Repository navigation
Realign4dAlgorithm memleaks? #398
Description
I have problems when using Realign4dAlgorithm, especially with large files in the nipype interface.
The memory seems no to be fully cleared then causing memory hog.
I tracked down with mprun down to cython function which seems to be the source.
_cspline_sample4d (3d/2d is certainly the same) to which an output array is given still take memory that is not released afterward.
for example running multiple times resampling (wrapped for memory_profiler to work) with the same value does take some memory (even if it stabilizes after a few). If I change the coordinates it again eat some more of the memory then stabilizes...
In [9]: %mprun -f memleaks.cspline_samp csampl=memleaks.cspline_samp(out,cdata,coords,coords,coords,coords)
Filename: /home/bpinsard/data/tests/memleaks.py
Line # Mem usage Increment Line Contents
================================================
7 896.8 MiB 0.0 MiB def cspline_samp(*args):
8 904.4 MiB 7.6 MiB nipy.algorithms.registration._registration._cspline_sample4d(*args)
('',)
In [10]: %mprun -f memleaks.cspline_samp csampl=memleaks.cspline_samp(out,cdata,coords,coords,coords,coords)
Filename: /home/bpinsard/data/tests/memleaks.py
Line # Mem usage Increment Line Contents
================================================
7 904.4 MiB 0.0 MiB def cspline_samp(*args):
8 904.5 MiB 0.1 MiB nipy.algorithms.registration._registration._cspline_sample4d(*args)
('',)
In [11]: %mprun -f memleaks.cspline_samp csampl=memleaks.cspline_samp(out,cdata,coords,coords,coords,coords)
Filename: /home/bpinsard/data/tests/memleaks.py
Line # Mem usage Increment Line Contents
================================================
7 904.5 MiB 0.0 MiB def cspline_samp(*args):
8 904.5 MiB 0.0 MiB nipy.algorithms.registration._registration._cspline_sample4d(*args)
('',)
@alexis-roche do you have any idea? Thanks
Activity
alexis-roche commentedon Mar 22, 2016
@bpinsard As far as I can see,
_cspline_sample4ddoes not allocate dynamic memory. The input arrays should be handled by Python's garbage collector.bpinsard commentedon Mar 22, 2016
Thanks for you quick reply.
I tried to force garbage collection using gc.collect() but is does not free any memory.
Can this problem be dependent on the version of python I am using?
Or even, because I run it in ipython?
yarikoptic commentedon Mar 22, 2016
I wonder if running your script through valgrind with an option to track origins would help to identify possibly outside of direct python code memory allocations which were left behind
On March 22, 2016 1:50:25 PM GMT+01:00, bpinsard notifications@github.com wrote:
Sent from a phone which beats iPhone.
bpinsard commentedon Mar 22, 2016
here is what I get:
I only copied the possibly lost blocks, I do not know if it helps.