README.md

Node distributed locking using redis with lua scripts. Compatible with redis >= 2.6.12. A better alternative to locking strategies based on SETNX or WATCH/MULTI. Refer to Implementation and Alternatives for details.

• Installation
• Overview
• Implementation
• Alternatives
• API
  • redislock.createLock(client, [options])
  • redislock.getAcquiredLocks()
  • redislock.LockAcquisitionError
  • redislock.LockReleaseError
  • redislock.LockExtendError
• Class: Lock
  • lock.acquire(key, [fn])
  • lock.release([fn])
  • lock.extend(time, [fn])
• Tests

Installation

Using npm, you can install redislock with npm install ioredis-lock -S.

Note: since version 3.4.0 it's possible to use this with node@4 once again

Overview

redislock offers both atomic acquire and release operations, avoiding race conditions among clients, as well as the need for lock-specific redis connections. Lock creation requires a node_redis client, and accepts an object specifying the following three options:

• timeout: Time in milliseconds before which a lock expires (default: 10000 ms)
• retries: Maximum number of retries in acquiring a lock if the first attempt failed (default: 0)
• delay: Time in milliseconds to wait between each attempt (default: 50 ms)

const Promise = require('bluebird');
const Redis = require('ioredis');
const client = new Redis();
const lock = require('ioredis-lock').createLock(client, {
  timeout: 20000,
  retries: 3,
  delay: 100,
});

// this uses bind feature of `bluebird`
Promise
  .bind(lock)
  .call('acquire', 'app:feature:lock')
  .catch(err => {
    // handle err
  })
  .call('release')
  .catch(err => {
    // handle err
  })
  .then(() => {
    // all good
  });
});

Supports promises, thanks to bluebird, out of the box:

const Redis = require('ioredis');
const client = new Redis();
const lock = require('ioredis-lock').createLock(client);

const LockAcquisitionError = redislock.LockAcquisitionError;
const LockReleaseError = redislock.LockReleaseError;

lock.acquire('app:feature:lock').then(() => {
  // Lock has been acquired
  return lock.release();
}).then(() => {
  // Lock has been released
}).catch(LockAcquisitionError, (err) => {
  // The lock could not be acquired
}).catch(LockReleaseError, (err) => {
  // The lock could not be released
});

And an example with co:

const co = require('co');
const Redis = require('ioredis');
const client = new Redis();
const lock = require('ioredis-lock').createLock(client);

co(function *(){
  try {
    yield lock.acquire('app:feature:lock');
  } catch (e) {
    // Failed to acquire the lock
  }

  try {
    yield lock.release();
  } catch (e) {
    // Failed to release
  }
})();

Implementation

Locking is performed using the following redis command:

SET key uuid PX timeout NX

If the SET returns OK, the lock has been acquired on the given key, and an expiration has been set. Then, releasing a lock uses the following redis script:

if redis.call('GET', KEYS[1]) == ARGV[1] then
  return redis.call('DEL', KEYS[1])
end
return 0

This ensures that the key is deleted only if it is currently holding the lock, by passing its UUID as an argument. Extending a lock is done with a similar lua script:

if redis.call('GET', KEYS[1]) == ARGV[1] then
  return redis.call('PEXPIRE', KEYS[1], ARGV[2])
end
return 0

Alternatives

Some alternative locking implementations do not use a random identifier, but instead simply invoke SETNX, assigning a timestamp. This has the problem of requiring synchronization of clocks between all instances to maintain timeout accuracy. Furthermore, freeing a lock with such an implementation may risk deleting a key set by a different lock.

Another technique used is to WATCH the key for changes when freeing, achieving a CAS-like operation, as described below:

WATCH key  # Begin watching the key for changes
GET key    # Retrieve its value, return an error if not equal to the lock's UUID
MULTI      # Start transaction
DEL key    # Delete the key
EXEC       # Execute the transaction, which will fail if the key had expired

However, this has the issue of requiring that you use a 1:1 mapping of redis clients to locks to ensure that a competing MULTI is not invoked, and that the release is unaffected by other watched keys.

In addition to the above, most locking libraries aren't compatible with promises by default, and due to their API, require "promisifying" individual locks. redislock avoids this issue by taking advantage of bluebird's nodeify function to offer an API that easily supports both callbacks and promises.

API

The module exports three functions for lock creation and management, as well as two errors for simplified error handling when using promises.

redislock.createLock(client, [options])

Creates and returns a new Lock instance, configured for use with the supplied redis client, as well as options, if provided. The options object may contain following three keys, as outlined at the start of the documentation: timeout, retries and delay.

var lock = redislock.createLock(client, {
  timeout: 10000,
  retries: 3,
  delay: 100
})

redislock.getAcquiredLocks()

Returns an array of currently active/acquired locks.

// Create 3 locks, but only acquire 2
redislock.createLock(client);

redislock.createLock(client).acquire('app:lock1', function(err) {
  redislock.createLock(client).acquire('app:lock2', function(err) {
    const locks = redislock.getAcquiredLocks(); // [lock, lock]
  });
});

redislock.LockAcquisitionError

The constructor for a LockAcquisitionError. Thrown or returned when a lock could not be acquired.

redislock.LockReleaseError

The constructor for a LockReleaseError. Thrown or returned when a lock could not be released.

redislock.LockExtendError

The constructor for a LockExtendError. Thrown or returned when a lock could not be extended.

Class: Lock

The lock class exposed by redislock. Each instance is assigned a UUID v1 string as an id, and is configured to work with the given redis client

lock.acquire[key, [fn]]

Attempts to acquire a lock, given a key, and an optional callback function. If the initial lock fails, additional attempts will be made for the configured number of retries, and padded by the delay. The callback is invoked with an error on failure, and returns a promise if no callback is supplied. If invoked in the context of a promise, it may throw a LockAcquisitionError.

const lock = redislock.createLock(client);
lock.acquire('example:lock', function(err) {
  if (err) return console.log(err.message); // 'Lock already held'
});

lock.release([fn])

Attempts to release the lock, and accepts an optional callback function. The callback is invoked with an error on failure, and returns a promise if no callback is supplied. If invoked in the context of a promise, it may throw a LockReleaseError.

const lock = redislock.createLock(client);
lock.acquire('app:lock', err => {
  if (err) return;

  setTimeout(() => {
    lock.release(err => {
      if (err) return console.log(err.message); // 'Lock on app:lock has expired'
    });
  }, 20000);
});

lock.extend(time, [fn])

Attempts to extend the timeout of a lock, and accepts an optional callback function. The callback is invoked with an error on failure, and returns a promise if no callback is supplied. If invoked in the context of a promise, it may throw a LockExtendError.

const lock = redislock.createLock(client);
lock.acquire('app:lock', function(err) {
  if (err) return;

  setTimeout(function() {
    lock.extend(20000, function(err) {
      if (err) return console.log(err.message); // 'Lock on app:lock has expired'
    });
  }, 20000)
});

Tests

Unit and functional tests are available in the base spec directory, and can be ran using npm test. Additional integration tests, which require an active redis-server configured on the default port and host, can be ran using mocha spec/integration/. Both tests suites are ran as part of the Travis CI build thanks to their support for services such as redis.