← Back to all sparks
P

Prometheus Alertmanager

INFRA · APIS
Velocity2.5

Alert routing, grouping, silencing and notification dispatcher for Prometheus

Alertmanager is rebuilding its internals in the open, and 0.34 makes operators update their dashboards.

alertingprometheusobservabilitynotification-routingevent-recordingbreaking-changes
Current state
Alertmanager handles deduplication, grouping, silencing and routing of alerts from Prometheus. The 0.34.0 release adds templatable labels on alert routes, server-side state filters on the silences API, and splits authError and rateLimited out of the generic clientError reason on the notification failure metric — a change that breaks any dashboard or alert matching reason="clientError" for 401, 403 or 429 responses.
Where it's heading
The last three minor releases show a project willing to break its own surfaces to fix its structure. Version 0.33.0 replaced the global alert marker with per-aggregation-group markers, removed the alertmanager_marked_alerts metric, and moved several types out of public packages; 0.34.0 continues by making a metric label more precise at the cost of existing queries. The structured event recorder, introduced behind a feature flag in 0.33.0, keeps accumulating output types and detail, which is the clearest signal of intent here — Alertmanager is building a first-class record of its own routing decisions.
Prediction
The event recorder is the thread most likely to advance next, given it has gained outputs and event detail in each of the last two releases and remains behind a feature flag. Expect it to stabilise before the flag comes off.

Recent moves

  1. 2d ago

    Notification failure reasons split; templatable route labels added

    Splits authError and rateLimited out of clientError on alertmanager_notifications_failed_total, which silently changes what existing dashboards match — the release notes flag it as requiring updates. Adds templatable labels on alert routes, server-side active, expired and pending filters on the silences API, SNS external_id support, and toDate template functions.

    View source ↗
  2. 1mo ago

    Silence snapshot legacy matchers field restored

    Three bug fixes, the notable one restoring the legacy matchers field in silence snapshots so older Alertmanager versions can read newer snapshots again. A compatibility repair on the 0.33 line.

    View source ↗
  3. 2mo ago

    Per-group alert markers replace the global marker; event recorder lands

    ⚡ SPARK

    Replaces the global alert marker with per-aggregation-group AlertMarkers, removing the alertmanager_marked_alerts metric and relocating marker types out of the public types package. The structured event recorder also arrives behind a feature flag with file, webhook and kafka outputs — the foundation the next release keeps building on.

    View source ↗
  4. 2mo ago

    Dispatcher goroutine leak on alertgroup swap fixed

    A single fix for dispatcher goroutine leaks when a destroyed alertgroup is swapped. One of the contention symptoms the per-group marker rework in 0.33.0 addresses structurally.

    View source ↗
  5. 3mo ago

    Contended route dispatch fixed; prebuilt UI assets shipped

    Fixes dispatching to a contended route and ships prebuilt UI assets in the release. Note this body is byte-identical to the 0.32.1-rc.0 entry below it — the release candidate and the final carry the same notes.

    View source ↗
  6. 3mo ago

    Release candidate carrying 0.32.1 notes verbatim

    The release candidate for 0.32.1, carrying notes identical to the final tag published a day later. Counted as one release, not two.

    View source ↗