Performance testing

Load test the requests you already have

Turn any request or flow into a load test, choose the concurrency and duration, and read latency percentiles, throughput, and error rate. Set thresholds so a regression fails the build, and watch the trend over time.

Commencer Get the desktop app

No separate load-test rig

The request you already send and the flow you already run are the load test. Point a performance run at them, and you are testing the exact calls your app makes, with the same auth, variables, and bodies.

There is nothing new to script and nothing to keep in sync with your functional tests.

  • Reuse existing requests and flows
  • Same auth, variables, and bodies
  • Nothing extra to script or maintain
A load test built from a request you already have. · click to enlarge

Concurrency and duration you control

Dial in how many requests run in parallel and for how long, and ramp up to find where things bend. Model a realistic peak or push until something breaks, on purpose.

  • Set parallelism and run duration
  • Ramp up to find the breaking point
  • Model a realistic peak load
Concurrency and duration tuned for the scenario. · click to enlarge

The numbers that matter

Averages hide the pain. Performance runs report p50, p95, and p99 latency alongside throughput and error rate, so you see what your slowest users actually experience, not just the mean.

Charts make the tail and the error spikes obvious at a glance.

  • p50, p95, and p99 latency, not just averages
  • Throughput and error rate
  • Charts that surface the tail and the spikes
Percentile latency, throughput, and error rate in one view. · click to enlarge

Thresholds that fail the build

Set a ceiling: p95 under 300ms, error rate under 1 percent. A run that breaches it fails, so a performance regression stops a release the same way a broken assertion does.

Thresholds run in CI, so slow code is caught in review, not by your users.

  • Pass or fail on latency and error-rate limits
  • Runs as a gate in CI
  • Catch regressions before release
A run that breached its p95 ceiling, failing the gate. · click to enlarge

Trend over time

Runs are stored in the cloud, so you can see whether last week's change made the p95 worse and when a slowdown crept in. Performance becomes a line you watch, not a one-off you forget.

  • History of every run in the cloud
  • Spot the release that slowed things down
  • Track p95 and error rate over time
p95 latency tracked across runs over time. · click to enlarge

How a load test runs

  1. 1

    Pick a request or flow

    Choose the calls to load test. They run with the same auth, variables, and bodies as your functional tests.

  2. 2

    Set load and limits

    Choose concurrency and duration, and set thresholds for p95 latency and error rate.

  3. 3

    Run locally or in CI

    Run from the CLI in a pipeline so a breach fails the build, or run ad hoc while you tune.

  4. 4

    Read and compare

    Read the percentiles and error rate, and compare against previous runs stored in the cloud.

Frequently asked questions

Do I have to write a separate load-test script?

No. A performance run reuses an existing request or flow, with the same auth, variables, and bodies, so there is nothing new to script or keep in sync.

Which metrics do I get?

p50, p95, and p99 latency, throughput, and error rate, shown as charts so the tail and any error spikes are easy to spot.

Can a slow result fail my pipeline?

Yes. Set thresholds such as p95 under 300ms or error rate under 1 percent, and a run that breaches them fails in CI like any other gate.

Can I compare runs over time?

Runs are stored in the cloud, so you can track p95 and error rate across releases and see exactly when a regression crept in.

Explore more

Testing

API testing →

Monitoring

API monitoring →

Mocking

API mocking →

Commencez gratuitement. Payez quand ça vous fait gagner du temps.

The desktop app is free and open source. Add the cloud when you want hosted mocks, monitoring, and a shared broker.