A p99 latency target is set
The service states how fast its slowest normal requests should be.
MediumChecked automatically
- Applies to:
- Every service
- Rule id:
- observability.latency-target
Why it matters
Averages hide the requests that hurt. The p99 is what one user in a hundred gets, and in a chain of calls most users hit somebody's p99. A stated target lets callers set their timeouts from a fact and lets you alert before "slow" turns into "down".
How to apply it
- Set the target on a high percentile (p99, or p95 for low-traffic services), per endpoint class if they differ a lot.
- Measure it at the server and, where you can, at the caller.
- Budget it: your target has to fit inside your callers' timeouts, and your dependencies' targets have to fit inside yours.
Example
slo.latencyP99: 300 ms for reads and 800 ms for writes, with the gateway timeout at 2 s.
References
Checklists with this rule
Related agent skill: performance-scalability