urimandujano / memleek

Python memory leak finder

Memleek

A Python in-code memory profiler built on top of tracemalloc

Usage

Once installed, the profiler is used as a contextmanager, inside of which the code to be profiled gets run:

from memleek import MemoryProfiler

def my_sus_func():
    ...

with MemoryProfiler() as profiler:
    my_sus_func()
    profiler.snap()

Sample output:

                                    Diff from start to end for 2 snapshots
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━┓
┃ File                                                                          ┃             Size ┃    Count ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━┩
│ /.pyenv/versions/3.8.17/lib/python3.8/sre_compile.py:780 │ 2112 B (+2112 B) │   5 (+5) │
│ /.pyenv/versions/3.8.17/lib/python3.8/sre_parse.py:529   │ 1456 B (+1456 B) │ 26 (+26) │
│ /.pyenv/versions/3.8.17/lib/python3.8/threading.py:244   │ 1248 B (+1248 B) │   4 (+4) │
└───────────────────────────────────────────────────────────────────────────────┴──────────────────┴──────────┘

You can also track how memory usage changes over repeated invocations:

from memleek import MemoryProfiler

def my_sus_func():
    ...

with MemoryProfiler() as profiler:
    while True:
        my_sus_func()
        profiler.snap()
        profiler.display_overall_stats()

Advanced usage

To control the number of stats that get displayed (aka the number of lines in the table that gets printed to the console), create your own StatsDisplay object and pass that into the MemoryProfiler:

from memleek import MemoryProfiler, StatsDisplay # 1. Import the class

def my_sus_func():
    ...

d = StatsDisplay(top_n=3) # 2. Customize the class

with MemoryProfiler(display=d) as profiler: # 3. Use the class
    while True
        my_sus_func()
        profiler.snap()

The tool attempts to filter out the noisiest Python modules (those that are consistently cluttering output to the console with data on builtins, not with data from your code). If you want to see all of the stats on memory usage used in your Python code, you can remove the filters like so:

from memleek import MemoryProfiler

def my_sus_func():
    ...

with MemoryProfiler(filters=[]) as profiler: # 1. Use a custom, empty filters
    while True
        my_sus_func()
        profiler.snap()

NOTE Similarly, you can customize the modules that get filtered out by creating your own instances of tracemalloc.Filter and passing those as filters to the MemoryProfiler on instantiation.