问题
How can I implement non-daemonic processes with pathos in python3 instead of with the multiprocessing module?
To be more specific, I am referring to: Python Process Pool non-daemonic?
The answer to this post implements non-daemonic processes via the multiprocessing module. Unfortunately, this module does not allow to pickle lambda functions among other objects, but pathos does in Python 2:
#import multiprocessing
#import multiprocessing.pool
import pathos
#class NoDaemonProcess(multiprocessing.Process):
class NoDaemonProcess(pathos.multiprocessing.Pool.Process):
def _get_daemon(self):
return False
def _set_daemon(self, value):
pass
daemon = property(_get_daemon, _set_daemon)
#class NoDaemonPool(multiprocessing.pool.Pool):
class NoDaemonPool(pathos.multiprocessing.Pool):
Process = NoDaemonProcess
def myproc(args):
i, max_workers = args
#pool = multiprocessing.Pool(max_workers)
pool = pathos.pools.ProcessPool(max_workers)
l_args = [j for j in range(i)]
mysubproc = lambda x : x
print("myproc", l_args, pool.map(mysubproc, l_args))
return i
max_workers = [2, 1]
executor = NoDaemonPool(max_workers[0])
#executor = pathos.multiprocessing.Pool(max_workers[0])
l_args = [(i, max_workers[1]) for i in range(10)]
print(executor.map(myproc, l_args))
output:
('myproc', [], [])
('myproc', [0, 1], [0, 1])
('myproc', [0], [0])
('myproc', [0, 1, 2], [0, 1, 2])
('myproc', [0, 1, 2, 3], [0, 1, 2, 3])
('myproc', [0, 1, 2, 3, 4, 5], [0, 1, 2, 3, 4, 5])
('myproc', [0, 1, 2, 3, 4], [0, 1, 2, 3, 4])
('myproc', [0, 1, 2, 3, 4, 5, 6], [0, 1, 2, 3, 4, 5, 6])
('myproc', [0, 1, 2, 3, 4, 5, 6, 7], [0, 1, 2, 3, 4, 5, 6, 7])
('myproc', [0, 1, 2, 3, 4, 5, 6, 7, 8], [0, 1, 2, 3, 4, 5, 6, 7, 8])
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
In Python 3, the pathos module has changed w.r.t. Python 2, e.g., pathos.multiprocessing.Pool.Process is not a class anymore, but a function, so one can no longer use it for inheritance (see above). - any pathos documentation I am missing?
How can I make the above code work in pathos in Python 3?
As a work-around for the above particular example, one can simply fall back to the multiprocessing NoDaemonPool implementation, and use pathos for the daemon sub-processes:
import multiprocessing
import multiprocessing.pool
import pathos
class NoDaemonProcess(multiprocessing.Process):
#class NoDaemonProcess(pathos.multiprocessing.Pool.Process):
def _get_daemon(self):
return False
def _set_daemon(self, value):
pass
daemon = property(_get_daemon, _set_daemon)
class NoDaemonPool(multiprocessing.pool.Pool):
#class NoDaemonPool(pathos.multiprocessing.Pool):
Process = NoDaemonProcess
def myproc(args):
i, max_workers = args
#pool = multiprocessing.Pool(max_workers)
pool = pathos.pools.ProcessPool(max_workers)
l_args = [j for j in range(i)]
mysubproc = lambda x : x
print("myproc", l_args, pool.map(mysubproc, l_args))
return i
max_workers = [2, 1]
executor = NoDaemonPool(max_workers[0])
#executor = pathos.multiprocessing.Pool(max_workers[0])
l_args = [(i, max_workers[1]) for i in range(10)]
print(executor.map(myproc, l_args))
However, this work-around is not a solution because (i) its imports both pathos and multiprocessing, and even more importantly (ii) it won't be able to pickle, e.g., if myproc instead is defined as
myproc = lambda x : x
Thank you very much, Best, Sebastian
回答1:
I just found the answer myself for Python 3 taking the original idea from
Python Process Pool non-daemonic?
and keeping a clean process error handling as suggested in
How to run nested, hierarchical pathos multiprocessing maps?
import pathos
import signal
import sys
import os
import time
# redefine process pool via inheritance
import multiprocess.context as context
class NoDaemonProcess(context.Process):
def _get_daemon(self):
return False
def _set_daemon(self, value):
pass
daemon = property(_get_daemon, _set_daemon)
class NoDaemonPool(pathos.multiprocessing.Pool):
def Process(self, *args, **kwds):
return NoDaemonProcess(*args, **kwds)
def get_pid_i(x):
return os.getpid()
def hard_kill_pool(pid_is, pool):
for pid_i in pid_is:
os.kill(pid_i, signal.SIGINT) # sending Ctrl+C
pool.terminate()
def myproc(args):
i, max_workers = args
l_args = [j for j in range(i)]
mysubproc = lambda x : x
pool = pathos.pools.ProcessPool(max_workers)
pool.restart(force=True)
pid_is = pool.map(get_pid_i, range(max_workers))
try:
l_traj_df = pool.amap(mysubproc, l_args)
counter_i = 0
while not l_traj_df.ready():
time.sleep(1)
if counter_i % 30 == 0:
print('Waiting for children running in pool.amap() in myproc( {} ) with PIDs: {}'.format(i, pid_is))
counter_i += 1
l_traj_df = l_traj_df.get()
pool.close()
pool.join()
except KeyboardInterrupt:
print('Ctrl+C received in myproc( {} ), attempting to terminate pool...').format(myens)
hard_kill_pool(pid_is, pool) # sending Ctrl+C
raise
except:
print('Attempting to close parallel after exception: {} in myproc( {} )'.format(sys.exc_info()[0], myens))
hard_kill_pool(pid_is, pool) # sending Ctrl+C
raise
#myproc = lambda x : x
max_workers = [2, 1]
pool = NoDaemonPool(max_workers[0])
#pool.restart(force=True)
pid_is = pool.map(get_pid_i, range(max_workers[0]))
try:
results = pool.map_async(myproc, l_args)
counter_i = 0
while not results.ready():
if counter_i % 30 == 0:
print('Waiting for children running in pool.map_async() in main() with PIDs: {}'.format(pid_is))
time.sleep(2)
counter_i += 1
results = results.get()
pool.close()
pool.join()
except KeyboardInterrupt:
print('Ctrl+C received in main(), attempting to terminate pool...')
hard_kill_pool(pid_is, pool) # sending Ctrl+C
raise
except:
print('Attempting to close parallel after exception: {} in main()'.format(_sys.exc_info()[0]))
hard_kill_pool(pid_is, pool) # sending Ctrl+C
raise
output:
Waiting for children running in pool.map_async() in main() with PIDs [15015, 15014]
Waiting for children running in pool.amap() in myproc( 2 ) with PIDs [15019]
Waiting for children running in pool.amap() in myproc( 1 ) with PIDs: [15020]
Waiting for children running in pool.amap() in myproc( 3 ) with PIDs [15021]
Waiting for children running in pool.amap() in myproc( 4 ) with PIDs [15022]
Waiting for children running in pool.amap() in myproc( 6 ) with PIDs [15024]
Waiting for children running in pool.amap() in myproc( 5 ) with PIDs [15023]
Waiting for children running in pool.amap() in myproc( 7 ) with PIDs [15025]
Waiting for children running in pool.amap() in myproc( 8 ) with PIDs [15026]
Waiting for children running in pool.amap() in myproc( 9 ) with PIDs [15028]
I used the following module versions:
python 3.6.0 0
pathos 0.2.1 py36_1 condo-forge
multiprocess 0.70.4 py36_0 http://conda.binstar.org/omnia
dill 0.2.7.1 py36h644ae93_0
pox 0.2.3 py36_0 conda-forge
ppft 1.6.4.7.1 py36_0 conda-forge
six 1.10.0 py36_0
来源：https://stackoverflow.com/questions/47574860/python-pathos-process-pool-non-daemonic