2013-02-10 03:22:12 +04:00
|
|
|
# worker.py - master-slave parallelism support
|
|
|
|
#
|
|
|
|
# Copyright 2013 Facebook, Inc.
|
|
|
|
#
|
|
|
|
# This software may be used and distributed according to the terms of the
|
|
|
|
# GNU General Public License version 2 or any later version.
|
|
|
|
|
2015-08-09 04:44:41 +03:00
|
|
|
from __future__ import absolute_import
|
|
|
|
|
|
|
|
import errno
|
|
|
|
import os
|
|
|
|
import signal
|
|
|
|
import sys
|
|
|
|
|
|
|
|
from .i18n import _
|
2016-11-15 02:12:11 +03:00
|
|
|
from . import (
|
2016-12-17 23:16:39 +03:00
|
|
|
encoding,
|
2016-11-15 02:12:11 +03:00
|
|
|
error,
|
2016-12-18 21:46:52 +03:00
|
|
|
pycompat,
|
2016-11-24 04:15:34 +03:00
|
|
|
scmutil,
|
2016-11-15 02:12:11 +03:00
|
|
|
util,
|
|
|
|
)
|
2013-02-10 03:22:12 +04:00
|
|
|
|
|
|
|
def countcpus():
|
|
|
|
'''try to count the number of CPUs on the system'''
|
2015-10-08 20:57:03 +03:00
|
|
|
|
|
|
|
# posix
|
|
|
|
try:
|
|
|
|
n = int(os.sysconf('SC_NPROCESSORS_ONLN'))
|
|
|
|
if n > 0:
|
|
|
|
return n
|
|
|
|
except (AttributeError, ValueError):
|
|
|
|
pass
|
|
|
|
|
|
|
|
# windows
|
2013-02-10 03:22:12 +04:00
|
|
|
try:
|
2016-12-17 23:16:39 +03:00
|
|
|
n = int(encoding.environ['NUMBER_OF_PROCESSORS'])
|
2015-10-08 20:57:03 +03:00
|
|
|
if n > 0:
|
|
|
|
return n
|
|
|
|
except (KeyError, ValueError):
|
|
|
|
pass
|
|
|
|
|
|
|
|
return 1
|
2013-02-10 03:51:26 +04:00
|
|
|
|
|
|
|
def _numworkers(ui):
|
|
|
|
s = ui.config('worker', 'numcpus')
|
|
|
|
if s:
|
|
|
|
try:
|
|
|
|
n = int(s)
|
|
|
|
if n >= 1:
|
|
|
|
return n
|
|
|
|
except ValueError:
|
2015-10-08 22:55:45 +03:00
|
|
|
raise error.Abort(_('number of cpus must be an integer'))
|
2013-02-10 03:51:26 +04:00
|
|
|
return min(max(countcpus(), 4), 32)
|
|
|
|
|
2016-12-18 21:46:52 +03:00
|
|
|
if pycompat.osname == 'posix':
|
2013-02-10 03:51:26 +04:00
|
|
|
_startupcost = 0.01
|
|
|
|
else:
|
|
|
|
_startupcost = 1e30
|
|
|
|
|
|
|
|
def worthwhile(ui, costperop, nops):
|
|
|
|
'''try to determine whether the benefit of multiple processes can
|
|
|
|
outweigh the cost of starting them'''
|
|
|
|
linear = costperop * nops
|
|
|
|
workers = _numworkers(ui)
|
|
|
|
benefit = linear - (_startupcost * workers + linear / workers)
|
|
|
|
return benefit >= 0.15
|
2013-02-10 03:51:32 +04:00
|
|
|
|
2013-02-10 03:51:32 +04:00
|
|
|
def worker(ui, costperarg, func, staticargs, args):
|
|
|
|
'''run a function, possibly in parallel in multiple worker
|
|
|
|
processes.
|
|
|
|
|
|
|
|
returns a progress iterator
|
|
|
|
|
|
|
|
costperarg - cost of a single task
|
|
|
|
|
|
|
|
func - function to run
|
|
|
|
|
|
|
|
staticargs - arguments to pass to every invocation of the function
|
|
|
|
|
|
|
|
args - arguments to split into chunks, to pass to individual
|
|
|
|
workers
|
|
|
|
'''
|
|
|
|
if worthwhile(ui, costperarg, len(args)):
|
|
|
|
return _platformworker(ui, func, staticargs, args)
|
|
|
|
return func(*staticargs + (args,))
|
|
|
|
|
|
|
|
def _posixworker(ui, func, staticargs, args):
|
|
|
|
rfd, wfd = os.pipe()
|
|
|
|
workers = _numworkers(ui)
|
2013-02-20 23:31:31 +04:00
|
|
|
oldhandler = signal.getsignal(signal.SIGINT)
|
|
|
|
signal.signal(signal.SIGINT, signal.SIG_IGN)
|
2016-11-15 05:10:40 +03:00
|
|
|
pids, problem = set(), [0]
|
2016-07-28 22:49:57 +03:00
|
|
|
def killworkers():
|
2016-11-17 14:44:05 +03:00
|
|
|
# unregister SIGCHLD handler as all children will be killed. This
|
|
|
|
# function shouldn't be interrupted by another SIGCHLD; otherwise pids
|
|
|
|
# could be updated while iterating, which would cause inconsistency.
|
|
|
|
signal.signal(signal.SIGCHLD, oldchldhandler)
|
2016-07-28 22:49:57 +03:00
|
|
|
# if one worker bails, there's no good reason to wait for the rest
|
|
|
|
for p in pids:
|
|
|
|
try:
|
|
|
|
os.kill(p, signal.SIGTERM)
|
|
|
|
except OSError as err:
|
|
|
|
if err.errno != errno.ESRCH:
|
|
|
|
raise
|
2016-07-28 22:57:07 +03:00
|
|
|
def waitforworkers(blocking=True):
|
2016-11-15 05:12:16 +03:00
|
|
|
for pid in pids.copy():
|
|
|
|
p = st = 0
|
|
|
|
while True:
|
|
|
|
try:
|
|
|
|
p, st = os.waitpid(pid, (0 if blocking else os.WNOHANG))
|
2016-11-17 15:43:01 +03:00
|
|
|
break
|
2016-11-15 05:12:16 +03:00
|
|
|
except OSError as e:
|
|
|
|
if e.errno == errno.EINTR:
|
|
|
|
continue
|
|
|
|
elif e.errno == errno.ECHILD:
|
2016-11-17 14:57:09 +03:00
|
|
|
# child would already be reaped, but pids yet been
|
|
|
|
# updated (maybe interrupted just after waitpid)
|
|
|
|
pids.discard(pid)
|
|
|
|
break
|
2016-11-15 05:12:16 +03:00
|
|
|
else:
|
|
|
|
raise
|
worker: ignore meaningless exit status indication returned by os.waitpid()
Before this patch, worker implementation assumes that os.waitpid()
with os.WNOHANG returns '(0, 0)' for still running child process. This
is explicitly specified as below in Python API document.
os.WNOHANG
The option for waitpid() to return immediately if no child
process status is available immediately. The function returns
(0, 0) in this case.
On the other hand, POSIX specification doesn't define the "stat_loc"
value returned by waitpid() with WNOHANG for such child process.
http://pubs.opengroup.org/onlinepubs/9699919799/functions/waitpid.html
CPython implementation for os.waitpid() on POSIX doesn't take any care
of this gap, and this may cause unexpected "exit status indication"
even on POSIX conformance platform.
For example, os.waitpid() with os.WNOHANG returns non-zero "exit
status indication" on FreeBSD. This implies os.kill() with own pid or
sys.exit() with non-zero exit code, even if no child process fails.
To ignore meaningless exit status indication returned by os.waitpid(),
this patch skips subsequent steps forcibly, if os.waitpid() returns 0
as pid.
This patch also arranges examination of 'p' value for readability.
FYI, there are some issues below about this behavior reported for
CPython.
https://bugs.python.org/issue21791
https://bugs.python.org/issue27808
2017-02-24 19:07:52 +03:00
|
|
|
if not p:
|
|
|
|
# skip subsequent steps, because child process should
|
|
|
|
# be still running in this case
|
|
|
|
continue
|
|
|
|
pids.discard(p)
|
|
|
|
st = _exitstatus(st)
|
2016-07-28 22:49:57 +03:00
|
|
|
if st and not problem[0]:
|
|
|
|
problem[0] = st
|
2016-11-12 06:07:22 +03:00
|
|
|
def sigchldhandler(signum, frame):
|
|
|
|
waitforworkers(blocking=False)
|
2016-11-17 15:08:58 +03:00
|
|
|
if problem[0]:
|
|
|
|
killworkers()
|
2016-11-12 06:07:22 +03:00
|
|
|
oldchldhandler = signal.signal(signal.SIGCHLD, sigchldhandler)
|
2017-03-28 20:21:38 +03:00
|
|
|
ui.flush()
|
2013-02-10 03:51:32 +04:00
|
|
|
for pargs in partition(args, workers):
|
|
|
|
pid = os.fork()
|
|
|
|
if pid == 0:
|
2013-02-20 23:31:31 +04:00
|
|
|
signal.signal(signal.SIGINT, oldhandler)
|
2016-11-12 06:07:22 +03:00
|
|
|
signal.signal(signal.SIGCHLD, oldchldhandler)
|
2016-11-24 04:15:34 +03:00
|
|
|
|
|
|
|
def workerfunc():
|
2013-02-10 03:51:32 +04:00
|
|
|
os.close(rfd)
|
|
|
|
for i, item in func(*(staticargs + (pargs,))):
|
|
|
|
os.write(wfd, '%d %s\n' % (i, item))
|
2016-11-24 04:15:34 +03:00
|
|
|
|
|
|
|
# make sure we use os._exit in all code paths. otherwise the worker
|
|
|
|
# may do some clean-ups which could cause surprises like deadlock.
|
|
|
|
# see sshpeer.cleanup for example.
|
|
|
|
try:
|
2017-02-25 06:48:50 +03:00
|
|
|
try:
|
|
|
|
scmutil.callcatch(ui, workerfunc)
|
|
|
|
finally:
|
|
|
|
ui.flush()
|
2013-02-10 03:51:32 +04:00
|
|
|
except KeyboardInterrupt:
|
|
|
|
os._exit(255)
|
2016-11-24 04:15:34 +03:00
|
|
|
except: # never return, therefore no re-raises
|
|
|
|
try:
|
|
|
|
ui.traceback()
|
2017-02-25 06:48:50 +03:00
|
|
|
ui.flush()
|
2016-11-24 04:15:34 +03:00
|
|
|
finally:
|
|
|
|
os._exit(255)
|
|
|
|
else:
|
|
|
|
os._exit(0)
|
2016-11-15 05:10:40 +03:00
|
|
|
pids.add(pid)
|
2013-02-10 03:51:32 +04:00
|
|
|
os.close(wfd)
|
2017-02-13 17:36:38 +03:00
|
|
|
fp = os.fdopen(rfd, pycompat.sysstr('rb'), 0)
|
2013-02-10 03:51:32 +04:00
|
|
|
def cleanup():
|
|
|
|
signal.signal(signal.SIGINT, oldhandler)
|
2016-11-12 06:06:07 +03:00
|
|
|
waitforworkers()
|
2016-11-12 06:07:22 +03:00
|
|
|
signal.signal(signal.SIGCHLD, oldchldhandler)
|
2013-02-20 23:31:34 +04:00
|
|
|
status = problem[0]
|
|
|
|
if status:
|
|
|
|
if status < 0:
|
|
|
|
os.kill(os.getpid(), -status)
|
|
|
|
sys.exit(status)
|
2013-02-10 03:51:32 +04:00
|
|
|
try:
|
2016-11-15 02:12:11 +03:00
|
|
|
for line in util.iterfile(fp):
|
2013-02-10 03:51:32 +04:00
|
|
|
l = line.split(' ', 1)
|
|
|
|
yield int(l[0]), l[1][:-1]
|
|
|
|
except: # re-raises
|
2013-02-20 23:31:34 +04:00
|
|
|
killworkers()
|
2013-02-10 03:51:32 +04:00
|
|
|
cleanup()
|
|
|
|
raise
|
|
|
|
cleanup()
|
|
|
|
|
2013-02-20 23:31:27 +04:00
|
|
|
def _posixexitstatus(code):
|
|
|
|
'''convert a posix exit status into the same form returned by
|
|
|
|
os.spawnv
|
|
|
|
|
|
|
|
returns None if the process was stopped instead of exiting'''
|
|
|
|
if os.WIFEXITED(code):
|
|
|
|
return os.WEXITSTATUS(code)
|
|
|
|
elif os.WIFSIGNALED(code):
|
|
|
|
return -os.WTERMSIG(code)
|
|
|
|
|
2016-12-18 21:46:52 +03:00
|
|
|
if pycompat.osname != 'nt':
|
2013-02-10 03:51:32 +04:00
|
|
|
_platformworker = _posixworker
|
2013-02-20 23:31:27 +04:00
|
|
|
_exitstatus = _posixexitstatus
|
2013-02-10 03:51:32 +04:00
|
|
|
|
2013-02-10 03:51:32 +04:00
|
|
|
def partition(lst, nslices):
|
2016-02-21 02:56:44 +03:00
|
|
|
'''partition a list into N slices of roughly equal size
|
|
|
|
|
|
|
|
The current strategy takes every Nth element from the input. If
|
|
|
|
we ever write workers that need to preserve grouping in input
|
|
|
|
we should consider allowing callers to specify a partition strategy.
|
2016-02-28 08:43:17 +03:00
|
|
|
|
|
|
|
mpm is not a fan of this partitioning strategy when files are involved.
|
|
|
|
In his words:
|
|
|
|
|
|
|
|
Single-threaded Mercurial makes a point of creating and visiting
|
|
|
|
files in a fixed order (alphabetical). When creating files in order,
|
|
|
|
a typical filesystem is likely to allocate them on nearby regions on
|
|
|
|
disk. Thus, when revisiting in the same order, locality is maximized
|
|
|
|
and various forms of OS and disk-level caching and read-ahead get a
|
|
|
|
chance to work.
|
|
|
|
|
|
|
|
This effect can be quite significant on spinning disks. I discovered it
|
|
|
|
circa Mercurial v0.4 when revlogs were named by hashes of filenames.
|
|
|
|
Tarring a repo and copying it to another disk effectively randomized
|
|
|
|
the revlog ordering on disk by sorting the revlogs by hash and suddenly
|
|
|
|
performance of my kernel checkout benchmark dropped by ~10x because the
|
|
|
|
"working set" of sectors visited no longer fit in the drive's cache and
|
|
|
|
the workload switched from streaming to random I/O.
|
|
|
|
|
|
|
|
What we should really be doing is have workers read filenames from a
|
|
|
|
ordered queue. This preserves locality and also keeps any worker from
|
|
|
|
getting more than one file out of balance.
|
2016-02-21 02:56:44 +03:00
|
|
|
'''
|
|
|
|
for i in range(nslices):
|
|
|
|
yield lst[i::nslices]
|