Concrete, cooling, then Kubernetes
Built from an empty room: A+B power, HVAC, structured cabling, VLANs, Proxmox clusters — and a public Arch Linux mirror on top.
This one started with a floor plan. Rack and stack, redundant A+B power with UPS and PDUs, HVAC and humidity control, structured cabling, then VLANs and L2/L3 routing before a single virtual machine existed.
Above that: Proxmox VE clusters running Windows Server alongside Arch and Debian, an on-prem Kubernetes footprint, and remote access locked behind Tailscale and WireGuard with Nginx terminating in front. Cloudflare proxies the edge; Sora is the load balancer and exit node.
Then the fun part — hosting an official Arch Linux mirror at high-volume uptime, with cron-driven backups landing in object storage and a secondary archive keeping seven rolling copies.
where the effort went
the numbers
UPS + PDU redundant feeds
cron → object storage
built with
delivery record
The studio needed an owned compute and network foundation that could support internal workloads and a public mirror without treating the physical room as someone else's problem.
Rack layout, A and B power feeds, UPS, PDUs, cooling, and humidity control
Structured cabling, VLANs, and routed network segments
Proxmox clusters for Windows, Arch, and Debian workloads
On-premises Kubernetes and secured remote administration
Cloudflare and Nginx edge path through the Sora load balancer
Public Arch Linux mirror with rolling off-site backups
The constraints mattered as much as the finished interface.
Named machines, as they appear in the diagram. Yes, they are all named after something.
Drawn as it was built. Open it full size — the labels are the interesting part.
after shipping
The cleanest cluster design cannot compensate for an unplanned physical failure domain.
Operating a public service is a practical test of capacity, observability, and restoration procedures.