perfscale load test action

Run a load test with perfscale — k6, locust, JMeter, or perfscale's own native step engine — and collect the metrics as a downloadable artifact, straight from a GitHub Actions workflow.

The action downloads the pinned perfscale release binary for the runner's OS/arch, verifies its sha256, runs your test, and packs the run summary plus any generated metrics files into a single zip. By default it also renders the metric table straight into the GitHub Actions job summary and writes a machine-readable perfscale-summary.json (requests, RPS, latency percentiles, error rate + engine/VUs/duration/timestamp metadata).

Usage

- uses: Perfscale/github-action@v1
  with:
    file: examples/hello.test.yaml
    config: examples/hello.config.yaml

Upload the produced artifact:

- uses: Perfscale/github-action@v1
  id: loadtest
  with:
    k6: script.js
- uses: actions/upload-artifact@v4
  with:
    name: perfscale-metrics
    path: ${{ steps.loadtest.outputs.output-file }}

Engines

Set exactly one of k6, locust, jmeter, or file.

# k6
- uses: Perfscale/github-action@v1
  with:
    k6: examples/hello.k6.js

# locust
- uses: Perfscale/github-action@v1
  with:
    locust: examples/hello.locust.py
    host: https://example.com

# JMeter — the plan owns the load shape; needs the `jmeter` binary on PATH
- uses: Perfscale/github-action@v1
  with:
    jmeter: examples/hello.jmx

# native step engine (requires config)
- uses: Perfscale/github-action@v1
  with:
    file: examples/hello.test.yaml
    config: examples/hello.config.yaml

k6, locust, and jmeter shell out to the respective binary on PATH — the action installs perfscale itself, not the engines. k6/locust are one-package installs; JMeter needs Java (preinstalled on GitHub-hosted runners) plus the distribution tarball:

- name: Install JMeter
  run: |
    curl -fsSL https://archive.apache.org/dist/jmeter/binaries/apache-jmeter-5.6.3.tgz \
      | tar -xz -C "$RUNNER_TEMP"
    echo "$RUNNER_TEMP/apache-jmeter-5.6.3/bin" >> "$GITHUB_PATH"

JMeter runs emit no latency percentiles — the final console summary = line is translated into the summary block with avg/min/max only, and the jmeter process exit code is the gate (no thresholds integration).

Inputs

Input	Default	Description
version	latest	perfscale release to install (tag like v0.2.0, or latest).
k6	—	Path to a k6 .js script (perfscale run --k6).
locust	—	Path to a locustfile (perfscale run --locust).
jmeter	—	Path to a JMeter .jmx test plan (perfscale run --jmeter). Requires jmeter on PATH.
file	—	Path to a native test.yaml (perfscale run -f). Requires config.
config	—	Path to config.yaml (-c). Required with file.
host	—	Target base URL for the locust engine (--host).
report	—	POST the summary to a perfscale serve instance (--report).
args	—	Extra raw arguments appended verbatim to perfscale run.
summary-export	perfscale-summary.json	Path for the machine-readable JSON summary (--summary-export). Empty string disables it.
job-summary	true	Render the metric table into $GITHUB_STEP_SUMMARY.
output	perfscale-output.zip	Path of the zip artifact holding the summary + metrics.
working-directory	.	Directory to run perfscale from.

summary-export and job-summary need a perfscale release that ships --summary-export; on older releases the action degrades to a warning and runs the test without them.

Outputs

Output	Description
summary-file	Path to the captured run summary (perfscale stdout).
summary-json	Path to the machine-readable JSON summary (empty when disabled/unsupported).
output-file	Path to the zip artifact with the summary and metrics files.
exit-code	Exit code returned by perfscale run.

Consuming the JSON summary

- uses: Perfscale/github-action@v1
  id: loadtest
  with:
    file: examples/hello.test.yaml
    config: examples/hello.config.yaml
- name: Gate on p95
  run: |
    p95=$(jq '.summary.p95_ms' "${{ steps.loadtest.outputs.summary-json }}")
    echo "p95 = ${p95} ms"
    awk "BEGIN { exit !(${p95} < 500) }" || { echo '::error::p95 over budget'; exit 1; }

Exit codes

Mirrors the CLI: 0 = run completed (even if requests/checks failed), 1 = the run could not execute (missing file, invalid YAML, engine binary not found, engine crashed before producing results), 2 = invalid arguments.

Platforms

Linux, macOS, and Windows runners (x64 and arm64) — matching the perfscale release matrix.

License

MIT