metrics

func RecordJobFailure ¶ added in v0.20.42

func RecordJobFailure(category, subcategory string)

RecordJobFailure increments the per-category failure counter. Should be called only after the failure event (JobFailedEvent, or ReturnLease for retryable pod issues) has been successfully reported, so failed sends do not inflate the counter.

An empty category indicates no classification happened (e.g. the feature flag is off or the classifier is nil); in that case no metric is emitted. An empty subcategory is allowed and indicates a matched rule with no subcategory set; it produces an empty-string label value.

func RecordPodForceDeleted ¶ added in v0.22.6

func RecordPodForceDeleted(pool string)

Counts escalations issued, not pods that overran their grace period: the escalation is gated on the repeat-deletion debounce, so a pod that overran its deadline but disappeared during that window is never force deleted and never counted. RecordPodTerminationOverdue measures pod behaviour.

func RecordPodTerminationOverdue ¶ added in v0.22.6

func RecordPodTerminationOverdue(pool string, overdue time.Duration)

A pod that terminated inside its grace period - the healthy case - is counted separately rather than observed as a negative value, which would make the histogram's _sum non-monotonic and so break rate() over it.

func RecordRuleEvaluationDuration ¶ added in v0.20.43

func RecordRuleEvaluationDuration(category, subcategory string, duration time.Duration)

RecordRuleEvaluationDuration records the time a single classification rule took to evaluate. Called for every rule regardless of match outcome. An empty category is a no-op to avoid an empty failure_category label.