Recipe

SLO Design Patterns

Proven patterns for defining, measuring, and maintaining service-level objectives that align with user expectations.

1. Request-Driven SLO

Define availability as the fraction of successful requests over a rolling window. Use status codes and latency thresholds as binary classifiers.

good_requests / total_requests ≥ 99.9%

2. Windowed Burn Rate

Track error budget consumption rate over short and long windows. Alert when burn rate exceeds sustainable thresholds before the budget is exhausted.

burn_rate = error_ratio / (budget / window)

3. Multi-Window Alerting

Combine a short window for fast detection with a long window to suppress false positives. Page on short-window burn, ticket on long-window exhaustion.

Short

5m · burn > 14.4x

Long

1h · burn > 1x

4. Client-Side SLO

Instrument the client to measure latency and success from the user's perspective. Aggregate percentiles at the edge before they reach your metrics pipeline.

Next: learn how to wire these patterns into Meridian's alerting pipeline.

Browse all recipes →