A Stack of Cheap Lenovos and a Raspberry Pi Turned Into the Perfect Homelab Rabbit Hole

# A Stack of Cheap Lenovos and a Raspberry Pi Turned Into the Perfect Homelab Rabbit Hole A pile of Lenovo mini machines, a Raspberry Pi 5, and one very dangerous question: what should this cluster run? That’s how every good homelab spiral starts. The setup wasn’t framed as production infrastructure or a polished rack build. It was a bench full of machines bought cheap from government auction sites, each with a 2TB NVMe and 4TB SSD, backed by a NAS that already handles storage and backups. The owner made it clear this cluster was for experimentation, learning, and finally putting idle hardware to work instead of letting half of it sit around like expensive desk decor. ## Cheap Hardware Changes the Whole Mood The best thing about this setup is that it doesn’t feel precious. Someone in the thread said cheap is the best way to go, explaining that their own three nodes were free from work or a school connection. That’s the real homelab energy: not chasing perfect enterprise gear, but turning whatever you can get into something useful. A cluster made from auction Lenovos isn’t just a budget move. It gives you permission to break things, rebuild them, and try ideas that would feel reckless on a machine running the whole house. That mattered because the owner wasn’t pretending to have a grand architecture plan. They’d done duplicate setups before with Syncthing, because they didn’t know much about clustering at the time, and now wanted to try the real thing. That honesty made the thread more interesting. This wasn’t a “validate my perfect design” post. It was someone standing at the edge of a new obsession, asking what rabbit hole was worth falling into first. ## The Raspberry Pi Wasn’t the Star, But It Had a Job The Raspberry Pi 5 in the setup became a small but important character. It wasn’t meant to be a heavy compute node. The owner described it more as an interface machine running Raspberry Pi OS with keyboard, mouse, and display. It would also serve as a secondary Tailscale, Unbound, Pi-hole, and DHCP box for a network separated from the main one. That’s a sensible role. Not glamorous, but extremely useful. Every lab needs something boring that keeps the boring stuff alive. There was also a bigger idea floating around: hardware diversity. Someone had suggested adding the spare Pi to diversify the cluster, and the owner had two more Pis waiting for repairs after accidentally breaking off resistors. That detail is painfully relatable. Homelabs are full of noble plans paused by tiny physical mistakes. The dream is high availability. The reality is sometimes a magnifying glass, a soldering iron, and a board that used to work before your hand slipped. ## High Availability Became the Real Goal The most important technical desire in the thread was clear: the owner wanted services to migrate from one node to another if a node went down. They’d tried a looser version before, and when one machine failed, things just got messy. Someone gave the right name for the dream: high availability. That’s the moment a homelab stops being just “a bunch of machines” and becomes a systems lesson. Running services is easy. Running services that survive failure is where the fun starts biting back. The advice was practical: shared storage matters. CEPH came up because it can use local SSD and NVMe across nodes, while the NAS was also suggested as a remote storage target. Another commenter suggested trying both CEPH and remote NAS storage because the whole setup is experimental anyway. That’s the perfect answer for this kind of lab. Don’t pick the one true architecture on day one. Build both, hurt yourself a little, learn why one feels elegant and the other feels like a maintenance tax. ## CloudStack Entered the Chat and Immediately Started a Fight No homelab thread stays simple for long, so CloudStack appeared. One commenter jokingly said to install CloudStack or Proxmox, then some agents, then mail them all the harvested RAM. The owner admitted they hadn’t thought about CloudStack when buying the machines but might consider it because people had recommended it. Then someone asked the obvious question: what even is CloudStack? Another asked how it compares to Proxmox. The answers were not exactly gentle. One person called CloudStack a behemoth compared to Proxmox, especially for smaller deployments. Another was even harsher, calling it an unnecessary layer. That split says a lot. CloudStack can make sense when you want cloud-style orchestration, self-service, tenant management, and a bigger abstraction layer. But for a small cluster of Lenovos in a learning lab, Proxmox is often the cleaner first step. The heavier the platform, the more time you spend feeding the platform instead of learning from the workloads. Sometimes the “enterprise” answer is just a bigger shovel for a smaller hole. ## The Best Suggestions Were Boring in the Best Way The strongest recommendations were not exotic. Run a mix of containers and VMs. Try Docker inside a container for the learning experience. Use Tailscale daily because it’s easy and useful. Add Pi-hole and Unbound. Use Nginx Proxy Manager once the number of services grows and the “Not Secure” browser warning becomes annoying. Pair it with local DNS so services get real names instead of a sad spreadsheet full of IP addresses. That’s not flashy, but it’s exactly how a lab becomes pleasant to use. That advice also matches the owner’s setup. They already had a NAS for backup and storage, a separate network, a Pi ready for DNS and DHCP duties, and enough local disk to play with distributed storage. The next step isn’t to install every trending tool at once. It’s to build a platform that makes experiments easy: DNS, certificates, remote access, backups, monitoring, and a clean way to spin up and tear down services without losing track of what lives where. ## This Is How a Homelab Turns Into a Skill Tree The charm of the whole thread was that nobody fully agreed on the destination. Some people pushed Proxmox. Some mentioned CloudStack. Some focused on CEPH and shared storage. Some talked about Nginx Proxy Manager, Pi-hole, Tailscale, containers, VMs, and Docker. The owner just seemed excited to finally use the hardware, admitting this was their current hyperfixation after finishing a laptop workstation setup and still trying to get a YubiKey working. That’s not a distraction. That’s the shape of learning infrastructure in real life. The best path for this cluster is probably not one big perfect deployment. It’s a series of experiments: build Proxmox, learn clustering, test HA, break CEPH, compare NAS storage, set up DNS, automate certificates, run a few services, kill a node, watch what survives, then rebuild smarter. The Lenovos don’t need to become a tiny enterprise datacenter overnight. They need to become a safe place to fail. That’s the whole point. Cheap nodes, spare Pi, too much storage, and too many ideas. Honestly, that’s not chaos. That’s a curriculum.