Profiling My Slow Django Templates

This is another techie post. If you have no interest in Django or Python then you may not be interested in this post either :-)

A Django app that I'm working on was running ridiculously slow. A quick Google search helped me to learn the basics of the Python profiler.

Add the following in order to profile my_function()


cProfile.runctx("my_function()", globals(), locals(), filename="profile.txt")


Run the code in question and then run the following in the Python interpreter in order to analyse the results:


import pstats
p = pstats.Stats("profile.txt")
p.sort_stats("cumulative").print_stats()


This will output results similar to the following, sorted in order of cumulative execution time:


Wed Nov 10 07:39:39 2010 profile.txt

11405 function calls (11319 primitive calls) in 0.459 CPU seconds

Ordered by: cumulative time

ncalls tottime percall cumtime percall filename:lineno(function)
1 0.000 0.000 0.458 0.458 /home/chris/MyProject/thingie.py:60(my_function)
1 0.001 0.001 0.458 0.458 /home/chris/MyProject/wizzle.py:32(thing_function)
1 0.000 0.000 0.358 0.358 /home/chris/MyProject/hoss.py:2(hussss)




In my case, I needed to see the call graph. The following draw a pretty .png picture of the call graph

gprof2dot -f pstats profile.txt | dot -Tpng -o out2.png

Armed with a visualisation of all this data, it became apparent that Django's template system was taking up most of the time. I had a line in a template, deep within a doubly nested for-loop that looked like this:

onclick="leftClick( {{cell.x}}, {{cell.y}} )"

Django was rendering this very slowly indeed. I ended up removing the for loop from the template and generating the html directly from within python code. This wasn't pretty but sped things up by nearly an order of magnitude.

I wonder if there is a more satisfactory solution to this problem, other than dumping Django templates altogether.