One of the questions I got is about what hardware I am actually running.
The short version is that most of it lives in a small open-frame rack under my desk. The rack has an MS-01, a 4-bay NAS, a switch + patch panel, and the usual power and cabling bits. I also have an old laptop and a desktop with a GPU sitting outside the rack. All of those machines are Proxmox hosts, and the Kubernetes nodes run in VMs provision in this Proxmox cluster.
The rack#
The setup started with one desktop that stayed on all day. It handled media, photos, files, apps, and a few development VMs. Later I added a NAS, and that was the point where backups and shared storage started to feel like part of the system rather than an afterthought.
After that, I added an old laptop and a new MS-01 and tied everything together as a Proxmox cluster. It worked, but it was a bit messy: cables everywhere, machines sitting wherever they happened to fit, and no real structure. Eventually I bought a small open-frame rack so the pile had somewhere proper to live.
Most of the hardware I care about is now in that frame. The laptop and desktop are still part of the Proxmox cluster, but they do not physically fit in the rack, so they stay outside it for now.
The frame is big enough to mount things cleanly, but small enough that it does not take over the room. Because it is open, I have not needed rack fans. If I want to move it, I can unplug one power cable and one network cable and lift the whole thing out.
I could stack another section later if I need more height. For now there is still some empty space, leaving room for airflow and future components, rather than filling every gap with gear I do not actually need.
The NAS sits on a shelf, and a small brush panel keeps the network cables managed. None of this is especially exciting, but that is kind of the point. It is much easier to maintain when I can see what is plugged in and move things around without fighting a knot of cables.
Compute nodes#
The MS-01 is the main machine in the rack. It has enough resources to run the cluster VMs while still leaving room for the occasional Linux or Windows guest when I need.
The laptop is an old daily driver that I no longer use. Its main job is to keep Proxmox quorum. It does not run any Kubernetes guests. It is quiet and low power, so it still earns its place in the setup.
The desktop is the third Proxmox node, mostly because it has a GPU. I use that for occasional local training or inference jobs. It is also the biggest machine in the setup, which is why it stays outside the rack.
In day-to-day use, the loudest part of the whole stack is not the MS-01. It is the NAS disks.
Kubernetes in Proxmox#
Each physical machine is a Proxmox host. The Kubernetes nodes are VMs, and they run on the MS-01 and the desktop. The laptop is part of the Proxmox cluster, but not part of the Kubernetes workload capacity.
Running Kubernetes this way keeps the hardware layer separate from the cluster itself. I can create, resize, or throw away a Kubernetes node without touching the physical machines. It also lets me run other VMs next to the cluster when I need a test machine or a temporary workstation.
The three hosts are not identical, and that is fine. Mixed hardware is normal in a Kubernetes cluster. Node selectors, taints, and labels exist for exactly this kind of thing. It works well enough, even if I would still like to replace the laptop and desktop with two more compact machines someday so the whole cluster fits neatly in the rack. That is a future upgrade, not a current problem.
Power consumption#
Power use is one of the reasons I like this setup. I track the rack with a smart plug, so I can see the actual wattage over time instead of guessing. The MS-01 is efficient for the amount of performance it gives me, and the NAS is designed to be reasonably power-friendly as well.
Compared with running a similar setup in the cloud, the ongoing cost is much easier to justify. Managed Kubernetes means paying for the control plane, nodes, storage, and traffic. At home, I already own the hardware, and the main recurring cost is electricity.
Of course, this is not the same as cloud infrastructure. I do not get managed control planes, multi-zone availability, or someone else replacing failed hardware at 3 AM. But for experimenting, and running personal services, this setup gives me a lot of flexibility for a cost that stays predictable.