def _start_service(self, config=_Config()):
c = MagicMock(config=config)
invocation_service = InvocationService(c, c._reactor)
self.service = invocation_service
invocation_service.init(c._internal_partition_service, c._connection_manager, c._listener_service)
invocation_service.start()
return c, invocation_service
def _start_service(self, config=_Config()):
c = MagicMock()
invocation_service = InvocationService(c, config, c._reactor)
self.service = invocation_service
invocation_service.init(
c._internal_partition_service,
c._connection_manager,
c._listener_service,
c.compact_schema_service,
)
invocation_service.start()
invocation_service.add_backup_listener()
return c, invocation_service
class HazelcastClient(object):
"""Hazelcast client instance to access and manipulate
distributed data structures on the Hazelcast clusters.
Keyword Args:
cluster_members (`list[str]`): Candidate address list that client
will use to establish initial connection.
By default, set to ``["127.0.0.1"]``.
cluster_name (str): Name of the cluster to connect to. The name is
sent as part of the the client authentication message and may
be verified on the member. By default, set to ``dev``.
client_name (str): Name of the client instance. By default,
set to ``hz.client_${CLIENT_ID}``, where ``CLIENT_ID`` starts
from ``0`` and it is incremented by ``1`` for each new client.
connection_timeout (float): Socket timeout value in seconds for
client to connect member nodes. Setting this to ``0`` makes
the connection blocking. By default, set to ``5.0``.
socket_options (`list[tuple]`): List of socket option tuples. The
tuples must contain the parameters passed into the
:func:`socket.setsockopt` in the same order.
redo_operation (bool): When set to ``True``, the client will redo
the operations that were executing on the server in case if the
client lost connection. This can happen because of network problems,
or simply because the member died. However it is not clear whether
the operation was performed or not. For idempotent operations this is
harmless, but for non idempotent ones retrying can cause to undesirable
effects. Note that the redo can be processed on any member. By default,
set to ``False``.
smart_routing (bool): Enables smart mode for the client instead of
unisocket client. Smart clients send key based operations to owner of
the keys. Unisocket clients send all operations to a single node. By
default, set to ``True``.
ssl_enabled (bool): If it is ``True``, SSL is enabled. By default, set
to ``False``.
ssl_cafile (str): Absolute path of concatenated CA certificates used to
validate server's certificates in PEM format. When SSL is enabled and
cafile is not set, a set of default CA certificates from default
locations will be used.
ssl_certfile (str): Absolute path of the client certificate in PEM format.
ssl_keyfile (str): Absolute path of the private key file for the client
certificate in the PEM format. If this parameter is ``None``, private
key will be taken from the certfile.
ssl_password (str|bytes|bytearray|function): Password for decrypting the
keyfile if it is encrypted. The password may be a function to call to
get the password. It will be called with no arguments, and it should
return a string, bytes, or bytearray. If the return value is a string
it will be encoded as UTF-8 before using it to decrypt the key.
Alternatively a string, bytes, or bytearray value may be supplied
directly as the password.
ssl_protocol (int|str): Protocol version used in SSL communication.
By default, set to ``TLSv1_2``. See the
:class:`hazelcast.config.SSLProtocol` for possible values.
ssl_ciphers (str): String in the OpenSSL cipher list format to set the
available ciphers for sockets. More than one cipher can be set by
separating them with a colon.
cloud_discovery_token (str): Discovery token of the Hazelcast Cloud cluster.
When this value is set, Hazelcast Cloud discovery is enabled.
async_start (bool): Enables non-blocking start mode of :class:`HazelcastClient`.
When set to ``True``, the client creation will not wait to connect to
cluster. The client instance will throw exceptions until it connects to
cluster and becomes ready. If set to ``False``, :class:`HazelcastClient`
will block until a cluster connection established and it is ready to use
the client instance. By default, set to ``False``.
reconnect_mode (int|str): Defines how a client reconnects to cluster after a
disconnect. By default, set to ``ON``. See the
:class:`hazelcast.config.ReconnectMode` for possible values.
retry_initial_backoff (float): Wait period in seconds after the first failure
before retrying. Must be non-negative. By default, set to ``1.0``.
retry_max_backoff (float): Upper bound for the backoff interval in seconds.
Must be non-negative. By default, set to ``30.0``.
retry_jitter (float): Defines how much to randomize backoffs. At each iteration
the calculated back-off is randomized via following method in pseudo-code
``Random(-jitter * current_backoff, jitter * current_backoff)``. Must be
in range ``[0.0, 1.0]``. By default, set to ``0.0`` (no randomization).
retry_multiplier (float): The factor with which to multiply backoff after a
failed retry. Must be greater than or equal to ``1``. By default,
set to ``1.05``.
cluster_connect_timeout (float): Timeout value in seconds for the client to
give up connecting to the cluster. Must be non-negative or
equal to `-1`. By default, set to `-1`. `-1` means that the client
will not stop trying to the target cluster. (infinite timeout)
portable_version (int): Default value for the portable version if the
class does not have the :func:`get_portable_version` method. Portable
versions are used to differentiate two versions of the
:class:`hazelcast.serialization.api.Portable` classes that have added
or removed fields, or fields with different types.
data_serializable_factories (dict[int, dict[int, class]]): Dictionary of
factory id and corresponding
:class:`hazelcast.serialization.api.IdentifiedDataSerializable` factories.
A factory is simply a dictionary with class id and callable class
constructors.
.. code-block:: python
FACTORY_ID = 1
CLASS_ID = 1
class SomeSerializable(IdentifiedDataSerializable):
# omitting the implementation
pass
client = HazelcastClient(data_serializable_factories={
FACTORY_ID: {
CLASS_ID: SomeSerializable
}
})
portable_factories (dict[int, dict[int, class]]): Dictionary of
factory id and corresponding
:class:`hazelcast.serialization.api.Portable` factories.
A factory is simply a dictionary with class id and callable class
constructors.
.. code-block:: python
FACTORY_ID = 2
CLASS_ID = 2
class SomeSerializable(Portable):
# omitting the implementation
pass
client = HazelcastClient(portable_factories={
FACTORY_ID: {
CLASS_ID: SomeSerializable
}
})
class_definitions (`list[hazelcast.serialization.portable.classdef.ClassDefinition]`):
List of all portable class definitions.
check_class_definition_errors (bool): When enabled, serialization system
will check for class definitions error at start and throw an
``HazelcastSerializationError`` with error definition. By default,
set to ``True``.
is_big_endian (bool): Defines if big-endian is used as the byte order
for the serialization. By default, set to ``True``.
default_int_type (int|str): Defines how the ``int``/``long`` type is
represented on the cluster side. By default, it is serialized
as ``INT`` (``32`` bits). See the
:class:`hazelcast.config.IntType` for possible values.
global_serializer (hazelcast.serialization.api.StreamSerializer):
Defines the global serializer. This serializer
is registered as a fallback serializer to handle all other objects
if a serializer cannot be located for them.
custom_serializers (dict[class, hazelcast.serialization.api.StreamSerializer]):
Dictionary of class and the corresponding custom serializers.
.. code-block:: python
class SomeClass(object):
# omitting the implementation
pass
class SomeClassSerializer(StreamSerializer):
# omitting the implementation
pass
client = HazelcastClient(custom_serializers={
SomeClass: SomeClassSerializer
})
near_caches (dict[str, dict[str, any]]): Dictionary of near cache
names and the corresponding near cache configurations as a dictionary.
The near cache configurations contains the following options.
When an option is missing from the configuration, it will be
set to its default value.
- **invalidate_on_change** (bool): Enables cluster-assisted invalidate
on change behavior. When set to ``True``, entries are invalidated
when they are changed in cluster. By default, set to ``True``.
- **in_memory_format** (int|str): Specifies in which format data will be
stored in the Near Cache. By default, set to ``BINARY``. See the
:class:`hazelcast.config.InMemoryFormat` for possible values.
- **time_to_live** (float): Maximum number of seconds that an entry can
stay in cache. When not set, entries won't be evicted due
to expiration.
- **max_idle** (float): Maximum number of seconds that an entry can stay
in the Near Cache until it is accessed. When not set, entries won't be
evicted due to inactivity.
- **eviction_policy** (int|str): Defines eviction policy configuration.
By default, set to ``LRU``. See the
:class:`hazelcast.config.EvictionPolicy` for possible values.
- **eviction_max_size** (int): Defines maximum number of entries kept in
the memory before eviction kicks in. By default, set to ``10000``.
- **eviction_sampling_count** (int): Number of random entries that are
evaluated to see if some of them are already expired. By default,
set to ``8``.
- **eviction_sampling_pool_size** (int): Size of the pool for eviction
candidates. The pool is kept sorted according to the eviction policy.
By default, set to ``16``.
load_balancer (hazelcast.util.LoadBalancer): Load balancer implementation
for the client
membership_listeners (`list[tuple[function, function]]`): List of membership
listener tuples. Tuples must be of size ``2``. The first element will
be the function to be called when a member is added, and the second
element will be the function to be called when the member is removed
with the :class:`hazelcast.core.MemberInfo` as the only parameter.
Any of the elements can be ``None``, but not at the same time.
lifecycle_listeners (`list[function]`): List of lifecycle listeners.
Listeners will be called with the new lifecycle state as the only
parameter when the client changes lifecycle states.
flake_id_generators (dict[str, dict[str, any]]): Dictionary of flake id
generator names and the corresponding flake id generator configurations
as a dictionary. The flake id generator configurations contains the
following options. When an option is missing from the configuration, it
will be set to its default value.
- **prefetch_count** (int): Defines how many IDs are pre-fetched on the
background when a new flake id is requested from the cluster. Should
be in the range ``1..100000``. By default, set to ``100``.
- **prefetch_validity** (float): Defines for how long the pre-fetched IDs
can be used. If this time elapsed, a new batch of IDs will be fetched.
Time unit is in seconds. By default, set to ``600`` (10 minutes).
The IDs contain timestamp component, which ensures rough global ordering
of IDs. If an ID is assigned to an object that was created much later,
it will be much out of order. If you don't care about ordering, set this
value to ``0`` for unlimited ID validity.
reliable_topics (dict[str, dict[str, any]]): Dictionary of reliable
topic names and the corresponding reliable topic configurations as
a dictionary. The reliable topic configurations contain the
following options. When an option is missing from the
configuration, it will be set to its default value.
- **overload_policy** (int|str): Policy to handle an overloaded
topic. By default, set to ``BLOCK``. See the
:class:`hazelcast.config.TopicOverloadPolicy` for possible values.
- **read_batch_size** (int): Number of messages the reliable topic
will try to read in batch. It will get at least one, but if
there are more available, then it will try to get more to
increase throughput. By default, set to ``10``.
labels (`list[str]`): Labels for the client to be sent to the cluster.
heartbeat_interval (float): Time interval between the heartbeats sent by the
client to the member nodes in seconds. By default, set to ``5.0``.
heartbeat_timeout (float): If there is no message passing between the client
and a member within the given time via this property in seconds, the
connection will be closed. By default, set to ``60.0``.
invocation_timeout (float): When an invocation gets an exception because
- Member throws an exception.
- Connection between the client and member is closed.
- Client's heartbeat requests are timed out.
Time passed since invocation started is compared with this property.
If the time is already passed, then the exception is delegated to the user.
If not, the invocation is retried. Note that, if invocation gets no exception
and it is a long running one, then it will not get any exception, no matter
how small this timeout is set. Time unit is in seconds. By default,
set to ``120.0``.
invocation_retry_pause (float): Pause time between each retry cycle of an
invocation in seconds. By default, set to ``1.0``.
statistics_enabled (bool): When set to ``True``, client statistics collection
is enabled. By default, set to ``False``.
statistics_period (float): The period in seconds the statistics run.
shuffle_member_list (bool): Client shuffles the given member list to prevent
all clients to connect to the same node when this property is set to
``True``. When it is set to ``False``, the client tries to connect to
the nodes in the given order. By default, set to ``True``.
backup_ack_to_client_enabled (bool): Enables client to get backup
acknowledgements directly from the member that backups are applied,
which reduces number of hops and increases performance for smart clients.
This option has no effect for unisocket clients. By default, set
to ``True`` (enabled).
operation_backup_timeout (float): If an operation has backups, defines
how long the invocation will wait for acks from the backup replicas
in seconds. If acks are not received from some backups, there won't
be any rollback on other successful replicas. By default, set to ``5.0``.
fail_on_indeterminate_operation_state (bool): When enabled, if an operation
has sync backups and acks are not received from backup replicas in time,
or the member which owns primary replica of the target partition leaves
the cluster, then the invocation fails with
:class:`hazelcast.errors.IndeterminateOperationStateError`. However,
even if the invocation fails, there will not be any rollback on other
successful replicas. By default, set to ``False`` (do not fail).
"""
_CLIENT_ID = AtomicInteger()
def __init__(self, **kwargs):
config = _Config.from_dict(kwargs)
self._config = config
self._context = _ClientContext()
client_id = HazelcastClient._CLIENT_ID.get_and_increment()
self._name = self._create_client_name(client_id)
self._reactor = AsyncoreReactor()
self._serialization_service = SerializationServiceV1(config)
self._near_cache_manager = NearCacheManager(
config, self._serialization_service)
self._internal_lifecycle_service = _InternalLifecycleService(config)
self._lifecycle_service = LifecycleService(
self._internal_lifecycle_service)
self._invocation_service = InvocationService(self, config,
self._reactor)
self._address_provider = self._create_address_provider()
self._internal_partition_service = _InternalPartitionService(self)
self._partition_service = PartitionService(
self._internal_partition_service, self._serialization_service)
self._internal_cluster_service = _InternalClusterService(self, config)
self._cluster_service = ClusterService(self._internal_cluster_service)
self._connection_manager = ConnectionManager(
self,
config,
self._reactor,
self._address_provider,
self._internal_lifecycle_service,
self._internal_partition_service,
self._internal_cluster_service,
self._invocation_service,
self._near_cache_manager,
)
self._load_balancer = self._init_load_balancer(config)
self._listener_service = ListenerService(self, config,
self._connection_manager,
self._invocation_service)
self._proxy_manager = ProxyManager(self._context)
self._cp_subsystem = CPSubsystem(self._context)
self._proxy_session_manager = ProxySessionManager(self._context)
self._transaction_manager = TransactionManager(self._context)
self._lock_reference_id_generator = AtomicInteger(1)
self._statistics = Statistics(
self,
config,
self._reactor,
self._connection_manager,
self._invocation_service,
self._near_cache_manager,
)
self._cluster_view_listener = ClusterViewListenerService(
self,
self._connection_manager,
self._internal_partition_service,
self._internal_cluster_service,
self._invocation_service,
)
self._shutdown_lock = threading.RLock()
self._invocation_service.init(self._internal_partition_service,
self._connection_manager,
self._listener_service)
self._internal_sql_service = _InternalSqlService(
self._connection_manager, self._serialization_service,
self._invocation_service)
self.sql = SqlService(self._internal_sql_service)
self._init_context()
self._start()
def _init_context(self):
self._context.init_context(
self,
self._config,
self._invocation_service,
self._internal_partition_service,
self._internal_cluster_service,
self._connection_manager,
self._serialization_service,
self._listener_service,
self._proxy_manager,
self._near_cache_manager,
self._lock_reference_id_generator,
self._name,
self._proxy_session_manager,
self._reactor,
)
def _start(self):
self._reactor.start()
try:
self._internal_lifecycle_service.start()
self._invocation_service.start()
membership_listeners = self._config.membership_listeners
self._internal_cluster_service.start(self._connection_manager,
membership_listeners)
self._cluster_view_listener.start()
self._connection_manager.start(self._load_balancer)
sync_start = not self._config.async_start
if sync_start:
self._internal_cluster_service.wait_initial_member_list_fetched(
)
self._connection_manager.connect_to_all_cluster_members(sync_start)
self._listener_service.start()
self._invocation_service.add_backup_listener()
self._load_balancer.init(self._cluster_service)
self._statistics.start()
except:
self.shutdown()
raise
_logger.info("Client started")
def get_executor(self, name):
"""Creates cluster-wide ExecutorService.
Args:
name (str): Name of the Executor proxy.
Returns:
hazelcast.proxy.executor.Executor: Executor proxy for the given name.
"""
return self._proxy_manager.get_or_create(EXECUTOR_SERVICE, name)
def get_flake_id_generator(self, name):
"""Creates or returns a cluster-wide FlakeIdGenerator.
Args:
name (str): Name of the FlakeIdGenerator proxy.
Returns:
hazelcast.proxy.flake_id_generator.FlakeIdGenerator: FlakeIdGenerator proxy for the given name
"""
return self._proxy_manager.get_or_create(FLAKE_ID_GENERATOR_SERVICE,
name)
def get_queue(self, name):
"""Returns the distributed queue instance with the specified name.
Args:
name (str): Name of the distributed queue.
Returns:
hazelcast.proxy.queue.Queue: Distributed queue instance with the specified name.
"""
return self._proxy_manager.get_or_create(QUEUE_SERVICE, name)
def get_list(self, name):
"""Returns the distributed list instance with the specified name.
Args:
name (str): Name of the distributed list.
Returns:
hazelcast.proxy.list.List: Distributed list instance with the specified name.
"""
return self._proxy_manager.get_or_create(LIST_SERVICE, name)
def get_map(self, name):
"""Returns the distributed map instance with the specified name.
Args:
name (str): Name of the distributed map.
Returns:
hazelcast.proxy.map.Map: Distributed map instance with the specified name.
"""
return self._proxy_manager.get_or_create(MAP_SERVICE, name)
def get_multi_map(self, name):
"""Returns the distributed MultiMap instance with the specified name.
Args:
name (str): Name of the distributed MultiMap.
Returns:
hazelcast.proxy.multi_map.MultiMap: Distributed MultiMap instance with the specified name.
"""
return self._proxy_manager.get_or_create(MULTI_MAP_SERVICE, name)
def get_pn_counter(self, name):
"""Returns the PN Counter instance with the specified name.
Args:
name (str): Name of the PN Counter.
Returns:
hazelcast.proxy.pn_counter.PNCounter: The PN Counter.
"""
return self._proxy_manager.get_or_create(PN_COUNTER_SERVICE, name)
def get_reliable_topic(self, name):
"""Returns the ReliableTopic instance with the specified name.
Args:
name (str): Name of the ReliableTopic.
Returns:
hazelcast.proxy.reliable_topic.ReliableTopic: The ReliableTopic.
"""
return self._proxy_manager.get_or_create(RELIABLE_TOPIC_SERVICE, name)
def get_replicated_map(self, name):
"""Returns the distributed ReplicatedMap instance with the specified name.
Args:
name (str): Name of the distributed ReplicatedMap.
Returns:
hazelcast.proxy.replicated_map.ReplicatedMap: Distributed ReplicatedMap instance with the specified name.
"""
return self._proxy_manager.get_or_create(REPLICATED_MAP_SERVICE, name)
def get_ringbuffer(self, name):
"""Returns the distributed Ringbuffer instance with the specified name.
Args:
name (str): Name of the distributed Ringbuffer.
Returns:
hazelcast.proxy.ringbuffer.Ringbuffer: Distributed RingBuffer instance with the specified name.
"""
return self._proxy_manager.get_or_create(RINGBUFFER_SERVICE, name)
def get_set(self, name):
"""Returns the distributed Set instance with the specified name.
Args:
name (str): Name of the distributed Set.
Returns:
hazelcast.proxy.set.Set: Distributed Set instance with the specified name.
"""
return self._proxy_manager.get_or_create(SET_SERVICE, name)
def get_topic(self, name):
"""Returns the Topic instance with the specified name.
Args:
name (str): Name of the Topic.
Returns:
hazelcast.proxy.topic.Topic: The Topic.
"""
return self._proxy_manager.get_or_create(TOPIC_SERVICE, name)
def new_transaction(self, timeout=120, durability=1, type=TWO_PHASE):
"""Creates a new Transaction associated with the current thread using default or given options.
Args:
timeout (int): The timeout in seconds determines the maximum lifespan of a transaction. So if a
transaction is configured with a timeout of 2 minutes, then it will automatically rollback if it hasn't
committed yet.
durability (int): The durability is the number of machines that can take over if a member fails during a
transaction commit or rollback.
type (int): The transaction type which can be ``TWO_PHASE`` or ``ONE_PHASE``.
Returns:
hazelcast.transaction.Transaction: New Transaction associated with the current thread.
"""
return self._transaction_manager.new_transaction(
timeout, durability, type)
def add_distributed_object_listener(self, listener_func):
"""Adds a listener which will be notified when a new distributed object is created or destroyed.
Args:
listener_func (function): Function to be called when a distributed object is created or destroyed.
Returns:
hazelcast.future.Future[str]: A registration id which is used as a key to remove the listener.
"""
is_smart = self._config.smart_routing
codec = client_add_distributed_object_listener_codec
request = codec.encode_request(is_smart)
def handle_distributed_object_event(name, service_name, event_type,
source):
event = DistributedObjectEvent(name, service_name, event_type,
source)
listener_func(event)
def event_handler(client_message):
return codec.handle(client_message,
handle_distributed_object_event)
return self._listener_service.register_listener(
request,
codec.decode_response,
client_remove_distributed_object_listener_codec.encode_request,
event_handler,
)
def remove_distributed_object_listener(self, registration_id):
"""Removes the specified distributed object listener.
Returns silently if there is no such listener added before.
Args:
registration_id (str): The id of registered listener.
Returns:
hazelcast.future.Future[bool]: ``True`` if registration is removed, ``False`` otherwise.
"""
return self._listener_service.deregister_listener(registration_id)
def get_distributed_objects(self):
"""Returns all distributed objects such as; queue, map, set, list, topic, lock, multimap.
Also, as a side effect, it clears the local instances of the destroyed proxies.
Returns:
list[hazelcast.proxy.base.Proxy]: List of instances created by Hazelcast.
"""
request = client_get_distributed_objects_codec.encode_request()
invocation = Invocation(request, response_handler=lambda m: m)
self._invocation_service.invoke(invocation)
local_distributed_object_infos = {
DistributedObjectInfo(dist_obj.service_name, dist_obj.name)
for dist_obj in self._proxy_manager.get_distributed_objects()
}
response = client_get_distributed_objects_codec.decode_response(
invocation.future.result())
for dist_obj_info in response:
local_distributed_object_infos.discard(dist_obj_info)
self._proxy_manager.get_or_create(dist_obj_info.service_name,
dist_obj_info.name,
create_on_remote=False)
for dist_obj_info in local_distributed_object_infos:
self._proxy_manager.destroy_proxy(dist_obj_info.service_name,
dist_obj_info.name,
destroy_on_remote=False)
return self._proxy_manager.get_distributed_objects()
def shutdown(self):
"""Shuts down this HazelcastClient."""
with self._shutdown_lock:
if self._internal_lifecycle_service.running:
self._internal_lifecycle_service.fire_lifecycle_event(
LifecycleState.SHUTTING_DOWN)
self._internal_lifecycle_service.shutdown()
self._proxy_session_manager.shutdown().result()
self._near_cache_manager.destroy_near_caches()
self._connection_manager.shutdown()
self._invocation_service.shutdown()
self._statistics.shutdown()
self._reactor.shutdown()
self._internal_lifecycle_service.fire_lifecycle_event(
LifecycleState.SHUTDOWN)
@property
def name(self):
"""str: Name of the client."""
return self._name
@property
def lifecycle_service(self):
"""LifecycleService: Lifecycle service allows you to check if the
client is running and add and remove lifecycle listeners.
"""
return self._lifecycle_service
@property
def partition_service(self):
"""PartitionService: Partition service allows you to get partition
count, introspect the partition owners, and partition ids of keys.
"""
return self._partition_service
@property
def cluster_service(self):
"""ClusterService: Cluster service allows you to get the list of
the cluster members and add and remove membership listeners.
"""
return self._cluster_service
@property
def cp_subsystem(self):
"""CPSubsystem: CP Subsystem offers set of in-memory linearizable
data structures.
"""
return self._cp_subsystem
def _create_address_provider(self):
config = self._config
cluster_members = config.cluster_members
address_list_provided = len(cluster_members) > 0
cloud_discovery_token = config.cloud_discovery_token
cloud_enabled = cloud_discovery_token is not None
if address_list_provided and cloud_enabled:
raise IllegalStateError(
"Only one discovery method can be enabled at a time. "
"Cluster members given explicitly: %s, Hazelcast Cloud enabled: %s"
% (address_list_provided, cloud_enabled))
if cloud_enabled:
connection_timeout = self._get_connection_timeout(config)
return HazelcastCloudAddressProvider(cloud_discovery_token,
connection_timeout)
return DefaultAddressProvider(cluster_members)
def _create_client_name(self, client_id):
client_name = self._config.client_name
if client_name:
return client_name
return "hz.client_%s" % client_id
@staticmethod
def _get_connection_timeout(config):
timeout = config.connection_timeout
return sys.maxsize if timeout == 0 else timeout
@staticmethod
def _init_load_balancer(config):
load_balancer = config.load_balancer
if not load_balancer:
load_balancer = RoundRobinLB()
return load_balancer