Skip to content

ENH: Explicit garbage collection in MultiProc #2315

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Dec 3, 2017
Merged

ENH: Explicit garbage collection in MultiProc #2315

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5573e54
ENH: Explicit garbage collection in MultiProc
effigies Nov 29, 2017
0edf2de
STY: Flake8 cleanup
effigies Nov 29, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
37 changes: 25 additions & 12 deletions nipype/pipeline/plugins/multiproc.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@
from multiprocessing import Process, Pool, cpu_count, pool
from traceback import format_exception
import sys
import gc

from copy import deepcopy
import numpy as np
Expand Down Expand Up @@ -121,13 +122,16 @@ def __init__(self, plugin_args=None):
non_daemon = self.plugin_args.get('non_daemon', True)
maxtasks = self.plugin_args.get('maxtasksperchild', 10)
self.processors = self.plugin_args.get('n_procs', cpu_count())
self.memory_gb = self.plugin_args.get('memory_gb', # Allocate 90% of system memory
get_system_total_memory_gb() * 0.9)
self.raise_insufficient = self.plugin_args.get('raise_insufficient', True)
self.memory_gb = self.plugin_args.get(
'memory_gb', # Allocate 90% of system memory
get_system_total_memory_gb() * 0.9)
self.raise_insufficient = self.plugin_args.get('raise_insufficient',
True)

# Instantiate different thread pools for non-daemon processes
logger.debug('MultiProcPlugin starting in "%sdaemon" mode (n_procs=%d, mem_gb=%0.2f)',
'non' * int(non_daemon), self.processors, self.memory_gb)
logger.debug('MultiProcPlugin starting in "%sdaemon" mode (n_procs=%d,'
'mem_gb=%0.2f)', 'non' * int(non_daemon), self.processors,
self.memory_gb)

NipypePool = NonDaemonPool if non_daemon else Pool
try:
Expand Down Expand Up @@ -204,12 +208,13 @@ def _send_procs_to_workers(self, updatehash=False, graph=None):
Sends jobs to workers when system resources are available.
"""

# Check to see if a job is available (jobs without dependencies not run)
# Check to see if a job is available (jobs with all dependencies run)
# See https://github.com/nipy/nipype/pull/2200#discussion_r141605722
jobids = np.nonzero(~self.proc_done & (self.depidx.sum(0) == 0))[1]

# Check available system resources by summing all threads and memory used
free_memory_gb, free_processors = self._check_resources(self.pending_tasks)
# Check available resources by summing all threads and memory used
free_memory_gb, free_processors = self._check_resources(
self.pending_tasks)

stats = (len(self.pending_tasks), len(jobids), free_memory_gb,
self.memory_gb, free_processors, self.processors)
Expand All @@ -228,7 +233,11 @@ def _send_procs_to_workers(self, updatehash=False, graph=None):
'be submitted to the queue. Potential deadlock')
return

jobids = self._sort_jobs(jobids, scheduler=self.plugin_args.get('scheduler'))
jobids = self._sort_jobs(jobids,
scheduler=self.plugin_args.get('scheduler'))

# Run garbage collector before potentially submitting jobs
gc.collect()

# Submit jobs
for jobid in jobids:
Expand Down Expand Up @@ -261,9 +270,10 @@ def _send_procs_to_workers(self, updatehash=False, graph=None):

free_memory_gb -= next_job_gb
free_processors -= next_job_th
logger.debug('Allocating %s ID=%d (%0.2fGB, %d threads). Free: %0.2fGB, %d threads.',
self.procs[jobid].fullname, jobid, next_job_gb, next_job_th,
free_memory_gb, free_processors)
logger.debug('Allocating %s ID=%d (%0.2fGB, %d threads). Free: '
'%0.2fGB, %d threads.', self.procs[jobid].fullname,
jobid, next_job_gb, next_job_th, free_memory_gb,
free_processors)

# change job status in appropriate queues
self.proc_done[jobid] = True
Expand Down Expand Up @@ -292,6 +302,9 @@ def _send_procs_to_workers(self, updatehash=False, graph=None):
free_processors += next_job_th
# Display stats next loop
self._stats = None

# Clean up any debris from running node in main process
gc.collect()
continue

# Task should be submitted to workers
Expand Down