Crush Test enables you to run high-performance load tests (using oha or k6) directly from AWS Lambda, in the same AWS Region as your deployed service. This is especially useful when running performance tests as part of CI/CD (e.g., GitHub Actions), allowing you to avoid the network variability and latency of running tests from external runners.
flowchart TD
GHA[GitHub Actions Runner<br>us-east-1] -- "aws lambda invoke" --> LAMBDA[Crush Test Lambda<br>us-east-2]
LAMBDA -- "HTTP(s) test" --> SERVICE[Tested Service<br>us-east-2]
- GitHub Actions Runner (in
us-east-1) uses the AWS CLI to invoke the Crush Test Lambda. - Crush Test Lambda (in
us-east-2) runs the load test tool in the same region as the service under test, ensuring the lowest and most consistent RTT.
| Resource | Location / Link |
|---|---|
| CDK Construct README | packages/cdk-construct/README.md |
| CDK Construct API Docs | packages/cdk-construct/API.md |
npm install @pwrdrvr/crush-test-cdk
# or
yarn add @pwrdrvr/crush-test-cdkYou can invoke the Lambda from a GitHub Actions runner (or any machine with AWS CLI access):
aws lambda invoke \
--cli-read-timeout 120 \
--function-name <your-lambda-name> \
--payload '{"tool":"k6","args":["run"],"env":{"TARGET_URL":"https://service.example.com/route"},"testProfile":{"base64Content":"<base64-encoded-k6-script>"},"labels":{"gha_run":"123","test":"api"}}' \
response-k6.json- tool:
"k6"or"oha" - args: Arguments for the tool (e.g.,
["run","-"]for k6) - stdin: For k6, provide the test script as a base64-encoded string (see below)
- labels: Any key-value pairs; echoed back in the response for easy identification
base64 -w 0 my-k6-script.jsThen use the output as the value for "stdin" in your payload.
aws lambda invoke \
--function-name <your-lambda-name> \
--payload '{"tool":"k6","args":["run"],"env":{"TARGET_URL":"https://service.example.com/route"},"testProfile":{"base64Content":"<base64-encoded-k6-script>"},"labels":{"gha_run":"123","test":"api"}}' \
response-k6.jsonaws lambda invoke \
--function-name <your-lambda-name> \
--payload '{"tool":"oha","args":["-n","1000","-c","20","https://service.example.com/route"],"env":{"TOKIO_WORKER_THREADS":"1"},"labels":{"gha_run":"123","test":"api"}}' \
response-oha.jsonAny labels you include in the labels field of your payload are echoed back in the Lambda's response. This makes it easy to correlate response files with the requests that generated them, especially when running multiple tests in parallel.
Important:
Take care to avoid hitting the CPU limit of the Lambda function. If the Lambda is throttled due to CPU exhaustion, the timing and accuracy of your performance tests will be impacted. For best results, allocate sufficient memory (and thus CPU) to your Lambda function for your expected test load.
Example response files for both oha and k6 are provided:
Example responses with body extracted and JSON.parsed are also provided:
The structure of these responses is defined by the output of:
oha -j(for oha)k6 --summary-export(for k6)
Caveats:
- Time values in k6's output are in milliseconds (ms).
- Time values in oha's output are in seconds.
You may find these scripts useful for parsing or post-processing the returned results:
MIT License. See LICENSE for details.
| Metric | Deploy Oha | Deploy K6 | Deploy Packaged Oha | Deploy Packaged K6 |
|---|---|---|---|---|
| Total Requests | 10 | 1,190 | 10 | 1,596 |
| Overall RPS | 2,135.9 | 396.3 | 2,059.9 | 531.3 |
| Min Response Time (ms) | 1.75 | 1.72 | 2.37 | 2.27 |
| p50 Response Time (ms) | 6.03 | 4.27 | 6.53 | 3.89 |
| p90 Response Time (ms) | 19.18 | 584.51 | 22.40 | 6.28 |
| p95 Response Time (ms) | 32.89 | 839.39 | 29.99 | 8.63 |