loadtest

When using TCP sockets with keep-alive and multiple cores, loadtest can achieve significant performance, reaching up to 100 krps (requests per second) locally. This allows it to compete with high-performance tools like ab, autocannon, and wrk.

Note: As of the current version, several options for TCP sockets are not yet implemented, such as secure connections via HTTPS.

If you want to access the loadtest API directly within your own JavaScript tests, install it as a development dependency in your project.

$ npm install --save-dev loadtest

To use loadtest as a command-line tool, install it globally. On Ubuntu or macOS, you may need to use sudo to grant the necessary permissions.

# npm install -g loadtest

# On Ubuntu or Mac OS X systems:
$ sudo npm install -g loadtest

To run the synthetic TCP performance benchmark on Ubuntu 22.04 LTS, first install Node.js via nodesource, then clone the repository and execute the benchmark script. Note that the benchmark bin/tcp-performance.js is designed to start a test server on up to three cores and run the loadtest on a single core to ensure the server can handle the load.

# Install Node.js
sudo apt install nodejs

# Download repo and run benchmark
git clone https://github.com/alexfernandez/loadtest
cd loadtest
npm install
node bin/tcp-performance.js

To achieve higher performance by bypassing the standard HTTP implementation and using raw TCP sockets, use the --tcp flag with the loadtest CLI. This mode is particularly effective for high-throughput benchmarking.

Note: When using --tcp, performance is significantly higher than the standard HTTP mode, but it may have different behaviors regarding header handling and connection management.

$ node bin/loadtest.js http://localhost:7357 --cores 1 --tcp

Unlike Apache ab, loadtest allows you to specify a fixed rate of requests per second using the --rps option. When using --rps, the --concurrency option is ignored as clients are created on demand to maintain the rate. Note that --rps is not supported for websockets.

loadtest -c 10 --rps 200 http://mysite.com/

To use the loadtest API for automated testing or performance verification within your own projects, install it as a development dependency using npm.

$ npm install --save-dev loadtest

To access custom data within the statusCallback, you can assign values to request.labels inside the requestGenerator. This data will then be available in the statusCallback via result.labels.

const options = {
    url: 'http://localhost:8000',
    requestGenerator: (params, options, client, callback) => {
        const randomInputData = Math.random().toString().substr(2, 8);
        const message = JSON.stringify({ randomInputData });
        const request = client(options, callback);
        
        // Attach custom data to labels
        request.labels = randomInputData;
        
        request.write(message);
        return request;
    },
    statusCallback: (error, result) => {
        // Access the custom data via result.labels
        console.log('Custom label:', result.labels);
    }
};

Use the loadtest command to simulate load on a URL. Supported protocols are http://, https://, and ws://.

Basic syntax: loadtest [-n requests] [-c concurrency] [-k] URL

Key parameters:

• -n, --maxRequests: Total number of requests to send.
• -c, --concurrency: Number of concurrent clients (default is 10).
• -k, --keepalive: Use keep-alive connections (recommended for most tests).
• -t, --maxSeconds: Maximum time to run the test in seconds (default is 10).

The testserver-loadtest command runs a simple server that returns 'OK' to help you measure the performance limits of loadtest itself. It provides real-time stats on requests per second and latency.

Available options:

• --delay ms: Wait specified milliseconds before answering.
• --error 5xx: Return a specific error code for every request.
• --percent yy: Return an error for only the specified % of requests.
• --cores number: Number of cores to use (defaults to half available).