Generated at:
Prometheus Rule Definition
Definition
An individual rule within a group. A rule is either a recording rule (has record) or
an alerting rule (has alert). Exactly one of record or alert must be set.
Recording rules pre-compute a PromQL expression and store the result as a new metric. Alerting rules evaluate a PromQL expression and fire alerts when the result is non-empty.
Name of the output metric for a recording rule. Must be a valid Prometheus metric name.
Mutually exclusive with alert.
Name of the alert for an alerting rule.
Mutually exclusive with record.
PromQL expression to evaluate.
For recording rules, the expression must satisfy these constraints (alerting rules are not subject to them):
- Maximum query window: the longest matrix-selector range (or subquery
range, plus
lookbackDelta) may not exceed 30 days + 15 minutes. - Minimum evaluation interval depends on the query window — see the
band table on the group's
intervalfield, including the RR-over-RR carve-out. - At most 7 distinct metric names may be referenced (only enforced when the query window exceeds 1h+10m).
- Each matrix selector must be bounded — at least one non-
__name__label matcher, an enclosing range function (rate,*_over_time, …), an enclosingby/withoutaggregation, or the inner metric must itself be an active recording-rule output in the same (organization, dataset). - At most 10
orarms (only enforced when the query window exceeds 1h+10m). - Self-reference is rejected: the rule's own
recordmay not appear in its own expression. - Vector matching must resolve to a clean, aligned match: a binary
operation matched with
on()and an empty label set is rejected when the one-side operand is not statically a single series (an unaggregated selector there fails at evaluation with a duplicate-series error). Aggregate the one-side to a single series (e.g.sum(...)) or match on explicit labels withon(<labels>).
Duration for which the expression must be continuously satisfied before an alert fires. Only applicable to alerting rules (e.g., "5m", "10m").
Duration for which an alert continues firing after the expression is no longer satisfied. Only applicable to alerting rules (e.g., "5m").
Additional labels to attach to the resulting metrics or alerts.
Annotations to attach to alerts. Only applicable to alerting rules.
Commonly used keys include summary and description.
Dash0 also recognizes:
dash0.com/notification-channel-ids: Comma-separated list of notification channels to notify when the alert fires. Each entry is a channel UUID or a channel origin, and may carry the:critical-onlysuffix to notify that channel only once the check reaches critical — for exampledash0.com/notification-channel-ids: fbc39ea2-d6da-4c8e-b77f-730821913a3c, pagerduty-prod:critical-only. The annotation declares the routing in full: a channel it does not name is unlinked, and a channel it names without the suffix notifies on every issue, including when the only-on-critical trigger had been set in the UI.dash0.com/on-no-data: The evaluation outcome when the expression returns no data. The values arecritical,degradedandignored. An absent key or an empty value meansignored. Any other value is rejected.
Used by 1
These schemas and endpoints reference this type directly.