ProfilingProxy

Name:

ProfilingProxy

Description:

ProfilingProxy is a wrapper class for profiling method calls of objects. A ProfilingProxy object is created with an existing object as an argument. It wraps the given object's attributes and method calls. When methods are called through the ProfilingProxy, it times how long method execution took. If an object method calls another method, ProfilingProxy keeps track of this, and can produce a breakdown of the total call time in or under a function by functions called.

ProfilingProxy keeps one set of statistics for each class it is initialised with. As such, different instances of a class created with ProfilingProxy within the same Python interpreter will share a set of statistics, which may be retrieved from any instance of the class. ProfilingProxy may be used for profiling any number of classes at the same time.

ProfilingProxy Methods:

ProfilingProxy(obj)

Creates a ProfilingProxy object wrapping obj. The created object appears to have all the methods and attributes of obj and should (in most cases) behave as a drop-in replacement for the object.

_profiledata()

Returns the ClassProfileData object for the class of the object in a ProspectProxy. The returned object can then be queried to retrieve profiling statistics.

ClassProfileData Methods:

reset()

Clears the statistics for this class.

getFlatStats()

Returns an iterable sequence of dictionaries, each representing an item from the 'flat' statistics, i.e., a list of what total time each method took (including time spent within methods it called). Each entry is a dictionary with the following keys:

• method - the method name
• calls - the number of times this method was called
• totaltime - the number of seconds in total spent within calls to this method (or any called by it); this is a float
• avgtime - the average number of seconds spent in any one call to this method

dumpflat()

Prints the flat statistics to stdout in a standard format.

getTreeStats()

Returns an iterable sequence of dictionaries, each representing an item from the tree statistics. Each line represents a function at one level of the tree; lines of a deeper level are returned immediately following their caller. Each line is represented as a dictionary with the following keys:

• level - the level of this cal. (Hint: using this as an indentation width will give you a nice display.)
• method - the method name. If this is '(body)', then this line represents the time spent in the calling function's code outside of any other method calls.
• calls - the number of times this method was called within this context (i.e., by the parent calling function, which was called by its parent, and so on)
• time - the number of seconds in total spent within calls to this method (or any called by it), within this context; this is a float
• percent - the percentage of the caller's total time which was spent within this method call.

dumptree()

Prints the tree statistics to stdout in a standard format.

Example:

>>> from profilingproxy import ProfilingProxy
>>> import time


# Define a class we want to profile
>>> class spam(object):
...    def a(self):
...        for i in range(10):
...            self.b()
...        time.sleep(0.1)
...        self.c()
...    def b(self):
...        time.sleep(0.05)
...    def c(self):
...        time.sleep(0.5)

# create and use some objects
>>> spam1 = ProfilingProxy(spam)
>>> spam1.a()

>>> spam2 = ProfilingProxy(spam)
>>> spam2.c()

# now get stats
>>> profile = spam1._profiledata()
>>> profile.dumptree()
Method                                   Time   #   %
a                                        1.12   1  69.1%
  b                                      0.52  10  46.3%
  c                                      0.50   1  44.7%
  (body)                                 0.10   1  9.0%
c                                        0.50   1  30.9%
>>> for line in profile.getFlatStats():
...    print "%(method)s took %(totaltime)f seconds over %(calls)d calls"%line
a took 1.119410 seconds over 1 calls
b took 0.518321 seconds over 10 calls
c took 0.999830 seconds over 2 calls
                

Side effects:

ProfilingProxy adds some overhead to method calls on the wrapped object; it should be disabled in a production environment.

A ProfilingProxy object reserves any attribute names beginning with _profile; if your object class has any such attributes, either it or ProfilingProxy may fail to work correctly.

Code which relies on object introspection or instanceof may not work correctly for objects wrapped with ProfilingProxy.

Version history:

• 1.0 - initial version.

Download:

ProfilingProxy-1.0.tar.gz

Author:

Andrew C. Bulhak

Copyright:

This code is copyright 2008 The pH Group Ltd., and has been released as free software.

License:

GPL 2 or later