← All projects
Telemetry service · 2026

Hardware Watcher

A telemetry service for the box that hosts everything here. It answers the question I actually care about — is anything about to go wrong — rather than drawing a wall of graphs nobody reads.

  • Reports the handful of numbers that predict trouble on a single-server setup: disk headroom, memory pressure, service liveness.
  • Runs as its own systemd unit on loopback, reached only through an authenticated route at the edge.
  • Deliberately small — no agent to install, no time-series database to operate, no second thing that can break.
  • Sits behind HTTP basic auth, because a status page is still an information leak.

Written after the specific, boring failure of running out of disk without noticing. Every monitoring product I looked at wanted an agent, an account, and a monthly fee to tell me a number I could read myself in one line of shell.

The restraint is the design. It watches a handful of things and shuts up about everything else, which is the only reason I still pay attention when it does say something.

Downloads
Nothing published for this project yet.