radargun

Easy to use benchmarking utility for Node.js. Provides high-precision execution time metrics for: average, min, and max.

Table of Contents

• Usage
• API
• CLI
• Benchmarking Very Fast Functions
• CI Pipelines
• Custom Reporter

Usage

1. Install radargun as a dev dependency in your app:

   $ npm install radargun -D

2. Create a .bench.js script with a series of functions to benchmark:

   benchmark/primes.bench.js

   The bench() function is a method added to the global scope by radargun.

   bench(
     [
       function sieveOfEratosthenes() {
         // calculate primes to 10000
       },
       function sieveOfSundaram() {
         // calculate primes to 10000
       }
     ],
     { runs: 1000 }
   );

3. Execute your script with radargun:

   $ radargun benchmark/primes.bench.js

4. Read the results that are printed to the terminal:

   ┌──────────────────────┬────────────┬────────────┬────────────┐
   │ NAME                 │ AVG        │ MIN        │ MAX        │
   ╞══════════════════════╪════════════╪════════════╪════════════╡
   │ sieveOfErathosthenes │ 1492392 ns │ 1337675 ns │ 5455338 ns │
   ├──────────────────────┼────────────┼────────────┼────────────┤
   │ sieveOfSundaram      │ 802019 ns  │ 688149 ns  │ 3883506 ns │
   └──────────────────────┴────────────┴────────────┴────────────┘

API

bench(functions [, options])

Runs a benchmark analysis for a given array of functions, and generates a report.

Parameters

• functions: (required) an array of functions for which to gather performance metrics. Each function generates its own set of metrics. Every entry in the array must be a function or an object with the following keys:

  • bind: (optional) the this context to use while executing the function. The default is null.

  • fn: (required) a reference to the function to execute.

  • label: (optional) the label to use for this function in the generated report. The default is the name of the function.

  • params: (optional) an array of values to pass into the function.

• options: (optional) an object containing keys to customize the behavior of bench(). This object can have the following keys:

  • reporter: (optional) a function that is responsible for generating the report. The default is the built-in reporter. See Custom Reporter for more information.

  • runs: (optional) the number of times to execute the function. The default is 10.

  • stream: (optional) a writable stream to which status and report information is written. The default is process.stdout.

  • thresholds: (optional) an object that sets performance thresholds for a target function. A failure to stay within these thresholds will cause the radargun utility to exit with 3.

    All metric thresholds are expressed as a percentage of performance, between the target function and all others (1 - target/other) that the target function must meet or exceed.

    • avg: (optional) a number specifying the percentage threshold for "average." If omitted, the target's "avg" metric must simply be less than or equal to all other functions.

    • max: (optional) a number specifying the percentage threshold for "max." If omitted, the target's "max" metric must simply be less than or equal to all other functions.

    • min: (optional) a number specifying the percentage threshold for "min." If omitted, the target's "min" metric must simply be less than or equal to all other functions.

    • target: (required) the target function who's benchmark metrics must fall within all specified performance thresholds. This value is a number specifying an index in the functions array.

Example

const comparisonModule = require('comparison-module');
const fs = require('fs');
 
const myModule = require('../lib');
 
const params = ['foo', 'bar'];
const stream = fs.createWriteStream('results.txt');
 
bench(
  [
    {
      fn: myModule,
      label: 'My Module',
      params,
    },
    {
      fn: comparisonModule,
      label: 'Comparison Module',
      params,
    },
  ],
  {
    runs: 1000,
    stream,
    thresholds: {
      avg: 0.5,
      max: 0.5,
      min: 0.5,
      target: 0,
    },
  }
);

In this example, we are comparing the performance of "My Module" against the performance of comparison-module. If My Module's performance is less than 50% greater than that of comparison-module, radargun exits with error code 3. Results are writen to a file "results.txt."

CLI

The radargun command line interface takes a single parameter, which is a glob that specifies how to find bench files to run.

$ radargun benchmark/**/*.bench.js

Benchmarking Very Fast Functions

In situations where you need to benchmark very fast functions, you get a more accurate understanding of performance by wrapping what it is you want to test in a function, and execute your function in a loop.

bench(
  [
    function () {
      for (let i = 0; i < 1000000) {
        myVeryFastFunction();
      }
    }
  ],
  { runs: 100 }
);

CI Pipelines

It's possible to use radargun in a continuous integration pipeline, and fail a build in the event that a new version of your code sees an unacceptable drop in performance.

Example

const myCurrentModule = require('my-module');
const myUpdatedModule = require('../lib');
 
bench(
  [
    {
      fn: myUpdatedModule,
      label: 'New Hotness',
      params: ['foo', 'bar'],
    },
    {
      fn: myCurrentModule,
      label: 'Currently Published',
      params: ['foo', 'bar'],
    },
  ],
  {
    runs: 1000,
    thresholds: {
      avg: -0.08,
      max: -0.08,
      min: -0.08,
      target: 0,
    },
  }
);

In this example, we are comparing the performance of the implementation of my-module in the local code base against the published version of my-module. Thresholds are set such that if there is a drop in performance of more than 8%, the radargun utility will exit with an error.

Keep in mind that benchmarking is non-deterministic. It's possible that, depending on conditions with the host, radargun could fail with a false negative. It is recommended that you do some experimentation before settling on thresholds to use in your CI pipeline.

Custom Reporter

A reporter is a function used to generate a benchmark report. It is called by the bench() method after it has completed gathering performance metrics. The radargun module ships with a built-in formater, which functions as the default. It is possible to provide a custom reporter function as an option to bench(). This function is called with the parameters:

• stream: a writable stream to which the report should be written.

• report: an array of data containing the performance metrics gathered by the bench() method. Each entry is an object with the keys:

  • avg: a number indicating the average execution time (in nanoseconds) of all runs for the function while benchmarking.

  • label: a string value indicating the label to use in the report. This is typically the name of the function

  • max: a number indicating the highest execution time (in nanoseconds) encountered for the function while benchmarking.

  • min: a number indicating the lowest execution time (in nanoseconds) encountered for the function while benchmarking.

• thresholds: if threshold configuration was provided as options to the bench() method, this value is passed to the reporter. Otherwise, this value is null.