Alerting

Two settings decide how an alert reaches you, and they are set in different places on purpose:

• A channel says how loud each severity is on it. A phone that must ring for a machine down but stay silent for a slow site, and a chat room that takes everything, are two channels with two loudness settings — not two copies of every rule.
• A subject says how much it matters. A machine, or one check on it, can raise or lower its importance: the shared hosting server is worth waking for, the name server that fails when somebody else's network breaks is not. Everything that departs from the default is listed in one place.

The chain is monitor → host → group → global: a subject that says nothing follows its machine, which follows its group, which follows the default. When no rule names a channel, alerts go to every enabled channel.

What is announced#

A fault, its recovery and the closing of a check are always announced; a recovery says what recovered. The first good answer of a new check never alerts, and a server restart re-announces nothing: what a check had is on file, and only a change speaks.

Maintenance windows silence what they cover, one-off or recurring (the nightly backup that slows a site needs a daily window, not a lower threshold). A window can carry a public note on the status page, and the alert held back during a window is delivered at its end if the fault outlived it.