Hey Warren, this is a known problem in 1.7. It is fixed in 1.8 (see https://code.djangoproject.com/ticket/23745). Unfortunately, the changes made to reduce the amount of get_models() and render() calls are too invasive to backport them to 1.7.
If you have the time, could you try running your project on the 1.8 alpha version and check the changes out and provide some feedback on how that turned out? Thanks in advance. /Markus On Thursday, January 22, 2015 at 9:47:22 PM UTC+1, Warren Smith wrote: > > I'm in the process of converting a large legacy app from Django 1.5 to > 1.7. I've got all the tests passing, so I'm researching an issue I noticed > during the course of my work: Test startup takes MUCH longer under Django > 1.7 than it did under 1.5. I've got a test that, when run by itself, took a > total of 5 seconds under 1.5. Under 1.7 it takes over a minute. > > To isolate the test startup, I setup a do-nothing test method within a > subclass of django.test.TestCase. I ran it under Django 1.5 and 1.7 with > similar results to the existing "real" test. > > I ran the do-nothing test within cProfile under Django 1.7 and found > something very interesting. According to pstats, the method > django.apps.config.AppConfig.get_models() was called 2.2+ MILLION times. > Even with 79 django apps in my INSTALLED_APPS setting, that seems excessive. > > I read the Django 1.7 release notes, especially the App Config stuff. I > noticed the warning about not importing models in the application root > module. I went through all of our internal apps and made sure they were not > importing models in the root module. I had to make some changes, moving > imports inside functions/methods. I also looked at my third-party > dependency apps. None of those were importing their models in their root > modules. > > After all that, though, I still see no change in behavior. > > I thought I might have some loop happening somewhere, though why it wasn't > infinite was interesting. It would be nice to know more about the context > of those get_models() calls. Perhaps I had a couple of mis-behaving apps > locked in a deadly embrace of some sort. > > I hacked on my locally installed django.apps.config.AppConfig class, > adding a class-level collections.Counter that gets incremented for > self.label by the get_models() method. > I dumped the results to a json file within my do-nothing test. It wouldn't > be comprehensive, but at least it would show me what had happened before > the test ran. > > I expected to see a few big numbers and a lot of small ones. As often > happens when profiling, my expectations were completely wrong. All the > counts were very similar, all in the 6000-7999 range. All 79 of my apps > were represented. > > Is there some sort of cartesian product going on here? > > Is this behavior normal and expected? > > -- You received this message because you are subscribed to the Google Groups "Django users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/django-users. To view this discussion on the web visit https://groups.google.com/d/msgid/django-users/147b99c8-a70f-4c27-82f5-be5e2710d4eb%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
