Uh oh, no it doesn't goes away. Check this (custom handler):
------ controller.py ---------------
import os
from time import time
from mod_python import apache
from mod_python import Session
def handler(req):
start_time = time()
req.sess = Session.FileSession(req)
req.content_type = 'text/html'
end_time = time()
totaltime = (end_time - start_time)*1000
req.write('%.0f ms' % ((totaltime)))
return apache.OK
------------------------------------
This takes 75ms on a 1Ghz with 1Gb RAM server. After some page refreshes the
memory load is over 70Mb. Is this right? (I'm asking for both the load time and
memory usage)
dharana wrote:
Please, ignore my previous email. The massive leak was being caused by
FileSession. I was using an earlier snapshot and after upgrading to:
http://svn.apache.org/snapshots/mod_python/mod_python_20050712224304.tar.gz
Everything went back normal. After checking everything works go upgrade
production. Load problem solved. I feel much better now, python handles
the heavy complexity I imposed to him with no problems.
How much is left for 3.2? I can't use an stable mp release right now
because I need several things included in 3.2 (segfaults for psp,
filesession, etc) and using svn HEAD in a prod server is scary.
dharana wrote:
look at this screenshot:
http://dharana.net/memory_concerns.png
It shows two screenshots of the same page of my webapp. The lower one
was shoot just after an apachectl start. The second one was shot after
6 consecutive refreshes where memory usage increased every time by a
lot of Mbs.
I use a lot of custom modules, from db calls to xslt transformations
(with libxml). But in the page shown in the screenshot it does little.
The strange thing is that after a few more page refreshes the memory
load goes back to 23Mb and then increases again.
I use some light module-level caching (I cache the 3 queries executed
in the lower page). But I don't think this is causing the heavy memory
usage because after it goes back to 23Mb it does query again (ie: the
caching mechanism still has it's cache).
I use apache prefork and Maxrequestperchild 0. After every page
refresh the unr obj var shown on the page doesn't changes. It's either
63 or 0.
mem, resident and stack come from this script:
http://aspn.activestate.com/ASPN/Cookbook/Python/Recipe/286222
unr obj is just the output of gc.collect() after everything else is
done (the debug print is printed just before return apache.OK.
What I am asking is for some help or pointers in how to track down who
is causing this or where to look for reducing the memory usage.
Thanks,
--
dharana
