Black Lives Matter. Please consider donating to the San Luis Obispo, CA Bail Fund or other bail funds so we can make the world a just and equitable place for all people of all colors.

Adding Pi's to a cluster

Zerg-rushing server workloads

Look at all of these single-board computers lying around. What are they even doing? Literally nothing. Let’s put them to good use.

Two Pi 3's, one Pi 2, and one Orange Pi one

Raspberry Pi SD Cards

I got the latest Raspbian Buster Lite image and extracted the image. To make things easier on myself, I decided to enable SSH-on-first-boot for the image so that I can just brainlessly dd everything onto SD cards.

First, I executed

losetup -P /dev/loop99 2021-01-11-raspios-buster-armhf-lite.img

to mount the image on a loop device. 99 was chosen because snapd makes a goddamn million loop devices.

lsblk returned the following:

loop0                   7:0    0 818.2M  1 loop /snap/android-studio/98
loop1                   7:1    0 373.8M  1 loop /snap/anbox/186
loop47                  7:47   0 135.7M  1 loop /snap/chromium/1466
loop99                  7:99   0   1.8G  0 loop
├─loop99p1            259:7    0   256M  0 part
└─loop99p2            259:8    0   1.5G  0 part

Perfect! Once that was done, it was a simple

mkdir -p /tmp/bootsd
mount /dev/loop99p1 /tmp/bootsd
touch /tmp/bootsd/ssh

and the SD card image was ready. To clean everything up:

umount /tmp/bootsd
losetup -d /dev/loop99

Finally, I plugged in the SD cards one by one and flashed them with

dd bs=64M of=/dev/sdb if=2021-01-11-raspios-buster-armhf-lite.img

As soon as I booted each of them, and they were assigned DHCP addresses, I SSH’ed into them to change the password because security.

Orange Pi SD Card

The Orange Pi One was set up in a similar fashion, except that I got the Armbian image from the Orange Pi website (scroll like, really far for the OPi One).

Getting k3s on them

I used ssh-copy-id to install my public key on all of the devices for passwordless login, and then used k3sup for quickly installing k3s on them:

k3sup join --sudo --server-user rancher --user pi --server-ip <my server's address> --ip <the pi's address>

Forking multiple instances of k3sup made life easier.

So now, I have 7 whole nodes on my kubernetes cluster! Fear my power!

$ kubectl get node
zerg-1       NotReady   <none>   2d1h    v1.19.7+k3s1  # rpi
k3os-28502   Ready      master   6d2h    v1.19.5+k3s2
orangepi     Ready      <none>   4d10h   v1.19.7+k3s1  # opi
zerg2        Ready      <none>   2d1h    v1.19.7+k3s1  # rpi
zerg3        Ready      <none>   2d1h    v1.19.7+k3s1  # rpi
k3os-10009   Ready      <none>   5d23h   v1.19.5+k3s2
k3os-3502    Ready      <none>   3h13m   v1.19.5+k3s2

Please ignore the inconsistent dashing scheme on the zerg* series.

Look at these SBC's!

Now what are they running? …unfortunately, still, absolutely nothing. Once again, my k3s cluster is a solution looking for a problem.

In my next post, I’ll talk about my Grafana + Prometheus setup.


Markdown formatting is supported.