app/django/core/handlers/profiler-hotshot.py
author Daniel Bentley <dbentley@google.com>
Sun, 12 Apr 2009 09:06:45 +0000
branchgae-fetch-limitation-fix
changeset 2313 c39a81bce1bd
parent 54 03e267d67478
permissions -rw-r--r--
Use offset_linkid instead of offset to scan >1000 entities. this is a first-cut. It works in all the ways I could make earlier versions fail. It passes link_id as URL parameters. It also has a new class LinkCreator which makes the main body of getListContents even easier to write. I wasn't sure if link_id's could have non alphanumeric characters; if so, they need to be URL encoded/decoded. I also need to go and remove any mention of raw offsets now, because we don't use them. I believe I've talked about this approach with a few of you and it sounded reasonable. Feel free to roll-back/fix/amend/comment-for-me-to-fix. This is my first big-logic-change to Melange. Patch by: Dan Bentley

import hotshot, time, os
from django.core.handlers.modpython import ModPythonHandler

PROFILE_DATA_DIR = "/var/log/cmsprofile"

def handler(req):
    '''
    Handler that uses hotshot to store profile data.

    Stores profile data in PROFILE_DATA_DIR.  Since hotshot has no way (that I
    know of) to append profile data to a single file, each request gets its own
    profile.  The file names are in the format <url>.<n>.prof where <url> is
    the request path with "/" replaced by ".", and <n> is a timestamp with
    microseconds to prevent overwriting files.

    Use the gather_profile_stats.py script to gather these individual request
    profiles into aggregated profiles by request path.
    '''
    profname = "%s.%.3f.prof" % (req.uri.strip("/").replace('/', '.'), time.time())
    profname = os.path.join(PROFILE_DATA_DIR, profname)
    prof = hotshot.Profile(profname)
    return prof.runcall(ModPythonHandler(), req)