K9

Load testing with k6's UX, a Rust engine and a Python API.

Status: v0.1 (alpha) (see the changelog). All of k6's executors, checks and thresholds work, from YAML plans or Python scripts, with HTTP/1.1, HTTP/2, HTTP/3, WebSockets and gRPC. Tests scale across cores and machines, send results to k6's outputs, and run inside pytest.

K9 lets you write load tests in Python, with the script shape, executors, checks, thresholds and metric names you know from k6. The part that sends requests and measures time is written in Rust. Python only describes the work, so most tests never pay for the GIL.

# A K9 script: k9 run script.py
import k9

options = {
    "vus": 50,
    "duration": "1m",
    "thresholds": {"http_req_duration": ["p(95)<300"]},
}

async def default():
    res = await k9.http.get("https://test.k6.io/")
    k9.check(res, {"status is 200": lambda r: r.status == 200})

Why another load tester?

• From k6: the script shape, the seven executors, checks, the threshold syntax, metric names and exit code 99. Your Grafana dashboards and CI configs carry over.
• From Goose: weighted transactions, coordinated-omission correction, live run control and the manager/worker distributed mode.
• Compared with Locust: Locust is the established Python load tester, and if it serves you well, keep using it. K9 targets cases where the load generator's own overhead or measurement accuracy matters. In Locust the Python process does the HTTP work. In K9 it doesn't. Measured on one machine (docs/benchmarks.md):
  • declarative plans send as many requests/s as oha, or more, and 1.4–1.5× k6's, for the least CPU per request;
  • one Python process gives 3–4× Locust's single process;
  • with --workers, Python scripts give 1.4× Locust's 8 processes at 256 VUs.

How Python is used: three tiers

k9 inspect tells you which tier a script runs in.

Quick start

$ pip install k9-load                                # or: uv tool install k9-load
$ k9 new --url http://localhost:3000/health          # a starter script.py for your service
$ k9 run script.py -u 5 -d 10s                       # run it, print k6's summary

Wheels cover Linux, macOS and Windows on Python 3.10 or newer. To build from source, see Development.

k9 new load.yaml writes a declarative plan instead, and k9 convert recording.har turns a browser recording into one.

Documentation

The full docs are at bakirrayan.github.io/k9_load_testing (sources in docs/; uv run --only-group docs mkdocs serve builds them locally):

Development

cargo test --workspace                      # Rust tests
uv sync && uv run pytest                    # Python tests (builds the extension)
uv run --only-group docs mkdocs serve            # the docs site, with the landing page

More in docs/development.md: the crate layout, benchmarks and the test server.

License

Dual-licensed under MIT or Apache-2.0, at your option.