Kubernetes Homelab
A 9-node Talos Linux Kubernetes cluster running around 38 services under GitOps, migrated from Docker Swarm without losing service. My working lab for the patterns real infrastructure teams use.
Role: Designed, built, and operate it
The short version
It started as a Plex server. It is now a 9-node Kubernetes cluster on Talos Linux, managed with ArgoCD, running around 38 services. I use it to learn the patterns real infrastructure teams use by operating them, not by reading about them.
What’s in it
- Control plane: 5 nodes, so losing a couple of them doesn’t take the cluster down.
- GPU workloads: scheduled across GPU nodes with NVIDIA MPS, so several workloads share a card instead of one hogging it.
- GitOps: everything reconciles from git through ArgoCD. A push to the repo is the deploy. Rollback is a git revert.
- Below the cluster: OpenTofu provisions the VMs and LXCs, Ansible configures the platform underneath. Infrastructure as code, all the way down.
- CI with teeth: security scanning runs in the pipeline and blocks commits that leak secrets or ship known-vulnerable images.
The migration
The whole thing used to run on Docker Swarm. I moved it to Kubernetes (I call it Evolution 3) without losing service along the way. That meant standing up the new cluster, porting each stack, cutting traffic over service by service, and decommissioning the old managers once nothing depended on them. GPU nodes were repurposed from the old Swarm workers.
Why it matters
Nobody vibe-codes a GitOps Kubernetes cluster into existence. Running this taught me the failure modes that slides never do: what happens at 2am when a node drops, how storage behaves under pressure, what a bad manifest does to a live service. That knowledge transfers straight to the enterprise Azure work I do by day.
The honest test I hold it to: can I fix it at 2am without wanting to throw my laptop. Most of the time now, yes.