A useful separation

Prometheus evaluates alert activity per label set produced by an alert expression. Rule labels are added to the resulting alert, while annotations provide informational fields for richer context. That division supports a practical design: let an application approve the few labels that distinguish or classify an alert, and put human-facing detail elsewhere.

The supplied Prometheus practice guidance also favors actionable, symptom-oriented alerts and recommends relevant consoles to help isolate the responsible component. Those recommendations support keeping diagnostic material available without automatically making every detail part of an alert’s identity.

Four decisions for alert information

This table separates documented mechanics from an application-owned placement decision. The last column is a proposed local policy, not a Prometheus-mandated implementation.

Four-case label and annotation decision table
CaseInformation purposeSource evidenceLabel-set consequenceLabel-or-annotation placementInference to rejectPolicy ownerBounded application-owned action
component identity distinguishing alert instancesSeparate alerts for an application-approved component or service boundary.Alert activity is associated with the label set of an expression result.A selected component value participates in the alert’s label set.Label, when the component boundary is intentionally part of identity.Every diagnostic attribute must distinguish an instance.Service owner with monitoring governance.Document the approved component dimensions and review a proposed new identity label before adoption.
owner-approved severity classificationConvey an agreed operational classification.Rule labels can add alert labels, and a conflicting label key is replaced.A configured severity label is attached to the alert.Label, if the application treats severity as an approved classification.Prometheus defines the application’s severity vocabulary.Incident-management or service owner.Maintain a locally approved severity vocabulary and assign its accountable owner.
console or runbook linkGive responders a route to investigation or response material.Annotations can carry descriptive information, including runbook links; practice guidance recommends relevant consoles.The link is contextual material rather than a required identity dimension in this proposed design.Annotation.A URL necessarily creates a distinct alert instance.Service owner responsible for the destination.Place the link in an annotation and define a local responsibility for keeping its destination useful.
current evaluated valueShow the numeric result that explains the present symptom.Templates can access an alert’s evaluated numeric result.If a rendered value changes in a label, the resulting label set changes too.Annotation.Because a value is template-accessible, it belongs in identity.Service owner with monitoring governance.Render the current value in an annotation; reserve labels for the approved stable distinctions.

The label-set consequence in the final row is a direct design deduction: a label set includes label values, so a different rendered value creates a different set. It is not a claim that Prometheus prescribes this placement policy.

Hypothetical failure walkthrough

This is an unexecuted scenario, not a test report. Suppose a team renders a changing evaluated value into an alert label. Under Prometheus’s label-set model, each changed rendering would produce a different alert label set; in the scenario, the team notices those changing identities.

  1. The team keeps the same symptom condition and retains its approved component and severity labels.
  2. It withdraws the value-as-label design after recognizing that the changing value is diagnostic context rather than an intended identity distinction.
  3. It renders the evaluated value in an annotation instead, preserving the value alongside the alert description for responders.
  4. It reviews whether the remaining labels still express the application’s approved identity and classification choices.

The conditional reasoning preserves the symptom evidence supplied by the alert condition while moving only its changing explanatory value. It does not establish notification-routing behavior, delivery behavior, or an operational result.

Operating policy and alternatives

A sensible local policy is to require an accountable owner for every identity dimension and severity term, use a consistent naming convention chosen by that organization, and review additions, removals, or renames before rule changes are accepted. These are application decisions rather than requirements imposed by the supplied documentation.

An alternative is context-heavy labeling, where measured or diagnostic details are placed in labels to make them conspicuous in label-oriented views. The supplied material shows that label values may use templates and that templates can expose the evaluated value. It does not turn that alternative into a recommended identity scheme. Use it only when its owner explicitly accepts the resulting distinctions and their review burden.

For alert content generally, favor a small set of actionable signals tied to user impact, allow room for short disturbances, and make diagnostic consoles easy to reach. Those are Prometheus practice recommendations; the exact taxonomy and approval workflow remain local choices.

Scope and limitations

This article is limited to the supplied Prometheus alert-rule and alerting-practice excerpts. The excerpts are related official Prometheus guidance on alerting rules and alerting practice. It does not infer undocumented Alertmanager routing semantics, choose thresholds, claim a benchmark or test outcome, or promise that a particular label policy prevents all alerting problems.

Before adopting the policy, identify the governance group that approves identity fields and severity terms, decide which component values are sufficiently stable for identity, and assign responsibility for runbook and console destinations. Those decisions require application context beyond the supplied excerpts.