Load testing
Your end-to-end tests already know how users use your product. A load test reuses one to find out what happens when many users do it at once.
sisyphus: 1name: Checkout under loadload: test: tests/checkout.sis.yaml target: https://staging.example.com allow_hosts: [staging.example.com, "*.api.example.com"] data: load/users.csv # one row per virtual user ui: { users: 10, ramp: 30s, iterations: 1 } # real browsers running the test replay: { users: 500, ramp: 60s, duration: 5m, think: 1s-3s } thresholds: { http_p95: 800ms, error_rate: 1%, ui_failures: 0 }sis load load/checkout.load.yamlIn Studio, the Load view edits and runs load tests with a live dashboard.
Two kinds of users
Section titled “Two kinds of users”- UI users (
ui:) are real browsers running your end-to-end test. They measure what people feel: how long steps take, and whether things break under load. They’re heavy, so you run tens, not thousands. - Replay users (
replay:) are lightweight: no browser, just the requests your test makes, each user with its own session. Hundreds or thousands of them put real pressure on your backend.
Use both together: replay users load the backend while UI users check the product still works for a person.
Thresholds
Section titled “Thresholds”thresholds decide pass or fail: the 95th-percentile response time (http_p95), the share of failed requests (error_rate), the number of UI users whose test failed (ui_failures). A load test that misses a threshold fails, like any other test, so it can run in CI.
Safety
Section titled “Safety”allow_hosts is required. Replay traffic only ever goes to the target and the hosts listed, and the run refuses to start if the target isn’t listed. UI users are real browsers and load whatever the page loads (analytics, CDNs, third-party APIs), so point them only at environments where that’s acceptable.
Full options: Load test reference.
