IssueTracker

When I try to run Django tests using the django 1.11 that comes with the AppEngine SDK it blows up with the following error:

File "/google-cloud-sdk/platform/google_appengine/lib/django-1.11/django/test/runner.py", line 288, in default_test_processes
    return multiprocessing.cpu_count()
NameError: global name 'multiprocessing' is not defined

although I'm not using the --parallel option.

I've checked the django1.11 code that comes with the AppEngineSDK and the problem is the following: the import of the multiprocessing module is removed. The first lines of /google-cloud-sdk/platform/google_appengine/lib/django-1.11/django/test/runner.py are the following:

#import ctypes  # Google: App Engine doesn't support ctypes
import itertools
import logging
#import multiprocessing  # Google: Don't parallelize tests under App Engine

The problem with removing this import is that the function default_test_processes (in the same module) uses the multiprocessing module to detect the number of CPUs, here is the code

def default_test_processes():
    """
    Default number of test processes when using the --parallel option.
    """
    # The current implementation of the parallel test runner requires
    # multiprocessing to start subprocesses with fork().
    # On Python 3.4+: if multiprocessing.get_start_method() != 'fork':
    if not hasattr(os, 'fork'):
        return 1
    try:
        return int(os.environ['DJANGO_TEST_PROCESSES'])
    except KeyError:
        return multiprocessing.cpu_count()

This function is used in the method add_arguments of the class DiscoverRunner which is invoked every time you run

from django.core.management import execute_from_command_line
execute_from_command_line(["tests"])

As a workaround, I've set the env variable DJANGO_TEST_PROCESSES equal to "1"

Note that this problem happens in Linux and MacOS but not in Windows, because it doesn't have fork.