potential memory leak

[Slyne]

Hi~

I use valgrind to test the below line:

import kaldifeat

The log is here: https://1drv.ms/t/s!AhtoTbISXXlSgSLtqDWjo8RP_ouA?e=Sn1cd0

Log summary:

==20510== LEAK SUMMARY:
==20510==    definitely lost: 0 bytes in 0 blocks
==20510==    indirectly lost: 0 bytes in 0 blocks
==20510==      possibly lost: 271,997 bytes in 129 blocks
==20510==    still reachable: 5,881,625 bytes in 22,065 blocks
==20510==                       of which reachable via heuristic:
==20510==                         stdstring          : 521,021 bytes in 6,954 blocks
==20510==         suppressed: 58,517 bytes in 5 blocks
==20510==
==20510== For lists of detected and suppressed errors, rerun with: -s
==20510== ERROR SUMMARY: 17745 errors from 434 contexts (suppressed: 14 from 2)

Reproduce

valgrind --leak-check=full --show-reachable=yes --trace-children=yes --track-origins=yes --trace-children=yes --log-file='log.txt' python3 test.py

test.py

import kaldifeat

Could you help check if these are potential issues? Thanks!

[csukuangfj]

Could you help check if these are potential issues? Thanks!

Please have a look at https://github.com/pytorch/pytorch/issues/39522 It is a known issue when using valgrind with PyTorch.

[Slyne]

Do you mind checking this issue: https://github.com/triton-inference-server/server/issues/3777#issuecomment-1027418452 ?

There's an invalid pointer issue related to kaldifeat + pybind.

[csukuangfj]

@Slyne

Please see the comment at https://github.com/triton-inference-server/server/issues/3777#issuecomment-1031408935

I would suggest you to file an issue at https://github.com/pytorch/pytorch/issues

You can replace import kaldifeat with import torch when posting the issue.

[yuekaizhang]

Do you mind checking this issue: triton-inference-server/server#3777 (comment) ?

There's an invalid pointer issue related to kaldifeat + pybind.

@Slyne Hi, I was wondering if you figure it out or posting the issue in pytorch.