After working through this I feel I need to follow through and
clarify what actually happened. As you say, Russ, there does not seem
to be any memory leak. However it is a lot fatter in it's memory
usage. I'm guessing this is due primarily to database caching. We will
be doing some capacity measurements to determine just what the
capacity impact is as we do have to reduce the number of simultaneous
processes we can launch to handle requests but, hopefully, the
increased memory usage comes with increased performance so less time
per request hopefully offsets and even improves our overall capacity.
This led me to believe that perhaps our postgres db library had an
issue but couldn't find anything pointing that direction and nothing
had changed in that regard. Finally, we discovered that the version of
mod_wsgi we were using had a known issue (
http://code.google.com/p/modwsgi/issues/detail?id=99&can=1&q=leak )
where it did not recover all its memory when apache was restarted. We
had actually eliminated immortal apache processes and had it
restarting quite often to offset the perceived memory leak - resulting
in actually introducing a memory leak on top of our increased memory
usage from Django 1.1. :-( Re-introducing immortal processes (albeit
fewer of them), got rid of the memory leak from restarts and then we
subsequently upgraded to the latest mod_wsgi (2.5) and the leak seems
to have disappeared.
Conclusion - so far no leaks known in Django 1.1 and it's working
fine in production now. It does use significantly more memory.
thanx,
-- Ben
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups
"Django users" group.
To post to this group, send email to django-users@googlegroups.com
To unsubscribe from this group, send email to
django-users+unsubscr...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/django-users?hl=en
-~----------~----~----~----~------~----~------~--~---
