Models.py not getting loaded on startup with uWSGI - django

Question

I have a system where I need full dynamic control of URLs, both before and after the request.
I am using signals for this, and for the pre-request signal (the one I'm having trouble with, I have a piece of middleware like this, which connects to the signal, allows it to check if the current request.path applies to it, and then goes with the first one it gets. This normally works fine, and is fairly elegant):
class PreRouteMiddleWare(object):
def process_request(self, request):
url = request.path.strip('/')
if url == '':
url = '/'
pre_routes = pre_route.send(sender=request, url=url)
for reciever, response in pre_routes:
if response:
return response
return None
Now, to register something that happens "pre" the Django routing stack, I do something like this in the app's models.py:
#receiver(pre_route)
def try_things(sender, url, **kwargs):
try:
thing= Thing.objects.get(url=url)
from myapp.views import myview
return myview(sender, some_args)
except Thing.DoesNotExist:
return False
Which also works great on my dev server.
However, the problem arises in production, where I use uWSGI. I start uWSGI (from upstart) like this:
sudo /usr/local/bin/uwsgi --emperor '/srv/*/uwsgi.ini' --enable-threads --single-interpreter
And my uwsgi.ini looks like this:
[uwsgi]
socket = /srv/new/uwsgi.sock
module = wsgi:app
chdir = /srv/new/myapp
virtualenv = /srv/new
env = DJANGO_SETTINGS_MODULE=myapp.settings
uid = wsgi_new
gid = www-data
chmod = 770
processes = 2
What seems to be happening is for each uWSGI process/thread, they only seem to load models.py on the first request, meaning that the first request for each process will fail to connect the signals. This means that I have n (where n is the number of processes) requests fail completely because models.py is not loaded at startup (as it is in development).
Am I configuring uWSGI wrong? Is there a better way to force signals to be connected at startup?

Answer 1

Django actually lazily loads stuff. Using the development server gives a false sense of security about how things will work in a real WSGI server because the loading of the management commands by the development server forces a lot of early initialisation that doesn't occur with a production server.
You might read:
http://blog.dscpl.com.au/2010/03/improved-wsgi-script-for-use-with.html
which explains the issue as it occurs in mod_wsgi. Same thing will happen for uWSGI.

Answer 2

Ok, turns out that I needed to make my middleware hook process_view as opposed to process_request:
class PreRouteMiddleWare(object):
def process_view(self, request, *args, **kwargs):
url = request.path.strip('/')
if url == '':
url = '/'
pre_routes = pre_route.send(sender=request, url=url)
for reciever, response in pre_routes:
if response:
return response
return None
And now it works great!