Documentation Plugins Prometheus Alerting Introduction Alert rule evaluation Alert rule state and health

Alert rule state and health

The plugin reports two independent things about a rule: its state, which is about the alerts it’s producing, and its health, which is about whether the rule itself is working.

They’re worth keeping separate. A rule can be perfectly healthy and producing no alerts, or badly broken and therefore silent, and those look identical if you only watch for firing alerts.

Alert state

State applies to alerting rules and describes what its alert instances are doing.

State is per instance, not per rule. One rule can have some instances firing while others are pending and others have gone inactive. The rule detail page shows this under Instances, with counts and a filter for All, Firing, and Pending.

Recording rules have no alert state. The plugin shows them as Recording.

Health

Health answers a different question: did the last evaluation succeed?

A rule in error health produces no alerts, which is the dangerous case: the thing it was watching could be broken and you’d hear nothing. Common causes are a query that references a metric that no longer exists, a syntax error introduced by an edit, or a query too expensive to complete.

For recording rules the plugin distinguishes this explicitly, showing Recording error when the recording rule itself failed.

Newly created rules

Rules take a moment to reach the ruler after you save them. During that window the plugin shows that the rule is being propagated to the monitoring system and refreshes on its own once it lands. This is normal and usually takes a few seconds. The same applies after deleting a rule.