Berlioz Queue Manager

Berlioz Queue Manager is responsible for processing jobs from a queue using a job handler. It supports advanced features like memory and time limits, signal handling, and customizable worker options.

For more information, and use of Berlioz Framework, go to website and online documentation : https://getberlioz.com

Installation

Composer

You can install Berlioz Queue Manager with Composer, it's the recommended installation.

$ composer require berlioz/queue-manager

Dependencies

• PHP ^8.2
• Packages:
  • berlioz/helpers
  • psr/clock
  • psr/container
  • psr/log

Usage

Jobs

JobDescriptorInterface

• Definition: Represents jobs that are ready to be pushed into a queue.
• Example: Defining the structure and payload of a task before queueing it.
• Note: A generic JobDescriptor class is available for creating new messages quickly. However, you can extend or override this class to provide additional control, such as custom payload validation or specific job behaviors.

JobInterface

• Definition: Manages jobs that have been consumed from a queue.
• Example: Handling retries, deleting jobs after processing, or releasing jobs back into the queue.

JobForQueue

• Definition: Ensures specific jobs are routed to designated queues.
• Example: Assigning priority tasks to a high-priority queue.

Jobs handlers

The JobHandlerManager is a central component for managing multiple job handlers in the Berlioz Queue Manager. It implements the JobHandlerInterface and acts as a dispatcher, delegating job processing to the appropriate handler based on the job's name.

use Berlioz\QueueManager\Job\JobHandlerManager;

$manager = new JobHandlerManager($container, $defaultHandler);
$manager->addHandler('email', EmailJobHandler::class);
$manager->addHandler('report', ReportJobHandler::class);

$job = new Job('email'); // Example job with name 'email'
$manager->handle($job); // Delegates to EmailJobHandler

The JobHandlerInterface defines the contract for handling jobs in the Berlioz Queue Manager. Implementing this interface allows you to define how specific jobs should be processed.

Below is an example implementation of a JobHandlerInterface for consuming and processing a job named "foo":

use Berlioz\QueueManager\Handler\JobHandlerInterface;
use Berlioz\QueueManager\Job\JobInterface;
use Berlioz\QueueManager\Exception\QueueManagerException;

class FooJobHandler implements JobHandlerInterface
{
    public function handle(JobInterface $job): void
    {
        if ($job->getName() !== 'foo') {
            throw new QueueManagerException('Invalid job name');
        }

        // Process the job
        $payload = $job->getPayload();
        echo "Processing job 'foo' with payload: " . json_encode($payload);
    }
}

TIP: JobHandlerManager accept a wildcard "*" at the end of job name.

Worker

The Worker class is the main part of the Berlioz Queue Manager and is responsible for processing jobs from a queue using a job handler.

use Berlioz\QueueManager\Queue\MemoryQueue;
use Berlioz\QueueManager\Worker;
use Berlioz\QueueManager\WorkerOptions;
use Berlioz\QueueManager\Handler\JobHandlerManager;
use Psr\Log\NullLogger;

// Create a Job Handler Manager
$jobHandler = new JobHandlerManager($container);

// Initialize the Worker
$worker = new Worker($jobHandler);

// Optionally, set a logger
$worker->setLogger(new NullLogger());

// Configure worker options
$options = new WorkerOptions(
    name: 'worker', // Worker name
    limit: 10, // Max jobs to execute
    memoryLimit: 128, // Memory limit in MB
    timeLimit: 60, // Time limit in seconds
    killFilePath: '/path/to/kill-file', // File to kill process
    sleep: 2, // Sleep time between jobs in seconds
    stopNoJob: true, // Stop if no job
    backoffTime: 0, // Time to wait before retry failed job
    backoffMultiplier: 1, // Multiplier for backoff time
);

// Create a queue instance
$queue = new MemoryQueue();

// Run the worker
$exitCode = $worker->run($queue, $options);

Backoff multiplier

Workers can also define retry behavior when a job fails. When using an exponential backoff, the delay increases for each retry based on the base delay and the multiplier.

The delay for retry n is computed as:

delay_n = base_delay × (multiplier ^ (n - 1))

Examples:

Base delay	Multiplier	Retry 1	Retry 2	Retry 3	Retry 4	Retry 5	Total before giving up
10	—	10	10	10	10	10	50s
2	2	2	4	8	16	32	62s
10	2	10	20	40	80	160	310s
1	3	1	3	9	27	81	121s
30	2	30	60	120	240	480	930s
60	2	60	120	240	480	960	1860s

Graceful Shutdown

The Worker handles PCNTL signals (SIGTERM, SIGQUIT) to ensure a clean exit. If a signal is received:

• The worker finishes the current job processing.
• It then exits with a specific code without starting a new job.

Note: The pcntl extension must be enabled for this feature.

Remote Stop (Kill File)

You can stop a worker remotely by creating a specific file. Configure the path in WorkerOptions:

$options = new WorkerOptions(
    killFilePath: '/tmp/stop_worker_1',
);

To stop the worker, just run: touch /tmp/stop_worker_1. The worker will check for this file's existence after each job.

Queues

DbQueue

The DbQueue is a durable implementation of a queue that uses a database to store jobs persistently. This ensures that jobs remain available even in the event of application or server restarts. By leveraging a database, the DbQueue provides reliability and durability, making it suitable for production environments where job data must not be lost.

Key Characteristics:

• Durable Storage: Jobs are stored in a relational or NoSQL database, ensuring persistence and fault tolerance.
• Transactional Guarantees: Can leverage database transactions to ensure that job insertion, processing, and deletion are atomic operations.
• Scalability: With proper indexing and optimization, the DbQueue can handle large volumes of jobs efficiently.
• Use Cases:
  • Applications that require guaranteed delivery and processing of jobs.
  • Scenarios where jobs must survive server or application crashes.
  • Environments where job metadata (e.g., retries, priorities) must be tracked over time.

While DbQueue offers durability and reliability, its performance may be impacted by database latency compared to in-memory queues. It is best suited for scenarios where persistence and fault tolerance are prioritized over low-latency operations.

use Berlioz\QueueManager\Queue\DbQueue;
use Hector\Connection\Connection;

$dbConnection = new Connection('mysql://localhost:3306');
$queue = new DbQueue(
    connection: $dbConnection, // Database connection
    name: 'default',           // Queue name
    tableName: 'queue_jobs',   // Name of MySQL table
    retryTime: 30,             // Time to wait after failed job
    maxAttempts: 5,            // Maximum attempts of a job
);

Example of schema for MySQL:

CREATE TABLE `queue_jobs`
(
    `job_id`            int unsigned NOT NULL AUTO_INCREMENT,
    `create_time`       timestamp    NOT NULL DEFAULT CURRENT_TIMESTAMP,
    `queue`             varchar(128) NOT NULL DEFAULT 'default',
    `availability_time` timestamp    NOT NULL,
    `attempts`          int unsigned NOT NULL DEFAULT '0',
    `lock_time`         timestamp    NULL     DEFAULT NULL,
    `payload`           json         NOT NULL,
    PRIMARY KEY (`job_id`),
    KEY `INDEX_job` (`queue`, `availability_time`, `lock_time`, `attempts`)
) ENGINE = InnoDB;

If you want to keep the jobs treated:

CREATE TABLE `queue_jobs_done`
(
    `job_id`            int unsigned NOT NULL,
    `create_time`       timestamp    NOT NULL,
    `queue`             varchar(128) NOT NULL,
    `availability_time` timestamp    NOT NULL,
    `attempts`          int unsigned NOT NULL,
    `lock_time`         timestamp    NOT NULL,
    `payload`           json         NOT NULL,
    PRIMARY KEY (`job_id`),
    KEY `INDEX_job` (`queue`, `lock_time`)
) ENGINE = InnoDB;

-- Trigger to make automatic insert into  `queue_jobs_done`
-- the deleted done jobs into `queue_jobs`.

DELIMITER $$
CREATE
    TRIGGER `queue_jobs_AFTER_DELETE`
    AFTER DELETE
    ON `queue_jobs`
    FOR EACH ROW
BEGIN
    INSERT INTO `queue_jobs_done`
    (`job_id`,
     `create_time`,
     `queue`,
     `availability_time`,
     `attempts`,
     `lock_time`,
     `payload`)
    VALUES (OLD.`job_id`,
            OLD.`create_time`,
            OLD.`queue`,
            OLD.`availability_time`,
            OLD.`attempts`,
            IFNULL(OLD.`lock_time`, CURRENT_TIMESTAMP),
            OLD.`payload`);
END;
$$
DELIMITER ;

Memory queue

The MemoryQueue is a lightweight, ephemeral implementation of a queue that stores jobs in memory for the duration of the script's execution. This queue is particularly useful for testing, development, or scenarios where persistent storage is not required. Since the jobs are stored in memory, they are lost when the script ends, making it unsuitable for production environments where job persistence is critical.

Key Characteristics:

• Ephemeral Nature: Jobs exist only during the script's runtime.
• Fast and Lightweight: No external dependencies or storage overhead.
• Use Cases:
  • Unit testing or local development.
  • Short-lived tasks that do not require durability.
  • Simulating job execution flows without external systems.

The MemoryQueue provides all the standard operations of a queue, such as pushing jobs, consuming jobs, and checking the size of the queue, while maintaining a simple in-memory data structure to manage these operations. However, since it lacks durability, it should be used with caution and only in scenarios where the transient nature of the data is acceptable.

use Berlioz\QueueManager\Queue\MemoryQueue;

$queue = new MemoryQueue(
    name: 'default', // Queue name
    retryTime: 30,   // Time to wait after failed job
);

AwsSqsQueue

The AwsSqsQueue is an implementation of a queue that integrates with Amazon Simple Queue Service (SQS), a fully managed message queuing service provided by AWS. This queue leverages the scalability, durability, and distributed nature of SQS to handle job storage and delivery in a reliable and fault-tolerant manner.

Key Characteristics:

• Fully Managed: Offloads the operational complexity of managing infrastructure, scaling, and maintenance.
• Highly Durable: Messages are redundantly stored across multiple data centers, ensuring data durability and availability.
• Scalable: Capable of handling an unlimited number of messages and automatically scaling to meet demand.
• Low Overhead: Removes the need for a dedicated queue server or database.
• Use Cases:
  • Distributed systems requiring reliable asynchronous communication.
  • Scenarios with high message throughput or unpredictable traffic spikes.
  • Cloud-native applications leveraging other AWS services like Lambda or EC2.

With features like visibility timeouts, message delays, and dead-letter queues, AwsSqsQueue provides robust mechanisms for handling complex workflows and ensuring job delivery. However, since it is a cloud-based service, its performance depends on network latency and AWS's regional availability.

use Aws\Sqs\SqsClient;
use Berlioz\QueueManager\Queue\AwsSqsQueue;

$queue = new AwsSqsQueue(
    sqsClient: new SqsClient(...), // Database connection
    name: 'default',               // Queue name
    queueUrl: '...',               // AWS queue URL
    retryTime: 30,                 // Time to wait after failed job
);

RedisQueue

The RedisQueue is a high-performance, in-memory implementation of a queue that uses Redis as its backend. By leveraging Redis’ fast data structures, RedisQueue enables quick enqueue and dequeue operations while providing optional durability through Redis persistence mechanisms. It is well-suited for environments requiring fast throughput and low latency.

Key Characteristics:

• High Performance: Uses Redis’ in-memory storage for rapid job management.
• Optional Durability: Jobs can survive restarts if Redis persistence (AOF or RDB) is enabled.
• Scalable: Easily supports distributed workers and large numbers of concurrent jobs.
• Atomic Operations: Utilizes Redis commands to guarantee atomic push/pop of jobs.
• Use Cases:
  • Real-time applications requiring low-latency job handling.
  • Scalable systems with multiple distributed consumers or producers.
  • Environments where Redis is already in use as a cache or data store.

The RedisQueue is ideal when you need both performance and a degree of durability, while maintaining a simple infrastructure. However, it is important to ensure your Redis instance is properly configured for persistence if job loss on crash is unacceptable.

use Redis;
use Berlioz\QueueManager\Queue\RedisQueue;

$redis = new Redis();
$redis->connect('127.0.0.1', 6379);

$queue = new RedisQueue(
  redis: $redis,    // Redis connection
  name: 'default',  // Queue name
);

Tips:

• For test isolation, use a dedicated Redis database and call $redis->flushDb() before/after your test suite.
• For production, monitor memory usage and persistence settings to prevent data loss.

AmqpQueue

The AmqpQueue is an advanced queue implementation based on the AMQP protocol, typically used with brokers like RabbitMQ. It provides high throughput, supports delayed jobs, priorities, and dead-lettering, and is suitable for distributed applications requiring reliable and scalable message processing.

Key Characteristics:

• AMQP/RabbitMQ Integration: Leverages a message broker (such as RabbitMQ) to provide asynchronous job distribution between producers and consumers.
• Delayed Jobs (without plugin): Uses per-delay queues with a Time-To-Live (TTL) and dead-letter exchange (DLX) to defer the execution of jobs without needing the x-delayed-message plugin.
• Prioritization: Supports message priorities, allowing urgent jobs to be processed before lower-priority ones.
• Dead-Letter Queue: Automatically routes messages that are expired, rejected, or exceed the maximum number of attempts to a dedicated dead-letter queue for later inspection or reprocessing.
• Scalability: Decouples job producers and consumers, allowing horizontal scaling of workers.
• Auto-Cleanup: Delayed queues can be auto-deleted when empty to avoid polluting the broker with unused queues.
• Use Cases:
  • Distributed or microservice architectures needing asynchronous background processing.
  • Workflows where job retries, delays, and prioritization are important.
  • Applications requiring monitoring and recovery of failed jobs.

How it works:

• Push with Delay: When a job is pushed with a delay, it is routed to a temporary queue with a TTL and DLX. Once the TTL expires, the broker moves the job into the main queue for processing.
• Retries: When a job fails, it can be retried with a delay and with a lower priority.
• Dead-Letter: If the number of attempts exceeds maxAttempts, the job is sent to the dead-letter queue for inspection.

use Berlioz\QueueManager\Queue\AmqpQueue;
use AMQPConnection;

$connection = new AMQPConnection([
  'host'     => 'localhost',
  'port'     => 5672,
  'login'    => 'guest',
  'password' => 'guest',
  'vhost'    => '/',
]);
$connection->connect();

$queue = new AmqpQueue(
  connection: $connection, // AMQP connection
  name: 'default',         // Queue name
  maxAttempts: 5,          // Maximum retry attempts
);

Filtering Queues

If you are using a QueueManager with many queues, you can filter them (e.g., to run a worker dedicated to a specific group) using wildcards:

$filteredManager = $queueManager->filter('notifications.*', 'emails');
// This will only process queues like 'notifications.sms', 'notifications.push' and 'emails'.

Rate limiter

The Rate Limiter allows you to control the flow of job processing by defining limits (e.g., "10 jobs per minute"). It can be applied at two levels: the Queue or the Worker.

Queue Level Rate Limiting

When a rate limit is defined on a specific queue, the QueueManager handles it gracefully. If a queue's limit is reached:

• The QueueManager does not wait for the rate to become available.
• It skips the current queue during its rotation and immediately attempts to consume from the next one.

This strategy ensures high availability by processing available jobs from other queues while one is throttled.

Worker Level Rate Limiting

Defining a rate limit on the worker (via WorkerOptions) applies a global constraint on the consumption speed, regardless of the source queue.

• Unlike the queue-level limit, the worker will actively wait (blocking sleep) if the limit is reached before consuming the next job.
• This limit is cumulative: a job will only be processed if both the worker and the specific queue allow it.

use Berlioz\QueueManager\RateLimiter\TimeRateLimiter;
use Berlioz\QueueManager\WorkerOptions;

$options = new WorkerOptions(
    rateLimiter: TimeRateLimiter::createFromString('100/1h') // Global worker limit
);

Creating Rate Limiters

Two factory methods are available in TimeRateLimiter and MultiRateLimiter to easily parse rate strings (e.g., "10/1m", "100/hour", "10/2hours").

Simple Rate Limit

Use TimeRateLimiter::createFromString() for a single constraint:

use Berlioz\QueueManager\RateLimiter\TimeRateLimiter;

// Limit to 10 jobs per minute
$limiter = TimeRateLimiter::createFromString('10/1m');

Multiple Rate Limits

Use MultiRateLimiter::createFromString() to combine multiple constraints (e.g., a burst limit and a daily limit):

use Berlioz\QueueManager\RateLimiter\MultiRateLimiter;

// Limit to 5 jobs per second AND 1000 jobs per day
$limiter = MultiRateLimiter::createFromString('5/1s', '1000/1d');

Supported units include s (seconds), m (minutes), h (hours), and d (days).