Fieldnotes
Notes on running small systems

Monitoring: start with two checks

14 October 2026

The first useful monitoring question is not "what are my metrics" but "how would I know right now that this is broken". For most small systems two checks answer that: is the site responding, and is the certificate valid.

Two checks

A request to the main page every five minutes, expecting a 200 and a known phrase. A TLS handshake, expecting more than fourteen days of validity. That is enough to catch the failures that actually happen: expired certificates, full disks, broken deploys.

Where to send alarms

Somewhere a human looks. A chat channel is fine. Email works if it is a channel someone reads. The alarm must contain what broke, since, and a link — not a metric name.

What not to do first

Dashboards and long retention are second-order. Start with the two checks, keep them boring, and add more only when you have been woken up by something they missed.

← All notes