Celery Once allows you to prevent multiple execution and queuing of celery tasks.
Installing celery_once
is simple with pip, just run:
pip install -U celery_once
- Celery. Built to run with Celery 4.0. Older versions may work, but are not officially supported.
To use celery_once
, your tasks need to inherit from an abstract base task called QueueOnce
.
Once installed, you'll need to configure a few options a ONCE
key in celery's conf.
from celery import Celery
from celery_once import QueueOnce
from time import sleep
celery = Celery('tasks', broker='amqp://guest@localhost//')
celery.conf.ONCE = {
'backend': 'celery_once.backends.Redis',
'settings': {
'url': 'redis://localhost:6379/0',
'default_timeout': 60 * 60
}
}
@celery.task(base=QueueOnce)
def slow_task():
sleep(30)
return "Done!"
The exact configuration, depends on which locking backend you want to use. See Backends.
Behind the scenes, this overrides apply_async
and delay
. It does not affect calling the tasks directly.
When running the task, celery_once
checks that no lock is in place (against a Redis key).
If it isn't, the task will run as normal. Once the task completes (or ends due to an exception) the lock will clear.
If an attempt is made to run the task again before it completes an AlreadyQueued
exception will be raised.
example.delay(10)
example.delay(10)
Traceback (most recent call last):
..
AlreadyQueued()
result = example.apply_async(args=(10))
result = example.apply_async(args=(10))
Traceback (most recent call last):
..
AlreadyQueued()
Optionally, instead of raising an AlreadyQueued
exception, the task can return None
if once={'graceful': True}
is set in the task's options or when run through apply_async
.
from celery_once import AlreadyQueued
# Either catch the exception,
try:
example.delay(10)
except AlreadyQueued:
pass
# Or, handle it gracefully at run time.
result = example.apply(args=(10), once={'graceful': True})
# or by default.
@celery.task(base=QueueOnce, once={'graceful': True})
def slow_task():
sleep(30)
return "Done!"
By default celery_once
creates a lock based on the task's name and its arguments and values.
Take for example, the following task below...
@celery.task(base=QueueOnce)
def slow_add(a, b):
sleep(30)
return a + b
Running the task with different arguments will default to checking against different locks.
slow_add(1, 1)
slow_add(1, 2)
If you want to specify locking based on a subset, or no arguments you can adjust the keys celery_once
looks at in the task's options with once={'keys': [..]}
@celery.task(base=QueueOnce, once={'keys': ['a']})
def slow_add(a, b):
sleep(30)
return a + b
example.delay(1, 1)
# Checks if any tasks are running with the `a=1`
example.delay(1, 2)
Traceback (most recent call last):
..
AlreadyQueued()
example.delay(2, 2)
@celery.task(base=QueueOnce, once={'keys': []})
def slow_add(a, b):
sleep(30)
return a + b
# Will enforce only one task can run, no matter what arguments.
example.delay(1, 1)
example.delay(2, 2)
Traceback (most recent call last):
..
AlreadyQueued()
As a fall back, celery_once
will clear a lock after 60 minutes.
This is set globally in Celery's configuration with ONCE_DEFAULT_TIMEOUT
but can be set for individual tasks using...
@celery.task(base=QueueOnce, once={'timeout': 60 * 60 * 10})
def long_running_task():
sleep(60 * 60 * 3)
By default, the lock is removed after the task has executed (using celery's after_return). This behaviour can be changed setting the task's option unlock_before_run
. When set to True
, the lock will be removed just before executing the task.
- Caveats:
- Any retry of the task won't re-enable the lock!
- This can only be set when defining the task, it cannot be passed dynamically to
apply_async
@celery.task(base=QueueOnce, once={'unlock_before_run': True})
def slow_task():
sleep(30)
return "Done!"
Requires:
- Redis is used as a distributed locking mechanism. Behind the scenes, it use redis-py's shared, distributed Lock.
Configuration:
backend
-celery_once.backends.Redis
settings
default_timeout
- how many seconds after a lock has been set before it should automatically timeout (defaults to 3600 seconds, or 1 hour).url
- should point towards a running Redis instance (defaults toredis://localhost:6379/0
). See below for the format options supportedblocking
(boolean value: defaultFalse
) - If set toTrue
, scheduling a task (by.delay/.apply_async
) will block for X seconds to acquire the lock (see:blocking_timeout
below). If no lock could be acquired after X seconds, will raise anAlreadyQueued
exception. This is a very specific use-case scenario and by default is disabled.blocking_timeout
(int or float value: default1
) - How many seconds the task will block trying to aquire the lock, ifblocking
is set toTrue
. Setting this toNone
set's no timeout (equivalent to infinite seconds).
The URL parser supports three patterns of urls:
redis://host:port[/db][?options]
: redis over TCPrediss://host:port[/db][?options]
: redis over TCP with SSL enabled.redis+socket:///path/to/redis.sock[?options]
: redis over a UNIX socketThe
options
query args are mapped to the StrictRedis keyword args. Examples: *redis://localhost:6379/1
redis://localhost:6379/1?ssl=true
rediss://localhost:6379/1
redis+socket:///var/run/redis/redis.sock?db=1
Example Configuration:
Minimal:
celery.conf.ONCE = {
'backend': 'celery_once.backends.Redis',
'settings': {
'url': 'redis://localhost:6379/0',
'default_timeout': 60 * 60
}
}
Advanced: Scheduling tasks blocks up to 30 seconds trying to acquire a lock before raising an exception.
celery.conf.ONCE = { 'backend': 'celery_once.backends.Redis', 'settings': { 'url': 'redis://localhost:6379/0', 'default_timeout': 60 * 60, 'blocking': True, 'blocking_timeout': 30 } }
Configuration:
backend
-celery_once.backends.File
settings
location
- directory where lock files will be located. Default is temporary directory.default_timeout
- how many seconds after a lock has been set before it should automatically timeout (defaults to 3600 seconds, or 1 hour).
Example Configuration:
celery.conf.ONCE = {
'backend': 'celery_once.backends.File',
'settings': {
'location': '/tmp/celery_once',
'default_timeout': 60 * 60
}
}
To avoid RuntimeError: Working outside of application context
errors when using celery_once
with Flask, you need to make the QueueOnce
task base class application context aware.
If you've implemented Celery following the Flask documentation you can extend it like so.
def make_celery(app): celery = Celery( app.import_name, backend=app.config['CELERY_RESULT_BACKEND'], broker=app.config['CELERY_BROKER_URL'] ) celery.conf.update(app.config) class ContextTask(celery.Task): def __call__(self, *args, **kwargs): with app.app_context(): return self.run(*args, **kwargs) celery.Task = ContextTask # Make QueueOnce app context aware. class ContextQueueOnce(QueueOnce): def __call__(self, *args, **kwargs): with app.app_context(): return super(ContextQueueOnce, self).__call__(*args, **kwargs) # Attach to celery object for easy access. celery.QueueOnce = ContextQueueOnce return celery
Now, when instead of importing the QueueOnce
base, you can use the context aware base on the celery
object.
celery = make_celery(app) @celery.task(base=celery.QueueOnce) def example_task(value): return
If you want to implement a custom locking backend, see BACKEND_GUIDE.rst.
- Tests are run against Python 2.7, 3.4 and 3.5. Other versions may work, but are not officially supported.
Contributions are welcome, and they are greatly appreciated! See contributing guide for more details.