Show More
@@ -1,188 +1,199 b'' | |||||
1 | # worker.py - master-slave parallelism support |
|
1 | # worker.py - master-slave parallelism support | |
2 | # |
|
2 | # | |
3 | # Copyright 2013 Facebook, Inc. |
|
3 | # Copyright 2013 Facebook, Inc. | |
4 | # |
|
4 | # | |
5 | # This software may be used and distributed according to the terms of the |
|
5 | # This software may be used and distributed according to the terms of the | |
6 | # GNU General Public License version 2 or any later version. |
|
6 | # GNU General Public License version 2 or any later version. | |
7 |
|
7 | |||
8 | from __future__ import absolute_import |
|
8 | from __future__ import absolute_import | |
9 |
|
9 | |||
10 | import errno |
|
10 | import errno | |
11 | import os |
|
11 | import os | |
12 | import signal |
|
12 | import signal | |
13 | import sys |
|
13 | import sys | |
14 | import threading |
|
14 | import threading | |
15 |
|
15 | |||
16 | from .i18n import _ |
|
16 | from .i18n import _ | |
17 | from . import ( |
|
17 | from . import ( | |
18 | error, |
|
18 | error, | |
19 | util, |
|
19 | util, | |
20 | ) |
|
20 | ) | |
21 |
|
21 | |||
22 | def countcpus(): |
|
22 | def countcpus(): | |
23 | '''try to count the number of CPUs on the system''' |
|
23 | '''try to count the number of CPUs on the system''' | |
24 |
|
24 | |||
25 | # posix |
|
25 | # posix | |
26 | try: |
|
26 | try: | |
27 | n = int(os.sysconf('SC_NPROCESSORS_ONLN')) |
|
27 | n = int(os.sysconf('SC_NPROCESSORS_ONLN')) | |
28 | if n > 0: |
|
28 | if n > 0: | |
29 | return n |
|
29 | return n | |
30 | except (AttributeError, ValueError): |
|
30 | except (AttributeError, ValueError): | |
31 | pass |
|
31 | pass | |
32 |
|
32 | |||
33 | # windows |
|
33 | # windows | |
34 | try: |
|
34 | try: | |
35 | n = int(os.environ['NUMBER_OF_PROCESSORS']) |
|
35 | n = int(os.environ['NUMBER_OF_PROCESSORS']) | |
36 | if n > 0: |
|
36 | if n > 0: | |
37 | return n |
|
37 | return n | |
38 | except (KeyError, ValueError): |
|
38 | except (KeyError, ValueError): | |
39 | pass |
|
39 | pass | |
40 |
|
40 | |||
41 | return 1 |
|
41 | return 1 | |
42 |
|
42 | |||
43 | def _numworkers(ui): |
|
43 | def _numworkers(ui): | |
44 | s = ui.config('worker', 'numcpus') |
|
44 | s = ui.config('worker', 'numcpus') | |
45 | if s: |
|
45 | if s: | |
46 | try: |
|
46 | try: | |
47 | n = int(s) |
|
47 | n = int(s) | |
48 | if n >= 1: |
|
48 | if n >= 1: | |
49 | return n |
|
49 | return n | |
50 | except ValueError: |
|
50 | except ValueError: | |
51 | raise error.Abort(_('number of cpus must be an integer')) |
|
51 | raise error.Abort(_('number of cpus must be an integer')) | |
52 | return min(max(countcpus(), 4), 32) |
|
52 | return min(max(countcpus(), 4), 32) | |
53 |
|
53 | |||
54 | if os.name == 'posix': |
|
54 | if os.name == 'posix': | |
55 | _startupcost = 0.01 |
|
55 | _startupcost = 0.01 | |
56 | else: |
|
56 | else: | |
57 | _startupcost = 1e30 |
|
57 | _startupcost = 1e30 | |
58 |
|
58 | |||
59 | def worthwhile(ui, costperop, nops): |
|
59 | def worthwhile(ui, costperop, nops): | |
60 | '''try to determine whether the benefit of multiple processes can |
|
60 | '''try to determine whether the benefit of multiple processes can | |
61 | outweigh the cost of starting them''' |
|
61 | outweigh the cost of starting them''' | |
62 | linear = costperop * nops |
|
62 | linear = costperop * nops | |
63 | workers = _numworkers(ui) |
|
63 | workers = _numworkers(ui) | |
64 | benefit = linear - (_startupcost * workers + linear / workers) |
|
64 | benefit = linear - (_startupcost * workers + linear / workers) | |
65 | return benefit >= 0.15 |
|
65 | return benefit >= 0.15 | |
66 |
|
66 | |||
67 | def worker(ui, costperarg, func, staticargs, args): |
|
67 | def worker(ui, costperarg, func, staticargs, args): | |
68 | '''run a function, possibly in parallel in multiple worker |
|
68 | '''run a function, possibly in parallel in multiple worker | |
69 | processes. |
|
69 | processes. | |
70 |
|
70 | |||
71 | returns a progress iterator |
|
71 | returns a progress iterator | |
72 |
|
72 | |||
73 | costperarg - cost of a single task |
|
73 | costperarg - cost of a single task | |
74 |
|
74 | |||
75 | func - function to run |
|
75 | func - function to run | |
76 |
|
76 | |||
77 | staticargs - arguments to pass to every invocation of the function |
|
77 | staticargs - arguments to pass to every invocation of the function | |
78 |
|
78 | |||
79 | args - arguments to split into chunks, to pass to individual |
|
79 | args - arguments to split into chunks, to pass to individual | |
80 | workers |
|
80 | workers | |
81 | ''' |
|
81 | ''' | |
82 | if worthwhile(ui, costperarg, len(args)): |
|
82 | if worthwhile(ui, costperarg, len(args)): | |
83 | return _platformworker(ui, func, staticargs, args) |
|
83 | return _platformworker(ui, func, staticargs, args) | |
84 | return func(*staticargs + (args,)) |
|
84 | return func(*staticargs + (args,)) | |
85 |
|
85 | |||
86 | def _posixworker(ui, func, staticargs, args): |
|
86 | def _posixworker(ui, func, staticargs, args): | |
87 | rfd, wfd = os.pipe() |
|
87 | rfd, wfd = os.pipe() | |
88 | workers = _numworkers(ui) |
|
88 | workers = _numworkers(ui) | |
89 | oldhandler = signal.getsignal(signal.SIGINT) |
|
89 | oldhandler = signal.getsignal(signal.SIGINT) | |
90 | signal.signal(signal.SIGINT, signal.SIG_IGN) |
|
90 | signal.signal(signal.SIGINT, signal.SIG_IGN) | |
91 | pids, problem = set(), [0] |
|
91 | pids, problem = set(), [0] | |
92 | def killworkers(): |
|
92 | def killworkers(): | |
93 | # if one worker bails, there's no good reason to wait for the rest |
|
93 | # if one worker bails, there's no good reason to wait for the rest | |
94 | for p in pids: |
|
94 | for p in pids: | |
95 | try: |
|
95 | try: | |
96 | os.kill(p, signal.SIGTERM) |
|
96 | os.kill(p, signal.SIGTERM) | |
97 | except OSError as err: |
|
97 | except OSError as err: | |
98 | if err.errno != errno.ESRCH: |
|
98 | if err.errno != errno.ESRCH: | |
99 | raise |
|
99 | raise | |
100 | def waitforworkers(blocking=True): |
|
100 | def waitforworkers(blocking=True): | |
101 | for pid in pids: |
|
101 | for pid in pids.copy(): | |
102 | p, st = os.waitpid(pid, 0 if blocking else os.WNOHANG) |
|
102 | p = st = 0 | |
|
103 | while True: | |||
|
104 | try: | |||
|
105 | p, st = os.waitpid(pid, (0 if blocking else os.WNOHANG)) | |||
|
106 | except OSError as e: | |||
|
107 | if e.errno == errno.EINTR: | |||
|
108 | continue | |||
|
109 | elif e.errno == errno.ECHILD: | |||
|
110 | break # ignore ECHILD | |||
|
111 | else: | |||
|
112 | raise | |||
103 | if p: |
|
113 | if p: | |
|
114 | pids.remove(p) | |||
104 | st = _exitstatus(st) |
|
115 | st = _exitstatus(st) | |
105 | if st and not problem[0]: |
|
116 | if st and not problem[0]: | |
106 | problem[0] = st |
|
117 | problem[0] = st | |
107 | killworkers() |
|
118 | killworkers() | |
108 | for pargs in partition(args, workers): |
|
119 | for pargs in partition(args, workers): | |
109 | pid = os.fork() |
|
120 | pid = os.fork() | |
110 | if pid == 0: |
|
121 | if pid == 0: | |
111 | signal.signal(signal.SIGINT, oldhandler) |
|
122 | signal.signal(signal.SIGINT, oldhandler) | |
112 | try: |
|
123 | try: | |
113 | os.close(rfd) |
|
124 | os.close(rfd) | |
114 | for i, item in func(*(staticargs + (pargs,))): |
|
125 | for i, item in func(*(staticargs + (pargs,))): | |
115 | os.write(wfd, '%d %s\n' % (i, item)) |
|
126 | os.write(wfd, '%d %s\n' % (i, item)) | |
116 | os._exit(0) |
|
127 | os._exit(0) | |
117 | except KeyboardInterrupt: |
|
128 | except KeyboardInterrupt: | |
118 | os._exit(255) |
|
129 | os._exit(255) | |
119 | # other exceptions are allowed to propagate, we rely |
|
130 | # other exceptions are allowed to propagate, we rely | |
120 | # on lock.py's pid checks to avoid release callbacks |
|
131 | # on lock.py's pid checks to avoid release callbacks | |
121 | pids.add(pid) |
|
132 | pids.add(pid) | |
122 | os.close(wfd) |
|
133 | os.close(wfd) | |
123 | fp = os.fdopen(rfd, 'rb', 0) |
|
134 | fp = os.fdopen(rfd, 'rb', 0) | |
124 | t = threading.Thread(target=waitforworkers) |
|
135 | t = threading.Thread(target=waitforworkers) | |
125 | t.start() |
|
136 | t.start() | |
126 | def cleanup(): |
|
137 | def cleanup(): | |
127 | signal.signal(signal.SIGINT, oldhandler) |
|
138 | signal.signal(signal.SIGINT, oldhandler) | |
128 | t.join() |
|
139 | t.join() | |
129 | status = problem[0] |
|
140 | status = problem[0] | |
130 | if status: |
|
141 | if status: | |
131 | if status < 0: |
|
142 | if status < 0: | |
132 | os.kill(os.getpid(), -status) |
|
143 | os.kill(os.getpid(), -status) | |
133 | sys.exit(status) |
|
144 | sys.exit(status) | |
134 | try: |
|
145 | try: | |
135 | for line in util.iterfile(fp): |
|
146 | for line in util.iterfile(fp): | |
136 | l = line.split(' ', 1) |
|
147 | l = line.split(' ', 1) | |
137 | yield int(l[0]), l[1][:-1] |
|
148 | yield int(l[0]), l[1][:-1] | |
138 | except: # re-raises |
|
149 | except: # re-raises | |
139 | killworkers() |
|
150 | killworkers() | |
140 | cleanup() |
|
151 | cleanup() | |
141 | raise |
|
152 | raise | |
142 | cleanup() |
|
153 | cleanup() | |
143 |
|
154 | |||
144 | def _posixexitstatus(code): |
|
155 | def _posixexitstatus(code): | |
145 | '''convert a posix exit status into the same form returned by |
|
156 | '''convert a posix exit status into the same form returned by | |
146 | os.spawnv |
|
157 | os.spawnv | |
147 |
|
158 | |||
148 | returns None if the process was stopped instead of exiting''' |
|
159 | returns None if the process was stopped instead of exiting''' | |
149 | if os.WIFEXITED(code): |
|
160 | if os.WIFEXITED(code): | |
150 | return os.WEXITSTATUS(code) |
|
161 | return os.WEXITSTATUS(code) | |
151 | elif os.WIFSIGNALED(code): |
|
162 | elif os.WIFSIGNALED(code): | |
152 | return -os.WTERMSIG(code) |
|
163 | return -os.WTERMSIG(code) | |
153 |
|
164 | |||
154 | if os.name != 'nt': |
|
165 | if os.name != 'nt': | |
155 | _platformworker = _posixworker |
|
166 | _platformworker = _posixworker | |
156 | _exitstatus = _posixexitstatus |
|
167 | _exitstatus = _posixexitstatus | |
157 |
|
168 | |||
158 | def partition(lst, nslices): |
|
169 | def partition(lst, nslices): | |
159 | '''partition a list into N slices of roughly equal size |
|
170 | '''partition a list into N slices of roughly equal size | |
160 |
|
171 | |||
161 | The current strategy takes every Nth element from the input. If |
|
172 | The current strategy takes every Nth element from the input. If | |
162 | we ever write workers that need to preserve grouping in input |
|
173 | we ever write workers that need to preserve grouping in input | |
163 | we should consider allowing callers to specify a partition strategy. |
|
174 | we should consider allowing callers to specify a partition strategy. | |
164 |
|
175 | |||
165 | mpm is not a fan of this partitioning strategy when files are involved. |
|
176 | mpm is not a fan of this partitioning strategy when files are involved. | |
166 | In his words: |
|
177 | In his words: | |
167 |
|
178 | |||
168 | Single-threaded Mercurial makes a point of creating and visiting |
|
179 | Single-threaded Mercurial makes a point of creating and visiting | |
169 | files in a fixed order (alphabetical). When creating files in order, |
|
180 | files in a fixed order (alphabetical). When creating files in order, | |
170 | a typical filesystem is likely to allocate them on nearby regions on |
|
181 | a typical filesystem is likely to allocate them on nearby regions on | |
171 | disk. Thus, when revisiting in the same order, locality is maximized |
|
182 | disk. Thus, when revisiting in the same order, locality is maximized | |
172 | and various forms of OS and disk-level caching and read-ahead get a |
|
183 | and various forms of OS and disk-level caching and read-ahead get a | |
173 | chance to work. |
|
184 | chance to work. | |
174 |
|
185 | |||
175 | This effect can be quite significant on spinning disks. I discovered it |
|
186 | This effect can be quite significant on spinning disks. I discovered it | |
176 | circa Mercurial v0.4 when revlogs were named by hashes of filenames. |
|
187 | circa Mercurial v0.4 when revlogs were named by hashes of filenames. | |
177 | Tarring a repo and copying it to another disk effectively randomized |
|
188 | Tarring a repo and copying it to another disk effectively randomized | |
178 | the revlog ordering on disk by sorting the revlogs by hash and suddenly |
|
189 | the revlog ordering on disk by sorting the revlogs by hash and suddenly | |
179 | performance of my kernel checkout benchmark dropped by ~10x because the |
|
190 | performance of my kernel checkout benchmark dropped by ~10x because the | |
180 | "working set" of sectors visited no longer fit in the drive's cache and |
|
191 | "working set" of sectors visited no longer fit in the drive's cache and | |
181 | the workload switched from streaming to random I/O. |
|
192 | the workload switched from streaming to random I/O. | |
182 |
|
193 | |||
183 | What we should really be doing is have workers read filenames from a |
|
194 | What we should really be doing is have workers read filenames from a | |
184 | ordered queue. This preserves locality and also keeps any worker from |
|
195 | ordered queue. This preserves locality and also keeps any worker from | |
185 | getting more than one file out of balance. |
|
196 | getting more than one file out of balance. | |
186 | ''' |
|
197 | ''' | |
187 | for i in range(nslices): |
|
198 | for i in range(nslices): | |
188 | yield lst[i::nslices] |
|
199 | yield lst[i::nslices] |
General Comments 0
You need to be logged in to leave comments.
Login now