Skip to content

Come have a look around the lab

The HPs do the everyday work. The Threadripper has the 3090. The shed talks to radios over a Wi-Fi bridge. TrueNAS holds the big files and backups, and a Pi runs Omni and DNS. Click a machine to see what is inside it, or explore the private, public and shed network paths below.

The interactive inventory is loading. The full written inventory is also available.

What I would change first

  1. Give the control plane a better disk. The SFF has plenty of capacity; the measured problem is how long durable writes take.
  2. Make the two HPs the everyday pair. Protect selected app data on both machines. Keep the Dell useful without depending on it long term.
  3. Let the Threadripper be the heavy-work machine. Keep its enterprise mirror and GPU. Gradually remove ordinary services' dependence on that host.
  4. Keep the NAS RAM. Its cache is doing useful work. We have not shown that reducing it would be worth the trade-off.

Argo's directory layout is worth keeping. The work is in disk placement, recovery, a few broad diff exceptions, and clearer rules about which jobs run where. It does not call for a new GitOps platform.

The full review has the reasoning and proposed PRs. The inventory links the application, route, volume and documentation CSVs. The recovery guide owns the actual procedures.

What this page knows

This is the September 5, 2026 audit snapshot, not a live monitoring screen. Host hardware and disks came from read-only SSH inspection. Node IPs and software versions were checked again at 2026-09-06 01:48 UTC after the Talos and Kubernetes upgrade. Pod placement and claims still use the earlier audit snapshot, which includes applications subsequently retired. The "what if" view explains dependencies; it does not switch machines off or prove a measured recovery time.

Physical-drive sizes use decimal GB/TB. VM disks and RAM use GiB/MiB. Disk bars compare capacity within a host; they do not claim to show used space. NAS pool bars show the recorded ZFS pool allocation, which differs from the dashboard's usable-dataset accounting. Device names identify the inspected layout and can change after reconnecting hardware.

The complete sanitized page inventory is downloadable inside the explorer. Its source is lab-inventory-data.js; it contains no disk serial numbers, credentials or raw diagnostic dumps. Update that snapshot and its date together after hardware moves; runtimeVersionsCheckedAt records the separate node-version check. Keep measured state separate from suggestions, and check the page at desktop and phone widths before publishing.