Celery broker and results backend implementation for Redis Sentinel

• Free software: MIT license
• GitHub: https://github.com/dealertrack/celery-redis-sentinel
• Documentation: https://celery-redis-sentinel.readthedocs.org.

Redis is a pretty awesome in-memory key-value data-store. Being in-memory makes it wickedly fast however at a cost of no-persistence. In business-critical applications (you know, which make company money) that makes stand-alone redis deployment non-practical. This is where Redis Sentinel comes in. It provides scalability and high availability to Redis 2.X (Redis 3.X comes with native-clustering which is different from Sentinel). As a result, Redis becomes a viable solution for solving some of business needs. As you can imagine from the project title, one use-case is using Redis Sentinel with celery. Unfortunately celery does not support Redis Sentinel by default hence this library which aims to provide non-official Redis Sentinel support as both celery broker and results backend.

Installation is super easy with pip:

$ pip install celery-redis-sentinel

Using this library is pretty simple. All you need to do is configure celery to use Redis Sentinel for broker and/or results backend. That is done with a couple of settings:

# celeryconfig.py
BROKER_URL = 'redis-sentinel://redis-sentinel:26379/0'
BROKER_TRANSPORT_OPTIONS = {
    'sentinels': [('192.168.1.1', 26379),
                  ('192.168.1.2', 26379),
                  ('192.168.1.3', 26379)],
    'service_name': 'master',
    'socket_timeout': 0.1,
}

CELERY_RESULT_BACKEND = 'redis-sentinel://redis-sentinel:26379/1'
CELERY_RESULT_BACKEND_TRANSPORT_OPTIONS = BROKER_TRANSPORT_OPTIONS

Some notes about the configuration:

• note the use of redis-sentinel schema within the URL for broker and results backend. In order to use that schema, which is not shipped with celery, where you create your celery application you must first need to register sentinel support:

  # tasks.py
  from celery_redis_sentinel import register
  
  # has to be called before creating celery app
  register()
  
  app = Celery('tasks')

• hostname and port are ignored within the actual URL. Sentinel uses transport options sentinels setting to create a Sentinel() instead of configuration URL.

Some considerations while using Redis Sentinel as a celery broker. While the failover is in progress, no tasks can be scheduled. Trying to schedule a task will either raise Timeout or ConnectionError. That is because other sentinel nodes within the cluster, depending on the configuration, have a timeout until they elect a new master. During that time, trying to schedule a task will attempt to store it in now-invalid master node hence the exception. If that is unacceptable within your application, this library comes with a small wrapper which allows to trigger tasks which will block the scheduling until new master will be elected:

from celery_redis_sentinel.redis_sentinel import ensure_redis_call
from tasks import add

# this will blow up during failover
add.delay(1, 2)
# this will keep retrying until it succeeds
ensure_redis_call(add.delay, 1, 2)

Alternatively you can use a supplied abstract celery task subclass which provides same retrying behavior in the task definition itself:

# tasks.py
from celery_redis_sentinel.tasks import EnsuredRedisTask

@app.task(base=EnsuredRedisTask)
def add(a, b):
    return a + b