Skip to content

Shows top suspects for memory leaks in your Python program.

Notifications You must be signed in to change notification settings

iyawnis/mem_top

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

mem_top

Shows top suspects for memory leaks in your Python program.

Usage:

pip install mem_top
from mem_top import mem_top

# From time to time:
logging.debug(mem_top()) # Or just print().

# Notice which counters keep increasing over time - they are the suspects.

Counters:

  • refs - number of direct references from this object to other objects, like keys and values of dict
  • bytes - size of this object in bytes
  • types - number of objects of this type still kept in memory after garbage collection

Real life example:

refs:
144997  <type 'collections.defaultdict'> defaultdict(<type 'collections.deque'>, {<GearmanJobRequest task='...', unique='.
144996  <type 'dict'> {'.:..............:.......': <GearmanJobRequest task='..................', unique='.................
18948   <type 'dict'> {...
1578    <type 'dict'> {...
968     <type 'dict'> {...
968     <type 'dict'> {...
968     <type 'dict'> {...
767     <type 'list'> [...
726     <type 'dict'> {...
608     <type 'dict'> {...

types:
292499  <type 'dict'>
217912  <type 'collections.deque'>
72702   <class 'gearman.job.GearmanJob'>
72702   <class 'gearman.job.GearmanJobRequest'>
12340   <type '...
3103    <type '...
1112    <type '...
855     <type '...
767     <type '...
532     <type '...
  • Noticed a leak of 6GB RAM and counting.
  • Added "mem_top" and let it run for a while.
  • When got the result above it became absolutely clear who is leaking here:
    the Python client of Gearman kept increasing its counters over time.
  • Found its known bug - Yelp/python-gearman#10
    leaking defaultdict of deques, and a dict of GearmanJobRequest-s,
    just as the "mem_top" showed.
  • Replaced "python-gearman" - long story: stale 2.0.2 at PyPI, broken 2.0.X at github, etc.
  • "mem_top" confirmed the leak is now completely closed.

UPDATES:

  • Pass e.g. verbose_types=[dict, list] to get their values sorted by repr length in verbose_file_name.
  • Added "bytes" top.

SEE ALSO:

Config defaults:

mem_top(
    limit=10, width=100, sep='\n',
    refs_format='{num}\t{type} {obj}', bytes_format='{num}\t {obj}', types_format='{num}\t {obj}',
    verbose_types=None, verbose_file_name='/tmp/mem_top',
)

mem_top version 0.1.7
Copyright (C) 2014-2018 by Denis Ryzhkov [email protected]
MIT License, see http:https://opensource.org/licenses/MIT

About

Shows top suspects for memory leaks in your Python program.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 99.0%
  • Shell 1.0%