I built a four-node Raspberry Pi cluster this month and I want to be honest about what it achieved: nothing. Nothing useful, anyway. There is no workload that needs it, nothing on it that could not run on a single Pi with room to spare, and the whole thing draws power to host a dashboard that mostly watches itself.
I do not regret it for a second.
The stack is four Pi 4s in one of those acrylic sandwich cases, a cheap unmanaged switch, and a USB hub that pretends to be a power supply and mostly succeeds. I put k3s on it because full Kubernetes on an ARM board is a special kind of self-harm. Getting the control plane up, the nodes joined, and a couple of pods scheduled across them was an evening of small satisfying wins.
What I actually got out of it was the muscle memory. I now know what a node looks like when it drops, what kubectl get nodes says when one of the SD cards is dying, and how long a rolling update really takes when the hardware is this weedy. That knowledge is worth something the next time a real cluster misbehaves at work, even if the lab itself does nothing.
So no, it taught me nothing useful, and it was one of the more enjoyable things I have built all year. Sometimes the point is the building.