README

Locksmith

PHP 8.3+ library which helps to manage execution in concurent situations (async code, parallel, distributed systems).

Installation

composer require mimatus/locksmith

Features

Mutual exclusion / exclusive locks / mutex / serialized execution / semaphore with capacity 1 - lot of buzzwords describing ensurance resource/code is accessed/executed only once
Semaphore / shared lock - limit access to resource N times at the same time
Deadlock prevention via max lock wait time and cooperative suspension points
Versioned locks - resource is locked only for higher/equal versions - prevents processing of outdated data/code
Async/concurrent-friendly - cooperative suspension points to allow other lock acquisition attempts or allow lock TTL checks for long running processes
TTL-based locks - locks are held only for specified time
In-memory semaphore implementation for single-process scenarios
Redis-based semaphore implementation for multi-process/distributed scenarios
Extendable via Semaphore interface - implement your own semaphore (e.g., Redis-based, database-based, etc.)
Deadlock prevention via max lock wait time and cooperative suspension points

Roadmap

Basic in-memory & Redis semaphore implementation
Redlock algorithm for Redis semaphore
Predis support for Redis semaphore
AMPHP Redis client support for Redis semaphore
First class support and tests for Redis 7 | Redis 8 | Valkey 9
Feedback and API stabilization
Redis Cluster support
Documentation improvements
MySQL/MariaDB/PostgreSQL semaphore implementation

Usage

Note

Project is still in early stages of development, so API is not stable yet and may change. Feedback is very welcome to help shape the API and make it more intuitive and easy to use.

In-Memory semaphore

For single-process scenarios you can use in-memory semaphore implementation. It allows to limit concurrent access to resource within single process (e.g., number of concurrent HTTP requests, background jobs, or other tasks).

It's suitable mainly for concurrent PHP - AMPHP, Swoole, ReactPHP, etc.

It's not suitable for multi-process scenarios (e.g., multiple PHP-FPM workers, multiple servers) as each process/server will have its own instance of in-memory semaphore. For multi-process scenarios you should use Redis-based semaphore implementation.

$locksmith = new Locksmith(
    semaphore: new InMemorySemaphore(maxConcurrentLocks: 1) // Single lock at a time -> mutex
);

$resource = new Resource(
    namespace: 'test-resource', // Namespace/identifier for resource
    version: 1, // Optional resource version
);

$locked = $locksmith->locked(
    $resource, 
    lockTTLNs: 1_000_000_000, // How long should be resource locked
    maxLockWaitNs: 500_000_000, // How long to wait for lock acquisition - error if exceeded
    minSuspensionDelayNs: 10_000 // Minimum delay between retries when lock acquisition fails
);

$locked(function (Closure $suspension): void {
    // Critical section - code executed under lock

    $suspension(); // Optional - cooperative suspension point to allow other lock acquisition attempts or allow lock TTL checks for long running processes
});

// Lock is released after callback execution

Redis semaphore

For distributed scenarios you can use Redis-based semaphore implementation.

Supported Redis servers:

Redis 7+
Valkey 9+

Supported Redis clients:

PhpRedis
Predis
AMPHP Redis client

$redis = new Redis();
$redis->connect('redis');
$phpRedisCleint = new PhpRedisClient($redis);

$semaphore = new RedisSemaphore(
    redisClient: $phpRedisCleint,
    maxConcurrentLocks: 3, // Max concurrent locks
);

$locksmith = new Locksmith(semaphore: $semaphore);


$resource = new Resource(
    namespace: 'test-resource', // Namespace/identifier for resource
    version: 1, // Optional resouce version
);

$locked = $locksmith->locked(
    $resource, 
    lockTTLNs: 1_000_000_000, // How long should be resource locked
    maxLockWaitNs: 500_000_000, // How long to wait for lock acquisition - error if exceeded
    minSuspensionDelayNs: 10_000 // Minimum delay between retries when lock acquisition fails
);

$locked(function (Closure $suspension): void {
    // Critical section - code executed under lock

    $suspension(); // Optional - cooperative suspension point to allow other lock acquisition attempts or allow lock TTL checks for long running processes
});
// Lock is released after callback execution

Distributed semaphore

Distributed semaphore allows to use multiple semaphore instances (e.g., multiple Redis instances) to achieve higher availability and fault tolerance. It uses quorum-based approach - single lock is successful only if the defined quorum of semaphores is reached.

Implementation of distributed semaphore is based on Redlock algorithm with some adjustments to fit the Semaphore interface and allow cooperative suspension points.

Note

It's important to note that while distributed semaphore can be used Redis instances, it does not have first class support for Redis Cluster or Sentinel. First class support for Redis Cluster is on the roadmap, but in the meantime you can use distributed semaphore with multiple independent Redis instances as a workaround.

$semaphores = new SemaphoreCollection([
    new RedisSemaphore(
        redisClient: $redisClient1,
        maxConcurrentLocks: 3,
    ),
    new RedisSemaphore(
        redisClient: $redisClient2,
        maxConcurrentLocks: 3,
    ),
    new RedisSemaphore(
        redisClient: $redisClient3,
        maxConcurrentLocks: 3,
    ),
]);

$locksmith = new Locksmith(
    semaphore: new DistributedSemaphore(
        semaphores: $semaphores,
        quorum: 2,
    ),
);
$resource = new Resource(
    namespace: 'test-resource', // Namespace/identifier for resource
    version: 1, // Optional resource version
);
$locked = $locksmith->locked(
    $resource, 
    lockTTLNs: 1_000_000_000, // How long should be resource locked
    maxLockWaitNs: 500_000_000, // How long to wait for lock acquisition - error if exceeded
    minSuspensionDelayNs: 10_000 // Minimum delay between retries when lock acquisition fails
);
$locked(function (Closure $suspension): void {
    // Critical section - code executed under lock

    $suspension(); // Optional - cooperative suspension point to allow other lock acquisition attempts or allow lock TTL checks for long running processes
});
// Lock is released after callback execution

Development

Commits

Project follows Conventional Commits for commit messages.

Helpers

All necessary commands are available via Makefile.

To see all available commands:

make help

The most important ones:

composer-install: Install PHP dependencies via Composer.
help: Show help for each of the Makefile recipes.
mago-analyze: Run static analysis via mago.
mago-format: Run code formatting via mago.
mago-lint-fix: Run linting with auto-fix via mago.
mago-lint: Run linting via mago.
run-tests: Run tests via PHPUnit.

License

MIT — see License.

mimatus / locksmith

Maintainers

Details