travaux-com / variantretriever

1.0.3 2020-11-09 11:06 UTC

Requires

• php: >=7.4

Requires (Dev)

• friendsofphp/php-cs-fixer: ^2.16
• pestphp/pest: ^0.2.4
• phpunit/phpunit: ^9.0

MIT

• Thomas P

Published on Packagist.org on 2020-11-20 07:47 UTC

README

VariantRetriever

VariantRetriever is a minimalist package for feature flagging. It's fast, database free and idempotent library to choose over a set of variants.

Getting Started

First of all, you need to define an Experiment with a name and it variants. Variant requires 2 arguments, a name and a rollout percentage (50% by default). Then create a variant retriever with this experiment and variants, and ask it to retrieve a variant for a resource (in the following code, it's for a user uuid).

$variantRetriever = new VariantRetriever();
$experiment = new Experiment('my-ab-test', ...[new Variant('control1'), new Variant('variant2')]);
$variantRetriever->addExperiment($experiment);

$affectedVariant = $variantRetriever->getVariantForExperiment(new Experiment('my-ab-test'), '77d8a1d5-97ba-42db-a4a7-3b9562f0ff22');

var_dump((string) $affectedVariant); // string(7) "variant2"

Running the Test Suite

VariantRetriever uses PHPUnit as testing framework. Once you have all dependencies installed via composer install, you can run the test suite with:

./vendor/bin/phpunit

To obtain the code coverage report, you'll need to have xdebug or pcov installed. Then, you can run:

./vendor/bin/phpunit --coverage-text

And this will give you detailed information about code coverage.

Running the test suite requires PHP 8.4+ (PHPUnit 13). The library itself still supports PHP 7.4+.

What about speed

VariantRetriever is fast. Variant allocations are built once when an experiment is registered, then each lookup is only a hash and an array access.

On a local machine, 100 000 lookups take about 40ms (about 3.7× faster than rebuilding a 50/50 allocation on every call, and about 11× faster with 10 variants). Local runs can do 500 000 lookups in well under a second. Tests also ensure that 10 000 lookups complete in less than 1 second.

What about randomness

Rows	Rollout	Rollout	Max Rollout difference
1k	50/50	46.5/46.5	3.5%
500k	50/50	49.87/49.87	0.13%
500k	10*10	9.88	0.12%
500k	10/10/80	9.88/9.88/79.92	0.12%