def set_alias(self, alias_name, index_name):
indices = IndicesClient(self.elastic)
old_index = alias_name + '_20*'
try:
indices.delete_alias(index=old_index, name=alias_name)
except NotFoundError:
pass
except Exception as e:
logger.error("Alias %s for %s not created: %s", alias_name,
index_name, str(e))
return
try:
indices.put_alias(index=index_name, name=alias_name, body={})
except Exception as e:
logger.error("Alias %s for %s not created: %s", alias_name,
index_name, str(e))
return
logger.info("Set alias %s -> %s", alias_name, index_name)
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--src-host", action="store", default="127.0.0.1", type=unicode, help="Source host [default: %(default)s]")
parser.add_argument("--src-port", action="store", default=9200, help="Source port [default: %(default)s]")
parser.add_argument("--src-index", action="store", default="", type=unicode, help="Source index")
parser.add_argument("--src-batch-size", action="store", type=int, default=5000, help="Source query batchsize [default: %(default)s]")
parser.add_argument("--src-scroll-interval", action="store", type=unicode, default="60m", help="Interval for source scroll query [default: %(default)s]")
parser.add_argument("--dest-host", action="store", default="127.0.0.1", type=unicode, help="Destination host [default: %(default)s]")
parser.add_argument("--dest-port", action="store", default=9200, help="Destination port [default: %(default)s]")
parser.add_argument("--dest-index", action="store", default="", type=unicode, help="Destination index")
parser.add_argument("--dest-batch-size", action="store", type=int, default=5000, help="Destination batchsize [default: %(default)s]")
parser.add_argument("--dest-alias", action="store", help="Destination index alias (to be set after we have finished populating)")
parser.add_argument("--dest-concurrency", action="store", type=int, default=4, help="Destination batchsize [default: %(default)s]")
parser.add_argument("--dest-delete-index", action="store_true", help="Delete destination index at before starting")
parser.add_argument("--query", action="store", type=unicode, default="", help="Query to use [if None is specified, a match_all will be used]")
args = parser.parse_args()
if args.src_index is None or len(args.src_index) == 0:
raise Exception("--src-index must be specified!")
if args.dest_index is None or len(args.dest_index) == 0:
raise Exception("--dest-index must be specified!")
dt_start = datetime.now()
# copy mapping
src_es_instance = get_elasticsearch(args.src_host, args.src_port)
dest_es_instance = get_elasticsearch(args.dest_host, args.dest_port)
# check if src_index exists
src_es_ic = IndicesClient(src_es_instance)
if not src_es_ic.exists(args.src_index):
raise Exception("--src-index %s does not exist!" % args.src_index)
# check if dest_index exists
dest_es_ic = IndicesClient(dest_es_instance)
if dest_es_ic.exists(args.dest_index):
if args.dest_delete_index:
dest_es_ic.delete(index=args.dest_index)
else:
raise Exception("--dest-index %s already exists! Use --dest-delete-index if you want to drop it" % args.dest_index)
log.info("Copying mapping...")
# copy mapping over to dest
src_index_information = src_es_ic.get(index=args.src_index)
dest_es_ic.create(index=args.dest_index, body=src_index_information.get(args.src_index, {}))
# set num_of_replicas to 0
dest_es_ic.put_settings(index=args.dest_index, body={"settings": {"index": {"number_of_replicas": 0}}})
# perform multiprocessing
log.info("Copying data...")
MAGIC_STRING = "%s:%s" % (str(uuid4()), str(uuid4()))
DEST_QUEUE = Queue()
DEST_COUNTER = Value('i', 0)
src_process = Process(target=src_worker, args=(args, DEST_QUEUE, MAGIC_STRING))
src_process.start()
dest_processes = [Process(target=dest_worker, args=(args, DEST_QUEUE, MAGIC_STRING, DEST_COUNTER)) for i in xrange(args.dest_concurrency)]
for i in dest_processes: i.start()
src_process.join()
for i in dest_processes: i.join()
log.info("[dest_worker] Total processed %s" % DEST_COUNTER.value)
if args.dest_alias is not None and len(args.dest_alias) > 0:
# we remove all existing mappings to this alias, then add it to the current dest_index
for idx_name, aliases_mapping in dest_es_ic.get_aliases().iteritems():
if args.dest_alias in aliases_mapping.get("aliases", {}):
dest_es_ic.delete_alias(index=idx_name, name=args.dest_alias)
dest_es_ic.put_alias(index=args.dest_index, name=args.dest_alias)
dest_es_ic.refresh(args.dest_index)
dt_end = datetime.now()
log.info("Time elapsed: %s" % (dt_end-dt_start, ))
class UpdateService(object):
"""
全量更新的service
"""
def __init__(self, taskName, handlerConfigPath):
self.handlerConfigPath = handlerConfigPath
self.redisClient = remote.getRedisClient()
self.esClient = remote.getElasticClient()
self.esIndexClient = IndicesClient(self.esClient)
# instance of RedisStatus
self.statusKey = taskName
self.status = None
# classname of handler
self.handlerClassName = 'modules.handlers.v1.CommonHandler'
def update(self):
"""
update action.
"""
# 检查status，看是否可以进行全量更新
if not self._checkStatus():
return False
# 更新status，标明准备开始全量更新
self._switchStatus()
# 将新的消费者的offset设置为 latest
self._copyKafkaOffset()
# 从mysql读取数据，同步到ES中
syncResult = self._syncFromMysqlToEs()
if syncResult:
# 从kafka队列中消费数据，直至赶上老的消费者
self._catchupwithCurrentConsumer()
# 移除老的es index的alias，增加新的es index的alias
self._setESIndexAlias()
# 修改status
self._resetStatus(True)
else:
# 修改status
self._resetStatus(False)
# 清理脏数据
self._cleanDirtyData()
return syncResult
def reset(self):
"""
重新设置status，清除脏数据
"""
status = RedisStatus(self.statusKey)
status.sync()
self.status = status
self._restoreESIndexAlias()
self._resetStatus(False)
self._cleanDirtyData()
def clean(self):
"""
清除脏数据
"""
self._cleanDirtyData()
def _cleanDirtyData(self):
"""
清除发生异常时导致的脏数据。
包括：无用的statusconfig, 无用的es index
"""
configKeys = self.redisClient.keys('CK_%s_*' % self.statusKey)
status = RedisStatus(self.statusKey)
status.sync()
liveKeys = [status.config, status.nextConfig, status.tmpConfig]
for key in configKeys:
if key not in liveKeys:
self._deleteStatusConfig(key, deleteIndices=True)
elif key == status.tmpConfig:
self._deleteStatusConfig(key, deleteIndices=False, deleteAlias=True)
def _fail(self, errinfo):
_logger.error(errinfo)
print(errinfo, file=sys.stderr)
def _checkStatus(self):
status = RedisStatus(self.statusKey)
status.sync()
self.status = status
if status.code != STATUS_INITIAL and status.code != STATUS_INCRE_UPDATING:
self._fail('status code [%s] MUST be in values array [%s, %s]' % (status, STATUS_INITIAL, STATUS_INCRE_UPDATING))
return False
else:
return True
def _loadConfig(self):
"""
从conf/handlers/目录中加载handlers相关的配置，
此外，还会生成一个新的consumer_group_id以及一个新的es_index,
这些数据都写入redis
"""
configKey = 'CK_%s_%s' % (self.statusKey, app.getUuid('update.loadconfig.gen.config_key'))
kafkaGroupId = app.getUuid('update.loadconfig.gen.kafka_group_id')
esIndexSuffix = str(int(time.time())) + '_' + str(random.randint(0, 100000))
handlerConfig = HandlerConfig()
handlerConfig.loadFromFile(self.handlerConfigPath)
statusConfig = RedisStatusConfig(configKey)
statusConfig.kafkaGroupId = kafkaGroupId
statusConfig.esIndexSuffix = esIndexSuffix
statusConfig.handlerConfig = handlerConfig
statusConfig.set()
_logger.info('load config: %s', statusConfig)
return statusConfig
def _acquireDLock(self):
return self.redisClient.acquireDLock(_DISTRIBUTE_REDIS_LOCK_PREFIX + self.statusKey)
def _releaseDLock(self, lock):
return self.redisClient.releaseDLock(lock)
def _switchStatus(self):
"""
更新status，标明准备开始全量更新
"""
lock = None
try:
# 创建全量更新时所需的statusConfig
statusConfig = self._loadConfig()
lock = self._acquireDLock()
# check the stauts again after get the lock
if not self._checkStatus():
return False
self.status.code = STATUS_FULL_UPDATING
self.status.nextConfig = statusConfig.key
self.status.update()
except Exception as e:
_logger.info('fail to switch status to full updating: %s', e)
raise
finally:
if lock:
self._releaseDLock(lock)
def _copyKafkaOffset(self):
"""
将新的消费者的offset设置为 latest
"""
# 首先要获取kafka topic的所有分区
topicName = config().get('kafka', 'topic')
if self.status.nextConfig:
nextStatusConfig = RedisStatusConfig(self.status.nextConfig, forceSync=True)
try:
nextConsumer = remote.getKafkaConsumer(
nextStatusConfig.kafkaGroupId,
autoCommit=False,
autoOffsetReset='latest'
)
_logger.debug('next kafka groupid is: %s', nextStatusConfig.kafkaGroupId)
clusterMetadata = nextConsumer.list_topics(topicName)
topicMetadata = clusterMetadata.topics.get(topicName, {})
partitions = topicMetadata.partitions
for pid in partitions.keys():
p = TopicPartition(topicName, pid)
nextConsumer.assign([p])
msg = nextConsumer.poll(10)
if msg:
offset = msg.offset() - 1
_logger.debug('pid[%s] topic[%s] offset[%s]', pid, topicName, offset)
if offset >= 0:
p.offset = offset
nextConsumer.commit(offsets=[p])
except Exception as e:
_logger.error('exception occurs when setting offset for new consumer: %s', Failure())
raise
finally:
if nextConsumer:
nextConsumer.close()
def _syncFromMysqlToEs(self):
nextStatusConfig = RedisStatusConfig(self.status.nextConfig, forceSync=True)
# 从mysql中全量更新
try:
handlerClass = utils.classForName(self.handlerClassName)
commonHandler = handlerClass(nextStatusConfig)
commonHandler.syncFromMySQL()
return True
except Exception as e:
_logger.error('exception occurs when syncFromMySQL: %s', Failure())
return False
def _catchupwithCurrentConsumer(self):
"""
取新的consumer_group_id，从kafka队列中消费数据。
直到新的消费者赶上旧的消费者。
"""
# TODO
def _setESIndexAlias(self):
"""
移除老的es index的alias，增加新的es index的alias
"""
configKey = self.status.config
if configKey:
statusConfig = RedisStatusConfig(configKey, forceSync=True)
self._removeESIndexAlias(statusConfig)
nextConfigKey = self.status.nextConfig
if nextConfigKey:
nextStatusConfig = RedisStatusConfig(nextConfigKey, forceSync=True)
self._addESIndexAlias(nextStatusConfig)
def _restoreESIndexAlias(self):
"""
恢复老的es index的alias，删除新的es index的alias
"""
nextConfigKey = self.status.nextConfig
if nextConfigKey:
nextStatusConfig = RedisStatusConfig(nextConfigKey, forceSync=True)
self._removeESIndexAlias(nextStatusConfig)
configKey = self.status.config
if configKey:
statusConfig = RedisStatusConfig(configKey, forceSync=True)
self._addESIndexAlias(statusConfig)
def _removeESIndexAlias(self, statusConfig):
"""
移除指定es index的alias
"""
esIndexSuffix = statusConfig.esIndexSuffix
indices = statusConfig.handlerConfig.indices()
for indexAlias in indices:
indexName = indexAlias + '_' + esIndexSuffix
try:
self.esIndexClient.delete_alias(
index=indexName,
name=indexAlias
)
except NotFoundError as e:
_logger.error('index[%s] not found: %s', indexName, e)
def _addESIndexAlias(self, statusConfig):
"""
移除指定es index的alias
"""
esIndexSuffix = statusConfig.esIndexSuffix
indices = statusConfig.handlerConfig.indices()
for indexAlias in indices:
indexName = indexAlias + '_' + esIndexSuffix
try:
self.esIndexClient.put_alias(
index=indexName,
name=indexAlias
)
except NotFoundError as e:
_logger.error('index[%s] not found: %s', indexName, e)
def _resetStatus(self, succeeded):
"""
重置status，标明全量更新完毕，开始进行增量更新
"""
try:
lock = self._acquireDLock()
if succeeded:
# after succeeded
self.status.code = STATUS_INCRE_UPDATING
self.status.tmpConfig = self.status.config
self.status.config = self.status.nextConfig
self.status.nextConfig = ''
self.status.update()
else:
# after failed
self.status.code = STATUS_INCRE_UPDATING
self.status.tmpConfig = self.status.nextConfig
self.status.nextConfig = ''
self.status.update()
# if failed, delete dirty data immediately
self._deleteStatusConfig(self.status.tmpConfig, deleteIndices=True)
except Exception as e:
_logger.info('fail to reset status to incre-updating: %s', Failure())
raise
finally:
if lock:
self._releaseDLock(lock)
def _deleteStatusConfig(self, configKey, deleteIndices=False, deleteAlias=False):
# 删除stats config
if configKey:
statusConfig = RedisStatusConfig(configKey, forceSync=True)
if deleteIndices:
handlerConfig = statusConfig.handlerConfig
for indexAlias in handlerConfig.indices():
indexName = indexAlias + '_' + statusConfig.esIndexSuffix
if self.esIndexClient.exists(index=indexName):
self.esIndexClient.delete(
index=indexName
)
elif deleteAlias:
# 只删除alias，不删除index
self._removeESIndexAlias(statusConfig)
# 在redis中删除status config
# 设置为1天后过期
statusConfig.delete(24 * 3600)
def _removeNextEsIndex(self):
"""
删除新的es index
"""
# TODO
def _removeNextStatusConfig(self):
"""
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--src-host",
action="store",
default="127.0.0.1",
type=str,
help="Source host [default: %(default)s]")
parser.add_argument("--src-port",
action="store",
default=9200,
help="Source port [default: %(default)s]")
parser.add_argument("--src-index",
action="store",
default="",
type=str,
help="Source index")
parser.add_argument("--src-batch-size",
action="store",
type=int,
default=5000,
help="Source query batchsize [default: %(default)s]")
parser.add_argument(
"--src-scroll-interval",
action="store",
type=str,
default="60m",
help="Interval for source scroll query [default: %(default)s]")
parser.add_argument("--dest-host",
action="store",
default="127.0.0.1",
type=str,
help="Destination host [default: %(default)s]")
parser.add_argument("--dest-port",
action="store",
default=9200,
help="Destination port [default: %(default)s]")
parser.add_argument("--dest-index",
action="store",
default="",
type=str,
help="Destination index")
parser.add_argument("--dest-batch-size",
action="store",
type=int,
default=5000,
help="Destination batchsize [default: %(default)s]")
parser.add_argument(
"--dest-alias",
action="store",
help=
"Destination index alias (to be set after we have finished populating)"
)
parser.add_argument("--dest-concurrency",
action="store",
type=int,
default=4,
help="Destination batchsize [default: %(default)s]")
parser.add_argument("--dest-delete-index",
action="store_true",
help="Delete destination index at before starting")
parser.add_argument(
"--query",
action="store",
type=str,
default="",
help="Query to use [if None is specified, a match_all will be used]")
args = parser.parse_args()
if args.src_index is None or len(args.src_index) == 0:
raise Exception("--src-index must be specified!")
if args.dest_index is None or len(args.dest_index) == 0:
raise Exception("--dest-index must be specified!")
dt_start = datetime.now()
# copy mapping
src_es_instance = get_elasticsearch(args.src_host, args.src_port)
dest_es_instance = get_elasticsearch(args.dest_host, args.dest_port)
# check if src_index exists
src_es_ic = IndicesClient(src_es_instance)
if not src_es_ic.exists(args.src_index):
raise Exception("--src-index %s does not exist!" % args.src_index)
# check if dest_index exists
dest_es_ic = IndicesClient(dest_es_instance)
if dest_es_ic.exists(args.dest_index):
if args.dest_delete_index:
dest_es_ic.delete(index=args.dest_index)
else:
raise Exception(
"--dest-index %s already exists! Use --dest-delete-index if you want to drop it"
% args.dest_index)
log.info("Copying mapping...")
# copy mapping over to dest
src_index_information = src_es_ic.get(index=args.src_index)
dest_es_ic.create(index=args.dest_index,
body=src_index_information.get(args.src_index, {}))
# set num_of_replicas to 0
dest_es_ic.put_settings(
index=args.dest_index,
body={"settings": {
"index": {
"number_of_replicas": 0
}
}})
# perform multiprocessing
log.info("Copying data...")
MAGIC_STRING = "%s:%s" % (str(uuid4()), str(uuid4()))
DEST_QUEUE = Queue()
DEST_COUNTER = Value('i', 0)
src_process = Process(target=src_worker,
args=(args, DEST_QUEUE, MAGIC_STRING))
src_process.start()
dest_processes = [
Process(target=dest_worker,
args=(args, DEST_QUEUE, MAGIC_STRING, DEST_COUNTER))
for i in xrange(args.dest_concurrency)
]
for i in dest_processes:
i.start()
src_process.join()
for i in dest_processes:
i.join()
log.info("[dest_worker] Total processed %s" % DEST_COUNTER.value)
if args.dest_alias is not None and len(args.dest_alias) > 0:
# we remove all existing mappings to this alias, then add it to the current dest_index
for idx_name, aliases_mapping in dest_es_ic.get_aliases().iteritems():
if args.dest_alias in aliases_mapping.get("aliases", {}):
dest_es_ic.delete_alias(index=idx_name, name=args.dest_alias)
dest_es_ic.put_alias(index=args.dest_index, name=args.dest_alias)
dest_es_ic.refresh(args.dest_index)
dt_end = datetime.now()
log.info("Time elapsed: %s" % (dt_end - dt_start, ))
