Run a load test

Send the same request many times at once and read the response-time spread.

In short

A load test sends the same saved request many times at once to see what happens under load: slowing answers, timeouts, rate limits and server errors. It runs in the background and reports throughput, the response-time spread and errors. Load testing is included in Pro, Business and Enterprise.

Before you start

Who: starting a load test needs Edit on the project. Anyone who can open it reads results.
Plan: load testing is part of the paid plans (Pro, Business and Enterprise); Plan & usage shows how many load tests you may run.
First: save the request, make sure it works once, and use a test environment.

Steps

Open "Test" › "Load tests" - or choose "Load test" in the project's … menu - and choose a saved request. Confirm it works once.
Pick a preset or "Custom".
Set "Timeout" and, if you want, "Retries".
Press "Run load test". The test starts at once and the page follows it live; you can keep working.

Result: a summary in plain words, then the numbers below.

Load is real traffic Start on a test environment with small numbers, and make sure you are allowed to create the traffic before you test another organization's system. Requests go out from the {{brand}} server.

Read the results

Example: 500 requests at 20 callers gave median 180 ms, p95 520 ms, p99 4,700 ms. Half were quicker than 180 ms, but one in a hundred took about 25 times as long. A gap that size usually means contention, such as a lock, a connection pool or a cache that sometimes misses.

Look beyond the average Compare p95 and p99 with p50, and compare only runs that used the same request, environment and concurrency. A timeline that climbs with flat errors usually means queuing; flat time with rising errors means a limit is being hit.

Paced runs

A load test against a site nobody proved stays small and goes out at a steady rate, at most 10 requests a second. So it can never flood someone else's server. Its result is labeled paced. Anything over 10 callers or 1,000 requests needs a proved site; prove it from Settings, then Verified sites: /help/api-testing/quality-view#ownership-proof.

Good to know

Closing the page does not stop a test; "Stop" does. A test that stops reporting for two minutes shows as interrupted.
You can have one load test running at a time, and only one against the same site at a time.
Ramp-up raises the number of callers gradually - up to five minutes.
"Stop when the target struggles" ends a run early when most of the last answers are server errors or timeouts. For a site you did not prove it always applies.
Every attempt is counted. A 429 answer is never retried, and it pauses the run for as long as the server's Retry-After asks (at most 60 seconds). Retries stay within a small budget.
"Show all numbers" adds the median. A run shorter than two seconds is too short to plot; choose Soak or a Custom run of 10 seconds or more.
What happens next: the result is kept below the form with who ran it, the load and the figures. Nobody is notified when it ends.

If something goes wrong

The run was kept small and paced: the site is not proved. Prove it in Settings, Verified sites.
Another load test is running: wait for it, or stop it.
Every request timed out: that is still a result - the API did not cope with that load. Try fewer callers.
"You have a load test running already. Wait for it to finish or stop it first."
A message that begins "A load test is running against" means another test is already running against the same site. Try again when it has finished.
"Concurrency must be between 1 and 100.": choose fewer callers. "Set either a total number of requests or a time to run for.": fill in only one of the two. "Total requests must be at most 20000.": lower the total.
"The time to run for must be at most 300 seconds.": shorten the run. "The timeout must be between 1 and 120 seconds." and "Retries must be between 0 and 5.": keep inside those ranges.
"This load test is not running.": it has already ended, so there is nothing to stop.

/help/api-testing/load-baselines
/help/settings/webhooks-and-site-ownership
/help/api-testing/time-bands