News
Justin Garrison: You learn Kubernetes by building it yourself
That Kubernetes and edge computing are ultimately about people, not technology, was the message delivered by Justin Garrison, Field CTO at Sidero Labs, during Edgecase 2026 in Hilversum. Building things yourself, experimenting in your garage, talking to others, and sharing your "LOL!" moments still matter. Learning from mistakes remains one of the most valuable skills, and according to Garrison, no AI can replace that creative process.
The title of his presentation, Gotta Build ’Em All, was a playful nod to Edgecase 2026's Pokémon-inspired theme, Gotta Catch ’Em All. The reason is simple, Justin Garrison loves building things. A self-confessed tinkerer and vintage hardware enthusiast, Garrison spends much of his free time experimenting with old technology. Among the collection in his garage are classic Mac IIs, but one of his favorite projects involved the transparent case of an Apple G4 Cube. During two separate weeks of isolation, he transformed the machine into a fully functioning Kubernetes cluster.
“I learned how to 3D print, how to laser-cut, and most importantly, how to be patient,” Garrison joked. “I also spent quite a lot of time swearing at it in my garage.”
The Cube still works today, although its practical value is limited. It is noisy, runs hot, and is hardly the ideal production platform.
From cat carrier to edge computing
His next project was even more unconventional.
Garrison built a portable Kubernetes cluster inside a cat carrier, which he dubbed the Petaflop Carrier. Fortunately, his cat seemed perfectly happy with the cardboard box it came in.
“The challenge was figuring out networking and how to make everything work in such a constrained environment,” he explained. “Those are exactly the kinds of challenges you encounter in edge computing.”
The project evolved into a stack of Raspberry Pis combined with a mini PC and external GPU. When preparing for Edgecase 2026, Garrison expanded the idea into an experiment exploring Kubernetes at every possible scale: from the smallest physical cluster imaginable to globally distributed clusters spanning multiple regions.
The question became simple: how small can Kubernetes get, and how large can it grow?
Edgecase 2026: Gotta read 'em all
All the speakers, sessions, and insights from Edgecase 2026 in one place.
Scaling Kubernetes
A Raspberry Pi alone was not interesting enough, according to Garrison.
Using CubeSolar, he demonstrated how Kubernetes can be stripped back to its essentials. With only a single node, for example, there is no need for a scheduler. The result is one of the smallest possible Kubernetes deployments.
But even that is not the limit.
“You can run CubeSolar on a smartwatch with 512 MB of RAM,” he said. “The only problem is that it doesn’t have Wi-Fi, so without another computer attached you can't do very much with it.”
From there, Garrison moved to the opposite extreme. By separating leader and worker nodes, he showed how pods can communicate across regions while maintaining resilience when parts of the infrastructure fail. For this type of deployment, Talos' KubeSpan offers an elegant solution, he argued, enabling connectivity between pods and nodes regardless of the networks or geographic regions they reside in.
Why failure still matters
Then, shortly before his presentation, disaster struck. Garrison accidentally spilled a full glass of water over his carefully assembled hardware collection. “My first thought was that everything was dead,” he said. “I tried all sorts of things, and eventually the hardware started coming back to life.”
For Garrison, that experience reinforced an important lesson. The process of diagnosing failures, understanding what went wrong, and bringing systems back online remains incredibly valuable, even in an age increasingly shaped by AI.
He closed his talk with a plea for curiosity and hands-on learning. “It’s important to think for yourself, build things yourself, and make your own mistakes,” he said. “There’s still something incredibly satisfying about breaking something and then figuring out how to make it work again.”
His takeaway was straightforward: don't be afraid to experiment. Whether you're squeezing Kubernetes onto a smartwatch, designing a globally distributed cluster, or simply connecting two servers together, the best way to learn is still by building.
