The Goose Bookopen source load testing in Rust

Have you ever been attacked by a goose?

Looks harmless, hits hard: given the same two cores and 1,000 users over HTTPS, Goose sent 21,400 requests per second, against 15,800 for k6 and 8,400 for Locust.

Goose is a load testing tool written in Rust. A load test is a small Rust program that depends on the goose crate: you build it with Cargo and run it against a host, choosing the users and the run time on the command line.

Get started Upgrading From 0.18

Goose, a Innovation

The Goose live dashboard about 150 seconds into a load test of 50 users against the Drupal Umami demo site: the KPI row with 50 of 50 users, requests per second, no failures, and the p95 and average response times; the requests per second chart, which climbs during the ramp up and then levels off; the active users chart, which ramps to 50 in 50 seconds and holds; and the response time heatmap with its p50, p95 and p99 lines — A 50 user run against a Drupal Umami demo site.

Run It Yourself

In the run above, users start one a second: the requests per second climb with them and level off, the tiles other than Users cover the whole run, and each column of the heatmap is one second of response times.

That run came from the Umami example in a checkout of Goose, against a copy of the Drupal Umami demo site, with its admin password in the ADMIN_PASSWORD environment variable. --no-reset-metrics keeps the ramp up in its numbers; without it, Goose clears the metrics when the ramp up ends:

cargo run --release --features dashboard \
  --example umami -- \
  --dashboard -H http://umami.example.com \
  -u 50 -r 1 -t 3m --no-reset-metrics

Why Use Goose

Goose is inspired by Locust. Where Locust has a user class with tasks, Goose has a scenario with transactions. HTTP requests go through the Reqwest client. A load test of another protocol, such as raw TCP, times each operation itself and records it in Goose's metrics.

One Process Uses Every CPU Core

A Goose load test runs its users as async tasks on Tokio's multi-threaded runtime, which spreads them across a server's cores. Locust needs a process per core to do the same.

Fills A Server's Uplink

In Tag1's A Goose In The Clouds (2021, an earlier Goose release), one 8 core server simulated 2,000 users at a sustained 2.5 Gbps, limited by its network uplink, and one 32 core server filled its 9 Gbps uplink.

Load Tests Are Rust Code

A load test can use any crate and builds to one binary, to run in CI or a container. See Creating A Load Test.

Results A Pipeline Can Act On

Thresholds fail a run on its metrics, and with a baseline the PDF, HTML and Markdown reports show each number's change since an earlier run.

Watching A Run And Its Reports

While It Runs

The live dashboard needs the dashboard feature, which is off by default; Watching a run shows how to turn it on. Goose then serves it at http://127.0.0.1:5118/ while a load test runs with --dashboard, and saves the run's report when it ends. With --dashboard-control and a dashboard token, the dashboard can also start and stop the test and change its users and settings; Turning it on explains the token, and --dashboard-no-auth, which turns authentication off.

When It Ends

Each --report-file <FILE> writes one report, its type set by the file's extension: PDF (from a build with the pdf-reports feature), HTML, Markdown or JSON. Writing reports shows each. Goose also prints its metrics to the console.

Goose 0.19 runs on one server: Gaggle, the distributed load test, was removed in Goose 0.17.0 and will be restored in Goose 0.20, before the 1.0 release. For a distributed load test today, use Goose 0.16.4.