My Opi5 Cluster

How it started

The cluster didn’t begin as a carefully planned infrastructure project. It started with 2 × Orange Pi 5s and 1 × Raspberry Pi 3 that I happened to have lying around.

At the time, the Orange Pis were the most powerful machines I had, so they naturally became the foundation of the cluster, and with that, the opi5-cluster was born.

From there, things gradually grew. More Orange Pis joined first, keeping the cluster firmly in ARM territory. Eventually, x86 machines found their way in too, as I started repurposing old laptops and mini-PCs that would otherwise have been gathering dust.

The hardware has changed quite a bit since those early days, but the opi5-cluster name stuck. These days, it's less a description of what's running in the cluster and more a reminder of where the whole thing started.

What is it?

A self-hosted K3s cluster running on a mix of ARM and x86 hardware, providing the foundation for infrastructure experiments, personal services, data platforms, AI workloads, and everything else I decide to build.

Design principles

Six ideas shape every decision in the cluster:

• Self-hosted first. If it can run at home, it runs at home. Cloud services are treated as an edge, not a foundation.
• Everything as code. Every application live in Git and are reconciled via ArgoCD. Nothing is hand-built and hoped for.
• Secrets out of Git. Vault is the single source of truth for credentials, synced into the cluster by External Secrets Operator.
• Heterogeneous by choice. ARM and x86 nodes coexist. Multi-arch images are a requirement, not an afterthought.
• Real services, real stakes. The cluster runs things I actually depend on: DNS, Media, Data, Bots. Uptime matters because the users are real.
• Serious, not bureaucratic. Operated with production discipline (GitOps, Observability, Backups) without meetings, tickets, or change-approval theatre.