https://github.com/scopatz/memleak

some xd tests
https://github.com/scopatz/memleak

Last synced: about 1 month ago
JSON representation

some xd tests

• Host: GitHub
• URL: https://github.com/scopatz/memleak
• Owner: scopatz
• Created: 2014-07-22T02:01:55.000Z (over 10 years ago)
• Default Branch: master
• Last Pushed: 2014-07-22T18:13:22.000Z (over 10 years ago)
• Last Synced: 2024-04-14T14:43:24.222Z (7 months ago)
• Language: Python
• Size: 149 KB
• Stars: 0
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: readme.rst

Awesome Lists containing this project

README

        
memory leak tests

=================

This tries a bunch of potential causes of a memory leak in gidden/cyclopts.

A plotting script is provided that can plot the results for any of the following

tests. For example, one can plot the results ``pttest.py`` by ::

    $ ./pttest.py > out

    $ ./plt.py out

    $ eog fig.png

An arbitrary number of files can be plotted against each other, e.g. ::

    $ ./pttest.py > out.pt

    $ ./pttest-np.py > out.np

    $ ./plt.py out.pt out.np

    $ eog fig.png

Test 1: xdress

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

This is not the cause.  To replicate run::

    $ xdress --debug

    $ ./setup.py build

    $ ./memtest.py

Play around with the ``N`` value in memtest and un/comment the gc.collect() call

inside of the loop.  If you are running h/top you'll see that the memory 

resources are cleaned up correctly.

Test 2: pytables row

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

This test definitely mem leaks. This uses the table row object and the 

memory leak seems to be fairly fast.  Run this test with::

    $ ./pttest.py

Test 3: pytables append

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

This test uses the table append() method rather than the row object. 

It also directly appends numpy arrays.  This may have a minor memory 

leak but it is much slower.  Run this test with::

    $ pttest-np.py

Conclusion

----------

The row attr should probably be avoided, not just because it is slow!