← Back to the chapter · Next detail →
Technical detail · Time-Series Anomalies: Real-Time Pipelines
Late windows and repeatable incidents
A watermark tracks event-time progress; the declared lateness policy controls scoring and revisions. Consider a 60-second event-time window with a 15-second lateness allowance. If the gateway has received 56 of 60 expected readings when that allowance expires, it can score provisionally, attach data_state=partial, and update the same incident if the four late readings arrive before the recomputation deadline. Silently replacing the first score erases the incomplete evidence on which the operator originally acted.
Reliability also needs explicit failure behavior. Dead-letter queues preserve malformed readings for inspection. Idempotency makes repeated writes update the same incident rather than create repeated pages. Exactly-once processing can help where supported, but edge retries, reconnects, and resends still make idempotent writes and duplicate suppression necessary in many IoT pipelines.
Incident key example site: plant-4 asset: compressor-17 detector: residual-ewma-v4 window_start: 2026-07-03T10:15:00Z window_end: 2026-07-03T10:20:00Z incident key: plant-4:compressor-17:residual-ewma-v4:2026-07-03T10:15:00Z Why it matters: If MQTT reconnects, Kafka retries, or the alert writer restarts, the same incident key can update the existing alert instead of paging again.
A stable key ties the revised window to the original alert. Keep its revision history and duplicate count so review preserves both what the operator first saw and what arrived later.