your new cluster.