self-hosted error monitor Open the panel

Alert rules

who gets told, about what, and when to go quiet open the screen

What it is for

Deciding who gets told about what, and when to shut up. With no rules at all, every channel that is on receives every event, and that is how an alert channel becomes a channel nobody reads.

The events

  • erro_novo: showed up now and already has volume.
  • regressao: a fixed error that came back on a later version.
  • em_alta: a known error that jumped above its own average.
  • recorrencia: an old error that piled up volume and still has no task.
  • batida_perdida: a job that stopped sending its signal.
  • url_fora and url_voltou.
  • entrada_muda: an input that is on and stopped bringing anything.
  • insight: a period summary written by an analyst.

What a rule decides

  • Channel: where it goes. A rule without a channel picks no destination: it only sets the threshold, and whoever already listens keeps listening.
  • Project, environment, class, minimum version: the slice.
  • Minimum and window: how many occurrences in how many minutes. That is what separates "an error" from "a problem".
  • Quiet period: how many minutes to wait before notifying about the same thing again. Without it, one error in a burst becomes five hundred messages.
  • Order: rules are evaluated in order, and the first match decides.

Who would be notified?

The dry run at the top of the screen answers that. A rule has order, threshold, window and quiet period, and the result of the set is not the sum of what each one says. Finding out in the middle of an incident is too late.

The dry run makes nothing happen

It records no firing and does not spend the rule quiet period. Asking ten times gives the same answer ten times.

What you should see

The list of rules in evaluation order, and the line saying which events are already under a rule. An event that is under no rule behaves as before: every channel that is on receives it.