[Bug] sub processes are kept running even after separating is done and are re-created even after kill

CoryXie commented 2 years ago

[X] I didn't find a similar issue already open.
[X] I read the documentation (README AND Wiki)
[X] I have installed FFMpeg
[X] My problem is related to Spleeter only, not a derivative product (such as Webapplication, or GUI provided by others)

Description

I am trying to use spleeter for vocal extraction in a song, with the following code. I found it is trying to spawn several sub processes for its work (and it works fine to separate what I wanted), but even after the separation is done, these sub processes are still kept running.

        separator = Separator('spleeter:2stems')
        separator.separate_to_file(filepath, self.filedirectory)
        separator = None

I even tried to kill these sub processes with below code, it seems the sub processes are being re-created after being killed.

        current_process = psutil.Process()
        children = current_process.children(recursive=True)
        for child in children:
            print('Child pid is {}'.format(child.pid))
            os.kill(child.pid, signal.SIGTERM)

See below for the debug screenshot before kill (after it was run).

See below for the debug screenshot after kill.

Each sub processes takes around 170MB of memory. See below for the memory usage even after kill (these sub processes are not created elsewhere).

Step to reproduce

Installed using pip install spleeter
Run as standalone or using VSC to debug

Output

Memory is kept high even after use and sub processes are kept running, even after killing they will be re-created.

Environment


OS	Windows
Installation type	pip
RAM available	8GB
Hardware spec	CPU

Additional context

ls-milkyway commented 2 years ago

hmmm.. this may be the reason why memory utilization is poor causing: [Bug] numpy.core._exceptions.MemoryError #728

Scylla2020 commented 2 years ago

Wow I never realised this but after checking I can also see multiple python processes are created and dont terminate. Looks like an issue with the separator.separate_to_file method as the multiprocessing pools are not being closed there. I referenced this to arrive to the solution.

So it means you could subclass Separator then implement an extra method that closes the pool. This seems to be able to close all processes at the end if synchronous is set to True in separate_to_file method, e.g.

from spleeter.separator import Separator

class MySeparator(Separator):
    def close_pool(self):
        if self._pool:
            self._pool.close()
            self._pool.terminate()

separator = MySeparator('spleeter:2stems')
separator.separate_to_file(filepath, self.filedirectory, synchronous=True)

# Wait for batch to finish and terminate pool.
separator.join()
separator.close_pool()

Scylla2020 commented 2 years ago

@ls-milkyway #728 is not related to this. Memory usage is just high by default and you need high specs for things to work properly.

deezer / spleeter