drewp for 2006 December 25 (entry 0)
< SHDH 14
Profiler talk at baypiggies >

[Trackback URL for this entry] 2006-12-25T16:01:06 Visualizing python profile results:

34 0.000 0.000 0.000 0.000 ConfigParser.py:335(optionxform)
6 0.000 0.000 0.000 0.000 ConfigParser.py:338(has_option)
28 0.001 0.000 0.001 0.000 ConfigParser.py:495(get)
28 0.000 0.000 0.000 0.000 ConfigParser.py:559(_interpolate)
1 0.000 0.000 0.000 0.000 ElementTree.py:1022(__init__)
1 0.000 0.000 0.000 0.000 ElementTree.py:1038(close)
3124 0.045 0.000 0.065 0.000 ElementTree.py:1043(_flush)
3123 0.019 0.000 0.019 0.000 ElementTree.py:1061(data)

As fun as those are to look at, I want to quickly find out where the bottlenecks are; whether they're in my code or not; and any other clues that will help me fix them. I couldn't find anything that performs further visualization on python profile output.

So I wrote a quick one with output like this. It parses profiler output back in for easiest use with "twistd -p", which means it only has base names of .py files. I attempted to color local files by their top dir, which helps a little.

The next big step is to use the profile output directly.

I also might make the report viewer into a full web server so it can do interesting sorts and filters. Or I might tag up the data well enough and do the UI in javascript. That would be harder to write, but it would mean that the report could be a standalone file to archive, share, etc. An example of "interesting filter" would be to show one function and all the deeper calls (as much as we can tell with the profile data).

Filed under: python

[Main]

Unless otherwise noted, all content licensed by Drew Perttula
under a Creative Commons License.