Easy to use benchmarking utility for Node.js. Provides high-precision execution time metrics for: average, min, and max.
Table of Contents
-
Install
radargunas a dev dependency in your app:$ npm install radargun -D
-
Create a
.bench.jsscript with a series of functions to benchmark:benchmark/primes.bench.js
The
bench()function is a method added to the global scope byradargun.bench( [ function sieveOfEratosthenes() { // calculate primes to 10000 }, function sieveOfSundaram() { // calculate primes to 10000 } ], { runs: 1000 } );
-
Execute your script with
radargun:$ radargun benchmark/primes.bench.js
-
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 │ └──────────────────────┴────────────┴────────────┴────────────┘
Runs a benchmark analysis for a given array of functions, and generates a report.
-
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) thethiscontext to use while executing the function. The default isnull. -
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 ofbench(). 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 is10. -
stream: (optional) a writable stream to which status and report information is written. The default isprocess.stdout. -
thresholds: (optional) an object that sets performance thresholds for a target function. A failure to stay within these thresholds will cause theradargunutility to exit with3.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 thefunctionsarray.
-
-
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."
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.jsIn 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 }
);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.
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 thebench()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 asoptionsto thebench()method, this value is passed to the reporter. Otherwise, this value isnull.