update itk requirement to >= 5.2

[wyli]

once itk 5.2 is released, requirements.txt needs to be updated and the warning message removed.

this is a follow up of

• ITKReader does not read NRRD05 images with 'measurement frame' field #1451
• Throw warning when reading NRRD05 files to upgrade to ITK >= 5.2 #1602
• Need to transpose array obtained from itk instead of reversing of itk.size() in ITKReader #1357 (comment)
• DICOM not loading #1713

see also:
https://github.com/InsightSoftwareConsortium/ITK/tags

[wyli]

itk 5.2.0 has been released, it is causing some issue:
for example, to replicate it (with monai da3240f):

docker run --gpus all --rm -ti --ipc=host --net=host -v ~/Documents/MONAI:/opt/monai nvcr.io/nvidia/pytorch:20.03-py3
# within the docker container run:
cd /opt/monai
python -m pip install --upgrade pip wheel
python -m pip uninstall -y torch torchvision
python -m pip install torch==1.7.1 torchvision==0.8.2
python -m pip install -r requirements-dev.txt
python -m tests.test_distributed_sampler    # run tests/test_distributed_sampler.py

outcome:

/opt/conda/lib/python3.6/multiprocessing/semaphore_tracker.py:143: UserWarning: semaphore_tracker: There appear to be 1 leaked semaphores to clean up at shutdown
  len(cache))
/opt/conda/lib/python3.6/multiprocessing/semaphore_tracker.py:143: UserWarning: semaphore_tracker: There appear to be 1 leaked semaphores to clean up at shutdown
  len(cache))
./opt/conda/lib/python3.6/multiprocessing/semaphore_tracker.py:143: UserWarning: semaphore_tracker: There appear to be 1 leaked semaphores to clean up at shutdown
  len(cache))
/opt/conda/lib/python3.6/multiprocessing/semaphore_tracker.py:143: UserWarning: semaphore_tracker: There appear to be 1 leaked semaphores to clean up at shutdown
  len(cache))
.
----------------------------------------------------------------------
Ran 2 tests in 12.074s

OK

example log: https://github.com/Project-MONAI/MONAI/runs/2266945686?check_suite_focus=true
eventually the integration tests become very slow with UserWarning: semaphore_tracker:

I had a quick look, the same code works fine with itk 5.1.2. so I suspect it's an issue in the way MONAI using itk

any idea @thewtex ?

[wyli]

a smaller example to reproduce (python3 + linux + pytorch 1.7/1.8):

from torch import multiprocessing

try:
    multiprocessing.set_start_method("spawn")
except RuntimeError:
    pass

import itk
import torch

if __name__ == "__main__":
    for d in torch.utils.data.DataLoader([torch.tensor(i) for i in range(10)], num_workers=4):
        pass

this works fine with itk 5.1.2 but has the warnings with itk5.2.0

perhaps this global RLock is causing the issue https://github.com/InsightSoftwareConsortium/ITK/blob/6dfbc9f0314b5fd64d4b2ae392ae3cc6dda17677/Wrapping/Generators/Python/itk/support/lazy.py#L29

cc @hjmjohnson

[thewtex]

@wyli thanks for the detailed, reproducible report 🥇

I verified that this patch addresses the issue:

InsightSoftwareConsortium/ITK#2552

[thewtex]

@wyli this has been addressed in itk-5.2.0.post3.

I will help with the remaining issues based on this package in the coming month.

[wyli]

thanks @thewtex!

[wyli]

it still gives the warning:

/opt/conda/lib/python3.8/multiprocessing/resource_tracker.py:216: UserWarning: resource_tracker: There appear to be 33 leaked semaphore objects to clean up at shutdown
  warnings.warn('resource_tracker: There appear to be %d '

https://github.com/Project-MONAI/MONAI/runs/2728004520?check_suite_focus=true#step:7:2534
using 5.2.0.post3 PR here: #2274

[thewtex]

@wyli thanks for the report. We will keep pushing on this! Tracked in InsightSoftwareConsortium/ITK#2603

[thewtex]

Following-up: itk-5.2.1 contains additional improvements.