annotate mercurial/worker.py @ 30421:47de34f79f93

worker: wait worker pid explicitly Before this patch, waitforworkers uses os.wait() to collect child workers, and only wait len(pids) processes. This can have serious issues if other code spawns new processes and does not reap them: 1. worker.py may get wrong exit code and kill innocent workers. 2. worker.py may continue without waiting for all workers to complete. This patch fixes the issue by using waitpid to wait worker pid explicitly. However, this patch introduces a new issue: worker failure may not be handled immediately. The issue will be addressed in next patches.
author Jun Wu <quark@fb.com>
date Thu, 28 Jul 2016 20:51:20 +0100
parents 7a5d6e2fd2d5
children 7bc25549e084
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
18635
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
1 # worker.py - master-slave parallelism support
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
2 #
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
3 # Copyright 2013 Facebook, Inc.
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
4 #
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
5 # This software may be used and distributed according to the terms of the
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
6 # GNU General Public License version 2 or any later version.
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
7
25992
2d76f8a2d831 worker: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents: 25660
diff changeset
8 from __future__ import absolute_import
2d76f8a2d831 worker: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents: 25660
diff changeset
9
2d76f8a2d831 worker: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents: 25660
diff changeset
10 import errno
2d76f8a2d831 worker: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents: 25660
diff changeset
11 import os
2d76f8a2d831 worker: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents: 25660
diff changeset
12 import signal
2d76f8a2d831 worker: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents: 25660
diff changeset
13 import sys
2d76f8a2d831 worker: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents: 25660
diff changeset
14 import threading
2d76f8a2d831 worker: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents: 25660
diff changeset
15
2d76f8a2d831 worker: use absolute_import
Gregory Szorc <gregory.szorc@gmail.com>
parents: 25660
diff changeset
16 from .i18n import _
30406
78a58dcf8853 worker: migrate to util.iterfile
Jun Wu <quark@fb.com>
parents: 28292
diff changeset
17 from . import (
78a58dcf8853 worker: migrate to util.iterfile
Jun Wu <quark@fb.com>
parents: 28292
diff changeset
18 error,
78a58dcf8853 worker: migrate to util.iterfile
Jun Wu <quark@fb.com>
parents: 28292
diff changeset
19 util,
78a58dcf8853 worker: migrate to util.iterfile
Jun Wu <quark@fb.com>
parents: 28292
diff changeset
20 )
18635
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
21
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
22 def countcpus():
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
23 '''try to count the number of CPUs on the system'''
26568
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
24
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
25 # posix
18635
fed06dd07665 worker: count the number of CPUs
Bryan O'Sullivan <bryano@fb.com>
parents:
diff changeset
26 try:
26568
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
27 n = int(os.sysconf('SC_NPROCESSORS_ONLN'))
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
28 if n > 0:
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
29 return n
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
30 except (AttributeError, ValueError):
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
31 pass
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
32
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
33 # windows
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
34 try:
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
35 n = int(os.environ['NUMBER_OF_PROCESSORS'])
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
36 if n > 0:
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
37 return n
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
38 except (KeyError, ValueError):
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
39 pass
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
40
c0501c26b05c worker: restore old countcpus code (issue4869)
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26063
diff changeset
41 return 1
18636
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
42
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
43 def _numworkers(ui):
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
44 s = ui.config('worker', 'numcpus')
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
45 if s:
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
46 try:
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
47 n = int(s)
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
48 if n >= 1:
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
49 return n
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
50 except ValueError:
26587
56b2bcea2529 error: get Abort from 'error' instead of 'util'
Pierre-Yves David <pierre-yves.david@fb.com>
parents: 26568
diff changeset
51 raise error.Abort(_('number of cpus must be an integer'))
18636
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
52 return min(max(countcpus(), 4), 32)
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
53
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
54 if os.name == 'posix':
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
55 _startupcost = 0.01
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
56 else:
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
57 _startupcost = 1e30
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
58
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
59 def worthwhile(ui, costperop, nops):
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
60 '''try to determine whether the benefit of multiple processes can
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
61 outweigh the cost of starting them'''
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
62 linear = costperop * nops
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
63 workers = _numworkers(ui)
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
64 benefit = linear - (_startupcost * workers + linear / workers)
dcb27c153a40 worker: estimate whether it's worth running a task in parallel
Bryan O'Sullivan <bryano@fb.com>
parents: 18635
diff changeset
65 return benefit >= 0.15
18637
ac4dbceeb14a worker: partition a list (of tasks) into equal-sized chunks
Bryan O'Sullivan <bryano@fb.com>
parents: 18636
diff changeset
66
18638
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
67 def worker(ui, costperarg, func, staticargs, args):
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
68 '''run a function, possibly in parallel in multiple worker
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
69 processes.
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
70
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
71 returns a progress iterator
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
72
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
73 costperarg - cost of a single task
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
74
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
75 func - function to run
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
76
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
77 staticargs - arguments to pass to every invocation of the function
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
78
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
79 args - arguments to split into chunks, to pass to individual
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
80 workers
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
81 '''
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
82 if worthwhile(ui, costperarg, len(args)):
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
83 return _platformworker(ui, func, staticargs, args)
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
84 return func(*staticargs + (args,))
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
85
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
86 def _posixworker(ui, func, staticargs, args):
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
87 rfd, wfd = os.pipe()
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
88 workers = _numworkers(ui)
18708
86524a70c0f6 worker: fix a race in SIGINT handling
Bryan O'Sullivan <bryano@fb.com>
parents: 18707
diff changeset
89 oldhandler = signal.getsignal(signal.SIGINT)
86524a70c0f6 worker: fix a race in SIGINT handling
Bryan O'Sullivan <bryano@fb.com>
parents: 18707
diff changeset
90 signal.signal(signal.SIGINT, signal.SIG_IGN)
18709
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
91 pids, problem = [], [0]
30420
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
92 def killworkers():
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
93 # if one worker bails, there's no good reason to wait for the rest
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
94 for p in pids:
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
95 try:
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
96 os.kill(p, signal.SIGTERM)
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
97 except OSError as err:
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
98 if err.errno != errno.ESRCH:
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
99 raise
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
100 def waitforworkers():
30421
47de34f79f93 worker: wait worker pid explicitly
Jun Wu <quark@fb.com>
parents: 30420
diff changeset
101 for pid in pids:
47de34f79f93 worker: wait worker pid explicitly
Jun Wu <quark@fb.com>
parents: 30420
diff changeset
102 st = _exitstatus(os.waitpid(pid, 0)[1])
30420
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
103 if st and not problem[0]:
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
104 problem[0] = st
7a5d6e2fd2d5 worker: move killworkers and waitforworkers up
Jun Wu <quark@fb.com>
parents: 30406
diff changeset
105 killworkers()
18638
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
106 for pargs in partition(args, workers):
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
107 pid = os.fork()
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
108 if pid == 0:
18708
86524a70c0f6 worker: fix a race in SIGINT handling
Bryan O'Sullivan <bryano@fb.com>
parents: 18707
diff changeset
109 signal.signal(signal.SIGINT, oldhandler)
18638
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
110 try:
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
111 os.close(rfd)
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
112 for i, item in func(*(staticargs + (pargs,))):
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
113 os.write(wfd, '%d %s\n' % (i, item))
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
114 os._exit(0)
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
115 except KeyboardInterrupt:
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
116 os._exit(255)
19408
c7ec39c1a381 worker: properly report errors from worker processes (issue3982)
Matt Mackall <mpm@selenic.com>
parents: 19406
diff changeset
117 # other exceptions are allowed to propagate, we rely
c7ec39c1a381 worker: properly report errors from worker processes (issue3982)
Matt Mackall <mpm@selenic.com>
parents: 19406
diff changeset
118 # on lock.py's pid checks to avoid release callbacks
18709
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
119 pids.append(pid)
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
120 pids.reverse()
18638
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
121 os.close(wfd)
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
122 fp = os.fdopen(rfd, 'rb', 0)
18709
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
123 t = threading.Thread(target=waitforworkers)
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
124 t.start()
18638
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
125 def cleanup():
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
126 signal.signal(signal.SIGINT, oldhandler)
18709
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
127 t.join()
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
128 status = problem[0]
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
129 if status:
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
130 if status < 0:
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
131 os.kill(os.getpid(), -status)
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
132 sys.exit(status)
18638
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
133 try:
30406
78a58dcf8853 worker: migrate to util.iterfile
Jun Wu <quark@fb.com>
parents: 28292
diff changeset
134 for line in util.iterfile(fp):
18638
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
135 l = line.split(' ', 1)
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
136 yield int(l[0]), l[1][:-1]
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
137 except: # re-raises
18709
9955fc5ee24b worker: handle worker failures more aggressively
Bryan O'Sullivan <bryano@fb.com>
parents: 18708
diff changeset
138 killworkers()
18638
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
139 cleanup()
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
140 raise
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
141 cleanup()
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
142
18707
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
143 def _posixexitstatus(code):
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
144 '''convert a posix exit status into the same form returned by
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
145 os.spawnv
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
146
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
147 returns None if the process was stopped instead of exiting'''
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
148 if os.WIFEXITED(code):
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
149 return os.WEXITSTATUS(code)
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
150 elif os.WIFSIGNALED(code):
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
151 return -os.WTERMSIG(code)
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
152
18638
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
153 if os.name != 'nt':
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
154 _platformworker = _posixworker
18707
d1a2b086d058 worker: on error, exit similarly to the first failing worker
Bryan O'Sullivan <bryano@fb.com>
parents: 18638
diff changeset
155 _exitstatus = _posixexitstatus
18638
047110c0e2a8 worker: allow a function to be run in multiple worker processes
Bryan O'Sullivan <bryano@fb.com>
parents: 18637
diff changeset
156
18637
ac4dbceeb14a worker: partition a list (of tasks) into equal-sized chunks
Bryan O'Sullivan <bryano@fb.com>
parents: 18636
diff changeset
157 def partition(lst, nslices):
28181
f8efc8a3a991 worker: change partition strategy to every Nth element
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26587
diff changeset
158 '''partition a list into N slices of roughly equal size
f8efc8a3a991 worker: change partition strategy to every Nth element
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26587
diff changeset
159
f8efc8a3a991 worker: change partition strategy to every Nth element
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26587
diff changeset
160 The current strategy takes every Nth element from the input. If
f8efc8a3a991 worker: change partition strategy to every Nth element
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26587
diff changeset
161 we ever write workers that need to preserve grouping in input
f8efc8a3a991 worker: change partition strategy to every Nth element
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26587
diff changeset
162 we should consider allowing callers to specify a partition strategy.
28292
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
163
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
164 mpm is not a fan of this partitioning strategy when files are involved.
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
165 In his words:
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
166
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
167 Single-threaded Mercurial makes a point of creating and visiting
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
168 files in a fixed order (alphabetical). When creating files in order,
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
169 a typical filesystem is likely to allocate them on nearby regions on
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
170 disk. Thus, when revisiting in the same order, locality is maximized
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
171 and various forms of OS and disk-level caching and read-ahead get a
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
172 chance to work.
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
173
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
174 This effect can be quite significant on spinning disks. I discovered it
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
175 circa Mercurial v0.4 when revlogs were named by hashes of filenames.
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
176 Tarring a repo and copying it to another disk effectively randomized
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
177 the revlog ordering on disk by sorting the revlogs by hash and suddenly
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
178 performance of my kernel checkout benchmark dropped by ~10x because the
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
179 "working set" of sectors visited no longer fit in the drive's cache and
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
180 the workload switched from streaming to random I/O.
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
181
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
182 What we should really be doing is have workers read filenames from a
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
183 ordered queue. This preserves locality and also keeps any worker from
3eb7faf6d958 worker: document poor partitioning scheme impact
Gregory Szorc <gregory.szorc@gmail.com>
parents: 28181
diff changeset
184 getting more than one file out of balance.
28181
f8efc8a3a991 worker: change partition strategy to every Nth element
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26587
diff changeset
185 '''
f8efc8a3a991 worker: change partition strategy to every Nth element
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26587
diff changeset
186 for i in range(nslices):
f8efc8a3a991 worker: change partition strategy to every Nth element
Gregory Szorc <gregory.szorc@gmail.com>
parents: 26587
diff changeset
187 yield lst[i::nslices]