Download the PHP package aternos/lock without Composer

PHP library for distributed locking

Distributed exclusive and shared resource locking based on etcd (using aternos/etcd).

About

This library was created to provide a highly available distributed locking solution for a file storage system, where a failed or incorrect lock could lead to potential data loss or corruption. Therefore it tries to achieve the best possible consistency while still providing simple usability, stability and scalability. There are lots of other locking libraries out there. Some of them don't allow shared locks, some of them implement complicated algorithms on the client-side which also didn't seem to fit our needs. Therefore we've decided to create our own solution and let etcd do most of the heavy lifting.

Etcd is a fast and reliable key-value store that allows transaction operations to ensure that no unexpected changes were made since the last read (see putIf() and deleteIf() in Aternos\Etcd\Client). This locking library uses those methods to ensure consistency. If two processes try to lock the same resource at the same time one of them will be denied by etcd and then retry its lock by either adding it to the other lock (shared) or waiting for the other lock to finish (exclusive). If more processes are fighting over the same lock, they will start to delay their retries in random intervals to find a consistent conclusion.

Timeouts (UnavailableExceptions) from etcd will also be detected and the operations retried after some delay to avoid problems because of short availability problems.

This library was extracted from a different project to be used in different places and not only for a file storage. Therefore we've already used this in production to create thousands of locks every minute in only a few milliseconds.

Installation

The gRPC PHP extension has to be installed to use this library. See aternos/etcd.

Usage

The most important class is the Lock class. Each instance of this class represents one lock on a resource that is identified by a key. There are also several static functions to set options for all locks, but all of them have default values, so you can just start by creating a lock:

Exclusive/shared locks

There can only be one exclusive lock at the same time, but there can be multiple shared locks. If there is any shared lock an exclusive lock is not possible. Exclusive locks can be useful for write operations while shared locks can be useful for read operations. By default all locks are shared, you can create an exclusive lock like this:

Locking time(out) and refreshing

You always have to set a timeout for your locks. The lock will be released automatically if the time runs out. This is necessary to avoid infinite dead locks. The timeout can be refreshed e.g. after writing or reading a chunk of data. You can also define a threshold for the refresh time. The lock will only be refreshed if the remaining locking time is below this threshold. This avoids spamming etcd with unnecessary queries.

Waiting for other locks

If the resource that you want to access is currently locked exclusively or you need an exclusive lock and there are still shared locks, you might want to wait some time for the locks to be released. You can specify a maximum time to wait for other locks:

Break the lock

Of course it's important to break the lock after finishing the operation to avoid running into timeouts. You can break a lock like this:

Identifier

You can identify yourself (the current process/request) by providing and identifier string. By default this library uses the uniqid() function to create a unique identifier. There aren't many reasons to provide a custom identifier. The default identifier is the same for all locks in the current process. Therefore all those locks are able to take over the other locks that were previously created in the same process. It also would be possible to see which other process (represented by its identifier) is currently holding a lock. There is currently no such function, but that would be easy to add.

There are two ways to provide a custom identifier. Either by setting the default for all locks or by overwriting the default identifier on a specific lock:

Settings

The default identifier above is already one of the static settings, which you can specify for all locks. All settings are stored in protected static fields in the Lock class (at the top) and have their own static setter functions. Below are only a few important ones, but you can read the PHPDoc function comments for further explanations of the other settings. Only change those if you know what you are doing.

Etcd client

You can set an etcd client object to specify the etcd connection details. See this for more information.

Etcd key prefix

You can set a prefix for all locking keys in etcd to avoid conflicts with other data in the same etcd cluster. The default prefix is lock/.