/
__init__.py
304 lines (269 loc) · 10.2 KB
/
__init__.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
#!/usr/bin/env python
# -*- coding: UTF-8 -*-
import multiprocessing, sys, cPickle, os, datetime
from multiprocessing.queues import SimpleQueue
from tempfile import NamedTemporaryFile, TemporaryFile
import heapq, itertools
class Job(object):
"""
Base class of the Job.
job.enumerate() should return an interator over ITEMS
for each INPUT
job.map(item, cb) is called
cb should be called for each key-value pair
all key values pairs are sorted, and for each key
job.reduce_key_start(key) is called
job.reduce_value(value) is called
job.reduce_key_stop(key) is called
"""
def map(self, item, cb):
cb (item)
def reduce_start(self):
pass
def reduce_key_start(self, key):
pass
def reduce_key_stop(self, key):
pass
def reduce_value(self, r):
pass
def reduce_stop(self):
pass
class WC(Job):
"Sample Word count parallel implementation"
lc = 0
wc = 0
bc = 0
def __init__(self, f):
self.file = f
def reduce_start(self):
self.lc = 0
self.wc = 0
self.bc = 0
def enumerate(self):
return enumerate(open(self.file))
def map(self, item, cb):
(pos, line) = item
cb((pos, (1, len(line.split()), len(line))))
def reduce_value(self, r):
(lc, wc, bc) = r
self.lc = self.lc + lc
self.wc = self.wc + wc
self.bc = self.bc + bc
def reduce_stop(self):
return (self.lc, self.wc, self.bc)
def debug_print(s):
print >> sys.stderr, "[%s] (pid %u) %s" % (datetime.datetime.now().strftime('%H:%M:%S'), os.getpid(), s)
class BaseRunner(object):
STOP_MSG = "##STOP_MSG##"
def __init__(self):
self.debug = False
pass
def reduce_loop(self, item_iterator):
job = self.job
job.reduce_start()
pkey = None
for (key, val) in item_iterator:
if pkey == None or pkey != key:
if not (pkey is None):
job.reduce_key_stop(pkey)
job.reduce_key_start(key)
pkey = key
job.reduce_value(val)
if not (pkey is None):
job.reduce_key_stop(pkey)
return job.reduce_stop()
class SingleThreadRunner(BaseRunner):
"""
Runner that executes a job in a single thread on a single process
"""
def __init__(self):
pass
def run(self, job):
self.job = job
buf = []
for elt in job.enumerate():
job.map(elt, buf.append)
buf.sort()
return self.reduce_loop(buf)
class BaseMultiprocessingRunner(BaseRunner):
def __init__(self):
super(BaseMultiprocessingRunner, self).__init__()
self.numprocs = max(multiprocessing.cpu_count() - 1, 1)
self.map_input_queue = SimpleQueue()
self.map_output_queue = SimpleQueue()
def run_map(self):
for item in iter(self.map_input_queue.get, self.STOP_MSG):
self.job.map(item, self.map_output_queue.put)
self.map_output_queue.put(self.STOP_MSG)
if self.debug:
debug_print("Output : STOP sent")
def run_enumerate(self):
for inp in self.job.enumerate():
self.map_input_queue.put(inp)
for work in range(self.numprocs):
self.map_input_queue.put(self.STOP_MSG)
if self.debug:
debug_print("Input: STOP sent")
def run(self, job):
self.job = job
# Process that reads the input file
self.enumeration_process = multiprocessing.Process(target=self.run_enumerate, args=())
self.mappers = [ multiprocessing.Process(target=self.run_map, args=())
for i in range(self.numprocs)]
self.enumeration_process.start()
for mapper in self.mappers:
mapper.start()
r = self.run_reduce()
self.enumeration_process.join()
for mapper in self.mappers:
mapper.join()
return r
class DiskBasedRunner(BaseMultiprocessingRunner):
def __init__(self, map_buffer_size = 10000, reduce_max_files = 10 ):
super(DiskBasedRunner, self).__init__()
self.item_buffer = {}
self.map_buffer_size = map_buffer_size
self.reduce_max_files = reduce_max_files
self.map_opened_files = []
def run_map(self):
self.item_buffer = []
for item in iter(self.map_input_queue.get, self.STOP_MSG):
self.job.map(item, self.item_buffer.append)
if len(self.item_buffer) > self.map_buffer_size:
self.map_buffer_clear()
self.map_buffer_clear()
self.map_output_queue.put(self.STOP_MSG)
if self.debug:
debug_print("Map done")
def map_buffer_clear(self):
self.item_buffer.sort()
f = NamedTemporaryFile() # We keep the file opened as it would close automatically
if self.debug:
debug_print('Temp file %s' % f.name)
for item in self.item_buffer:
cPickle.dump(item, f, cPickle.HIGHEST_PROTOCOL)
f.flush()
self.map_opened_files.append(f)
self.map_output_queue.put(f.name)
del self.item_buffer[:]
def get_next_file(self):
while self.stopped_received < self.numprocs:
filename = self.map_output_queue.get()
if filename == self.STOP_MSG:
self.stopped_received = self.stopped_received + 1
if self.debug:
debug_print("Reduced received complete output from %u mappers" % self.stopped_received)
continue
else:
if self.debug:
debug_print('Reading %s' % filename)
yield open(filename, 'r')
if self.debug:
debug_print('All files from mappers received')
def iter_on_file(self, stream):
try:
while True:
yield cPickle.load(stream)
except EOFError:
stream.close()
if hasattr(stream, "name"):
os.remove(stream.name)
def run_reduce(self):
self.stopped_received = 0
self.merged_files = []
merged_iterator = None
while True:
# Iterate and merge files until all jobs are processed
get_next = self.get_next_file()
files = get_next
#itertools.islice(get_next, self.reduce_max_files)
all_files = [file for file in files]
iterables = [self.iter_on_file(file) for file in all_files]
merged_iterator = heapq.merge(*iterables)
if self.stopped_received < self.numprocs:
if self.debug:
debug_print("Performing intermediate merge on %u files" % len(iterables))
f = TemporaryFile()
self.merged_files.append(f)
for m in merged_iterator:
cPickle.dump(m, f, cPickle.HIGHEST_PROTOCOL)
f.seek(0)
f.flush()
else:
break
if len(self.merged_files) > 0:
if self.debug:
debug_print("Final merge")
# Final merge if required
merged_iterator = heapq.merge(*([self.iter_on_file(stream) for stream in self.merged_files]+[merged_iterator]))
if self.debug:
debug_print("Reduce loop")
result = self.reduce_loop(merged_iterator)
return result
class SedLikeJobRunner(BaseMultiprocessingRunner):
"""
Runner optimzed for jobs that outputs key values of the form (i, value) where i are consecutive integer
starting at '0'
"""
def __init__(self):
super(SedLikeJobRunner, self).__init__()
def run_reduce(self):
cur = 0
buffer = {}
self.job.reduce_start()
for mappers in range(self.numprocs):
for msg in iter(self.map_output_queue.get, self.STOP_MSG):
(i, val) = msg
# verify rows are in order, if not save in buffer
if i != cur:
buffer[i] = val
else:
self.job.reduce_key_start(i)
self.job.reduce_value(val)
self.job.reduce_key_stop(i)
cur += 1
while cur in buffer:
self.job.reduce_key_start(cur)
self.job.reduce_value(buffer[cur])
self.job.reduce_key_stop(cur)
del buffer[cur]
cur += 1
if self.debug:
debug_print("Mapper done %u" % mappers)
return self.job.reduce_stop()
class WCLikeJobRunner(BaseMultiprocessingRunner):
"""
Runner optimized for jobs that outputs always the same key, and perform only a global reduce over all values
"""
def __init__(self):
super(WCLikeJobRunner, self).__init__()
def run_reduce(self):
self.job.reduce_start()
for mappers in range(self.numprocs):
for msg in iter(self.map_output_queue.get, self.STOP_MSG):
(key, val) = msg
self.job.reduce_value(val)
return self.job.reduce_stop()
class RambasedRunner(BaseMultiprocessingRunner):
def __init__(self):
super(RambasedRunner, self).__init__()
def run_reduce(self):
self.job.reduce_start()
buf = []
for mappers in range(self.numprocs):
for msg in iter(self.map_output_queue.get, self.STOP_MSG):
buf.append(msg)
buf.sort()
return self.reduce_loop(buf)
if __name__ == "__main__":
runners = []
runners.append(SingleThreadRunner())
runners.append(RambasedRunner())
runners.append(WCLikeJobRunner())
runners.append(SedLikeJobRunner())
runners.append(DiskBasedRunner())
for runner in runners:
runner.debug = True
for argv in sys.argv[1:]:
(lc, wc, bc) = runner.run(WC(argv))
print "(%s)\t%u\t%u\t%u\t%s" % (runner.__class__.__name__, lc, wc, bc, argv)