I ran Kubernetes at home for the best part of a year, and I am here to report that it was, mostly, a mistake. Three nodes of old hardware, a control plane that used more of my evenings than the services it hosted, and a persistent-volume story that turned every reboot into a small adventure. I was self-hosting about six things. Six. You do not need a fleet orchestrator to keep six containers alive; you need docker-compose up and the discipline to write the file once.
The "mostly" is doing real work in that sentence, though, because it was not all waste. What Kubernetes actually taught me, by making me suffer for every shortcut, was to treat my setup as declarative and reproducible. Once everything was YAML in a git repo, rebuilding a node was boring instead of frightening, and "what is even running on this box" had an answer I could read. That habit outlived the cluster.
I tore it down in the end and moved the lot to a single host with Compose and a tidy reverse proxy. Less magic, more sleep. But I kept the git repo and the declarative mindset, which is the bit that was worth the year. If you want to learn Kubernetes, the homelab is a fine place to do it, just don't confuse learning it with needing it.