Profiling Memory Usage in Python

Learn what Python tools you can use to track memory usage in your application and how to lower memory usage in this full Python tutorial. Read and learn TODAY!

Apr 17, 2019 • 4 Minute Read

Subscribe to the newsletter

Introduction

One of the ways Python makes development fast and easier than languages like C and C++ is memory management. In Python it's simple, the language handles memory management for you. However, this doesn't mean memory should be forgotten. Good developers will want to track the memory usage of their application and look to lower memory usage. This post will explain common tools for doing this.

Profiling Size of Individual Objects

The lowest layer of memory profiling involves looking at a single object in memory. You can do this by opening up a shell and doing something like the following:

          >>> import sys
    >>> sys.getsizeof({})
    136
    >>> sys.getsizeof([])
    32
    >>> sys.getsizeof(set())
    112
    

The above snippet illustrates the overhead associated with a list object. A list is 32 bytes (on a 32-bit machine running Python 2.7.3). This style of profiling is useful when determining what type of data type to use.

Profiling a Single Function or Method

The easiest way to profile a single method or function is the open source memory-profiler package. It's similar to line_profiler which I've written about before.

You can use a memory profiling by putting the @profile decorator around any function or method and running python -m memory_profiler myscript. You'll see line-by-line memory usage once your script exits.

This is extremely useful if you're wanting to profile a section of memory-intensive code, but it won't help much if you have no idea where the biggest memory usage is. In that case, a higher-level approach for profiling is needed first.

Profiling an Entire Application

There are a number of ways to profile an entire Python application. You can use the standard unix tools top and ps. A more Python specific way is guppy.

Guppy

To use guppy, you drop something like the following in your code:

          from guppy import hpy
    h = hpy()
    print h.heap()
    

This will print you a nice table of usage grouped by object type. Here's an example of an PyQt4 application that I've been working on:

          Partition of a set of 235760 objects. Total size = 19909080 bytes.
    Index  Count   %     Size   % Cumulative  % Kind (class / dict of class)
        0  97264  41  8370996  42   8370996  42 str
        1  47430  20  1916788  10  10287784  52 tuple
        2    937   0  1106440   6  11394224  57 dict of PyQt4.QtCore.pyqtWrapperType
        3    646   0  1033648   5  12427872  62 dict of module
        4  11683   5   841176   4  13269048  67 types.CodeType
        5  11684   5   654304   3  13923352  70 function
        6   1200   1   583872   3  14507224  73 dict of type
        7    782   0   566768   3  15073992  76 dict (no owner)
        8   1201   1   536512   3  15610504  78 type
        9   1019   0   499124   3  16109628  81 unicode
    

This type of profiling can be difficult if you have a large application using a relatively small number of object types.

mprof

Finally, I recently discovered that memory-profiler comes with a script called mprof, which can show you memory usage over the lifetime of your application. This can be useful if you want to see if your memory is getting cleaned up and released periodically.

Using mprof is easy, just run mprof run script script_args in your shell of choice. mprof will automatically create a graph of your script's memory usage over time, which you can view by running mprof plot. Be aware that plotting requires matplotlib.