FIFO leak when using many processes under pypy

fifoleak.py

import multiprocessing
from subprocess import run, PIPE
import os
import gc

def worker(x):
    return x * x

def n_files_open():
    return len(files_open())

def is_fifo(line):
    return 'FIFO' in line

def files_open():
    return run(['lsof', '-p', str(os.getpid())], stdout=PIPE, universal_newlines=True, check=True).stdout.split('\n')

def n_fifos_open():
    return len(list((filter(is_fifo, files_open()))))

print(n_files_open(), n_fifos_open())
with multiprocessing.Pool(maxtasksperchild=1) as p:
    for result in p.imap_unordered(worker, range(10 ** 2)):
        pass
print(n_files_open(), n_fifos_open())
gc.collect()
print(n_files_open(), n_fifos_open())

"""
→ python3 --version
Python 3.6.5
→ python3 fifoleak.py
26 1
62 17
62 17
→ pypy3 --version
Python 3.5.3 (fdd60ed87e941677e8ea11acf9f1819466521bf2, Apr 27 2018, 15:39:57)
[PyPy 6.0.0 with GCC 7.2.0]
→ pypy3 fifoleak.py
28 1
150 113
54 17
"""

maxtasksperchild=1 spawns a new process each time, imap_unordered is only used here b/c by default it has a chunksize of 1, so we truly are spawning one child per iterated element. the parent process doesn't trigger a gc in a simple loop (not uncommon for coordinating process) so the FIFO used in the SimpleQueue in the Pool don't get closed promptly