-
Notifications
You must be signed in to change notification settings - Fork 0
/
new_leader.py
877 lines (786 loc) · 34.7 KB
/
new_leader.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
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
from datetime import datetime
import server
from server import encode_command, decode_command, encode_ping_ack, decode_ping_ack
import threading
import socket
import os
import time
import struct
import json
import numpy as np
from math import ceil
from queue import Queue, PriorityQueue
from keras.preprocessing.image import image, load_img, img_to_array
from tensorflow.keras.applications.resnet50 import ResNet50
from tensorflow.keras.applications.mobilenet import MobileNet
from tensorflow.keras.applications.imagenet_utils import preprocess_input, decode_predictions
BUFFER_SIZE = 4096
MASTER_HOST = INTRODUCER_HOST = socket.gethostbyname('fa22-cs425-3601.cs.illinois.edu')
STANDBY_HOST = socket.gethostbyname('fa22-cs425-3602.cs.illinois.edu')
MACHINE_NUM = int(socket.gethostname()[13:15])
LOG_FILEPATH = f'machine.{MACHINE_NUM}.log'
PING_PORT = 20240
MEMBERSHIP_PORT = 20241
PING_INTERVAL = 2.5
PING_TIMEOUT = 2
MASTER_PORT = 20086
FILE_PORT = 10086
GET_ADDR_PORT = 10087
def send_file(conn: socket.socket, localfilepath, sdfsfileid, timestamp):
header_dic = {
'sdfsfileid': sdfsfileid, # 1.txt
'timestamp': timestamp,
'file_size': os.path.getsize(localfilepath)
}
header_json = json.dumps(header_dic)
header_bytes = header_json.encode()
conn.send(struct.pack('i', len(header_bytes)))
conn.send(header_bytes)
with open(localfilepath, 'rb') as f:
for line in f:
conn.send(line)
def receive_file(conn: socket.socket):
obj = conn.recv(4)
header_size = struct.unpack('i', obj)[0]
header_bytes = conn.recv(header_size)
header_json = header_bytes.decode()
header_dic = json.loads(header_json)
total_size, sdfsfileid, timestamp = header_dic['file_size'], header_dic['sdfsfileid'], header_dic['timestamp']
data = b''
recv_size = 0
while recv_size < total_size:
line = conn.recv(BUFFER_SIZE)
data += line
recv_size += len(line)
return data, sdfsfileid, timestamp
class FServer(server.Node):
class FileTable:
def __init__(self):
self.file_lookup = {}
self.file_lookup_lock = threading.Lock()
# insert into arr, timestamp, in order
def _insert(self, arr, e):
l, r = 0, len(arr)
while l<r:
mid=(l+r)//2
if arr[mid] >= e:
r = mid
else:
l = mid + 1
arr.insert(l, e)
# insert filename-timestamp
def insert_file(self, file, sdfsfileid, timestamp):
self.file_lookup_lock.acquire()
self.file_lookup.setdefault(sdfsfileid, [])
self._insert(self.file_lookup[sdfsfileid], timestamp)
self.file_lookup_lock.release()
sdfsfilename = sdfsfileid + '-' + str(timestamp)
with open(sdfsfilename, 'wb') as f:
f.write(file)
# remove file from filesystem
def delete_file(self, sdfsfileid):
self.file_lookup_lock.acquire()
self.file_lookup.pop(sdfsfileid)
self.file_lookup_lock.release()
# delete file
# check if file is in filesystem
def check_file(self, sdfsfileid):
if sdfsfileid not in self.file_lookup:
return None
return self.file_lookup[sdfsfileid][-1]
# print files
def show_file(self):
print('files stored at this machine:')
self.file_lookup_lock.acquire()
for i in self.file_lookup.keys():
print(' ', i)
self.file_lookup_lock.release()
# return the first n versions
def get_n_versions(self, sdfsfileid, n):
self.file_lookup_lock.acquire()
n = min(n, len(self.file_lookup[sdfsfileid]))
timestamps = [i for i in self.file_lookup[sdfsfileid][-n:]]
self.file_lookup_lock.release()
data = []
for t in timestamps:
with open(sdfsfileid+'-'+str(t), 'rb') as f:
data.append(f.read())
return data
def ping_thread(self):
# generate ping_id
ping_id = self.id + '-' + str(self.ping_count)
self.ping_count += 1
# encode ping
encoded_ping = encode_ping_ack(ping_id, {'type': 'ping', 'member_id': self.id})
# initialize cache for the ping_id
self.ack_cache_lock.acquire()
self.ack_cache[ping_id] = set()
self.ack_cache_lock.release()
# transmit ping, get the ids of the member that's been pinged
for i in range(4):
ids = self.transmit_message(encoded_ping, self.ping_port)
# wait for some time to receive ack
time.sleep(self.ping_timeout)
# get the received ack
self.ack_cache_lock.acquire()
ack_cache_for_this_ping_id = self.ack_cache[ping_id]
self.ack_cache.pop(ping_id)
self.ack_cache_lock.release()
# check all the acks that's received
fail_ip = []
for id in ids:
if id not in ack_cache_for_this_ping_id: # if an ack is not received
fail_ip.append(id.split(':')[0]) # add to failed
new_membership_list = self.update_membership_list(0, id) # get updated membership_list by deleting the member that's missing
# assign unique command id
self.command_lock.acquire()
command_id = self.id + '-' + str(self.command_count)
self.command_count += 1
self.commands.add(command_id)
self.command_lock.release()
# encode command
command_content = {'type': 'failed', 'content': id}
encoded_command_tosend = encode_command(command_id, command_content)
self.mode_lock.acquire()
if self.debug:
self.mode_lock.release()
print("haven't receiving ack from ", id)
print('sending command ', command_content) # print statement for debugging
else:
self.mode_lock.release()
# transmit message, using old membership_list
self.transmit_message(encoded_command_tosend, self.membership_port)
# update membership list
self.membership_lock.acquire()
self.membership_list = new_membership_list
self.membership_lock.release()
self.log_generate(id, 'failed', self.membership_list)
if fail_ip: # inform master of failed node
with socket.socket(socket.AF_INET, socket.SOCK_DGRAM) as s:
s.sendto(json.dumps({'command_type': 'fail_notice', 'command_content': fail_ip}).encode(), (self.master_ip, self.master_port))
def __init__(self, ping_port: int, membership_port: int, ping_timeout: int, ping_interval: int, log_filepath: str, file_port: int, master_port: int, master_host: str):
super().__init__(ping_port, membership_port, ping_timeout, ping_interval, log_filepath)
self.file_port = file_port
self.file_cache = {}
self.file_lock = threading.Lock()
self.file_table = self.FileTable()
self.put_lock = threading.Lock()
self.get_lock = threading.Lock()
self.put_ack_cache = {}
self.get_ack_cache = {}
self.ls_cache = {}
self.ls_lock = threading.Lock()
self.master_port = master_port
self.master_ip = socket.gethostbyname(master_host)
# jobs priority queue([priority, batch])
self.job_queue = PriorityQueue()
# batch_queue [(model, [batch])]
self.batch_queue = Queue()
# running_batches map(node -> [batch])
self.running_batches = {}
self.models = {}
self.total_batches = 0
# initialize locks
self.seen_lock = threading.Lock()
self.leader_lock = threading.Lock()
self.members_lock = threading.Lock()
self.batches_lock = threading.Lock()
self.finished_lock = threading.Lock()
self.finished = False
self.seen_jobs = set()
self.conf = ""
# retrieves ip
def get_ip(self, sdfsfileid):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
try:
s.connect((self.master_ip, GET_ADDR_PORT))
except socket.error as e:
return
s.send(sdfsfileid.encode())
ips = json.loads(s.recv(4096).decode())
return ips
# utilities
def filehash(self, sdfsfilename):
self.membership_lock.acquire()
index = hash(sdfsfilename) % len(self.membership_list)
self.membership_lock.release()
return index
# get the appropriate pred and successors replicas are stored at
def getAllReplicas(self, index):
indexes = set()
self.membership_lock.acquire()
l = len(self.membership_list)
self.membership_lock.release()
for i in range(0, 4):
indexes.add((index + i) % l)
self.membership_lock.acquire()
res = [self.membership_list[i].split(':')[0] for i in indexes]
self.membership_lock.release()
return res
def fileServerBackground(self):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
s.bind((self.host, self.file_port))
s.listen()
while True:
conn, addr = s.accept()
t = threading.Thread(target=self.requestHandleThread, args=(conn, ))
t.start()
def requestHandleThread(self, conn: socket.socket):
global MASTER_HOST
command = conn.recv(BUFFER_SIZE).decode()
if command == 'put':
t = threading.Thread(target=self.handle_put_request, args=(conn,))
t.start()
elif command == 'delete':
t = threading.Thread(target=self.handle_delete_request, args=(conn,))
t.start()
elif command == 'get':
t = threading.Thread(target=self.handle_get_request, args=(conn,))
t.start()
elif command == 'ls':
t = threading.Thread(target=self.handle_ls_request, args=(conn,))
t.start()
elif command == 'repair':
t = threading.Thread(target=self.handle_repair_request, args=(conn,))
t.start()
elif command == 'replicate':
t = threading.Thread(target=self.handle_replicate_request, args=(conn,))
t.start()
elif command == 'multiget':
t = threading.Thread(target=self.handle_multiple_get_request, args=(conn,))
t.start()
# runs a batch by calling handle_execute
elif command == 'executeBatch':
t = threading.Thread(target=self.handle_execute, args=(conn,))
t.start()
# when finishedBatch is called, call handle_finished to update structures
elif command == 'finishedBatch':
t = threading.Thread(target=self.handle_finished, args=(conn,))
t.start()
# handle beginning of a job
elif command == 'beginningJob':
t = threading.Thread(target=self.handle_beginning, args=conn)
# assign the new leader
elif command == 'newLeader':
conn.send(b'1')
decoded_command = conn.recv(BUFFER_SIZE).decode()
MASTER_HOST = decoded_command[1]
print("new leader:", MASTER_HOST)
# handle the beginning of the inference
elif command == 'beginningInference':
t = threading.Thread(target=self.handle_beginning, args=(conn,))
t.start()
# when a job is finished, add it to the jobs that are done
elif command == 'finishedJob':
conn.send(b'1')
if not self.job_queue.empty:
self.seen_jobs.add(self.job_queue.get())
# check if the all the sent ips are in the replica set, if not, handle_replicate
def handle_repair_request(self, conn: socket.socket):
conn.send(b'1')
encoded_command = conn.recv(BUFFER_SIZE)
decoded_command = json.loads(encoded_command.decode())
sdfsfileid, ips = decoded_command['sdfsfileid'], decoded_command['ips']
if not self.file_table.check_file(sdfsfileid):
conn.send(b'0')
self.membership_lock.acquire()
index = self.membership_list.index(self.id)
for i in range(1, 4):
ip = self.membership_list[(i+index)%len(self.membership_list)].split(':')[0]
if ip not in ips:
self.handle_replicate(sdfsfileid, ip)
conn.send(b'1')
break
self.membership_lock.release()
conn.send(b'0')
return
# send a put command to the master for putting the file into the replica
def handle_replicate(self, sdfsfileid, ip):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
try:
s.connect((ip, self.file_port))
except socket.error as e:
return
s.send(b'replicate')
s.recv(1)
timestamp = self.file_table.check_file(sdfsfileid)
localfilepath = sdfsfileid + '-' + str(timestamp)
send_file(s, localfilepath, sdfsfileid, timestamp)
s.recv(1)
with socket.socket(socket.AF_INET, socket.SOCK_DGRAM) as s:
s.sendto(json.dumps({'command_type': 'put_notice', 'command_content': [sdfsfileid, ip]}).encode(), (self.master_ip, self.master_port))
def handle_replicate_request(self, conn):
conn.send(b'1')
data, sdfsfileid, timestamp = receive_file(conn)
self.file_table.insert_file(data, sdfsfileid, timestamp)
conn.send(b'1')
def handle_put(self, localfilepath, sdfsfileid, ip, timestamp):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
try:
s.connect((ip, self.file_port))
except socket.error as e:
return
s.send(b'put')
s.recv(1) # for ack
send_file(s, localfilepath, sdfsfileid, timestamp)
s.recv(1) # for ack
command_id = sdfsfileid + '-' + str(timestamp)
self.put_lock.acquire()
self.put_ack_cache.setdefault(command_id, 0)
self.put_ack_cache[command_id] += 1
self.put_lock.release()
with socket.socket(socket.AF_INET, socket.SOCK_DGRAM) as s:
s.sendto(json.dumps({'command_type': 'put_notice', 'command_content': [sdfsfileid, ip]}).encode(), (self.master_ip, self.master_port))
# put the file data into filesystem with entry sdfsfileid
def handle_put_request(self, conn: socket.socket):
conn.send(b'1')
data, sdfsfileid, timestamp = receive_file(conn)
self.file_table.insert_file(data, sdfsfileid, timestamp)
conn.send(b'1')
# gets the file from the filesystem if timestamp is correct
def handle_get(self, sdfsfileid, ip):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
try:
s.connect((ip, self.file_port))
except socket.error as e:
return
s.send(b'get')
s.recv(1) # for ack
s.send(sdfsfileid.encode())
data, sdfsfileid, timestamp = receive_file(s)
self.file_lock.acquire()
self.file_cache.setdefault(sdfsfileid, [None, 0])
if timestamp > self.file_cache[sdfsfileid][1]:
self.file_cache[sdfsfileid] = [data, timestamp]
self.file_lock.release()
self.get_lock.acquire()
self.get_ack_cache.setdefault(sdfsfileid, 0)
self.get_ack_cache[sdfsfileid] += 1
self.get_lock.release()
# Handles get request
def handle_get_request(self, conn: socket.socket):
conn.send(b'1')
sdfsfileid = conn.recv(BUFFER_SIZE).decode()
timestamp = self.file_table.check_file(sdfsfileid)
sdfsfilename = sdfsfileid + '-' + str(timestamp)
send_file(conn, sdfsfilename, sdfsfileid, timestamp)
return
# handles send
def handle_send(self, command, ip):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
try:
s.connect((ip, self.file_port))
except socket.error as e:
return
s.send(command[0].encode())
s.recv(1) # for ack
s.send(json.dumps(command).encode())
# handles delete
def handle_delete(self, sdfsfileid, ip):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
try:
s.connect((ip, self.file_port))
except socket.error as e:
return
s.send(b'delete')
s.recv(1) # for ack
s.send(sdfsfileid.encode())
# handles delete
def handle_delete_request(self, conn: socket.socket):
conn.send(b'1')
sdfsfileid = conn.recv(BUFFER_SIZE).decode()
self.file_table.delete_file(sdfsfileid)
with socket.socket(socket.AF_INET, socket.SOCK_DGRAM) as s:
s.sendto(json.dumps({'command_type': 'delete_notice', 'command_content': [sdfsfileid, socket.gethostbyname(socket.gethostname())]}).encode(),
(self.master_ip, self.master_port))
return
# handles ls
def handle_ls(self, sdfsfileid, ip):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
try:
s.connect((ip, self.file_port))
except socket.error as e:
return
s.send(b'ls')
s.recv(1) # for ack
s.send(sdfsfileid.encode())
res = s.recv(1).decode()
if res == '1':
self.ls_lock.acquire()
self.ls_cache.setdefault(sdfsfileid, [])
self.ls_cache[sdfsfileid].append(id)
self.ls_lock.release()
# handles ls request
def handle_ls_request(self, conn: socket.socket):
conn.send(b'1')
sdfsfileid = conn.recv(BUFFER_SIZE).decode()
exist = self.file_table.check_file(sdfsfileid)
if exist:
conn.send(b'1')
else:
conn.send(b'0')
def handle_multiple_get(self, sdfsfileid, ip, n):
with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
try:
s.connect((ip, self.file_port))
except socket.error as e:
return
s.send(b'multiget')
s.recv(1) # for ack
s.send(json.dumps({'sdfsfileid': sdfsfileid, 'n': n}).encode())
obj = s.recv(4)
header_size = struct.unpack('i', obj)[0]
header_bytes = s.recv(header_size)
header_json = header_bytes.decode()
header_dic = json.loads(header_json)
total_size, latest_t = header_dic['file_size'], header_dic['latest_t']
data = b''
recv_size = 0
while recv_size < total_size:
frag = s.recv(BUFFER_SIZE)
data += frag
recv_size += len(frag)
k = sdfsfileid + '-' + str(n)
self.file_lock.acquire()
self.file_cache.setdefault(k, [None, 0])
if self.file_cache[k][1] < latest_t:
self.file_cache[k][0] = data
self.file_lock.release()
self.get_lock.acquire()
self.get_ack_cache.setdefault(k, 0)
self.get_ack_cache[k] += 1
self.get_lock.release()
# handles multiple get
def handle_multiple_get_request(self, conn: socket.socket):
conn.send(b'1')
decoded = json.loads(conn.recv(BUFFER_SIZE).decode())
sdfsfileid, n = decoded['sdfsfileid'], decoded['n']
data = self.file_table.get_n_versions(sdfsfileid, n)
delimiter = b'\n\n-----------------------\n\n'
data = delimiter.join(data)
header_dic = {
'file_size': len(data),
'latest_t': self.file_table.check_file(sdfsfileid)
}
header_json = json.dumps(header_dic)
header_bytes = header_json.encode()
conn.send(struct.pack('i', len(header_bytes)))
conn.send(header_bytes)
conn.send(data)
# executes the batch, called when "executeBranch" is called
def handle_execute(self, conn: socket.socket):
conn.send(b'1')
command, input_file, model_name, indices = json.loads(conn.recv(BUFFER_SIZE).decode())
start, end = indices
# update seen_jobs with input_file
with self.seen_lock:
if input_file not in self.seen_jobs:
self.seen_jobs.add(input_file)
with open(input_file, "r") as i_f:
mode = "a"
if model_name not in self.models:
# initialize models
if model_name == 'resnet':
self.models[model_name] = ResNet50(weights='imagenet')
elif model_name == 'mobilenet':
self.models[model_name] = MobileNet(weights='imagenet')
model = self.models[model_name]
predictions = []
with open(f"{self.host}-{input_file}", mode) as o_f:
for _ in range(start-1):
i_f.readline()
start_time = datetime.now()
for _ in range(start, end, 1):
line = i_f.readline().strip().split(",")
dim_ct = int(float(line[0])//1)
dims = []
for i in range(dim_ct):
dims.append(int(float(line[i+1])//1))
data = np.array(line[dim_ct+1:]).astype(np.float64)
data = np.reshape(data, tuple(dims))
x = preprocess_input(np.array([data]))
preds = model.predict(x)
predictions.append(decode_predictions(preds, top=1)[0][0][1])
# when the batch has finished executing, send out a finished batch command
cmd = ["finishedBatch", self.host, predictions, int((1000 * (datetime.now() - start_time).total_seconds())), end - start]
self.handle_send(cmd, MASTER_HOST)
print("finished with local batch!")
# when a node is done working on its batch, remove it from running_batches and call reassign
def handle_finished(self, conn: socket.socket):
conn.send(b'1')
_, finished_ip, predictions, time, query_size = json.loads(conn.recv(BUFFER_SIZE).decode())
with self.batches_lock:
self.total_batches += 1
c = "\n"
print(f"output from {finished_ip}:{predictions}\naverage ms per query: {time/query_size}. total batches processed = {self.total_batches}")
if finished_ip in self.running_batches:
self.running_batches.pop(finished_ip)
self.reassign()
# handle node failure and reassign jobs if a node fails; assign batches to nodes
def reassign(self):
with self.members_lock:
if len(self.membership_list) == 0:
print("no workers found! exiting")
exit(1)
# handle failure
failed = set()
with self.batches_lock:
working_nodes = set(self.running_batches.keys())
alive_nodes = set(map(lambda x: x.split(":")[0], self.membership_list))
failed = working_nodes - alive_nodes
for i in failed:
self.batch_queue.put(self.running_batches[i])
self.running_batches.pop(i)
# assign batches to nodes
for m in self.membership_list:
if m not in self.running_batches:
host = m.split(":")[0]
# don't assign a batch to a node if that node is the master or hotstandby
if host == self.master_ip or host == STANDBY_HOST:
continue
if not self.batch_queue.empty():
input_file, model_name, indices = self.batch_queue.get()
self.running_batches[host] = (input_file, model_name, indices)
self.handle_send(["executeBatch", input_file, model_name, indices], host)
else:
with self.finished_lock:
finished = True
# handles the inference, gets each value passed into the input, adds values to batch_queue to be processed, reassigns the batches
def handle_inference(self, config):
print("in inference!")
with open(config, "r") as f:
jobs = json.loads(f.read())["data"]
if self.host != STANDBY_HOST:
self.handle_send(["beginningInference", config], STANDBY_HOST)
# handle input requirements
for job in jobs:
name = job["name"]
enabled = job["enabled"]
priority = job.get("priority", 1000)
batch_size = job.get("batch-size", 1)
num_queries = job.get("num-queries", 1)
model_name = job.get("model", None)
input_source = job.get("input-source", None)
if not enabled or not input_source or not model_name or num_queries == 0 or batch_size == 0:
continue
self.job_queue.put((priority, name, job))
print("put em all!")
while not self.job_queue.empty():
with self.finished_lock:
self.finished = False
priority, name, job = self.job_queue.get()
print(name)
if name in self.seen_jobs:
if self.job_queue.empty():
print("finished!")
exit(1)
priority, name, job = self.job_queue.get()
print(f"running {job}")
batch_size = job.get("batch-size", 1)
num_queries = job.get("num-queries", 1)
input_source = job.get("input-source", None)
model_name = job.get("model", None)
num_batches = ceil(num_queries/batch_size)
# update batch_queue
with self.batches_lock:
for start in range(num_batches):
idx = (start * batch_size, min((start + 1) * batch_size, num_queries))
self.batch_queue.put((input_source, model_name, idx))
# reassign batches to nodes
self.reassign()
finished = False
while not finished:
time.sleep(1)
with self.finished_lock:
finished = self.finished
print("finished!")
# gets the conf command
def handle_beginning(self, conn:socket.socket):
conn.send(b'1')
decoded_command = conn.recv(BUFFER_SIZE).decode()
self.conf = decoded_command[1]
# put sdfsfileid localfilepath
def put(self, localfilepath, sdfsfileid):
ips = self.get_ip(sdfsfileid)
if not ips:
index = self.filehash(sdfsfileid)
ips = self.getAllReplicas(index)
print(ips)
timestamp = time.time()
for ip in ips:
t = threading.Thread(target=self.handle_put, args = (localfilepath, sdfsfileid, ip, timestamp))
t.start()
i = 0
command_id = sdfsfileid + '-' + str(timestamp)
while i < 100:
self.put_lock.acquire()
self.put_ack_cache.setdefault(command_id, 0)
cnt = self.put_ack_cache[command_id]
self.put_lock.release()
if cnt >= min(3, len(ips)):
break
time.sleep(2)
i += 1
print("done")
return
def get(self, localfilepath, sdfsfileid):
ips = self.get_ip(sdfsfileid)
print(len(ips))
for ip in ips:
t = threading.Thread(target=self.handle_get, args=(sdfsfileid, ip))
t.start()
i = 0
while i < 10:
self.get_lock.acquire()
self.get_ack_cache.setdefault(sdfsfileid, 0)
cnt = self.get_ack_cache[sdfsfileid]
self.get_lock.release()
if cnt >= min(3, len(ips)):
break
time.sleep(2)
i += 1
if i == 10:
print('get failed.')
else:
self.file_lock.acquire()
data = self.file_cache[sdfsfileid][0]
self.file_lock.release()
with open(localfilepath, 'wb') as f:
f.write(data)
print('get complete.')
# mulitcasts out the leader
def multicast_leader(self):
for member in self.membership_list:
host = member.split(":")[0]
t = threading.Thread(target=self.handle_send, args=(["newLeader", self.host],host))
t.start()
# checks if the leader has failed by checking if the master exists in the membership list
def check_leader(self):
global MASTER_HOST
while True:
with self.members_lock:
if MASTER_HOST not in set(map(lambda x: x.split(":")[0], self.membership_list)):
print("MASTER FAILED!")
threading.Thread(target=self.multicast_leader).start()
MASTER_HOST = self.host
t = threading.Thread(target=self.handle_inference, args=(self.conf,))
t.start()
break
time.sleep(1)
return
def run(self):
self.join()
t1 = threading.Thread(target=self.fileServerBackground)
t1.start()
if self.host == STANDBY_HOST:
t2 = threading.Thread(target=self.check_leader)
t2.start()
while True:
command = input('>')
parsed_command = command.split()
start_time = time.time()
if parsed_command[0] == 'put':
localfilepath, sdfsfileid = parsed_command[1], parsed_command[2]
self.put(localfilepath, sdfsfileid)
elif parsed_command[0] == 'get':
sdfsfileid, localfilepath = parsed_command[1], parsed_command[2]
self.get(localfilepath, sdfsfileid)
elif parsed_command[0] == 'delete':
sdfsfileid = parsed_command[1]
ips = self.get_ip(sdfsfileid)
for ip in ips:
t = threading.Thread(target=self.handle_delete, args=(sdfsfileid, ip))
t.start()
elif parsed_command[0] == 'store':
self.file_table.show_file()
elif parsed_command[0] == 'ls':
sdfsfileid = parsed_command[1]
ips = self.get_ip(sdfsfileid)
print('the file exists in:')
for ip in ips:
print(' ', ip)
elif parsed_command[0] == 'get_versions':
sdfsfileid, num_versions, localfilepath = parsed_command[1], int(parsed_command[2]), parsed_command[3]
ips = self.get_ip(sdfsfileid)
for ip in ips:
self.handle_multiple_get(sdfsfileid, ip, num_versions)
i = 0
while i < 10:
k = sdfsfileid + '-' + str(num_versions)
self.get_lock.acquire()
self.get_ack_cache.setdefault(k, 0)
cnt = self.get_ack_cache[k]
self.get_lock.release()
if cnt >= 3:
break
time.sleep(2)
i += 1
if i == 10:
print('get failed.')
else:
self.file_lock.acquire()
data = self.file_cache[k][0]
self.file_lock.release()
with open(localfilepath, 'wb') as f:
f.write(data)
print('get complete.')
# command to run the inference
elif parsed_command[0] == 'inference':
config = parsed_command[1]
t = threading.Thread(target=self.handle_inference, args=(config,))
t.start()
elif command == 'leave':
# create command id
self.command_lock.acquire()
command_id = self.id + '-' + str(self.command_count)
self.command_count += 1
self.commands.add(command_id)
self.command_lock.release()
# encode command
command_content = {'type': 'leave', 'content': self.id}
encoded_command_tosend = encode_command(command_id, command_content)
self.transmit_message(encoded_command_tosend, self.membership_port)
self.membership_lock.acquire()
self.membership_list = []
self.membership_lock.release()
self.log_generate(command_id[:-2], 'leave', self.membership_list)
print('Leaving...')
break
elif command == 'list_mem':
print('isIntroducer: ', self.isIntroducer)
self.membership_lock.acquire()
print(f'there are {len(self.membership_list)} member in membership_list: ')
for member in self.membership_list:
print(' ', member)
self.membership_lock.release()
self.command_lock.acquire()
print(f'{len(self.commands)} commands have been executed')
self.command_lock.release()
elif command == 'debug':
if self.debug:
print('debug mode off')
else:
print('debug mode on')
self.log_generate(self.id, command, self.membership_list)
self.debug = not self.debug
elif command == 'list_self':
print(self.id)
elif command == 'bandwidth':
self.bytes_lock.acquire()
b = self.bytes
self.bytes_lock.release()
print(b / (time.time()-self.start_time))
elif command == 'reset_time':
self.bytes_lock.acquire()
self.bytes = 0
self.start_time = time.time()
self.bytes_lock.release()
else:
print('command not found!')
end_time = time.time()
print('time consumed: ', end_time - start_time)
if __name__ == '__main__':
# def __init__(self, ping_port: int, membership_port: int, ping_timeout: int, ping_interval: int, log_filepath: str,
# file_port: int):
server = FServer(PING_PORT, MEMBERSHIP_PORT, PING_TIMEOUT, PING_INTERVAL, LOG_FILEPATH, FILE_PORT, MASTER_PORT, MASTER_HOST)
server.run()