Skip to main content
Back to blog

How to Set Latency SLOs for an HTTP API

2 min read
Loadcurl
Loadcurl
Loadcurl Team

An SLO turns “the API should feel fast” into a number you can test. For HTTP APIs, that number should almost always be a latency percentile under a defined load — not an average from a quiet afternoon.

Pick the user-visible interaction​

Examples:

  • POST /checkout p95 < 300 ms at 150 RPS
  • GET /search p99 < 500 ms at 400 RPS
  • Auth token exchange p95 < 150 ms at expected peak

Write the endpoint, percentile, threshold, and RPS context together. An SLO without a load context is incomplete.

Prefer p95 or p99 over the mean​

Averages hide tails. Clients abandon and retry on the tail. Background: What Is p95 and p99 Latency?.

Common patterns:

  • User-facing read APIs: p95
  • Payments / auth / fan-out critical paths: p99
  • Internal mesh hops: tighter budgets so end-to-end still fits

Separate correctness from speed​

A latency SLO should sit next to an availability / error objective:

  • Success rate ≥ 99.5% at the same RPS profile
  • Timeout rate under a small budget

Fast 500s are not success. Loadcurl reports make you look at both status mix and percentiles.

Derive thresholds from product reality​

Sources, in order of honesty:

  1. Current production percentiles at known RPS (APM)
  2. Competitor/UX research budgets
  3. Dependency math (your p99 must fit inside the caller’s timeout)

Then add headroom for regressions. Do not invent “50 ms p99 at 10k RPS” if production never saw that load.

Turn SLOs into load-test pass criteria​

Before Start in Loadcurl:

Duration 600s, ramp-up 60s, target 150 RPS. Pass if successful p95 < 300 ms, errors < 0.5%, average RPS within 10% of target.

After the run, read the report scopes (successful vs all) so timeouts do not silently invent a fake p99. Guide: How to Read an API Load Test Report.

Keep SLOs testable​

  • One primary percentile per endpoint
  • Same load profile when comparing releases
  • Document staging vs production scale factor
  • Re-run after risky deploys; attach PDF

Anti-patterns​

  • SLO on average latency only
  • SLO without RPS or concurrency context
  • Tightening timeouts instead of fixing queues
  • Different payloads every run so trends are meaningless
Browse all

  • Metrics & SLOs · latency

    What Is p95 and p99 Latency?

    Understand p50, p95, and p99 latency percentiles for API load testing — why averages lie, how tails form, and how to read them in a load test report.

  • Metrics & SLOs · p99

    What Causes High p99 Latency in APIs?

    Common causes of high API p99 latency under load — queues, pools, GC, locks, dependencies — and how to confirm them with load-test reports.

  • Metrics & SLOs · RPS

    How Many Requests Do I Need for a Meaningful Load Test?

    Estimate how many total requests make a load test meaningful using duration × RPS, warmup, percentile stability, and Loadcurl quota holds.