Live Profiling of Python Server

Question

I want to know where the python interpreter spends the most time. I use it on a live django application, but it should work for all long running python processes.

I answer my own question.

Answer 1

import os, re, sys, time, datetime, collections, thread, threading, atexit, traceback 

u'''

debug_live.start(seconds_float) starts a monitor thread which print 
the stacktrace of all threads into a logfile.
You can report which lines are executed the most with this script:

app_foo_d@server:~$ python djangotools/utils/debug_live.py -h
usage: debug_live.py [-h] [--most-common N] {sum-all-frames,sum-last-frame}

Read stacktrace log

positional arguments:
  {sum-all-frames,sum-last-frame}

optional arguments:
  -h, --help            show this help message and exit
  --most-common N       Display the N most common lines in the stacktraces

---------------------------------

You can start the watching thread your django middleware like this:

class FOOMiddleware:
    def __init__(self):
        u'This code gets executed once after the start of the wsgi worker process. Not for every request!'
        seconds=getattr(settings, 'debug_live_interval', None)
        if seconds:
            seconds=float(seconds)
            from djangotools.utils import debug_live
            debug_live.start(seconds)

# settings.py
debug_live_interval=0.3 # ever 0.3 second

# Inspired by http://code.google.com/p/modwsgi/wiki/DebuggingTechniques

You can get a simple report of the log file of stacktraces like below. The lines
which are not from django are marked with "<====". That's most likely your code
and this could be a bottle neck.

python ..../debug_live.py read
 1971 File: "/home/foo_bar_p/django/core/handlers/wsgi.py", line 272, in __call__
      response = self.get_response(request)
 1812 File: "/home/foo_bar_p/django/core/handlers/base.py", line 111, in get_response
      response = callback(request, *callback_args, **callback_kwargs)
 1725 File: "/home/foo_bar_p/django/db/backends/postgresql_psycopg2/base.py", line 44, in execute
      return self.cursor.execute(query, args)
 1724 File: "/home/foo_bar_p/django/db/models/sql/compiler.py", line 735, in execute_sql
      cursor.execute(sql, params)
 1007 File: "/home/foo_bar_p/django/db/models/sql/compiler.py", line 680, in results_iter
      for rows in self.execute_sql(MULTI):
  796 File: "/home/foo_bar_p/django/db/models/query.py", line 273, in iterator
      for row in compiler.results_iter():
  763 File: "/home/foo_bar_p/foo/utils/ticketutils.py", line 135, in __init__      <====
      filter=type_filter(root_node=self.root_node)
  684 File: "/home/foo_bar_p/django/db/models/query.py", line 334, in count
      return self.query.get_count(using=self.db)
  679 File: "/home/foo_bar_p/django/db/models/sql/query.py", line 367, in get_aggregation
      result = query.get_compiler(using).execute_sql(SINGLE)
  677 File: "/home/foo_bar_p/django/db/models/sql/query.py", line 401, in get_count
      number = obj.get_aggregation(using=using)[None]


'''


from django.conf import settings

outfile = os.path.expanduser('~/tmp/debug_live.log')

other_code=re.compile(r'/(django|python...)/')


def stacktraces():
    code=[]
    now=datetime.datetime.now()
    pid=os.getpid()
    my_thread_id=thread.get_ident()
    for thread_id, stack in sys._current_frames().items(): 
        if thread_id==my_thread_id:
            continue # Don't print this monitor thread
        code.append("\n\n#START date: %s\n# ProcessId: %s\n# ThreadID: %s" % (now, pid, thread_id))
        for filename, lineno, name, line in traceback.extract_stack(stack): 
            code.append('File: "%s", line %d, in %s' % (filename, lineno, name)) 
            if line: 
                code.append("  %s" % (line.strip()))
        code.append('#END')
    if not code:
        return
    fd=open(outfile, 'at')
    fd.write('\n'.join(code))
    fd.close()

def monitor(interval):
    while monitor_thread:
        stacktraces()
        time.sleep(interval)

monitor_thread=None

def exiting():
    global monitor_thread
    monitor_thread=None


def start(interval):
    global monitor_thread
    if monitor_thread:
        return
    assert not os.path.islink(outfile), outfile # well known temporary name.... symlink attack...
    monitor_thread = threading.Thread(target=monitor, args=[interval])
    monitor_thread.setDaemon(True)
    atexit.register(exiting)
    monitor_thread.start()

def read_logs(args):
    # The outfile can be huge, don't read the whole file into memory.
    counter=collections.Counter()
    cur_stack=[]
    py_line=''
    code_line=''
    if args.action=='sum-all-frames':
        sum_all_frames=True
    else:
        sum_all_frames=False
    for line in open(outfile):
        if line.startswith('#END'):
            if sum_all_frames:
                frames=cur_stack
            else:
                frames=cur_stack[-1:]
            counter.update(frames)
            cur_stack=[]
            continue
        if line[0] in '\n#':
            continue
        if line.startswith('File:'):
            py_line=line.rstrip()
            continue
        if line.startswith(' '):
            code_line=line.rstrip()
            if not (py_line, code_line) in cur_stack:
                # If there is a recursion, count the line only once per stacktrace
                cur_stack.append((py_line, code_line))
            continue
        print 'ERROR unparsed', line
    for (py, code), c in counter.most_common(args.most_common):
        if not other_code.search(py):
            py='%s      <====' % py
        print '% 5d %s\n    %s' % (c, py, code)

def main():
    import argparse
    parser=argparse.ArgumentParser(description='Read stacktrace log')
    parser.add_argument('action', choices=['sum-all-frames', 'sum-last-frame'])
    parser.add_argument('--most-common', metavar='N', default=30, type=int, help='Display the N most common lines in the stacktraces')
    args=parser.parse_args()
    return read_logs(args)

if __name__=='__main__':
    main()