master · Module 3.6
Google SRE Multi-Window Burn-Rate Alerting
Learning Objectives
- Define an SLO, error budget, and burn rate.
- Explain why multi-window, multi-burn-rate alerts balance fast detection with low false positives.
- Order burn-rate thresholds so shorter windows require higher burn rates.
Key Concepts
SLO, error budget, burn rate
An SLO target (e.g., 99.9%) implies an error budget = (1 - target) x window. Burn rate is how fast you are consuming that budget relative to the sustainable pace (1x burns it exactly over the window).
Multi-window, multi-burn-rate
The Google SRE workbook pairs a fast window (catch severe burns quickly) with a slower confirmation window (avoid alerting on brief blips). A page fires only when both agree, balancing sensitivity and precision.
Threshold ordering
For a fixed budget fraction, a shorter alert window requires a higher burn-rate threshold than a longer window — the sandbox enforces this ordering.
Insight
Burn-rate alerting is traffic-independent: it pages on budget consumption rate, so a popular match doesn't trigger false pages the way raw error-count alerts do.
Info
The SLO sandbox computes allowable downtime = (1 - target/100) x window and rejects targets outside (0, 100].
Knowledge Check
1. What is the error budget for a 99.9% SLO over a 30-day window?
2. Why use a multi-window, multi-burn-rate alert instead of a single-window one?
3. For a fixed budget fraction, how do burn-rate thresholds order across windows?
4. Why is burn-rate alerting preferable to raw error-count alerting for a live match?