-
Notifications
You must be signed in to change notification settings - Fork 20
/
rebalance_rackaware.py
1289 lines (1096 loc) · 52 KB
/
rebalance_rackaware.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
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
"""
Rebalance Kafka partition replicas to achieve HA (Fault Domain/Update Domain awareness). The tool distributes
replicas of partitions of a topic across brokers in a manner such that each replica is in a separate fault domain and
update domain. The tool also distributes the leaders such that each broker has approximately the same number of
leaders for partitions. Rebalance can be executed for one or more topics.
PRE-REQS:
=========
sudo apt-get install -y libffi-dev libssl-dev
sudo pip install --upgrade requests[security] PyOpenSSL ndg-httpsclient pyasn1 kazoo retry pexpect
RUNNING THE SCRIPT:
===================
1) Copy the script to /usr/hdp/current/kafka-broker/bin on your cluster
2) Run this script with sudo privilege due to permission issues on some python packages: sudo python rebalance_new.py
"""
import argparse
import errno
import json
import logging
import os
import os.path
import pexpect
import random
import re
import requests
import socket
import subprocess
import sys
import tempfile
import traceback
from logging import StreamHandler
from logging.handlers import RotatingFileHandler
from logging.handlers import SysLogHandler
from operator import itemgetter
from hdinsight_common.AmbariHelper import AmbariHelper
from kazoo.client import KazooClient
from kazoo.client import KazooState
from retry import retry
# LOGGING
logger = logging.getLogger(__name__)
log_file = "rebalance_log"
log_dir = "/var/log/kafka/"
SIMPLE_FORMATTER = logging.Formatter('%(asctime)s - %(filename)s [%(process)d] %(name)s - %(levelname)s - %(message)s')
SYSLOG_FORMAT_STRING = ' %(filename)s [%(process)d] - %(name)s - %(levelname)s - %(message)s'
SYSLOG_FORMATTER = logging.Formatter(SYSLOG_FORMAT_STRING)
MIN_LOG_LEVEL = logging.DEBUG
MAX_RETRIES = 3
RETRY_INTERVAL_DELAY = 1
RETRY_INTERVAL_BACKOFF = 2
'''Filters (lets through) all messages with level < LEVEL'''
class LogFilter(logging.Filter):
def __init__(self, level):
self.level = level
def filter(self, log_record):
return log_record.levelno < self.level
# LOG_LOCAL0 - belongs to hdinsight-agent
# LOG_LOCAL1 - belongs to ambari-agent
# LOG_LOCAL2 - belongs to syslog catch all
# We want to log to LOG_LOCAL2
def initialize_logger(logger, log_file, syslog_facility=SysLogHandler.LOG_LOCAL2):
logger.setLevel(MIN_LOG_LEVEL)
if not len(logger.handlers):
add_console_handler(logger)
add_file_handler(logger, log_file)
add_syslog_handler(logger, syslog_facility)
'''Given a logger, we attach a console handler that will log only error messages'''
def add_console_handler(logger):
stdout_handler = StreamHandler(sys.stdout)
stderr_handler = StreamHandler(sys.stderr)
log_filter = LogFilter(logging.WARNING)
stdout_handler.addFilter(log_filter)
stdout_handler.setLevel(logging.INFO)
stdout_handler.setFormatter(SIMPLE_FORMATTER)
stderr_handler.setLevel(max(MIN_LOG_LEVEL, logging.WARNING))
stderr_handler.setFormatter(SIMPLE_FORMATTER)
logger.addHandler(stdout_handler)
logger.addHandler(stderr_handler)
'''Given a logger, we attach a rotating file handler that will log to the specified output file'''
def add_file_handler(logger, log_file_name):
if not log_file_name.endswith('.log'):
log_file_name = log_file_name + '.log'
try:
os.makedirs(log_dir)
except OSError as exc:
if exc.errno == errno.EEXIST and os.path.isdir(log_dir):
pass
else:
raise_error('Unable to create log dir: {0}'.format(log_dir))
log_file_path = os.path.join(log_dir, log_file_name)
file_handler = RotatingFileHandler(filename=log_file_path, maxBytes=104857600, backupCount=100)
file_handler.setLevel(logging.DEBUG)
file_handler.setFormatter(SIMPLE_FORMATTER)
logger.addHandler(file_handler)
# add syslog handler if we are on linux. We need to add retry on this because /dev/log might not be created by
# rsyslog yet
@retry(exceptions=BaseException, tries=MAX_RETRIES, delay=RETRY_INTERVAL_DELAY, backoff=RETRY_INTERVAL_BACKOFF)
def add_syslog_handler(logger, syslog_facility):
try:
mds_syslog_handler = SysLogHandler(address="/dev/log",
facility=syslog_facility)
mds_syslog_handler.setLevel(MIN_LOG_LEVEL)
mds_syslog_handler.setFormatter(SYSLOG_FORMATTER)
logger.addHandler(mds_syslog_handler)
return True
except Exception:
raise_error('Exception occurred when adding syslog handler: ' + traceback.format_exc())
return False
# Constants
KAFKA_BROKER = "kafka-broker"
ASSIGNMENT_JSON_FILE = "kafkaRebalancePlan.json"
ZOOKEEPER_PORT = ":2181"
BROKER_PORT = ":9092"
ZOOKEEPER_HOSTS = None
BROKERS_ID_PATH = "brokers/ids"
KAFKA_HOME = "/usr/hdp/current/kafka-broker/"
KAFKA_LIBS_PATH = KAFKA_HOME + "libs/"
KAFKA_TOPICS_TOOL_PATH = KAFKA_HOME + "bin/kafka-topics.sh"
KAFKA_REASSIGN_PARTITIONS_TOOL_PATH = KAFKA_HOME + "bin/kafka-reassign-partitions.sh"
FQDN = "fqdn"
BROKER_ID = "brokerId"
FAULT_DOMAIN = "faultDomain"
UPDATE_DOMAIN = "updateDomain"
FAULT_DOMAIN_SHORT = "FD"
UPDATE_DOMAIN_SHORT = "UD"
RACK = "rack"
VM_ID = "vmId"
PARTITION = "partition"
REPLICAS = "replicas"
LEADER = "leader"
LEADERS = "leaders"
FOLLOWERS = "followers"
TOPICS = "topics"
ISR = "isr"
FREE_DISK_SPACE = "fds"
PARTITION_SIZE = "partition_size"
ASSIGNED = "assigned"
DEFAULT_REBALANCE_THROTTLE_RATE_BPS = "50000000"
ALL_TOPICS_STRING = "all"
'''
Get information of Zookeeper Hosts
'''
def get_zookeeper_connect_string():
ah = AmbariHelper()
hosts = ah.get_host_components()
zkHosts = ""
for item in hosts["items"]:
if item["HostRoles"]["component_name"] == "ZOOKEEPER_SERVER":
zkHosts += item["HostRoles"]["host_name"]
zkHosts += ZOOKEEPER_PORT
zkHosts += ","
if len(zkHosts) > 2:
return zkHosts[:-1]
else:
raise_error("Failed to get Zookeeper information from Ambari!")
'''
Get information of Broker Hosts
'''
def get_broker_connect_string():
ah = AmbariHelper()
hosts = ah.get_host_components()
brokerHosts = ""
for item in hosts["items"]:
if item["HostRoles"]["component_name"] == "KAFKA_BROKER":
brokerHosts += item["HostRoles"]["host_name"]
brokerHosts += BROKER_PORT
brokerHosts += ","
if len(brokerHosts) > 2:
return brokerHosts[:-1]
else:
raise_error("Failed to get Broker information from Ambari!")
'''
Returns a list of all topics in Kafka by executing the Kafka-topics tool
'''
def get_topic_list():
try:
kafka_version, hdp_version = get_kafka_hdp_version()
if kafka_version >= '3.2.0':
s = subprocess.check_output([
KAFKA_TOPICS_TOOL_PATH,
"--bootstrap-server",
get_broker_connect_string(),
"--list"
])
else:
s = subprocess.check_output([
KAFKA_TOPICS_TOOL_PATH,
"--zookeeper",
get_zookeeper_connect_string(),
"--list"
])
if len(s) > 0:
logger.info("List of all topics: %s", s.split())
return s.split()
else:
return []
except Exception:
raise_error('Exception occurred when calling Kafka topics tool: ' + traceback.format_exc())
sys.exit()
'''
Uses AmbariHelper from hdinsight-common to get the cluster manifest and parses it to get the cluster topology JSON object.
'''
def get_cluster_topology_json():
ambariHelper = AmbariHelper()
cluster_manifest = ambariHelper.get_cluster_manifest()
settings = cluster_manifest.settings
if "cluster_topology_json_url" in settings:
json_url = settings["cluster_topology_json_url"]
logger.info("Retrieved Cluster Topology JSON document.")
logger.info("URL: %s", json_url)
r = requests.get(json_url)
topology_info = r.text
logger.debug("Cluster Topology: %s", topology_info)
return topology_info
else:
raise_error("Failed to get cluster_topology_json_url from cluster manifest.")
'''
Parses the cluster topology JSON doc and returns Host information
Returns: A list of dictionaries.
host_info = [
{
"vmId": 1
"FD": 0
"UD": 0
"fqdn": 'wn0-k09v3'
"brokerId": 1024,
"rack": FD0UD0,
"free_disk_space": 0
},
]
'''
def parse_topo_info(cluster_topology_json, brokers_info, dead_hosts=None):
logger.info("Parsing topology info to retrieve information about hosts.")
workernode_info = json.loads(cluster_topology_json)["hostGroups"]["workernode"]
host_info = []
for node in workernode_info:
host = {
VM_ID: node[VM_ID],
FAULT_DOMAIN: str(node[FAULT_DOMAIN]),
UPDATE_DOMAIN: str(node[UPDATE_DOMAIN]),
FQDN: node[FQDN],
BROKER_ID: brokers_info[node[FQDN]] if node[FQDN] in brokers_info else None,
RACK: FAULT_DOMAIN_SHORT + str(node[FAULT_DOMAIN]) + UPDATE_DOMAIN_SHORT + str(node[UPDATE_DOMAIN]),
FREE_DISK_SPACE: 0
}
if dead_hosts:
if not node[FQDN] in dead_hosts:
host_info.append(host)
else:
host_info.append(host)
return host_info
'''
Call the Kafka topic tool to get partition info about a topic. The return format is:
[
'Topic:dummyTopic PartitionCount:13 ReplicationFactor:3 Configs:segment.bytes=104857600,cleanup.policy=compact,compression.type=uncompressed',
' Topic: dummyTopic Partition: 0 Leader: 1026 Replicas: 1026,1028,1014 Isr: 1026,1028,1014',
' Topic: dummyTopic Partition: 1 Leader: 1020 Replicas: 1020,1014,1017 Isr: 1020,1014,1017',
..
]
'''
def get_topic_info(topic):
kafka_version, hdp_version = get_kafka_hdp_version()
if kafka_version >= '3.2.0':
topicInfo = subprocess.check_output([
KAFKA_TOPICS_TOOL_PATH,
"--bootstrap-server",
get_broker_connect_string(),
"--describe",
"--topic",
topic
])
else:
topicInfo = subprocess.check_output([
KAFKA_TOPICS_TOOL_PATH,
"--zookeeper",
get_zookeeper_connect_string(),
"--describe",
"--topic",
topic
])
if topicInfo is None or len(topicInfo) == 0:
raise_error("Failed to get Kafka partition info for topic " + topic)
return topicInfo
def get_replica_count_topic(topic):
topicInfo = get_topic_info(topic)
topicInfo_lines = topicInfo.split('\n')
if len(topicInfo_lines) < 2:
raise_error("Failed to parse Kafka topic info")
summaryWithSpaceFixed = topicInfo_lines[0].replace(": ", ":")
summary = summaryWithSpaceFixed.split()
replica_count = int(summary[2].split(":")[1])
return replica_count
'''
Parses through the output of the Kafka Topic tools and returns info about partitions for a given topic.
Return format: partitions_info = [
{
"partition": partition,
"leader": leader,
"replicas": replicas,
"isr": isr
},
...
]
'''
def get_partition_info(topic, topicInfo_lines, partition_sizes):
logger.info("Retrieving partition information for topic: %s", topic)
partitions_info = []
for i in range(1, len(topicInfo_lines)):
if len(topicInfo_lines[i].strip()) == 0:
continue
partition = int(topicInfo_lines[i].split('Partition: ')[1].split()[0])
leader = int(topicInfo_lines[i].split('Leader: ')[1].split()[0])
replicas = map(int, topicInfo_lines[i].split('Replicas: ')[1].split()[0].split(','))
isr = map(int, topicInfo_lines[i].split('Isr: ')[1].split()[0].split(','))
partition_name = str(topic + "-" + str(partition))
partition_size = 0
if partition_sizes:
if partition_name in partition_sizes:
partition_size = int(partition_sizes[partition_name])
partition_info = {
PARTITION: partition,
LEADER: leader,
REPLICAS: replicas,
ISR: isr,
PARTITION_SIZE: partition_size,
ASSIGNED: None
}
partitions_info.append(partition_info)
# Return the list sorted by increasing partition size so that we rebalance the smaller partitions first
return sorted(partitions_info, key=itemgetter(PARTITION_SIZE))
# Connect to Zookeeper
@retry(exceptions=BaseException, tries=MAX_RETRIES, delay=RETRY_INTERVAL_DELAY, backoff=RETRY_INTERVAL_BACKOFF,
logger=logger)
def connect(zk_quorum):
logger.info('Connecting to zookeeper quorum at: {0}'.format(zk_quorum))
zk = KazooClient(hosts=zk_quorum)
zk.start()
zk.add_listener(connection_lost)
return zk
def connection_lost(state):
if state == KazooState.LOST or state == KazooState.SUSPENDED:
raise RuntimeError("Fatal error lost connection to zookeeper.")
'''
Get broker ID to Host mapping from zookeeper.
Returns a dictionary:
brokers_info = {
'wn30-foobar': '1017',
'wn25-foobar': '1016',
'wn7-foobar': '1008',
..
}
'''
def get_brokerhost_info(zookeeper_client):
logger.info("Associating brokers to hosts...")
zk_brokers_ids = zookeeper_client.get_children(BROKERS_ID_PATH)
brokers_info = {}
for zk_broker_id in zk_brokers_ids:
zk_broker_id_data, stat = zookeeper_client.get('{0}/{1}'.format(BROKERS_ID_PATH, zk_broker_id))
zk_broker_info = json.loads(zk_broker_id_data)
zk_broker_host = _get_fqdn(zk_broker_info['host'])
brokers_info[zk_broker_host] = zk_broker_id
return brokers_info
def _valid_ipv4_address(host):
sub_parts = host.split('.')
if len(sub_parts) != 4:
return False
try:
return all(0 <= int(part) < 256 for part in sub_parts)
except ValueError:
return False
def _get_fqdn(zk_broker_host):
if _valid_ipv4_address(zk_broker_host):
zk_broker_host = socket.getfqdn(zk_broker_host)
return zk_broker_host.split('.')[0]
def generate_fd_list_ud_list(host_info):
# Get set of UDs & FDs
ud_set = set()
for val in host_info:
ud_set.add(val[UPDATE_DOMAIN])
ud_list = sorted(ud_set)
fd_set = set()
for val in host_info:
fd_set.add(val[FAULT_DOMAIN])
fd_list = sorted(fd_set)
return fd_list, ud_list
def check_brokers_up(host_info):
for host in host_info:
if not host[BROKER_ID]:
logger.error(
"VM %s with FQDN: %s has no brokers assigned. Ensure that all brokers are up! It is not recommended "
"to perform replica rebalance when brokers are down.", host[VM_ID], host[FQDN])
return
return True
'''
Determine the free space available on the brokers along with the sizes of the partitions hosted on them.
Returns a dictionary of the format: {"topic-partition_number, size"}
'''
def get_storage_info(host_info):
logger.info("Querying partition sizes from all brokers... This operation can take a few minutes.")
global_partition_sizes = {}
for host in host_info:
free_disk_space, partition_sizes = get_partition_sizes(host[FQDN])
host[FREE_DISK_SPACE] = int(free_disk_space)
for i in range(0, len(partition_sizes)):
partitions = partition_sizes[i].split(';')
for p in partitions[1:-1]:
p_name = str(p.split(',')[1].split('/')[2])
p_size = int(p.split(',')[0])
if p_name in global_partition_sizes and not (global_partition_sizes[p_name] is None):
global_partition_sizes[p_name] = max(global_partition_sizes[p_name], p_size)
else:
global_partition_sizes[p_name] = p_size
return global_partition_sizes
'''
Generate a replica reassignment JSON file to be passed to the Kafka Replica reassignment tool.
The method parses the cluster manifest to retrieve the topology information about hosts, including the fault & update domains. These are passed to the ReassignmentGenerator class which checks if each topic is already balanced and generates a reassignment plan if not.
Return value (reassignment json) is of the format:
{"partitions":
[{"topic": "foo",
"partition": 1,
"replicas": [1,2,3] }
],
"version":1
}
Sample Cluster Topology JSON:
{
"hostGroups": {
"gateway": [
{
...
},
.
.
],
"headnode": [
{
...
},
.
.
],
"workernode": [
{
"vmId": 0,
"fqdn": "wn0-k09v3",
"state": "Succeeded",
"faultDomain": 0,
"updateDomain": 2,
"availabilitySetId": "/subscriptions/abe48551-c98b-4263-97b3-098a4c35bc08/resourcegroups/rg0-d373d1ab2fb94339ad55b18da21bb049resourcegroup/providers/Microsoft.Compute/availabilitySets/workernode-0"
},
.
.
],
"zookeepernode": [
{
...
}
]
}
}
'''
def generate_reassignment_plan(plan_directory, topics, brokers_info, compute_storage_cost=False, dead_hosts=None,
force_rebalance=False):
logger.info("Starting tool execution...")
ret = None
# Retrieve Cluster topology
cluster_topology_json = get_cluster_topology_json()
# Parse JSON to retrieve information about hosts
host_info = parse_topo_info(cluster_topology_json, brokers_info, dead_hosts)
partitions_sizes = []
if compute_storage_cost:
partitions_sizes = get_storage_info(host_info)
logger.debug("Partition Sizes: %s", str(partitions_sizes))
logger.debug("\nHost Information: \n%s", host_info)
logger.debug("\n\nHost Rack Information: \n%s\n", "\n".join(
list(map(lambda datum: "Broker: {0} Rack: {1}".format(str(datum['brokerId']), datum['rack']), host_info))))
fd_list, ud_list = generate_fd_list_ud_list(host_info)
fd_count, ud_count = len(fd_list), len(ud_list)
logger.info("Checking if all brokers are up.")
if check_brokers_up(host_info):
# Keep track of number of replicas we assign to each broker. This count is across all topics
brokers_replica_count = []
for host in host_info:
b = {
BROKER_ID: host[BROKER_ID],
LEADERS: 0,
FOLLOWERS: 0,
}
brokers_replica_count.append(b)
# Keep track of already balanced partitions across topics. This is so that we can verify # of leaders across
# brokers at the end.
global_balanced_partitions = []
for topic in topics:
logger.info("Getting topic information for Topic: %s", topic)
# Get topic info using the Kafka topic tool
topicInfo = get_topic_info(topic)
replica_count_topic = get_replica_count_topic(topic)
logger.info("Replica count for topic %s is %s", topic, replica_count_topic)
logger.debug("Info for topic %s: \n%s", topic, topicInfo)
topicInfo_lines = topicInfo.split('\n')
if len(topicInfo_lines) < 2:
raise_error("Failed to parse Kafka topic info for topic: %s", topic)
partition_info = get_partition_info(topic, topicInfo_lines, partitions_sizes)
logger.debug("Partition info for topic %s : \n%s", topic, partition_info)
reassignment_generator = ReassignmentGenerator(host_info, topic, partition_info, compute_storage_cost)
# Generate Rack alternated list
fd_ud_list = reassignment_generator._generate_fd_ud_list()
rack_alternated_list = reassignment_generator._generate_alternated_fd_ud_list(fd_ud_list, fd_list, ud_list)
logger.debug("Generated Rack alternated list: %s", str(rack_alternated_list))
# Variables to keep track of which rack in the alternated list is the next one to be assigned a replica.
rand_rack = random.randrange(0, len(rack_alternated_list))
logger.debug("Rand_rack: %s", str(rand_rack))
next_Leader = (int(len(ud_list)) * rand_rack) % len(rack_alternated_list)
logger.debug("Start with position in Rack Alternated List: %s", str(next_Leader))
reassignment_plan, balanced_partitions_for_topic = reassignment_generator._generate_reassignment_plan_for_topic(
replica_count_topic, next_Leader, rack_alternated_list, fd_count, ud_count, brokers_replica_count,
force_rebalance)
logger.debug("\nAlready balanced partitions for the topic %s are: \n%s", topic,
balanced_partitions_for_topic)
for partition in balanced_partitions_for_topic:
global_balanced_partitions.append(partition)
if reassignment_plan is not None:
logger.info("\n\nReassignment plan generated for topic: %s\n%s\n", topic, reassignment_plan)
if ret is not None:
ret["partitions"] += reassignment_plan["partitions"]
else:
ret = reassignment_plan
verify_plan = reassignment_generator._verify_reassignment_plan(reassignment_plan, topic,
replica_count_topic, fd_count, ud_count)
verify_leaders_distributed(host_info, ret, global_balanced_partitions)
if ret is not None:
verify_leader_count_balanced = verify_leaders_distributed(host_info, ret, global_balanced_partitions)
# save the reassignment plan in ASSIGNMENT_JSON_FILE
ret = json.dumps(ret)
f = open(os.path.join(plan_directory, ASSIGNMENT_JSON_FILE), "w")
f.write(ret)
f.close()
logger.info("\n\nComplete Reassignment plan for all topics! \n%s\n", ret)
else:
# remove contents from ASSIGNMENT_JSON_FILE
open(os.path.join(plan_directory, ASSIGNMENT_JSON_FILE), "w").close()
return ret
class ReassignmentGenerator:
def __init__(self, host_info, topic, partition_info, compute_storage_cost):
self.host_info = host_info
self.topic = topic
self.partition_info = partition_info
self.partitions_count = len(partition_info)
self.compute_storage_cost = compute_storage_cost
def _generate_fd_ud_list(self):
# Get set of FD+UDs
fd_ud_set = set()
for val in self.host_info:
fd_ud_set.add(val[RACK])
return sorted(fd_ud_set)
def _get_fd_rack(self, rack):
domains = re.findall(r'\d+', rack)
return int(domains[0])
def _get_ud_rack(self, rack):
domains = re.findall(r'\d+', rack)
return int(domains[1])
def _gcd(self, a, b):
while b:
a, b = b, a % b
return a
'''Generates a list of alternated FD + UD combinations. List = [ (fd1,ud1) , (fd2,ud2), ... ] Example with 3 FDs
and 3 UDs : ['FD0UD0', 'FD1UD1', 'FD2UD2', 'FD0UD1', 'FD1UD2', 'FD2UD0', 'FD0UD2', 'FD1UD0', 'FD2UD1'] '''
def _generate_alternated_fd_ud_list(self, fd_ud_list, fd_list, ud_list):
alternated_list = []
# Find largest FD# & UD#. This is required because there could be gaps and we need to know the largest # to
# compute the possible FD x UD matrix. Missing combinations of (FD,UD) in the VMs allocated are not added to
# the final list.
fd_length = max(map(int, fd_list)) + 1
ud_length = max(map(int, ud_list)) + 1
i = 0
j = 0
k = 1
# Traverse matrix in diagonal slices
while True:
current_rack = FAULT_DOMAIN_SHORT + str(i % fd_length) + UPDATE_DOMAIN_SHORT + str(j % ud_length)
if current_rack in fd_ud_list:
if not current_rack in alternated_list:
alternated_list.append(current_rack)
# If FD + UD combo is already present in alternated_list, we are revisiting this the second time. Hence,
# break out of the loop.
else:
break
i += 1
# If matrix inputs are of form (n,n) or (m,m), add a shift to UD index so that we get a different
# diagonal slice. To get the next adjacent diagonal slice, we add a shift by ud_length - 1.
if self._gcd(ud_length, fd_length) > 1 and i % fd_length == 0:
j = k
k += 1
else:
j += 1
return alternated_list
'''
Conditions required for a partition to be eligible for ReassignmentGenerator
1> MIN(len(ISR)) >= 1
2> Leader is in ISR
3> Leader is assigned (not -1)
4> Replicas are present
5> Number of replicas for partition should be equal to replica count for topic
'''
def _is_partition_eligible_reassignment(self, partition, replica_count_topic):
does_not_meet_criteria_msg = "Partition for topic does not meet criteria for rebalance. Skipping."
partition[ASSIGNED] = False
if not len(partition[ISR]) >= 1:
logger.warning(
"%s - Topic: %s, Partition: %s. Criteria not met: 'There should be at least one replica in the ISR'.",
does_not_meet_criteria_msg, self.topic, partition[PARTITION])
return False
if not partition[LEADER] in partition[ISR]:
logger.warning("%s - Topic: %s, Partition: %s. Criteria not met: 'The leader should be in the ISR'.",
does_not_meet_criteria_msg, self.topic, partition[PARTITION])
return False
if not int(partition[LEADER]) != -1:
logger.warning(
"%s - Topic: %s, Partition: %s. Criteria not met: 'There should be an assigned leader. Leader cannot "
"be -1'.",
does_not_meet_criteria_msg, self.topic, partition[PARTITION])
return False
if not partition[REPLICAS]:
logger.warning("%s - Topic: %s, Partition: %s. Criteria not met: 'Replicas cannot be null'.",
does_not_meet_criteria_msg, self.topic, partition[PARTITION])
return False
partition[ASSIGNED] = True
return True
def _get_brokers_in_rack(self, rack):
return [element for element in self.host_info if element[RACK] == rack]
def _get_broker_info(self, b_id):
retList = [element for element in self.host_info if int(element[BROKER_ID]) == b_id]
if retList:
return retList[0]
else:
logger.warn("Cannot retrieve host associated with broker with ID: %s", b_id)
return None
def _get_count_replicas_in_broker(self, broker_id, broker_replica_count):
retList = [element for element in broker_replica_count if element[BROKER_ID] == broker_id]
if retList:
return retList[0]
else:
return None
def _increment_count_replicas_in_broker(self, broker_id, broker_replica_count, type_of_count):
retList = [element for element in broker_replica_count if element[BROKER_ID] == broker_id]
if retList:
e = retList[0]
e[type_of_count] += 1
else:
logger.warning("%s not found in broker_replica_count!")
def _get_weighted_count_replicas_in_rack(self, broker_replica_count, rack_alternated_list, rack_index,
type_of_replica):
count = 0
brokers_in_rack = self._get_brokers_in_rack(rack_alternated_list[rack_index])
if brokers_in_rack:
for broker in brokers_in_rack:
count += self._get_count_replicas_in_broker(broker[BROKER_ID], broker_replica_count)[type_of_replica]
return count / float(len(brokers_in_rack))
else:
logger.error("No brokers were found for rack %s. Please verify brokers are up!",
rack_alternated_list[rack_index])
return
'''Determines the rack (FD+UD combination) for the replica. Once determined, there could be multiple brokers that
meet the criteria. We choose the broker which has less number of replicas assigned to it. (distribute the load) '''
def _assign_replica_for_partition(self, rack_alternated_list, broker_replica_count, next_rack, type_of_replica):
eligible_brokers = self._get_brokers_in_rack(rack_alternated_list[next_rack])
if eligible_brokers:
new_broker = eligible_brokers[0]
for broker in eligible_brokers:
a = self._get_count_replicas_in_broker(broker[BROKER_ID], broker_replica_count)[type_of_replica]
b = self._get_count_replicas_in_broker(new_broker[BROKER_ID], broker_replica_count)[type_of_replica]
if a < b:
new_broker = broker
self._increment_count_replicas_in_broker(new_broker[BROKER_ID], broker_replica_count, type_of_replica)
return new_broker[BROKER_ID]
else:
logger.error("No eligible brokers found for rack: %s", rack_alternated_list[next_rack])
return
'''
This method reassigns the replicas for the given partition. The algorithm for assignment is as follows:
1> Iterate through the rack alternated list and look at sets of size replica_count.
For 3 x 3: the list is: (0,0), (1,1), (2,2), (0,1), (1,2), (2,0), (0,2), (1,0), (2,1)
In first iteration we look at: (0,0) (1,1) (2,2) if replica count is 3.
Each of these represent racks for which there could be multiple brokers.
2> Determine which of the racks has the least number of leaders.
3> Assign this rack as the leader for the partition.
4> Determine all eligible brokers within this rack. Assign the broker with the least number of leaders within the rack as the leader for this partition.
5> Assign the remaining replicas to the 2 other racks in the set. These are follower replicas.
This is to ensure we will not always get the same set of sequences.
6> Look at the next set of 3 Racks and repeat from 1>
'''
def _scan_partition_for_reassignment(self, index, brokers_replica_count, rack_alternated_list, start_index,
ud_count, replica_count_topic):
reassignment = {"topic": self.topic,
PARTITION: int(self.partition_info[index][PARTITION]),
REPLICAS: []
}
replica_count = int(replica_count_topic)
rack_count = len(rack_alternated_list)
'''
Re-assign replicas for the PARTITION.
Replicas will be distributed across following racks: start_index, start_index + 1, ...., start_index + replica_count - 1.
'''
# Determine which rack has fewest LEADERS
current_min = sys.maxint
relative_rack_index = 0
for i in range(0, replica_count):
leaders_in_current_rack = self._get_weighted_count_replicas_in_rack(brokers_replica_count,
rack_alternated_list,
(start_index + i) % rack_count, LEADERS)
if leaders_in_current_rack < current_min:
current_min = leaders_in_current_rack
relative_rack_index = i
rack_index = (start_index + relative_rack_index) % rack_count
# Do the actual assignment of leader
leader_broker_id = self._assign_replica_for_partition(rack_alternated_list, brokers_replica_count, rack_index,
LEADERS)
# Check if there is sufficient space on the broker, if not set the "ASSIGNED" property of partition to False
# to indicate that it was not assigned
if self.compute_storage_cost:
logger.debug("Checking if there is sufficient disk space on broker.")
host_for_broker = [element for element in self.host_info if element[BROKER_ID] == leader_broker_id][0]
free_disk_space = host_for_broker[FREE_DISK_SPACE]
if free_disk_space < self.partition_info[index][PARTITION_SIZE]:
logger.warning(
"Not sufficient disk space on elected leader: %s with broker ID: %s. Skipping rebalance for "
"partition: %s",
host_for_broker[FQDN], host_for_broker[BROKER_ID], self.partition_info[index][PARTITION])
self.partition_info[index][ASSIGNED] = False
return None, start_index % rack_count
else:
# Since we are assigning the partition to the broker, reduce the available free space by the size of
# the partition
host_for_broker[FREE_DISK_SPACE] -= self.partition_info[index][PARTITION_SIZE]
reassignment[REPLICAS].append(int(leader_broker_id))
# Assign replicas
p_size = "N/A"
for follower_index in range(0, replica_count):
if follower_index != relative_rack_index:
follower_broker_id = self._assign_replica_for_partition(rack_alternated_list, brokers_replica_count,
(start_index + follower_index) % rack_count,
FOLLOWERS)
reassignment[REPLICAS].append(int(follower_broker_id))
if self.compute_storage_cost:
host_for_broker = \
[element for element in self.host_info if element[BROKER_ID] == follower_broker_id][0]
host_for_broker[FREE_DISK_SPACE] -= self.partition_info[index][PARTITION_SIZE]
p_size = self.partition_info[index][PARTITION_SIZE]
logger.debug("Topic: %s Reassigning Partition: %s of SIZE: %s from %s --> %s", self.topic,
self.partition_info[index][PARTITION], p_size, self.partition_info[index][REPLICAS],
reassignment[REPLICAS])
self.partition_info[index][ASSIGNED] = True
start_index += min(ud_count, replica_count)
return reassignment, start_index % rack_count
'''
Iterate through all replicas of a topic to determine if it is balanced:
1) Add the UDs of the replicas to a list - fd_list
1) Add the UDs of the replicas to a list - fd_list
2) Verify that number of domains the replicas are in is equal min(#replicas, #domains). This ensures that all replicas are in separate UDs and separate FDs.
'''
def _check_if_partition_balanced(self, partition, replica_count, fd_count, ud_count, brokers_replica_count,
balanced_partitions):
logger.debug("Checking if Partition %s is balanced.", partition)
if len(partition[REPLICAS]) != replica_count:
logger.warning(
"The replica count for the partition is not the same as the replica count for the topic. Rebalance "
"recommended.")
return False
fd_list, ud_list = [], []
for replica in partition[REPLICAS]:
# Get the rack associated with the replica and add to list
host = self._get_broker_info(int(replica))
# If host was removed from the rack the above will return null. In this case the partition is not balanced
if not host:
return False
current_FD = host[FAULT_DOMAIN]
current_UD = host[UPDATE_DOMAIN]
fd_list.append(current_FD)
ud_list.append(current_UD)
if len(set(fd_list)) == min(fd_count, replica_count) and len(set(ud_list)) == min(ud_count, replica_count):
if brokers_replica_count:
# Update brokers_replica_count to keep track of number of leaders, followers across brokers
self._increment_count_replicas_in_broker(str(partition[REPLICAS][0]), brokers_replica_count, LEADERS)
for i in range(1, len(partition[REPLICAS])):
self._increment_count_replicas_in_broker(str(partition[REPLICAS][i]), brokers_replica_count,
FOLLOWERS)
balanced_partitions.append(partition)
logger.debug("Partition is balanced across available fault and update domains!")
return True
logger.debug("Partition needs to be balanced.")
return False
def _generate_reassignment_plan_for_topic(self, replica_count_topic, next_Leader, rack_alternated_list, fd_count,
ud_count, brokers_replica_count, force_rebalance):
ret = None
reassignment = {"partitions": [], "version": 1}
reassignment_changes = []
retained_assignment = {"partitions": []}
balanced_partitions = []
# Check if #replicas is less than 3 if #FD==3/#FD==1 or #replica is less than 4 if #FD is 2
if (fd_count == 1 and replica_count_topic < 3) or (fd_count == 2 and replica_count_topic < 4) or (
fd_count == 3 and replica_count_topic < 3):
logger.warning("""
There are not as many upgrade/fault domains as the replica count for the topic %s.\n
Replica Count: %s, Number of Fault Domains: %s, Number of Update Domains: %s.\n
The recommendation is to have at least 3 replicas if number of fault domains in the region is 3,
and 4 replicas if number of fault domains is 2.
""", self.topic, replica_count_topic, fd_count, ud_count)
if not force_rebalance:
logger.error(
"Rebalance with HA not possible! Skipping rebalance for the topic. If you like to rebalance "
"regardless, please run the tool with -force flag.")
return ret, balanced_partitions
else:
logger.info("Proceeding with generation of reassignment plan since -force flag was specified.")
# Check if there is a valid number of replicas for the topic
if replica_count_topic <= 1:
logger.warning(
"Invalid number of replicas for topic %s. Rebalance with HA guarantee not possible! The tool will try "
"to do whats possible.",
self.topic)
return ret, balanced_partitions
logger.info("Checking if Topic: %s needs to be re-balanced.", self.topic)
# Keep track of numbers of replicas assigned to each broker
# Iterate through all partitions and check whether they need to be re-balanced
for i in range(0, len(self.partition_info)):
if (not self._check_if_partition_balanced(self.partition_info[i], replica_count_topic, fd_count, ud_count,
brokers_replica_count, balanced_partitions)):
if self._is_partition_eligible_reassignment(self.partition_info[i], replica_count_topic):
r, next_Leader = self._scan_partition_for_reassignment(i, brokers_replica_count,
rack_alternated_list, next_Leader, ud_count,
replica_count_topic)
if r is not None:
reassignment["partitions"].append(r)
ret = reassignment
reassignment_changes.append(
"Partition {0} {1} --> {2}".format(int(self.partition_info[i][PARTITION]),
self.partition_info[i][REPLICAS], r[REPLICAS]))
else:
# Partition is already balanced. Add the existing assignment to the rebalance plan.
current_partition_assignment = {
"topic": self.topic,
PARTITION: self.partition_info[i][PARTITION],
REPLICAS: self.partition_info[i][ISR]
}
retained_assignment["partitions"].append(current_partition_assignment)
# If some partitions need to be rebalanced (ret is not None) then append the retained assignment to the
# reassignment partitions and update ret
if len(retained_assignment["partitions"]) > 0 and ret is not None:
logger.info("Topic: %s Partitions Already Balanced: %s Partitions To Be Rebalanced: %s", self.topic,
list(map(lambda datum: datum['partition'], retained_assignment["partitions"])),
list(map(lambda datum: datum['partition'], ret["partitions"])))
for j in retained_assignment["partitions"]:
reassignment["partitions"].append(j)
ret = reassignment
if len(reassignment_changes) > 0:
logger.info(
"\n\nReassignment changes: Topic: {0} \n{1}\n".format(self.topic, "\n".join(reassignment_changes)))
return ret, balanced_partitions