AMQP

This is an AMQP client implementing the version 0.9 of the protocol.

The goal of this implementation is to provide a PHP land implementation (for ease of use and readability) with a clear separation between the AMQP Model, transport layer and user API.

Note: This implementation couldn't have been done without php-amqplib that helped a lot to figure out the details of the transport layer.

Important: If you are using RabbitMQ be aware that it doesn't implemented the specification completely, Qos and Recover methods are not implemented. And if you find yourself using Value implementations note that ShortString, SignedLongLongInteger and SignedShortInteger generate server errors on some methods (like using them as message headers).

Documentation

Important: you must use vimeo/psalm to make sure you use this library correctly.

Installation

composer require innmind/amqp

Usage

use Innmind\AMQP\{
    Factory,
    Command\DeclareExchange,
    Command\DeclareQueue,
    Command\Bind,
    Command\Publish,
    Model\Basic\Message,
    Model\Exchange\Type,
};
use Innmind\Socket\Internet\Transport;
use Innmind\TimeContinuum\Earth\ElapsedPeriod;
use Innmind\OperatingSystem\Factory as OSFactory;
use Innmind\Url\Url;
use Innmind\Immutable\Str;

$os = OSFactory::build();
$client = Factory::of($os)
    ->make(
        Transport::tcp(),
        Url::of('amqp://guest:guest@localhost:5672/'),
        new ElapsedPeriod(1000), // timeout
    )
    ->with(DeclareExchange::of('crawler', Type::direct))
    ->with(DeclareQueue::of('parser'))
    ->with(Bind::of('crawler', 'parser'))
    ->with(Publish::one(Message::of(Str::of('https://github.com')))->to('crawler'))
    ->run(null)
    ->match(
        static fn() => null, // success
        static fn($failure) => throw new \RuntimeException($failure::class),
    );

The above example will declare an exchange named crawler and queue parser that will receive messages from our exchange. Finally it will publish a message with the payload http://github.com/ to crawler (and the server will route it to parser).

And to consume the messages you have 2 approaches:

use Innminq\AMQP\{
    Command\Get,
    Command\Consume,
    Consumer\Continuation,
    Model\Basic\Message,
};

$state = $client
    ->with(Get::of('parser')->handle(static function($state, Message $message, Continuation $continuation) {
        $state = $message->body()->toString();

        return $continuation->ack($state);
    }))
    ->run(null) // <- this argument will passed as the state to the handler above
    ->match(
        static fn($state) => $state,
        static fn($failure) => throw new \RuntimeException($failure::class),
    );
echo $state; // will print "http://github.com/"
// or
$client
    ->with(Consume::of('crawler')->handle(static function($state, Message $message, Continuation $continuation) {
        doStuff($message);

        return $continuation->reject($state); // to reject the message
        return $continuation->requeue($state); // put the message back in the queue so it can be redelivered
        return $continuation->cancel($state); // instruct to stop receiving messages (current will be acknowledged first)
    }))
    ->run(null)
    ->match(
        static fn() => null, // in this case only reachable when you cancel the consumer
        static fn($failure) => throw new \RuntimeException($failure::class),
    );

reject() and requeue() can also be used in the get callback.

Feel free to look at the Command namespace to explore all capabilities.

Benchmarks

make benchmark run on a MacBookPro18,2 (M1 Max, 32Gb RAM) with a RabbitMQ running in a container (via docker for mac) produces this result:

make benchmark
Publishing 4000 msgs with 1KB of content:
php benchmark/producer.php 4000
0.48978996276855
Consuming 4000:
php benchmark/consumer.php
Pid: 701, Count: 4000, Time: 2.3580

By comparison, the php-amqplib produces this result:

Publishing 4000 msgs with 1KB of content:
php benchmark/producer.php 4000
0.15483689308167
Consuming 4000:
php benchmark/consumer.php
Pid: 46862, Count: 4000, Time: 0.2366

So it appears pure functions come at a cost!

Note: both benchmarks use manual acknowledgement of messages