Laravel queue rate limiting or throttling

Viewed 12129

I am working on an app that requires fetching data from a third-party server and that server allows max 1 request per seconds.

Now, all request send as job and I am trying to implement Laravel "Rate Limiting" to release 1 job per second but unable to figure out why it should be implemented and there is no real-life example in the web.

Did anyone implement it?

Any hint of this?

5 Answers

I'm the author of mxl/laravel-queue-rate-limit Composer package.

It allows you to rate limit jobs on specific queue without using Redis.

  1. Install it with:

    $ composer require mxl/laravel-queue-rate-limit:^1.0
    
  2. This package is compatible with Laravel 5.5+ and uses auto-discovery feature to add MichaelLedin\LaravelQueueRateLimit\QueueServiceProvider::class to providers.

  3. Add rate limit settings to config/queue.php:

    'rateLimit' => [
        'mail' => [
            'allows' => 1,
            'every' => 5
        ]
    ]
    

    These settings allow to run 1 job every 5 seconds on mail queue. Make sure that default queue driver (default property in config/queue.php) is set to any value except sync.

  4. Run queue worker with --queue mail option:

    $ php artisan queue:work --queue mail
    

    You can run worker on multiple queues, but only queues referenced in rateLimit setting will be rate limited:

    $ php artisan qeueu:work --queue mail,default
    

    Jobs on default queue will be executed without rate limiting.

  5. Queue some jobs to test rate limiting:

    SomeJob::dispatch()->onQueue('mail');
    SomeJob::dispatch()->onQueue('mail');
    SomeJob::dispatch()->onQueue('mail');
    SomeJob::dispatch();
    

Assuming you have only single worker you can do something like this:

  • do what has to be done
  • get time (with microseconds)
  • sleep time that is 1s minus difference between finish time and start time

so basically:

doSomething()
$time = microtime(true);
usleep(1000 - ($time - LARAVEL_START));

If you need "throttling" and are not using Redis as your queue driver you can try to use the following code:

public function throttledJobDispatch( $delayInSeconds = 1 ) 
{
   $lastJobDispatched = Cache::get('lastJobDispatched');

   if( !$lastJobDispatched ) {
      $delay_until = now();
   } else { 
      if ($lastJobDispatched->addSeconds($delayInSeconds) < now()) {
         $delay_until = now();
      } else {
         $delay_until = $lastJobDispatched->addSeconds($delayInSeconds);
      }
   }
   Job::dispatch()->onQueue('YourQueue')->delay($delay_until);
   Cache::put('lastJobDispatched', $delay_until, now()->addYears(1) );
}

What this code does is release a job to the queue and set the start time X seconds after the last dispatched job's start time. I successully tested this with database as queue-driver and file as cache driver.

There are two minor problems I have encountered so far:

1) When you use only 1 second as a delay, depending on your queue worker - the queue worker may actually only "wake up" once every couple of seconds. So, if it wakes up every 3 seconds, it will perform 3 jobs at once and then "sleep" 3 seconds again. But on average you will still only perform one job every second.

2) In Laravel 5.7 it is not possible to use Carbon to set the job delay to less than a second because it does not support milli- or microseconds yet. That should be possible with Laravel 5.8 - just use addMilliseconds instead of addSeconds.

spatie/laravel-rate-limited-job-middleware

This is a nice package if you are using laravel 6 or above. Nice thing is you can configure middleware in the job.

Install

composer require spatie/laravel-rate-limited-job-middleware

You could use this package to use rate limiting with Redis or another source, like a file. Uses settings to set bucket size and rate as fractions of the time limit, so very small storage.

composer require bandwidth-throttle/token-bucket

https://github.com/bandwidth-throttle/token-bucket

It allows you to wrap the check in an if, so it will wait for a free token to be available, 1 a minute in your example. In effect, it makes the service sleep for the required amount of time until a new minute.

Related