When the homelab became a platform

Why this mattered

The first milestone was getting nodes joined. The more useful milestone was having a place where data services, model serving, traces, and public ingress could coexist.

This belongs in the development timeline because RPi Kubernetes is not a single feature. It is a hybrid k3s homelab with an Ubuntu control plane, four Raspberry Pi 5 workers, Cloudflare Tunnel, and a data platform made from Kafka, Flink, Redis Stack, MinIO, DataHub, Airbyte, Polaris, and observability services. The project only became useful once its infrastructure decisions were written down well enough to be repeated.

Design decision

The important decision was to treat every add-on as part of a platform contract: namespaces, storage, observability, and documented recovery steps came before another shiny service.

The practical stack around this decision includes k3s, Kustomize, Helm, Strimzi Kafka, Flink Operator, Redis Stack, RAGFlow, DataHub, Airbyte, Polaris, MinIO, Prometheus, Grafana, Loki, OpenTelemetry, Cloudflare Tunnel, FastAPI, Next.js. I try to keep the interfaces small: configuration describes intent, runtime code owns behavior, and operational notes explain what a future maintainer should check first.

What I would repeat

Small clusters punish casual overcommitment, so the design keeps heavyweight workloads on the Ubuntu control plane and lets the Pi workers prove the ARM path.

The repeatable pattern is to make the boring path explicit. For this project that means clear repository boundaries, documented setup, predictable deployment commands, and enough observability to know whether the system is healthy or merely quiet.

Reader takeaway

If you are building something similar, start with the workflow you need to repeat every week. Then add only the platform pieces that make that workflow easier to recover, explain, and extend.