Overview

Lock context manager implemented via redis SETNX/BLPOP.

• Free software: BSD 2-Clause License

Interface targeted to be exactly like threading.Lock.

Usage

Because we don't want to require users to share the lock instance across processes you will have to give them names.

from redis import Redis
conn = Redis()

import redis_lock
lock = redis_lock.Lock(conn, "name-of-the-lock")
if lock.acquire(blocking=False):
    print("Got the lock.")
    lock.release()
else:
    print("Someone else has the lock.")

Locks as Context Managers

conn = StrictRedis()
with redis_lock.Lock(conn, "name-of-the-lock"):
    print("Got the lock. Doing some work ...")
    time.sleep(5)

You can also associate an identifier along with the lock so that it can be retrieved later by the same process, or by a different one. This is useful in cases where the application needs to identify the lock owner (find out who currently owns the lock).

import socket
host_id = "owned-by-%s" % socket.gethostname()
lock = redis_lock.Lock(conn, "name-of-the-lock", id=host_id)
if lock.acquire(blocking=False):
    assert lock.locked() is True
    print("Got the lock.")
    lock.release()
else:
    if lock.get_owner_id() == host_id:
        print("I already acquired this in another process.")
    else:
        print("The lock is held on another machine.")

Avoid dogpile effect in django

The dogpile is also known as the thundering herd effect or cache stampede. Here's a pattern to avoid the problem without serving stale data. The work will be performed a single time and every client will wait for the fresh data.

To use this you will need django-redis, however, python-redis-lock provides you a cache backend that has a cache method for your convenience. Just install python-redis-lock like this:

pip install "python-redis-lock[django]"

Now put something like this in your settings:

CACHES = {
    'default': {
        'BACKEND': 'redis_lock.django_cache.RedisCache',
        'LOCATION': 'redis://127.0.0.1:6379/1',
        'OPTIONS': {
            'CLIENT_CLASS': 'django_redis.client.DefaultClient'
        }
    }
}

This backend just adds a convenient .lock(name, expire=None) function to django-redis's cache backend.

You would write your functions like this:

from django.core.cache import cache

def function():
    val = cache.get(key)
    if not val:
        with cache.lock(key):
            val = cache.get(key)
            if not val:
                # DO EXPENSIVE WORK
                val = ...
                cache.set(key, value)
    return val

Troubleshooting

In some cases, the lock remains in redis forever (like a server blackout / redis or application crash / an unhandled exception). In such cases, the lock is not removed by restarting the application. One solution is to turn on the auto_renewal parameter in combination with expire to set a time-out on the lock, but let Lock() automatically keep resetting the expire time while your application code is executing:

# Get a lock with a 60-second lifetime but keep renewing it automatically
# to ensure the lock is held for as long as the Python process is running.
with redis_lock.Lock(conn, name='my-lock', expire=60, auto_renewal=True):
    # Do work....

Another solution is to use the reset_all() function when the application starts:

# On application start/restart
import redis_lock
redis_lock.reset_all()

Alternatively, you can reset individual locks via the reset method.

Use these carefully, if you understand what you do.

Features

• based on the standard SETNX recipe
• optional expiry
• optional timeout
• optional lock renewal (use a low expire but keep the lock active)
• no spinloops at acquire

Implementation

redis_lock will use 2 keys for each lock named ``:

• lock: - a string value for the actual lock
• lock-signal: - a list value for signaling the waiters when the lock is released

This is how it works:

Documentation

Development

To run the all tests run:

tox

Requirements

OS
Any

Runtime
Python 2.7, 3.3 or later, or PyPy

Services
Redis 2.6.12 or later.

Similar projects

• bbangert/retools - acquire does spinloop
• distributing-locking-python-and-redis - acquire does polling
• cezarsa/redis_lock - acquire does not block
• andymccurdy/redis-py - acquire does spinloop
• mpessas/python-redis-lock - blocks fine but no expiration
• brainix/pottery - acquire does spinloop

Changelog

4.0.0 (2022-10-17)

• Dropped support for Python 2.7 and 3.6.
• Switched from Travis to GitHub Actions.
• Made logging messages more consistent.
• Replaced the redis_lock.refresh.thread.* loggers with a single redis_lock.refresh.thread logger.
• Various testing cleanup (mainly removal of hardcoded tmp paths).

3.7.0 (2020-11-20)

• Made logger names more specific. Now can have granular filtering on these new logger names:

  • redis_lock.acquire (emits DEBUG messages)
  • redis_lock.acquire (emits WARN messages)
  • redis_lock.acquire (emits INFO messages)
  • redis_lock.refresh.thread.start (emits DEBUG messages)
  • redis_lock.refresh.thread.exit (emits DEBUG messages)
  • redis_lock.refresh.start (emits DEBUG messages)
  • redis_lock.refresh.shutdown (emits DEBUG messages)
  • redis_lock.refresh.exit (emits DEBUG messages)
  • redis_lock.release (emits DEBUG messages)

  Contributed by Salomon Smeke Cohen in 80.

• Fixed few CI issues regarding doc checks. Contributed by Salomon Smeke Cohen in 81.

3.6.0 (2020-07-23)

• Improved timeout/expire validation so that:
  • timeout and expire are converted toNoneif they are falsy. Previously onlyNone`` disabled these options, other falsy values created buggy situations.
  • Using timeout greater than expire is now allowed, if auto_renewal is set to True. Previously a TimeoutTooLarge error was raised. See 74.
  • Negative timeout or expire are disallowed. Previously such values were allowed, and created buggy situations. See 73.
• Updated benchmark and examples.
• Removed the custom script caching code. Now the register_script method from the redis client is used. This will fix possible issue with redis clusters in theory, as the redis client has some specific handling for that.

3.5.0 (2020-01-13)

• Added a locked method. Contributed by Artem Slobodkin in 72.

3.4.0 (2019-12-06)

• Fixed regression that can cause deadlocks or slowdowns in certain configurations. See: 71.

3.3.1 (2019-01-19)

• Fixed failures when running python-redis-lock 3.3 alongside 3.2. See: 64.

3.3.0 (2019-01-17)

• Fixed deprecated use of warnings API. Contributed by Julie MacDonell in 54.
• Added auto_renewal option in RedisCache.lock (the Django cache backend wrapper). Contributed by c in 55.
• Changed log level for "%(script)s not cached" from WARNING to INFO.
• Added support for using decode_responses=True. Lock keys are pure ascii now.

3.2.0 (2016-10-29)

• Changed the signal key cleanup operation do be done without any expires. This prevents lingering keys around for some time. Contributed by Andrew Pashkin in 38.
• Allow locks with given id to acquire. Previously it assumed that if you specify the id then the lock was already acquired. See 44 and 39.
• Allow using other redis clients with a strict=False. Normally you're expected to pass in an instance of redis.StrictRedis.
• Added convenience method locked_get_or_set to Django cache backend.

3.1.0 (2016-04-16)

• Changed the auto renewal to automatically stop the renewal thread if lock gets garbage collected. Contributed by Andrew Pashkin in 33.

3.0.0 (2016-01-16)

• Changed release so that it expires signal-keys immediately. Contributed by Andrew Pashkin in 28.
• Resetting locks (reset or reset_all) will release the lock. If there's someone waiting on the reset lock now it will acquire it. Contributed by Andrew Pashkin in 29.
• Added the extend method on Lock objects. Contributed by Andrew Pashkin in 24.
• Documentation improvements on release method. Contributed by Andrew Pashkin in 22.
• Fixed acquire(block=True) handling when expire option was used (it wasn't blocking indefinitely). Contributed by Tero Vuotila in 35.
• Changed release to check if lock was acquired with he same id. If not, NotAcquired will be raised. Previously there was just a check if it was acquired with the same instance (self._held). BACKWARDS INCOMPATIBLE
• Removed the force option from release - it wasn't really necessary and it only encourages sloppy programming. See 25. BACKWARDS INCOMPATIBLE
• Dropped tests for Python 2.6. It may work but it is unsupported.

2.3.0 (2015-09-27)

• Added the timeout option. Contributed by Victor Torres in 20.

2.2.0 (2015-08-19)

• Added the auto_renewal option. Contributed by Nick Groenen in 18.

2.1.0 (2015-03-12)

• New specific exception classes: AlreadyAcquired and NotAcquired.
• Slightly improved efficiency when non-waiting acquires are used.

2.0.0 (2014-12-29)

• Rename Lock.token to Lock.id. Now only allowed to be set via constructor. Contributed by Jardel Weyrich in 11.

1.0.0 (2014-12-23)

• Fix Django integration. (reported by Jardel Weyrich)
• Reorganize tests to use py.test.
• Add test for Django integration.
• Add reset_all functionality. Contributed by Yokotoka in 7.
• Add Lock.reset functionality.
• Expose the Lock.token attribute.

0.1.2 (2013-11-05)

• ?

0.1.1 (2013-10-26)

• ?

0.1.0 (2013-10-26)

• ?

0.0.1 (2013-10-25)

• First release on PyPI.