This document describes the current stable version of Celery (3.1). For development docs, go here.
For long-running Task‘s, it can be desirable to support aborting during execution. Of course, these tasks should be built to support abortion specifically.
The AbortableTask serves as a base class for all Task objects that should support abortion by producers.
The necessary intermediate communication is dealt with by the AbortableTask implementation.
In the consumer:
from __future__ import absolute_import
from celery.contrib.abortable import AbortableTask
from celery.utils.log import get_task_logger
from proj.celery import app
logger = get_logger(__name__)
@app.task(bind=True, base=AbortableTask)
def long_running_task(self):
results = []
for i in range(100):
# check after every 5 iterations...
# (or alternatively, check when some timer is due)
if not i % 5:
if self.is_aborted():
# respect aborted state, and terminate gracefully.
logger.warning('Task aborted')
return
value = do_something_expensive(i)
results.append(y)
logger.info('Task complete')
return results
In the producer:
from __future__ import absolute_import
import time
from proj.tasks import MyLongRunningTask
def myview(request):
# result is of type AbortableAsyncResult
result = long_running_task.delay()
# abort the task after 10 seconds
time.sleep(10)
result.abort()
After the result.abort() call, the task execution is not aborted immediately. In fact, it is not guaranteed to abort at all. Keep checking result.state status, or call result.get(timeout=) to have it block until the task is finished.
Note
In order to abort tasks, there needs to be communication between the producer and the consumer. This is currently implemented through the database backend. Therefore, this class will only work with the database backends.
Represents a abortable result.
Specifically, this gives the AsyncResult a abort() method, which sets the state of the underlying Task to ‘ABORTED’.
A celery task that serves as a base class for all Task‘s that support aborting during execution.
All subclasses of AbortableTask must call the is_aborted() method periodically and act accordingly when the call evaluates to True.
Checks against the backend whether this AbortableAsyncResult is ABORTED.
Always return False in case the task_id parameter refers to a regular (non-abortable) Task.
Be aware that invoking this method will cause a hit in the backend (for example a database query), so find a good balance between calling it regularly (for responsiveness), but not too often (for performance).