OBSERVERTC / Concepts

Detector evaluation and issue lifecycles

Understand when evidence becomes an active condition.

A detector checks a particular condition using available measurements and state. Some detectors report observations; others open an issue and later resolve it. Their output is evidence about that condition, not a universal diagnosis of the call.

Evaluate sustained evidence

Many conditions need several observations to distinguish a persistent problem from a short fluctuation. Detection windows collect recent values; recovery windows help determine when a condition has settled. Other detectors use explicit elapsed-time thresholds or immediate state transitions.

Changing collection cadence changes how much time a window covers. It does not necessarily change every detector’s independent timer. See configuration for window sizes and tuning consequences.

Follow a stateful issue

Stateful issue lifecycle: observations, confirmation, raise, evidence updates, recovery, resolutionEligible observationsMeasurements and stateCondition confirmedDetector criteria metKeyed issue raisedAn active occurrence beginsEvidence updatedThe issue remains activeRecovery confirmedRecovery criteria metIssue resolvedThe active occurrence endsThe same issue key links an occurrence from raise to resolution.
Copyable ASCII version
Eligible observations → condition confirmed → keyed issue raised
                                             ↓
                                      evidence updated
                                             ↓
                                      recovery confirmed
                                             ↓
                                        issue resolved

The issue key identifies the active occurrence. Several tracks can have the same issue type, so the type alone is not a unique key. One-off issue records do not have this ongoing lifecycle.

Distinguish absent evidence

If required statistics are unavailable, or the application intentionally pauses a track, evaluation can be gated. An inactive detector or missing input is not equivalent to a healthy measurement. Read each detector’s guards and limitations before treating its output as a guarantee.

Use monitor issues for lifecycle handling and server ingestion for mirrored client issues.

Version coverage varies by reference. Source revisions · Markdown source