Error State Bypasses NoData Handling in Grafana Unified Alerting Rule Evaluation

Context

Grafana unified alerting (v9+) distinguishes between NoData state (query returns zero points) and Error state (query execution fails). NoData behavior is configurable per rule: Alerting, KeepLast, or Resolve. Error state, however, always fires an alert unless intercepted by a catch-all rule, creating an asymmetric handling path that operators must account for when designing noise-resistant notification pipelines.

Goal

Define a consistent strategy for handling both NoData and Error states across a fleet of alert rules so that transient query failures do not generate notification storms while genuine signal loss remains visible.

Constraints and Uncertainty

The KeepLast option for NoData preserves the Alerting state across data gaps, which can mask real recoveries if queries intermittently fail. Error state offers no equivalent "keep last" knob—it unconditionally transitions to Alerting. Inhibition rules and mute timings operate on firing alerts, but they cannot distinguish whether an alert originated from NoData or Error, limiting downstream suppression logic. Contact point templates also lack access to the originating state type, preventing enrichment-based routing.

Questions

1. What pattern ensures Error-state alerts are routed to a triage queue without paging, while NoData alerts follow the standard severity path?
2. Can a single catch-all rule reliably intercept Error-state firings across all namespaces without duplicating notification policies?
3. Does the KeepLast behavior for NoData interact with group_wait timing in a way that delays recovery notifications beyond the configured interval?