Self-Hosting Everything on a Three-Node k3s Cluster

A home lab that runs real apps, restores its own backups and tells you when it breaks. Twelve chapters, each one a thing that runs on three small VMs right now, with the mistakes left in. Made for people who run (or want to run) k3s at home, have more ideas than evenings, and would like their lab to be boring in the good way. What's inside Why Kubernetes at home, and when Docker Compose is the better answer. The hardware: why a NAS with an N100 crashed the cluster, and the one-template way to build every VM. One cluster and Flux: the repository is the cluster, and moving the whole thing in 40 minutes of downtime. Networking: names inside, one tunnel out, zero open ports, and the IPv6 setting that made Chrome forget every name. The data layer: Postgres, MariaDB and S3 in one shared namespace, and the plugin that broke pages without a single error. Observability in 1.5 GiB of RAM: the diet that took Prometheus from 204 rules to 46. Backups that restore: the backup that never ran, a real point-in-time restore, and restore tests that found bugs in themselves. Upgrades without fear: the update checker that couldn't count past 1,000, and the majors night. Losing a node: any one of three nodes can go and the cluster carries on (new in 1.1). Real apps, security hygiene, maintenance as a habit. Most of these chapters started as something going wrong on my cluster, and I wrote down what I did about it. The mistakes stayed in because they're the useful part. Read three chapters free first Chapters 1, 6 and 7 are free to read on rttf.dev. If you like those, the rest reads the same. If you don't, you've just saved €19, and I'll count that as useful feedback. Format PDF (A5) and EPUB, version 1.2.1. About 21,000 words, every number from a real run.