Every question, one engine.
Shapes are presets: any scenario runs with any shape. One flag in the CLI, one choice in the UI, one slash command in the console.
Traffic shapes
| Shape | Question | What it does | Command |
|---|---|---|---|
| smoke | Do the journeys work at all? | 2 users or 1/s for a minute | /run smoke |
| baseline | How does it behave at normal traffic? | ramp to normal, hold | /run baseline |
| stress | What happens past the expected peak? | peak, then twice the peak | /run stress |
| spike | Does it survive a sudden jump? | tenfold jump in 10 seconds, hold, drop back | /run spike |
| soak | Do leaks or slow degradation show over hours? | hold for hours | /run soak |
| breakpoint | At exactly what load does it fail? | steps up; stops at the first level that misses a target | /run breakpoint |
| steps | How does each increment change latency? | steps up, never stops early | /run steps |
| recovery | How fast does it return to normal after overload? | normal, threefold overload, normal | /run recovery |
| wave | Does it cope with repeated peaks and troughs? | cycles between levels | /run wave |
The breakpoint and the knee
A breakpoint run checks your targets at the end of each level and stops at the first one that fails, naming the last load that held. For any run whose load changes, the report also finds the knee: the last level where extra load still bought extra throughput.
Conditions
Slower networks
network: { profile: 3g } adds round-trip time, jitter, bandwidth limits and packet loss to every user's connections: slow-3g, 3g, 4g, slow-wifi, or your own values.
Auto-abort
Stop a run when errors or p95 stay above a limit for a set time. The server applies a floor to every run.
Contention and spikes
Last-item races and flash-sale spikes ship in the e-commerce pack; seat-lock races and on-sale rushes in the ticketing pack; login storms, rate-limit bursts, reconnect storms and noisy tenants in the identity, public API, chat and SaaS packs; duplicate-payment races, viral spikes, breaking news, live events, exam starts, results days, dinner rushes, push storms and cold-start bursts in the fintech, social, content, streaming, edtech, government, delivery, mobile and serverless packs.
Replay
mode: replay sends the requests of an access log or HAR file at their recorded times, sped up if you like, with latency per endpoint.
Fault injection
stampede agent slows, throttles or cuts a dependency's connections, pauses containers or scales deployments on a timeline, and reverts everything at the end.
Large payloads
A generate feeder with text(5MB) builds a body of that size once and sends it on every request, to find upload and memory limits.
Planned
Connection floods and slow clients.