def get_debug_test_context(is_standalone=False, manager_port=4670, egg_port=20001, transfer_port=20002, session_id='testing'):
manager_port = manager_port
egg_ports = [egg_port]
egg_transfer_ports = [transfer_port]
self_server_node_id = 2
options = {}
if is_standalone:
options[SessionConfKeys.CONFKEY_SESSION_DEPLOY_MODE] = "standalone"
options[TransferConfKeys.CONFKEY_TRANSFER_SERVICE_HOST] = "127.0.0.1"
options[TransferConfKeys.CONFKEY_TRANSFER_SERVICE_PORT] = str(transfer_port)
options[ClusterManagerConfKeys.CONFKEY_CLUSTER_MANAGER_PORT] = str(manager_port)
options[NodeManagerConfKeys.CONFKEY_NODE_MANAGER_PORT] = str(manager_port)
egg = ErProcessor(id=1,
server_node_id=self_server_node_id,
processor_type=ProcessorTypes.EGG_PAIR,
status=ProcessorStatus.RUNNING,
command_endpoint=ErEndpoint("127.0.0.1", egg_ports[0]),
transfer_endpoint=ErEndpoint("127.0.0.1",
egg_transfer_ports[0]))
roll = ErProcessor(id=1,
server_node_id=self_server_node_id,
processor_type=ProcessorTypes.ROLL_PAIR_MASTER,
status=ProcessorStatus.RUNNING,
command_endpoint=ErEndpoint("127.0.0.1", manager_port))
session = ErSession(session_id,
processors=[egg, roll],
options=options)
context = RollPairContext(session)
return context
def get_option(role,
self_party_id=None,
conf_file=default_props_file,
deploy_mode=DeployModes.CLUSTER):
print(f'conf file: {conf_file}')
configs = configparser.ConfigParser()
configs.read(conf_file)
eggroll_configs = configs['eggroll']
options = {}
if self_party_id is None:
party_id = eggroll_configs[
RollSiteConfKeys.EGGROLL_ROLLSITE_PARTY_ID.key]
else:
party_id = self_party_id
options['self_party_id'] = party_id
options['self_role'] = role
options['proxy_endpoint'] = \
ErEndpoint(host=eggroll_configs[RollSiteConfKeys.EGGROLL_ROLLSITE_HOST.key],
port=int(eggroll_configs[RollSiteConfKeys.EGGROLL_ROLLSITE_PORT.key]))
options[RollSiteConfKeys.EGGROLL_ROLLSITE_DEPLOY_MODE.key] = deploy_mode
return options
def __init__(self, options: dict = None):
if options is None:
options = {}
super().__init__(options)
er_partition = options['er_partition']
self.partition = er_partition
self.store_locator = er_partition._store_locator
self.partition_id = er_partition._id
self.namespace = self.store_locator._namespace
#_store_type = StoreTypes.ROLLPAIR_ROLLSITE
# self._store_locator = meta_pb2.StoreLocator(storeType=_store_type,
# namespace=self.namespace,
# name=self.store_locator._name,
# partitioner=self.store_locator._partitioner,
# serdes=self.store_locator._serdes,
# totalPartitions=self.store_locator._total_partitions)
self.roll_site_header_string = options.get('roll_site_header', None)
self.is_writable = False
if self.roll_site_header_string:
self.roll_site_header = ErRollSiteHeader.from_proto_string(
self.roll_site_header_string.encode(stringify_charset))
self.proxy_endpoint = ErEndpoint.from_proto_string(
options['proxy_endpoint'].encode(stringify_charset))
self.obj_type = options['obj_type']
self.is_writable = True
L.info(
f"writable RollSiteAdapter: {self.namespace}, {self.partition_id}. proxy_endpoint: {self.proxy_endpoint}, partition: {self.partition}"
)
def init_roll_site_context(runtime_conf, session_id):
from eggroll.roll_site.roll_site import RollSiteContext
from eggroll.roll_pair.roll_pair import RollPairContext
LOGGER.info("init_roll_site_context runtime_conf: {}".format(runtime_conf))
session_instance = FateSession.get_instance()._eggroll.get_session()
rp_context = RollPairContext(session_instance)
role = runtime_conf.get("local").get("role")
party_id = str(runtime_conf.get("local").get("party_id"))
_path = file_utils.get_project_base_directory(
) + "/arch/conf/server_conf.json"
server_conf = file_utils.load_json_conf(_path)
host = server_conf.get('servers').get('proxy').get("host")
port = server_conf.get('servers').get('proxy').get("port")
options = {
'self_role': role,
'self_party_id': party_id,
'proxy_endpoint': ErEndpoint(host, int(port))
}
rs_context = RollSiteContext(session_id,
rp_ctx=rp_context,
options=options)
LOGGER.info("init_roll_site_context done: {}".format(rs_context.__dict__))
return rp_context, rs_context
def __init__(self,
roll_site_session_id,
rp_ctx: RollPairContext,
options: dict = None):
if options is None:
options = {}
self.roll_site_session_id = roll_site_session_id
self.rp_ctx = rp_ctx
self.push_session_enabled = RollSiteConfKeys.EGGROLL_ROLLSITE_PUSH_SESSION_ENABLED.get_with(
options)
if self.push_session_enabled:
# create session for push roll_pair and object
self._push_session = ErSession(
session_id=roll_site_session_id + "_push",
options=rp_ctx.get_session().get_all_options())
self._push_rp_ctx = RollPairContext(session=self._push_session)
L.info(
f"push_session={self._push_session.get_session_id()} enabled")
def stop_push_session():
self._push_session.stop()
else:
self._push_session = None
self._push_rp_ctx = None
self.role = options["self_role"]
self.party_id = str(options["self_party_id"])
self._options = options
self._registered_comm_types = dict()
self.register_comm_type('grpc', RollSiteGrpc)
endpoint = options["proxy_endpoint"]
if isinstance(endpoint, str):
splitted = endpoint.split(':')
self.proxy_endpoint = ErEndpoint(host=splitted[0].strip(),
port=int(splitted[1].strip()))
elif isinstance(endpoint, ErEndpoint):
self.proxy_endpoint = endpoint
else:
raise ValueError("endpoint only support str and ErEndpoint type")
self.is_standalone = RollSiteConfKeys.EGGROLL_ROLLSITE_DEPLOY_MODE.get_with(
options) == "standalone"
# if self.is_standalone:
# self.stub = None
# else:
# channel = self.grpc_channel_factory.create_channel(self.proxy_endpoint)
# self.stub = proxy_pb2_grpc.DataTransferServiceStub(channel)
self.pushing_latch = CountDownLatch(0)
self.rp_ctx.get_session().add_exit_task(self._wait_push_complete)
if self.push_session_enabled:
self.rp_ctx.get_session().add_exit_task(stop_push_session)
self._wait_push_exit_timeout = int(
RollSiteConfKeys.EGGROLL_ROLLSITE_PUSH_OVERALL_TIMEOUT_SEC.
get_with(options))
L.info(f"inited RollSiteContext: {self.__dict__}")
def __init__(self, options: dict = None):
if options is None:
options = {}
self.__endpoint = ErEndpoint(
options[NodeManagerConfKeys.CONFKEY_NODE_MANAGER_HOST],
int(options[NodeManagerConfKeys.CONFKEY_NODE_MANAGER_PORT]))
if 'serdes_type' in options:
self.__serdes_type = options['serdes_type']
else:
self.__serdes_type = SerdesTypes.PROTOBUF
self.__command_client = CommandClient()
def __init__(self, options: dict = None):
if options is None:
options = {}
super().__init__(options)
er_partition = options['er_partition']
self.partition = er_partition
self.store_locator = er_partition._store_locator
self.partition_id = er_partition._id
self.namespace = self.store_locator._namespace
#_store_type = StoreTypes.ROLLPAIR_ROLLSITE
# self._store_locator = meta_pb2.StoreLocator(storeType=_store_type,
# namespace=self.namespace,
# name=self.store_locator._name,
# partitioner=self.store_locator._partitioner,
# serdes=self.store_locator._serdes,
# totalPartitions=self.store_locator._total_partitions)
self.roll_site_header_string = options.get('roll_site_header', None)
self.is_writable = False
if self.roll_site_header_string:
self.roll_site_header = ErRollSiteHeader.from_proto_string(
self.roll_site_header_string.encode(stringify_charset))
self.roll_site_header._options['partition_id'] = self.partition_id
self.proxy_endpoint = ErEndpoint.from_proto_string(
options['proxy_endpoint'].encode(stringify_charset))
self.obj_type = options['obj_type']
self.is_writable = True
L.trace(
f"writable RollSiteAdapter: {self.namespace}, partition_id={self.partition_id}. proxy_endpoint={self.proxy_endpoint}, partition={self.partition}"
)
self.unarycall_max_retry_cnt = int(
RollSiteConfKeys.EGGROLL_ROLLSITE_UNARYCALL_CLIENT_MAX_RETRY.
get_with(options))
self.push_max_retry_cnt = int(
RollSiteConfKeys.EGGROLL_ROLLSITE_PUSH_CLIENT_MAX_RETRY.get_with(
options))
self.push_overall_timeout = int(
RollSiteConfKeys.EGGROLL_ROLLSITE_OVERALL_TIMEOUT_SEC.get_with(
options))
self.push_completion_wait_timeout = int(
RollSiteConfKeys.EGGROLL_ROLLSITE_COMPLETION_WAIT_TIMEOUT_SEC.
get_with(options))
self.push_packet_interval_timeout = int(
RollSiteConfKeys.EGGROLL_ROLLSITE_PACKET_INTERVAL_TIMEOUT_SEC.
get_with(options))
def get_option(role, conf_file=default_props_file):
print(f'conf file: {conf_file}')
configs = configparser.ConfigParser()
configs.read(conf_file)
eggroll_configs = configs['eggroll']
options = {}
party_id = eggroll_configs[RollSiteConfKeys.EGGROLL_ROLLSITE_PARTY_ID.key]
options['self_party_id'] = party_id
options['self_role'] = role
options['proxy_endpoint'] = \
ErEndpoint(host=eggroll_configs[RollSiteConfKeys.EGGROLL_ROLLSITE_HOST.key],
port=int(eggroll_configs[RollSiteConfKeys.EGGROLL_ROLLSITE_PORT.key]))
return options
def test_send(self):
transfer_client = TransferClient()
broker = FifoBroker()
broker.put(b'hello')
broker.put(b'world')
broker.put(b'this')
broker.put(b'is')
broker.put(b'a')
broker.put(b'test')
broker.signal_write_finish()
future = transfer_client.send(broker=broker,
endpoint=ErEndpoint(host='localhost',
port=transfer_port),
tag='test')
future.result()
def __init__(self, rp_ctx, rs_session_id, party, proxy_endpoint):
LOGGER.debug(f"[federation.eggroll]init federation: "
f"rp_session_id={rp_ctx.session_id}, rs_session_id={rs_session_id}, "
f"party={party}, proxy_endpoint={proxy_endpoint}")
from eggroll.core.meta_model import ErEndpoint
if isinstance(proxy_endpoint, str):
splited = proxy_endpoint.split(':')
proxy_endpoint = ErEndpoint(host=splited[0].strip(), port=int(splited[1].strip()))
options = {
'self_role': party.role,
'self_party_id': party.party_id,
'proxy_endpoint': proxy_endpoint
}
self._rsc = RollSiteContext(rs_session_id, rp_ctx=rp_ctx, options=options)
LOGGER.debug(f"[federation.eggroll]init federation context done")
def __init__(self,
roll_site_session_id,
rp_ctx: RollPairContext,
options: dict = None):
if options is None:
options = {}
self.roll_site_session_id = roll_site_session_id
self.rp_ctx = rp_ctx
self.role = options["self_role"]
self.party_id = str(options["self_party_id"])
self._options = options
endpoint = options["proxy_endpoint"]
if isinstance(endpoint, str):
splitted = endpoint.split(':')
self.proxy_endpoint = ErEndpoint(host=splitted[0].strip(),
port=int(splitted[1].strip()))
elif isinstance(endpoint, ErEndpoint):
self.proxy_endpoint = endpoint
else:
raise ValueError("endpoint only support str and ErEndpoint type")
self.is_standalone = RollSiteConfKeys.EGGROLL_ROLLSITE_DEPLOY_MODE.get_with(
options) == "standalone"
if self.is_standalone:
self.stub = None
else:
channel = self.grpc_channel_factory.create_channel(
self.proxy_endpoint)
self.stub = proxy_pb2_grpc.DataTransferServiceStub(channel)
self.pushing_latch = CountDownLatch(0)
self.rp_ctx.get_session().add_exit_task(self._wait_push_complete)
self._wait_push_exit_timeout = int(
RollSiteConfKeys.EGGROLL_ROLLSITE_PUSH_OVERALL_TIMEOUT_SEC.
get_with(options))
L.info(f"inited RollSiteContext: {self.__dict__}")
def __init__(self, options=None):
if options is None:
options = {}
static_er_conf = get_static_er_conf()
host = options.get(
ClusterManagerConfKeys.CONFKEY_CLUSTER_MANAGER_HOST,
static_er_conf.get(
ClusterManagerConfKeys.CONFKEY_CLUSTER_MANAGER_HOST, None))
port = options.get(
ClusterManagerConfKeys.CONFKEY_CLUSTER_MANAGER_PORT,
static_er_conf.get(
ClusterManagerConfKeys.CONFKEY_CLUSTER_MANAGER_PORT, None))
if not host or not port:
raise ValueError(
f'failed to load host or port in creating cluster manager client. host: {host}, port: {port}'
)
self.__endpoint = ErEndpoint(host, int(port))
if 'serdes_type' in options:
self.__serdes_type = options['serdes_type']
else:
self.__serdes_type = SerdesTypes.PROTOBUF
self.__command_client = CommandClient()
options['self_party_id'] = party_id
options['self_role'] = role
options['proxy_endpoint'] = \
ErEndpoint(host=eggroll_configs[RollSiteConfKeys.EGGROLL_ROLLSITE_HOST.key],
port=int(eggroll_configs[RollSiteConfKeys.EGGROLL_ROLLSITE_PORT.key]))
return options
host_ip = 'localhost'
guest_ip = 'localhost'
host_options = {
'self_role': 'host',
'self_party_id': 10001,
'proxy_endpoint': ErEndpoint(host=host_ip, port=9395),
}
guest_options = {
'self_role': 'guest',
'self_party_id': 10002,
'proxy_endpoint': ErEndpoint(host=guest_ip, port=9396),
}
ER_STORE1 = ErStore(store_locator=ErStoreLocator(
store_type=StoreTypes.ROLLPAIR_LEVELDB, namespace="namespace",
name="name"))
roll_site_session_id = f'atest'
def serve(args):
prefix = 'v1/egg-pair'
set_data_dir(args.data_dir)
CommandRouter.get_instance().register(
service_name=f"{prefix}/runTask",
route_to_module_name="eggroll.roll_pair.egg_pair",
route_to_class_name="EggPair",
route_to_method_name="run_task")
max_workers = int(
RollPairConfKeys.
EGGROLL_ROLLPAIR_EGGPAIR_SERVER_EXECUTOR_POOL_MAX_SIZE.get())
executor_pool_type = CoreConfKeys.EGGROLL_CORE_DEFAULT_EXECUTOR_POOL.get()
command_server = grpc.server(
create_executor_pool(canonical_name=executor_pool_type,
max_workers=max_workers,
thread_name_prefix="eggpair-command-server"),
options=
[("grpc.max_metadata_size",
int(CoreConfKeys.
EGGROLL_CORE_GRPC_SERVER_CHANNEL_MAX_INBOUND_METADATA_SIZE.get())
),
('grpc.max_send_message_length',
int(CoreConfKeys.
EGGROLL_CORE_GRPC_SERVER_CHANNEL_MAX_INBOUND_MESSAGE_SIZE.get())
),
('grpc.max_receive_message_length',
int(CoreConfKeys.
EGGROLL_CORE_GRPC_SERVER_CHANNEL_MAX_INBOUND_MESSAGE_SIZE.get())
),
('grpc.keepalive_time_ms',
int(CoreConfKeys.CONFKEY_CORE_GRPC_CHANNEL_KEEPALIVE_TIME_SEC.get())
* 1000),
('grpc.keepalive_timeout_ms',
int(CoreConfKeys.
CONFKEY_CORE_GRPC_SERVER_CHANNEL_KEEPALIVE_TIMEOUT_SEC.get()) *
1000),
('grpc.keepalive_permit_without_calls',
int(CoreConfKeys.
CONFKEY_CORE_GRPC_SERVER_CHANNEL_KEEPALIVE_WITHOUT_CALLS_ENABLED.
get())),
('grpc.per_rpc_retry_buffer_size',
int(CoreConfKeys.CONFKEY_CORE_GRPC_SERVER_CHANNEL_RETRY_BUFFER_SIZE.
get())), ('grpc.so_reuseport', False)])
command_servicer = CommandServicer()
command_pb2_grpc.add_CommandServiceServicer_to_server(
command_servicer, command_server)
transfer_servicer = GrpcTransferServicer()
port = args.port
transfer_port = args.transfer_port
port = command_server.add_insecure_port(f'[::]:{port}')
if transfer_port == "-1":
transfer_server = command_server
transfer_port = port
transfer_pb2_grpc.add_TransferServiceServicer_to_server(
transfer_servicer, transfer_server)
else:
transfer_server_max_workers = int(
RollPairConfKeys.
EGGROLL_ROLLPAIR_EGGPAIR_DATA_SERVER_EXECUTOR_POOL_MAX_SIZE.get())
transfer_server = grpc.server(
create_executor_pool(canonical_name=executor_pool_type,
max_workers=transfer_server_max_workers,
thread_name_prefix="transfer_server"),
options=
[('grpc.max_metadata_size',
int(CoreConfKeys.
EGGROLL_CORE_GRPC_SERVER_CHANNEL_MAX_INBOUND_METADATA_SIZE.
get())),
('grpc.max_send_message_length',
int(CoreConfKeys.
EGGROLL_CORE_GRPC_SERVER_CHANNEL_MAX_INBOUND_MESSAGE_SIZE.
get())),
('grpc.max_receive_message_length',
int(CoreConfKeys.
EGGROLL_CORE_GRPC_SERVER_CHANNEL_MAX_INBOUND_MESSAGE_SIZE.
get())),
('grpc.keepalive_time_ms',
int(CoreConfKeys.
CONFKEY_CORE_GRPC_SERVER_CHANNEL_KEEPALIVE_WITHOUT_CALLS_ENABLED
.get()) * 1000),
('grpc.keepalive_timeout_ms',
int(CoreConfKeys.
CONFKEY_CORE_GRPC_SERVER_CHANNEL_KEEPALIVE_TIMEOUT_SEC.get())
* 1000),
('grpc.keepalive_permit_without_calls',
int(CoreConfKeys.
CONFKEY_CORE_GRPC_SERVER_CHANNEL_KEEPALIVE_WITHOUT_CALLS_ENABLED
.get())),
('grpc.per_rpc_retry_buffer_size',
int(CoreConfKeys.
CONFKEY_CORE_GRPC_SERVER_CHANNEL_RETRY_BUFFER_SIZE.get())),
('grpc.so_reuseport', False)])
transfer_port = transfer_server.add_insecure_port(
f'[::]:{transfer_port}')
transfer_pb2_grpc.add_TransferServiceServicer_to_server(
transfer_servicer, transfer_server)
transfer_server.start()
pid = os.getpid()
L.info(
f"starting egg_pair service, port: {port}, transfer port: {transfer_port}, pid: {pid}"
)
command_server.start()
cluster_manager = args.cluster_manager
myself = None
cluster_manager_client = None
if cluster_manager:
session_id = args.session_id
server_node_id = int(args.server_node_id)
static_er_conf = get_static_er_conf()
static_er_conf['server_node_id'] = server_node_id
if not session_id:
raise ValueError('session id is missing')
options = {SessionConfKeys.CONFKEY_SESSION_ID: args.session_id}
myself = ErProcessor(id=int(args.processor_id),
server_node_id=server_node_id,
processor_type=ProcessorTypes.EGG_PAIR,
command_endpoint=ErEndpoint(host='localhost',
port=port),
transfer_endpoint=ErEndpoint(host='localhost',
port=transfer_port),
pid=pid,
options=options,
status=ProcessorStatus.RUNNING)
cluster_manager_host, cluster_manager_port = cluster_manager.strip(
).split(':')
L.info(f'egg_pair cluster_manager: {cluster_manager}')
cluster_manager_client = ClusterManagerClient(
options={
ClusterManagerConfKeys.CONFKEY_CLUSTER_MANAGER_HOST:
cluster_manager_host,
ClusterManagerConfKeys.CONFKEY_CLUSTER_MANAGER_PORT:
cluster_manager_port
})
cluster_manager_client.heartbeat(myself)
if platform.system() == "Windows":
t1 = threading.Thread(target=stop_processor,
args=[cluster_manager_client, myself])
t1.start()
L.info(f'egg_pair started at port={port}, transfer_port={transfer_port}')
run = True
def exit_gracefully(signum, frame):
nonlocal run
run = False
L.info(
f'egg_pair {args.processor_id} at port={port}, transfer_port={transfer_port}, pid={pid} receives signum={signal.getsignal(signum)}, stopping gracefully.'
)
signal.signal(signal.SIGTERM, exit_gracefully)
signal.signal(signal.SIGINT, exit_gracefully)
while run:
time.sleep(1)
L.info(f'sending exit heartbeat to cm')
if cluster_manager:
myself._status = ProcessorStatus.STOPPED
cluster_manager_client.heartbeat(myself)
GrpcChannelFactory.shutdown_all_now()
L.info(f'closing RocksDB open dbs')
#todo:1: move to RocksdbAdapter and provide a cleanup method
from eggroll.core.pair_store.rocksdb import RocksdbAdapter
for path, db in RocksdbAdapter.db_dict.items():
del db
gc.collect()
L.info(f'system metric at exit: {get_system_metric(1)}')
L.info(
f'egg_pair {args.processor_id} at port={port}, transfer_port={transfer_port}, pid={pid} stopped gracefully'
)
# See the License for the specific language governing permissions and
# limitations under the License.
import queue
import threading
import unittest
from concurrent.futures import ThreadPoolExecutor
from eggroll.core.conf_keys import TransferConfKeys
from eggroll.core.datastructure.broker import FifoBroker, BrokerClosed
from eggroll.core.meta_model import ErEndpoint
from eggroll.core.transfer.transfer_service import TransferService, \
GrpcTransferService, TransferClient
transfer_port = 20002
transfer_endpont = ErEndpoint('localhost', transfer_port)
class TestTransfer(unittest.TestCase):
def setUp(self) -> None:
self.__executor_pool = ThreadPoolExecutor(max_workers=5)
def test_recv(self):
def start_server():
transfer_service = GrpcTransferService()
options = {
TransferConfKeys.CONFKEY_TRANSFER_SERVICE_PORT: transfer_port
}
transfer_service.start(options=options)
def serve(args):
prefix = 'v1/egg-pair'
set_data_dir(args.data_dir)
CommandRouter.get_instance().register(
service_name=f"{prefix}/runTask",
route_to_module_name="eggroll.roll_pair.egg_pair",
route_to_class_name="EggPair",
route_to_method_name="run_task")
command_server = grpc.server(
futures.ThreadPoolExecutor(max_workers=500,
thread_name_prefix="grpc_server"),
options=[("grpc.max_metadata_size", 32 << 20),
(cygrpc.ChannelArgKey.max_send_message_length, 2 << 30 - 1),
(cygrpc.ChannelArgKey.max_receive_message_length, 2 << 30 - 1)
])
command_servicer = CommandServicer()
command_pb2_grpc.add_CommandServiceServicer_to_server(
command_servicer, command_server)
transfer_servicer = GrpcTransferServicer()
port = args.port
transfer_port = args.transfer_port
port = command_server.add_insecure_port(f'[::]:{port}')
if transfer_port == "-1":
transfer_server = command_server
transfer_port = port
transfer_pb2_grpc.add_TransferServiceServicer_to_server(
transfer_servicer, transfer_server)
else:
transfer_server = grpc.server(
futures.ThreadPoolExecutor(max_workers=500,
thread_name_prefix="transfer_server"),
options=[
(cygrpc.ChannelArgKey.max_send_message_length, 2 << 30 - 1),
(cygrpc.ChannelArgKey.max_receive_message_length, 2 << 30 - 1),
('grpc.max_metadata_size', 32 << 20)
])
transfer_port = transfer_server.add_insecure_port(
f'[::]:{transfer_port}')
transfer_pb2_grpc.add_TransferServiceServicer_to_server(
transfer_servicer, transfer_server)
transfer_server.start()
L.info(
f"starting egg_pair service, port:{port}, transfer port: {transfer_port}"
)
command_server.start()
cluster_manager = args.cluster_manager
myself = None
cluster_manager_client = None
if cluster_manager:
session_id = args.session_id
if not session_id:
raise ValueError('session id is missing')
options = {SessionConfKeys.CONFKEY_SESSION_ID: args.session_id}
myself = ErProcessor(id=int(args.processor_id),
server_node_id=int(args.server_node_id),
processor_type=ProcessorTypes.EGG_PAIR,
command_endpoint=ErEndpoint(host='localhost',
port=port),
transfer_endpoint=ErEndpoint(host='localhost',
port=transfer_port),
pid=os.getpid(),
options=options,
status=ProcessorStatus.RUNNING)
cluster_manager_host, cluster_manager_port = cluster_manager.strip(
).split(':')
L.info(f'cluster_manager: {cluster_manager}')
cluster_manager_client = ClusterManagerClient(
options={
ClusterManagerConfKeys.CONFKEY_CLUSTER_MANAGER_HOST:
cluster_manager_host,
ClusterManagerConfKeys.CONFKEY_CLUSTER_MANAGER_PORT:
cluster_manager_port
})
cluster_manager_client.heartbeat(myself)
L.info(f'egg_pair started at port {port}, transfer_port {transfer_port}')
run = True
def exit_gracefully(signum, frame):
nonlocal run
run = False
signal.signal(signal.SIGTERM, exit_gracefully)
signal.signal(signal.SIGINT, exit_gracefully)
import time
while run:
time.sleep(1)
if cluster_manager:
myself._status = ProcessorStatus.STOPPED
cluster_manager_client.heartbeat(myself)
L.info(
f'egg_pair at port {port}, transfer_port {transfer_port} stopped gracefully'
)
