Course contentsShow
← Cloud and Infrastructure
Lesson 2121 of 2,98945. Observability - Dashboards and SLOsFree lesson

Multi-Window and Multi-Burn-Rate Alerting

Implement sophisticated alerting that detects both fast and slow error budget consumption.

Multi-Window and Multi-Burn-Rate Alerting

You'll learn: How to create smart alerts that catch both sudden incidents and slow degradations before your error budget runs out.

The Problem with Single-Window Alerts

Traditional alerts fire when a threshold is crossed in one fixed time window (e.g., "error rate > 5% for 5 minutes"). This approach misses two critical scenarios:

  1. Fast burns: A catastrophic failure might consume 10% of your monthly error budget in 10 minutes
  2. Slow burns: A subtle regression might eat 2% per day, unnoticed until it's too late

Single-window alerts either fire too often (alert fatigue) or too late (budget exhausted).

How Multi-Window, Multi-Burn-Rate Works

This approach monitors error budget consumption at multiple time scales simultaneously:

  • Short window + high burn rate: Catches fast-moving incidents (1 hour at 14.4x burn rate)
  • Long window + moderate burn rate: Detects slow degradations (6 hours at 6x burn rate)

Think of it like home security: motion sensors catch intruders (fast), while a water leak detector catches slow flooding (slow). You need both.

Each alert layer combines:

  • A short window for immediate signal
  • A longer window for trend confirmation (reduces false positives)

Example Configuration Logic

For a 99.9% monthly SLO (43.2 minutes of allowed downtime):

  • Page immediately: 2% budget consumed in 1 hour (14.4x burn) + confirmed over 5 minutes
  • Page during business hours: 5% budget consumed in 6 hours (6x burn) + confirmed over 30 minutes
  • Ticket for investigation: 10% consumed in 3 days (2x burn) + confirmed over 6 hours

This layered approach gives you early warning for slow problems while still reacting instantly to outages.

Key Takeaway: Multi-window, multi-burn-rate alerting protects your error budget from both sudden incidents and gradual degradations by monitoring consumption at multiple time scales, dramatically reducing both false positives and delayed detection.