I am trying to write a program which applies a certain function hundreds of times. In order to speed up the process, instead of doing it sequentially, I am trying to parallelize execution of the function. What I am doing, is the following:
import multiprocessing
def start_process():
logger.debug('Starting {0}'.format(multiprocessing.current_process().name))
pool_size = len(inputs) if len(inputs) < multiprocessing.cpu_count() - 1 else multiprocessing.cpu_count() - 1
with multiprocessing.Pool(processes=pool_size, initializer=start_process) as pool:
for o, ipt in pool.imap_unordered(train_func, inputs):
output[(ipt[0], ipt[2])] = o
In the above code, I am using the argument initializer
in order to be able to keep track of the number of processes that should be spawned. Moreover, the function train_func
is a function that runs an optimization.
I am running the code on a server that has a maximum of 32 processors available at any time. Even though I would expect the number of "spawned" processes to reach a maximum of 31, I can see that more than 200-300 processes are spawned, and the program eventually crashes.
Moreover, I get the following error:
ERROR; return code from pthread_create() is 11
ERROR; return code from pthread_create() is 11
Error detail: Resource temporarily unavailable
Error detail: Resource temporarily unavailable
OMP: Error #34: System unable to allocate necessary resources for OMP thread:
OMP: System error #11: Resource temporarily unavailable
OMP: Hint Try decreasing the value of OMP_NUM_THREADS.
/bin/sh: fork: retry: No child processes
ERROR; return code from pthread_create() is 11
Error detail: Resource temporarily unavailable
/bin/sh: fork: retry: No child processes
OMP: Error #34: System unable to allocate necessary resources for OMP thread:
OMP: System error #11: Resource temporarily unavailable
OMP: Hint Try decreasing the value of OMP_NUM_THREADS.
Could you please provide any hint as to how I can indeed limit the number of processes spawned?
You should probably use the min
builtin function.
Also, the initialization code must be wrapped in if __name__ == '__main__':
, as described in the multiprocessing docs.
So all in all something like
import multiprocessing
def start_process():
logger.debug(
"Starting {0}".format(multiprocessing.current_process().name)
)
def train_func(*args):
pass
def main():
pool_size = min(len(inputs), multiprocessing.cpu_count() - 1)
with multiprocessing.Pool(
processes=pool_size, initializer=start_process
) as pool:
for o, ipt in pool.imap_unordered(train_func, inputs):
output[(ipt[0], ipt[2])] = o
if __name__ == "__main__":
main()
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.