celery: daemonic processes are not allowed to have children

Question

In Python (2.7) I try to create processes (with multiprocessing) in a celery task (celery 3.1.17) but it gives the error:

daemonic processes are not allowed to have children

Googling it, I found that most recent versions of billiard fix the "bug" but I have the most recent version (3.3.0.20) and the error is still happening. I also tried to implement this workaround in my celery task but it gives the same error.

Does anybody know how to do it? Any help is appreciated, Patrick

EDIT: snippets of code

Task:

from __future__ import absolute_import
from celery import shared_task
from embedder.models import Embedder

@shared_task
def embedder_update_task(embedder_id):
    embedder = Embedder.objects.get(pk=embedder_id)
    embedder.test()

Artificial test function (from here):

def sleepawhile(t):
    print("Sleeping %i seconds..." % t)
    time.sleep(t)
    return t    

def work(num_procs):
    print("Creating %i (daemon) workers and jobs in child." % num_procs)
    pool = mp.Pool(num_procs)

    result = pool.map(sleepawhile,
        [randint(1, 5) for x in range(num_procs)])

    # The following is not really needed, since the (daemon) workers of the
    # child's pool are killed when the child is terminated, but it's good
    # practice to cleanup after ourselves anyway.
    pool.close()
    pool.join()
    return result

def test(self):
    print("Creating 5 (non-daemon) workers and jobs in main process.")
    pool = MyPool(5)

    result = pool.map(work, [randint(1, 5) for x in range(5)])

    pool.close()
    pool.join()
    print(result)

My real function:

import mulitprocessing as mp

def test(self):
    self.init()
    for saveindex in range(self.start_index,self.start_index+self.nsaves):
        self.create_storage(saveindex)
        # process creation:
        procs = [mp.Process(name="Process-"+str(i),target=getattr(self,self.training_method),args=(saveindex,)) for i in range(self.nproc)]
        for p in procs: p.start()
        for p in procs: p.join()
    print "End of task"

The init function defines a multiprocessing array and an object that share the same memory so that all my processes can update this same array at the same time:

mp_arr = mp.Array(c.c_double, np.random.rand(1000000)) # example
self.V = numpy.frombuffer(mp_arr.get_obj()) #all the processes can update V

Error generated when task is called:

[2015-06-04 09:47:46,659: INFO/MainProcess] Received task: embedder.tasks.embedder_update_task[09f8abae-649a-4abc-8381-bdf258d33dda]
[2015-06-04 09:47:47,674: WARNING/Worker-5] Creating 5 (non-daemon) workers and jobs in main process.
[2015-06-04 09:47:47,789: ERROR/MainProcess] Task embedder.tasks.embedder_update_task[09f8abae-649a-4abc-8381-bdf258d33dda]     raised unexpected: AssertionError('daemonic processes are not allowed to have children',)
Traceback (most recent call last):
  File "/usr/local/lib/python2.7/dist-packages/celery/app/trace.py", line 240, in trace_task
   R = retval = fun(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/celery/app/trace.py", line 438, in __protected_call__
   return self.run(*args, **kwargs)
  File "/home/patrick/django/entite-tracker-master/entitetracker/embedder/tasks.py", line 21, in embedder_update_task
    embedder.test()
  File "/home/patrick/django/entite-tracker-master/entitetracker/embedder/models.py", line 475, in test
    pool = MyPool(5)
  File "/usr/lib/python2.7/multiprocessing/pool.py", line 159, in __init__
self._repopulate_pool()
  File "/usr/lib/python2.7/multiprocessing/pool.py", line 223, in _repopulate_pool
    w.start()
  File "/usr/lib/python2.7/multiprocessing/process.py", line 124, in start
'daemonic processes are not allowed to have children'
AssertionError: daemonic processes are not allowed to have children

Answer 1

billiard and multiprocessing are different libraries - billiard is the Celery project's own fork of multiprocessing. You will need to import billiard and use it instead of multiprocessing

However the better answer is probably that you should refactor your code so that you spawn more Celery tasks instead of using two different ways of distributing your work.

You can do this using Celery canvas

from celery import group

@app.task
def sleepawhile(t):
    print("Sleeping %i seconds..." % t)
    time.sleep(t)
    return t    

def work(num_procs):
    return group(sleepawhile.s(randint(1, 5)) for x in range(num_procs)])

def test(self):
    my_group = group(work(randint(1, 5)) for x in range(5))
    result = my_group.apply_async()
    result.get()

I've attempted to make a working version of your code that uses canvas primitives instead of multiprocessing. However since your example was quite artificial it's not easy to come up with something that makes sense.

Update:

Here is a translation of your real code that uses Celery canvas:

tasks.py:

@shared_task
run_training_method(saveindex, embedder_id):
    embedder = Embedder.objects.get(pk=embedder_id)
    embedder.training_method(saveindex)

models.py:

from tasks import run_training_method
from celery import group

class Embedder(Model):

    def embedder_update_task(self):
        my_group = []

        for saveindex in range(self.start_index, self.start_index + self.nsaves):
            self.create_storage(saveindex)
            # Add to list
            my_group.extend([run_training_method.subtask((saveindex, self.id)) 
                         for i in range(self.nproc)])

        result = group(my_group).apply_async()