def __init__(self, listen_port, peer_addr, master_addr, rank_id, etcd_name,
etcd_base_dir, etcd_addrs, options):
master_channel = make_insecure_channel(
master_addr,
ChannelType.INTERNAL,
options=[('grpc.max_send_message_length', 2**31 - 1),
('grpc.max_receive_message_length', 2**31 - 1)])
self._master_client = dj_grpc.DataJoinMasterServiceStub(master_channel)
self._rank_id = rank_id
etcd = EtcdClient(etcd_name, etcd_addrs, etcd_base_dir,
options.use_mock_etcd)
data_source = self._sync_data_source()
self._data_source_name = data_source.data_source_meta.name
self._listen_port = listen_port
self._server = grpc.server(futures.ThreadPoolExecutor(max_workers=10))
peer_channel = make_insecure_channel(
peer_addr,
ChannelType.REMOTE,
options=[('grpc.max_send_message_length', 2**31 - 1),
('grpc.max_receive_message_length', 2**31 - 1)])
peer_client = dj_grpc.DataJoinWorkerServiceStub(peer_channel)
self._data_join_worker = DataJoinWorker(peer_client,
self._master_client, rank_id,
etcd, data_source, options)
dj_grpc.add_DataJoinWorkerServiceServicer_to_server(
self._data_join_worker, self._server)
self._role_repr = "leader" if data_source.role == \
common_pb.FLRole.Leader else "follower"
self._server.add_insecure_port('[::]:%d' % listen_port)
self._server_started = False
def __init__(self, listen_port, peer_addr, master_addr, rank_id, etcd_name,
etcd_base_dir, etcd_addrs, options):
master_channel = make_insecure_channel(master_addr,
ChannelType.INTERNAL)
master_client = dj_grpc.DataJoinMasterServiceStub(master_channel)
etcd = EtcdClient(etcd_name, etcd_addrs, etcd_base_dir)
data_source = self.sync_data_source(master_client)
self._data_source_name = data_source.data_source_meta.name
self._listen_port = listen_port
self._rank_id = rank_id
self._server = grpc.server(futures.ThreadPoolExecutor(max_workers=10))
peer_channel = make_insecure_channel(peer_addr, ChannelType.REMOTE)
if data_source.role == common_pb.FLRole.Leader:
self._role_repr = "leader"
peer_client = dj_grpc.DataJoinFollowerServiceStub(peer_channel)
self._diw = data_join_leader.DataJoinLeader(
peer_client, master_client, rank_id, etcd, data_source,
options)
dj_grpc.add_DataJoinLeaderServiceServicer_to_server(
self._diw, self._server)
else:
assert data_source.role == common_pb.FLRole.Follower
self._role_repr = "follower"
peer_client = dj_grpc.DataJoinLeaderServiceStub(peer_channel)
self._diw = data_join_follower.DataJoinFollower(
peer_client, master_client, rank_id, etcd, data_source,
options)
dj_grpc.add_DataJoinFollowerServiceServicer_to_server(
self._diw, self._server)
self._server.add_insecure_port('[::]:%d' % listen_port)
self._server_started = False
def __init__(self, master_addr):
self._master_addr = master_addr
channel = make_insecure_channel(master_addr, ChannelType.INTERNAL)
self._master_cli = dj_grpc.DataJoinMasterServiceStub(channel)
self._data_source = None
self._raw_date_ctl = None
self._raw_data_updated_datetime = {}
def __init__(self, listen_port, peer_addr, data_source_name, etcd_name,
etcd_base_dir, etcd_addrs):
channel = make_insecure_channel(peer_addr, ChannelType.REMOTE)
peer_client = dj_grpc.DataJoinMasterServiceStub(channel)
self._data_source_name = data_source_name
self._listen_port = listen_port
self._server = grpc.server(futures.ThreadPoolExecutor(max_workers=10))
self._dim = DataJoinMaster(peer_client, data_source_name, etcd_name,
etcd_addrs, etcd_base_dir)
dj_grpc.add_DataJoinMasterServiceServicer_to_server(
self._dim, self._server)
self._server.add_insecure_port('[::]:%d' % listen_port)
self._server_started = False
def _launch_masters(self):
self._master_addr_l = 'localhost:4061'
self._master_addr_f = 'localhost:4062'
master_options = dj_pb.DataJoinMasterOptions(use_mock_etcd=True)
self._master_l = data_join_master.DataJoinMasterService(
int(self._master_addr_l.split(':')[1]), self._master_addr_f,
self._data_source_name, self._db_database, self._db_base_dir_l,
self._db_addr, self._db_username_l, self._db_password_l,
master_options)
self._master_f = data_join_master.DataJoinMasterService(
int(self._master_addr_f.split(':')[1]), self._master_addr_l,
self._data_source_name, self._db_database, self._db_base_dir_f,
self._db_addr, self._db_username_f, self._db_password_f,
master_options)
self._master_f.start()
self._master_l.start()
channel_l = make_insecure_channel(self._master_addr_l,
ChannelType.INTERNAL)
self._master_client_l = dj_grpc.DataJoinMasterServiceStub(channel_l)
channel_f = make_insecure_channel(self._master_addr_f,
ChannelType.INTERNAL)
self._master_client_f = dj_grpc.DataJoinMasterServiceStub(channel_f)
while True:
req_l = dj_pb.DataSourceRequest(
data_source_meta=self._data_source_l.data_source_meta)
req_f = dj_pb.DataSourceRequest(
data_source_meta=self._data_source_f.data_source_meta)
dss_l = self._master_client_l.GetDataSourceStatus(req_l)
dss_f = self._master_client_f.GetDataSourceStatus(req_f)
self.assertEqual(dss_l.role, common_pb.FLRole.Leader)
self.assertEqual(dss_f.role, common_pb.FLRole.Follower)
if dss_l.state == common_pb.DataSourceState.Processing and \
dss_f.state == common_pb.DataSourceState.Processing:
break
else:
time.sleep(2)
logging.info("masters turn into Processing state")
def __init__(self, listen_port, peer_addr, data_source_name, kvstore_type,
options):
channel = make_insecure_channel(
peer_addr,
ChannelType.REMOTE,
options=[('grpc.max_send_message_length', 2**31 - 1),
('grpc.max_receive_message_length', 2**31 - 1)])
peer_client = dj_grpc.DataJoinMasterServiceStub(channel)
self._data_source_name = data_source_name
self._listen_port = listen_port
self._server = grpc.server(futures.ThreadPoolExecutor(max_workers=10))
self._data_join_master = DataJoinMaster(peer_client, data_source_name,
kvstore_type, options)
dj_grpc.add_DataJoinMasterServiceServicer_to_server(
self._data_join_master, self._server)
self._server.add_insecure_port('[::]:%d' % listen_port)
self._server_started = False
def setUp(self):
etcd_name = 'test_etcd'
etcd_addrs = 'localhost:2379'
etcd_base_dir_l = 'byefl_l'
etcd_base_dir_f = 'byefl_f'
data_source_name = 'test_data_source'
etcd_l = EtcdClient(etcd_name, etcd_addrs, etcd_base_dir_l, True)
etcd_f = EtcdClient(etcd_name, etcd_addrs, etcd_base_dir_f, True)
etcd_l.delete_prefix(
common.data_source_etcd_base_dir(data_source_name))
etcd_f.delete_prefix(
common.data_source_etcd_base_dir(data_source_name))
data_source_l = common_pb.DataSource()
self.raw_data_pub_dir_l = './raw_data_pub_dir_l'
data_source_l.raw_data_sub_dir = self.raw_data_pub_dir_l
data_source_l.role = common_pb.FLRole.Leader
data_source_l.state = common_pb.DataSourceState.Init
data_source_l.data_block_dir = "./data_block_l"
data_source_l.raw_data_dir = "./raw_data_l"
data_source_l.example_dumped_dir = "./example_dumped_l"
data_source_f = common_pb.DataSource()
self.raw_data_pub_dir_f = './raw_data_pub_dir_f'
data_source_f.role = common_pb.FLRole.Follower
data_source_f.raw_data_sub_dir = self.raw_data_pub_dir_f
data_source_f.state = common_pb.DataSourceState.Init
data_source_f.data_block_dir = "./data_block_f"
data_source_f.raw_data_dir = "./raw_data_f"
data_source_f.example_dumped_dir = "./example_dumped_f"
data_source_meta = common_pb.DataSourceMeta()
data_source_meta.name = data_source_name
data_source_meta.partition_num = 2
data_source_meta.start_time = 0
data_source_meta.end_time = 100000000
data_source_l.data_source_meta.MergeFrom(data_source_meta)
common.commit_data_source(etcd_l, data_source_l)
data_source_f.data_source_meta.MergeFrom(data_source_meta)
common.commit_data_source(etcd_f, data_source_f)
master_options = dj_pb.DataJoinMasterOptions(use_mock_etcd=True)
master_addr_l = 'localhost:4061'
master_addr_f = 'localhost:4062'
master_l = data_join_master.DataJoinMasterService(
int(master_addr_l.split(':')[1]),
master_addr_f,
data_source_name,
etcd_name,
etcd_base_dir_l,
etcd_addrs,
master_options,
)
master_l.start()
master_f = data_join_master.DataJoinMasterService(
int(master_addr_f.split(':')[1]), master_addr_l, data_source_name,
etcd_name, etcd_base_dir_f, etcd_addrs, master_options)
master_f.start()
channel_l = make_insecure_channel(master_addr_l, ChannelType.INTERNAL)
master_client_l = dj_grpc.DataJoinMasterServiceStub(channel_l)
channel_f = make_insecure_channel(master_addr_f, ChannelType.INTERNAL)
master_client_f = dj_grpc.DataJoinMasterServiceStub(channel_f)
while True:
req_l = dj_pb.DataSourceRequest(
data_source_meta=data_source_l.data_source_meta)
req_f = dj_pb.DataSourceRequest(
data_source_meta=data_source_f.data_source_meta)
dss_l = master_client_l.GetDataSourceStatus(req_l)
dss_f = master_client_f.GetDataSourceStatus(req_f)
self.assertEqual(dss_l.role, common_pb.FLRole.Leader)
self.assertEqual(dss_f.role, common_pb.FLRole.Follower)
if dss_l.state == common_pb.DataSourceState.Processing and \
dss_f.state == common_pb.DataSourceState.Processing:
break
else:
time.sleep(2)
self.master_client_l = master_client_l
self.master_client_f = master_client_f
self.master_addr_l = master_addr_l
self.master_addr_f = master_addr_f
self.etcd_l = etcd_l
self.etcd_f = etcd_f
self.data_source_l = data_source_l
self.data_source_f = data_source_f
self.master_l = master_l
self.master_f = master_f
self.data_source_name = data_source_name,
self.etcd_name = etcd_name
self.etcd_addrs = etcd_addrs
self.etcd_base_dir_l = etcd_base_dir_l
self.etcd_base_dir_f = etcd_base_dir_f
self.raw_data_publisher_l = raw_data_publisher.RawDataPublisher(
self.etcd_l, self.raw_data_pub_dir_l)
self.raw_data_publisher_f = raw_data_publisher.RawDataPublisher(
self.etcd_f, self.raw_data_pub_dir_f)
if gfile.Exists(data_source_l.data_block_dir):
gfile.DeleteRecursively(data_source_l.data_block_dir)
if gfile.Exists(data_source_l.example_dumped_dir):
gfile.DeleteRecursively(data_source_l.example_dumped_dir)
if gfile.Exists(data_source_l.raw_data_dir):
gfile.DeleteRecursively(data_source_l.raw_data_dir)
if gfile.Exists(data_source_f.data_block_dir):
gfile.DeleteRecursively(data_source_f.data_block_dir)
if gfile.Exists(data_source_f.example_dumped_dir):
gfile.DeleteRecursively(data_source_f.example_dumped_dir)
if gfile.Exists(data_source_f.raw_data_dir):
gfile.DeleteRecursively(data_source_f.raw_data_dir)
self.worker_options = dj_pb.DataJoinWorkerOptions(
use_mock_etcd=True,
raw_data_options=dj_pb.RawDataOptions(raw_data_iter='TF_RECORD',
compressed_type=''),
example_id_dump_options=dj_pb.ExampleIdDumpOptions(
example_id_dump_interval=1, example_id_dump_threshold=1024),
example_joiner_options=dj_pb.ExampleJoinerOptions(
example_joiner='STREAM_JOINER',
min_matching_window=64,
max_matching_window=256,
data_block_dump_interval=30,
data_block_dump_threshold=1000),
batch_processor_options=dj_pb.BatchProcessorOptions(
batch_size=512, max_flying_item=2048),
data_block_builder_options=dj_pb.WriterOptions(
output_writer='TF_RECORD'))
self.total_index = 1 << 13
def test_api(self):
etcd_name = 'test_etcd'
etcd_addrs = 'localhost:2379'
etcd_base_dir_l = 'byefl_l'
etcd_base_dir_f = 'byefl_f'
data_source_name = 'test_data_source'
etcd_l = EtcdClient(etcd_name, etcd_addrs, etcd_base_dir_l)
etcd_f = EtcdClient(etcd_name, etcd_addrs, etcd_base_dir_f)
etcd_l.delete_prefix(data_source_name)
etcd_f.delete_prefix(data_source_name)
data_source_l = common_pb.DataSource()
data_source_l.role = common_pb.FLRole.Leader
data_source_l.state = common_pb.DataSourceState.Init
data_source_l.data_block_dir = "./data_block_l"
data_source_l.raw_data_dir = "./raw_data_l"
data_source_l.example_dumped_dir = "./example_dumped_l"
data_source_f = common_pb.DataSource()
data_source_f.role = common_pb.FLRole.Follower
data_source_f.state = common_pb.DataSourceState.Init
data_source_f.data_block_dir = "./data_block_f"
data_source_f.raw_data_dir = "./raw_data_f"
data_source_f.example_dumped_dir = "./example_dumped_f"
data_source_meta = common_pb.DataSourceMeta()
data_source_meta.name = data_source_name
data_source_meta.partition_num = 1
data_source_meta.start_time = 0
data_source_meta.end_time = 100000000
data_source_meta.min_matching_window = 32
data_source_meta.max_matching_window = 1024
data_source_meta.data_source_type = common_pb.DataSourceType.Sequential
data_source_meta.max_example_in_data_block = 1000
data_source_l.data_source_meta.MergeFrom(data_source_meta)
etcd_l.set_data(os.path.join(data_source_name, 'master'),
text_format.MessageToString(data_source_l))
data_source_f.data_source_meta.MergeFrom(data_source_meta)
etcd_f.set_data(os.path.join(data_source_name, 'master'),
text_format.MessageToString(data_source_f))
master_addr_l = 'localhost:4061'
master_addr_f = 'localhost:4062'
master_l = data_join_master.DataJoinMasterService(
int(master_addr_l.split(':')[1]), master_addr_f, data_source_name,
etcd_name, etcd_base_dir_l, etcd_addrs)
master_l.start()
master_f = data_join_master.DataJoinMasterService(
int(master_addr_f.split(':')[1]), master_addr_l, data_source_name,
etcd_name, etcd_base_dir_f, etcd_addrs)
master_f.start()
channel_l = make_insecure_channel(master_addr_l, ChannelType.INTERNAL)
client_l = dj_grpc.DataJoinMasterServiceStub(channel_l)
channel_f = make_insecure_channel(master_addr_f, ChannelType.INTERNAL)
client_f = dj_grpc.DataJoinMasterServiceStub(channel_f)
while True:
rsp_l = client_l.GetDataSourceState(data_source_l.data_source_meta)
rsp_f = client_f.GetDataSourceState(data_source_f.data_source_meta)
self.assertEqual(rsp_l.status.code, 0)
self.assertEqual(rsp_l.role, common_pb.FLRole.Leader)
self.assertEqual(rsp_l.data_source_type,
common_pb.DataSourceType.Sequential)
self.assertEqual(rsp_f.status.code, 0)
self.assertEqual(rsp_f.role, common_pb.FLRole.Follower)
self.assertEqual(rsp_f.data_source_type,
common_pb.DataSourceType.Sequential)
if (rsp_l.state == common_pb.DataSourceState.Processing
and rsp_f.state == common_pb.DataSourceState.Processing):
break
else:
time.sleep(2)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
join_example=dj_pb.JoinExampleRequest(partition_id=-1))
rdrsp = client_l.RequestJoinPartition(rdreq)
self.assertTrue(rdrsp.status.code == 0)
self.assertFalse(rdrsp.HasField('manifest'))
self.assertFalse(rdrsp.HasField('finished'))
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
sync_example_id=dj_pb.SyncExampleIdRequest(partition_id=-1))
rdrsp = client_l.RequestJoinPartition(rdreq)
self.assertEqual(rdrsp.status.code, 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.state, dj_pb.Syncing)
self.assertEqual(rdrsp.manifest.allocated_rank_id, 0)
self.assertEqual(rdrsp.manifest.partition_id, 0)
rdrsp = client_l.RequestJoinPartition(rdreq)
self.assertEqual(rdrsp.status.code, 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.state, dj_pb.Syncing)
self.assertEqual(rdrsp.manifest.allocated_rank_id, 0)
self.assertEqual(rdrsp.manifest.partition_id, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
sync_example_id=dj_pb.SyncExampleIdRequest(partition_id=0))
rdrsp = client_f.RequestJoinPartition(rdreq)
self.assertEqual(rdrsp.status.code, 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.state, dj_pb.Syncing)
self.assertEqual(rdrsp.manifest.allocated_rank_id, 0)
self.assertEqual(rdrsp.manifest.partition_id, 0)
frreq = dj_pb.FinishRawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
sync_example_id=dj_pb.SyncExampleIdRequest(partition_id=0))
frrsp = client_l.FinishJoinPartition(frreq)
self.assertEqual(frrsp.code, 0)
rdrsp = client_l.FinishJoinPartition(rdreq)
self.assertEqual(frrsp.code, 0)
rdreq = dj_pb.FinishRawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
sync_example_id=dj_pb.SyncExampleIdRequest(partition_id=0))
frrsp = client_f.FinishJoinPartition(rdreq)
self.assertEqual(frrsp.code, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
join_example=dj_pb.JoinExampleRequest(partition_id=-1))
rdrsp = client_l.RequestJoinPartition(rdreq)
self.assertEqual(rdrsp.status.code, 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.state, dj_pb.Joining)
self.assertEqual(rdrsp.manifest.allocated_rank_id, 0)
self.assertEqual(rdrsp.manifest.partition_id, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
join_example=dj_pb.JoinExampleRequest(partition_id=0))
rdrsp = client_f.RequestJoinPartition(rdreq)
self.assertEqual(rdrsp.status.code, 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.state, dj_pb.Joining)
self.assertEqual(rdrsp.manifest.allocated_rank_id, 0)
self.assertEqual(rdrsp.manifest.partition_id, 0)
frreq = dj_pb.FinishRawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
join_example=dj_pb.JoinExampleRequest(partition_id=0))
frrsp = client_l.FinishJoinPartition(rdreq)
self.assertEqual(frrsp.code, 0)
frrsp = client_l.FinishJoinPartition(rdreq)
self.assertEqual(frrsp.code, 0)
frreq = dj_pb.FinishRawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
join_example=dj_pb.JoinExampleRequest(partition_id=0))
frrsp = client_f.FinishJoinPartition(rdreq)
self.assertEqual(frrsp.code, 0)
while True:
rsp_l = client_l.GetDataSourceState(data_source_l.data_source_meta)
rsp_f = client_f.GetDataSourceState(data_source_l.data_source_meta)
self.assertEqual(rsp_l.status.code, 0)
self.assertEqual(rsp_l.role, common_pb.FLRole.Leader)
self.assertEqual(rsp_l.data_source_type,
common_pb.DataSourceType.Sequential)
self.assertEqual(rsp_f.status.code, 0)
self.assertEqual(rsp_f.role, common_pb.FLRole.Follower)
self.assertEqual(rsp_f.data_source_type,
common_pb.DataSourceType.Sequential)
if (rsp_l.state == common_pb.DataSourceState.Finished
and rsp_f.state == common_pb.DataSourceState.Finished):
break
else:
time.sleep(2)
master_l.stop()
master_f.stop()
help='the addr(uuid) of local data_join_master')
parser.add_argument('partition_id', type=int,
help='the partition to control')
parser.add_argument('--files', type=str, nargs='+',
help='the need raw data fnames')
parser.add_argument('--src_dir', type=str,
help='the directory of input raw data. The input '\
'file sequence is sorted by file name and rank '\
'after raw data input by --files')
parser.add_argument('--dedup', action='store_true',
help='dedup the input files, otherwise, '\
'error if dup input files')
args = parser.parse_args()
master_channel = make_insecure_channel(args.master_addr,
ChannelType.INTERNAL)
master_cli = dj_grpc.DataJoinMasterServiceStub(master_channel)
data_src = master_cli.GetDataSource(empty_pb2.Empty())
rdc = RawDataController(data_src, master_cli)
if args.cmd == 'add':
all_fpaths = []
if args.files is not None:
for fp in args.files:
all_fpaths.append(fp)
if args.src_dir is not None:
dir_fpaths = \
[path.join(args.src_dir, f)
for f in gfile.ListDirectory(args.src_dir)
if not gfile.IsDirectory(path.join(args.src_dir, f))]
dir_fpaths.sort()
all_fpaths += dir_fpaths
if not all_fpaths:
def _inner_test_round(self, start_index):
for i in range(self.data_source_l.data_source_meta.partition_num):
self.generate_raw_data(
start_index, self.etcd_l, self.raw_data_publisher_l,
self.data_source_l, self.raw_data_dir_l, i, 2048, 64,
'leader_key_partition_{}'.format(i) + ':{}',
'leader_value_partition_{}'.format(i) + ':{}'
)
self.generate_raw_data(
start_index, self.etcd_f, self.raw_data_publisher_f,
self.data_source_f, self.raw_data_dir_f, i, 4096, 128,
'follower_key_partition_{}'.format(i) + ':{}',
'follower_value_partition_{}'.format(i) + ':{}'
)
master_addr_l = 'localhost:4061'
master_addr_f = 'localhost:4062'
master_options = dj_pb.DataJoinMasterOptions(use_mock_etcd=True,
batch_mode=True)
master_l = data_join_master.DataJoinMasterService(
int(master_addr_l.split(':')[1]), master_addr_f,
self.data_source_name, self.etcd_name, self.etcd_base_dir_l,
self.etcd_addrs, master_options,
)
master_l.start()
master_f = data_join_master.DataJoinMasterService(
int(master_addr_f.split(':')[1]), master_addr_l,
self.data_source_name, self.etcd_name, self.etcd_base_dir_f,
self.etcd_addrs, master_options
)
master_f.start()
channel_l = make_insecure_channel(master_addr_l, ChannelType.INTERNAL)
master_client_l = dj_grpc.DataJoinMasterServiceStub(channel_l)
channel_f = make_insecure_channel(master_addr_f, ChannelType.INTERNAL)
master_client_f = dj_grpc.DataJoinMasterServiceStub(channel_f)
while True:
try:
req_l = dj_pb.DataSourceRequest(
data_source_meta=self.data_source_l.data_source_meta
)
req_f = dj_pb.DataSourceRequest(
data_source_meta=self.data_source_f.data_source_meta
)
dss_l = master_client_l.GetDataSourceStatus(req_l)
dss_f = master_client_f.GetDataSourceStatus(req_f)
self.assertEqual(dss_l.role, common_pb.FLRole.Leader)
self.assertEqual(dss_f.role, common_pb.FLRole.Follower)
if dss_l.state == common_pb.DataSourceState.Processing and \
dss_f.state == common_pb.DataSourceState.Processing:
break
except Exception as e:
pass
time.sleep(2)
worker_addr_l = 'localhost:4161'
worker_addr_f = 'localhost:4162'
worker_l = data_join_worker.DataJoinWorkerService(
int(worker_addr_l.split(':')[1]),
worker_addr_f, master_addr_l, 0,
self.etcd_name, self.etcd_base_dir_l,
self.etcd_addrs, self.worker_options
)
worker_f = data_join_worker.DataJoinWorkerService(
int(worker_addr_f.split(':')[1]),
worker_addr_l, master_addr_f, 0,
self.etcd_name, self.etcd_base_dir_f,
self.etcd_addrs, self.worker_options
)
th_l = threading.Thread(target=worker_l.run, name='worker_l')
th_f = threading.Thread(target=worker_f.run, name='worker_f')
th_l.start()
th_f.start()
while True:
try:
req_l = dj_pb.DataSourceRequest(
data_source_meta=self.data_source_l.data_source_meta
)
req_f = dj_pb.DataSourceRequest(
data_source_meta=self.data_source_f.data_source_meta
)
dss_l = master_client_l.GetDataSourceStatus(req_l)
dss_f = master_client_f.GetDataSourceStatus(req_f)
self.assertEqual(dss_l.role, common_pb.FLRole.Leader)
self.assertEqual(dss_f.role, common_pb.FLRole.Follower)
if dss_l.state == common_pb.DataSourceState.Ready and \
dss_f.state == common_pb.DataSourceState.Ready:
break
except Exception as e: #xx
pass
time.sleep(2)
th_l.join()
th_f.join()
master_l.stop()
master_f.stop()
def test_api(self):
logging.getLogger().setLevel(logging.DEBUG)
etcd_name = 'test_etcd'
etcd_addrs = 'localhost:2379'
etcd_base_dir_l = 'byefl_l'
etcd_base_dir_f = 'byefl_f'
data_source_name = 'test_data_source'
etcd_l = EtcdClient(etcd_name, etcd_addrs, etcd_base_dir_l, True)
etcd_f = EtcdClient(etcd_name, etcd_addrs, etcd_base_dir_f, True)
etcd_l.delete_prefix(
common.data_source_etcd_base_dir(data_source_name))
etcd_f.delete_prefix(
common.data_source_etcd_base_dir(data_source_name))
data_source_l = common_pb.DataSource()
data_source_l.role = common_pb.FLRole.Leader
data_source_l.state = common_pb.DataSourceState.Init
data_source_l.output_base_dir = "./ds_output_l"
data_source_f = common_pb.DataSource()
data_source_f.role = common_pb.FLRole.Follower
data_source_f.state = common_pb.DataSourceState.Init
data_source_f.output_base_dir = "./ds_output_f"
data_source_meta = common_pb.DataSourceMeta()
data_source_meta.name = data_source_name
data_source_meta.partition_num = 1
data_source_meta.start_time = 0
data_source_meta.end_time = 100000000
data_source_l.data_source_meta.MergeFrom(data_source_meta)
common.commit_data_source(etcd_l, data_source_l)
data_source_f.data_source_meta.MergeFrom(data_source_meta)
common.commit_data_source(etcd_f, data_source_f)
master_addr_l = 'localhost:4061'
master_addr_f = 'localhost:4062'
options = dj_pb.DataJoinMasterOptions(use_mock_etcd=True)
master_l = data_join_master.DataJoinMasterService(
int(master_addr_l.split(':')[1]), master_addr_f, data_source_name,
etcd_name, etcd_base_dir_l, etcd_addrs, options)
master_l.start()
master_f = data_join_master.DataJoinMasterService(
int(master_addr_f.split(':')[1]), master_addr_l, data_source_name,
etcd_name, etcd_base_dir_f, etcd_addrs, options)
master_f.start()
channel_l = make_insecure_channel(master_addr_l, ChannelType.INTERNAL)
client_l = dj_grpc.DataJoinMasterServiceStub(channel_l)
channel_f = make_insecure_channel(master_addr_f, ChannelType.INTERNAL)
client_f = dj_grpc.DataJoinMasterServiceStub(channel_f)
while True:
req_l = dj_pb.DataSourceRequest(
data_source_meta=data_source_l.data_source_meta)
req_f = dj_pb.DataSourceRequest(
data_source_meta=data_source_f.data_source_meta)
dss_l = client_l.GetDataSourceStatus(req_l)
dss_f = client_f.GetDataSourceStatus(req_f)
self.assertEqual(dss_l.role, common_pb.FLRole.Leader)
self.assertEqual(dss_f.role, common_pb.FLRole.Follower)
if dss_l.state == common_pb.DataSourceState.Processing and \
dss_f.state == common_pb.DataSourceState.Processing:
break
else:
time.sleep(2)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
partition_id=-1,
join_example=empty_pb2.Empty())
rdrsp = client_f.RequestJoinPartition(rdreq)
self.assertTrue(rdrsp.status.code == 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.join_example_rep.rank_id, 0)
self.assertEqual(rdrsp.manifest.join_example_rep.state,
dj_pb.JoinExampleState.Joining)
self.assertEqual(rdrsp.manifest.partition_id, 0)
#check idempotent
rdrsp = client_f.RequestJoinPartition(rdreq)
self.assertTrue(rdrsp.status.code == 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.join_example_rep.rank_id, 0)
self.assertEqual(rdrsp.manifest.join_example_rep.state,
dj_pb.JoinExampleState.Joining)
self.assertEqual(rdrsp.manifest.partition_id, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
join_example=empty_pb2.Empty())
rdrsp = client_l.RequestJoinPartition(rdreq)
self.assertTrue(rdrsp.status.code == 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.join_example_rep.rank_id, 0)
self.assertEqual(rdrsp.manifest.join_example_rep.state,
dj_pb.JoinExampleState.Joining)
self.assertEqual(rdrsp.manifest.partition_id, 0)
#check idempotent
rdrsp = client_l.RequestJoinPartition(rdreq)
self.assertTrue(rdrsp.status.code == 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.join_example_rep.rank_id, 0)
self.assertEqual(rdrsp.manifest.join_example_rep.state,
dj_pb.JoinExampleState.Joining)
self.assertEqual(rdrsp.manifest.partition_id, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=1,
partition_id=-1,
sync_example_id=empty_pb2.Empty())
rdrsp = client_l.RequestJoinPartition(rdreq)
self.assertEqual(rdrsp.status.code, 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.sync_example_id_rep.rank_id, 1)
self.assertEqual(rdrsp.manifest.sync_example_id_rep.state,
dj_pb.SyncExampleIdState.Syncing)
self.assertEqual(rdrsp.manifest.partition_id, 0)
#check idempotent
rdrsp = client_l.RequestJoinPartition(rdreq)
self.assertEqual(rdrsp.status.code, 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.sync_example_id_rep.rank_id, 1)
self.assertEqual(rdrsp.manifest.sync_example_id_rep.state,
dj_pb.SyncExampleIdState.Syncing)
self.assertEqual(rdrsp.manifest.partition_id, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=1,
partition_id=0,
sync_example_id=empty_pb2.Empty())
rdrsp = client_f.RequestJoinPartition(rdreq)
self.assertEqual(rdrsp.status.code, 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.sync_example_id_rep.rank_id, 1)
self.assertEqual(rdrsp.manifest.sync_example_id_rep.state,
dj_pb.SyncExampleIdState.Syncing)
self.assertEqual(rdrsp.manifest.partition_id, 0)
#check idempotent
rdrsp = client_f.RequestJoinPartition(rdreq)
self.assertEqual(rdrsp.status.code, 0)
self.assertTrue(rdrsp.HasField('manifest'))
self.assertEqual(rdrsp.manifest.sync_example_id_rep.rank_id, 1)
self.assertEqual(rdrsp.manifest.sync_example_id_rep.state,
dj_pb.SyncExampleIdState.Syncing)
self.assertEqual(rdrsp.manifest.partition_id, 0)
rdreq1 = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=1,
partition_id=0,
sync_example_id=empty_pb2.Empty())
try:
rsp = client_l.FinishJoinPartition(rdreq1)
except Exception as e:
self.assertTrue(True)
else:
self.assertTrue(False)
rdreq2 = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
join_example=empty_pb2.Empty())
try:
rsp = client_l.FinishJoinPartition(rdreq2)
except Exception as e:
self.assertTrue(True)
else:
self.assertTrue(False)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
)
manifest_l = client_l.QueryRawDataManifest(rdreq)
self.assertTrue(manifest_l is not None)
self.assertFalse(manifest_l.finished)
self.assertEqual(manifest_l.next_process_index, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
)
rsp = client_l.GetRawDataLatestTimeStamp(rdreq)
self.assertEqual(rsp.status.code, 0)
self.assertTrue(rsp.HasField('timestamp'))
self.assertEqual(rsp.timestamp.seconds, 0)
self.assertEqual(rsp.timestamp.nanos, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
added_raw_data_metas=dj_pb.AddedRawDataMetas(
dedup=False,
raw_data_metas=[
dj_pb.RawDataMeta(
file_path='a',
timestamp=timestamp_pb2.Timestamp(seconds=3))
]))
rsp = client_l.AddRawData(rdreq)
self.assertEqual(rsp.code, 0)
manifest_l = client_l.QueryRawDataManifest(rdreq)
self.assertTrue(manifest_l is not None)
self.assertFalse(manifest_l.finished)
self.assertEqual(manifest_l.next_process_index, 1)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
)
rsp = client_l.GetRawDataLatestTimeStamp(rdreq)
self.assertEqual(rsp.status.code, 0)
self.assertTrue(rsp.HasField('timestamp'))
self.assertEqual(rsp.timestamp.seconds, 3)
self.assertEqual(rsp.timestamp.nanos, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
added_raw_data_metas=dj_pb.AddedRawDataMetas(
dedup=False,
raw_data_metas=[
dj_pb.RawDataMeta(
file_path='b',
timestamp=timestamp_pb2.Timestamp(seconds=5))
]))
rsp = client_l.AddRawData(rdreq)
self.assertEqual(rsp.code, 0)
manifest_l = client_l.QueryRawDataManifest(rdreq)
self.assertTrue(manifest_l is not None)
self.assertFalse(manifest_l.finished)
self.assertEqual(manifest_l.next_process_index, 2)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
)
rsp = client_l.GetRawDataLatestTimeStamp(rdreq)
self.assertEqual(rsp.status.code, 0)
self.assertTrue(rsp.HasField('timestamp'))
self.assertEqual(rsp.timestamp.seconds, 5)
self.assertEqual(rsp.timestamp.nanos, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
added_raw_data_metas=dj_pb.AddedRawDataMetas(
dedup=True,
raw_data_metas=[
dj_pb.RawDataMeta(
file_path='b',
timestamp=timestamp_pb2.Timestamp(seconds=5)),
dj_pb.RawDataMeta(
file_path='a',
timestamp=timestamp_pb2.Timestamp(seconds=3))
]))
rsp = client_l.AddRawData(rdreq)
self.assertEqual(rsp.code, 0)
manifest_l = client_l.QueryRawDataManifest(rdreq)
self.assertTrue(manifest_l is not None)
self.assertFalse(manifest_l.finished)
self.assertEqual(manifest_l.next_process_index, 2)
rsp = client_l.GetRawDataLatestTimeStamp(rdreq)
self.assertEqual(rsp.status.code, 0)
self.assertTrue(rsp.HasField('timestamp'))
self.assertEqual(rsp.timestamp.seconds, 5)
self.assertEqual(rsp.timestamp.nanos, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
partition_id=0,
)
manifest_f = client_f.QueryRawDataManifest(rdreq)
self.assertTrue(manifest_f is not None)
self.assertFalse(manifest_f.finished)
self.assertEqual(manifest_f.next_process_index, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
)
rsp = client_f.GetRawDataLatestTimeStamp(rdreq)
self.assertEqual(rsp.status.code, 0)
self.assertTrue(rsp.HasField('timestamp'))
self.assertEqual(rsp.timestamp.seconds, 0)
self.assertEqual(rsp.timestamp.nanos, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
partition_id=0,
added_raw_data_metas=dj_pb.AddedRawDataMetas(
dedup=False,
raw_data_metas=[
dj_pb.RawDataMeta(
file_path='a',
timestamp=timestamp_pb2.Timestamp(seconds=1))
]))
rsp = client_f.AddRawData(rdreq)
self.assertEqual(rsp.code, 0)
manifest_f = client_f.QueryRawDataManifest(rdreq)
self.assertTrue(manifest_f is not None)
self.assertFalse(manifest_f.finished)
self.assertEqual(manifest_f.next_process_index, 1)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
)
rsp = client_f.GetRawDataLatestTimeStamp(rdreq)
self.assertEqual(rsp.status.code, 0)
self.assertTrue(rsp.HasField('timestamp'))
self.assertEqual(rsp.timestamp.seconds, 1)
self.assertEqual(rsp.timestamp.nanos, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
partition_id=0,
added_raw_data_metas=dj_pb.AddedRawDataMetas(
dedup=False,
raw_data_metas=[
dj_pb.RawDataMeta(
file_path='b',
timestamp=timestamp_pb2.Timestamp(seconds=2))
]))
rsp = client_f.AddRawData(rdreq)
self.assertEqual(rsp.code, 0)
manifest_f = client_f.QueryRawDataManifest(rdreq)
self.assertTrue(manifest_f is not None)
self.assertFalse(manifest_f.finished)
self.assertEqual(manifest_f.next_process_index, 2)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
)
rsp = client_f.GetRawDataLatestTimeStamp(rdreq)
self.assertEqual(rsp.status.code, 0)
self.assertTrue(rsp.HasField('timestamp'))
self.assertEqual(rsp.timestamp.seconds, 2)
self.assertEqual(rsp.timestamp.nanos, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
partition_id=0,
added_raw_data_metas=dj_pb.AddedRawDataMetas(
dedup=True,
raw_data_metas=[
dj_pb.RawDataMeta(
file_path='a',
timestamp=timestamp_pb2.Timestamp(seconds=1)),
dj_pb.RawDataMeta(
file_path='b',
timestamp=timestamp_pb2.Timestamp(seconds=2))
]))
rsp = client_f.AddRawData(rdreq)
self.assertEqual(rsp.code, 0)
manifest_f = client_f.QueryRawDataManifest(rdreq)
self.assertTrue(manifest_f is not None)
self.assertFalse(manifest_f.finished)
self.assertEqual(manifest_f.next_process_index, 2)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
)
rsp = client_f.GetRawDataLatestTimeStamp(rdreq)
self.assertEqual(rsp.status.code, 0)
self.assertTrue(rsp.HasField('timestamp'))
self.assertEqual(rsp.timestamp.seconds, 2)
self.assertEqual(rsp.timestamp.nanos, 0)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
finish_raw_data=empty_pb2.Empty())
rsp = client_l.FinishRawData(rdreq)
self.assertEqual(rsp.code, 0)
#check idempotent
rsp = client_l.FinishRawData(rdreq)
self.assertEqual(rsp.code, 0)
manifest_l = client_l.QueryRawDataManifest(rdreq)
self.assertTrue(manifest_l is not None)
self.assertTrue(manifest_l.finished)
self.assertEqual(manifest_l.next_process_index, 2)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_l.data_source_meta,
rank_id=0,
partition_id=0,
added_raw_data_metas=dj_pb.AddedRawDataMetas(
dedup=False,
raw_data_metas=[
dj_pb.RawDataMeta(
file_path='x',
timestamp=timestamp_pb2.Timestamp(seconds=4))
]))
try:
rsp = client_l.AddRawData(rdreq)
except Exception as e:
self.assertTrue(True)
else:
self.assertTrue(False)
try:
rsp = client_f.FinishJoinPartition(rdreq2)
except Exception as e:
self.assertTrue(True)
else:
self.assertTrue(False)
rsp = client_l.FinishJoinPartition(rdreq1)
self.assertEqual(rsp.code, 0)
#check idempotent
rsp = client_l.FinishJoinPartition(rdreq1)
self.assertEqual(rsp.code, 0)
rsp = client_f.FinishJoinPartition(rdreq1)
self.assertEqual(rsp.code, 0)
#check idempotent
rsp = client_f.FinishJoinPartition(rdreq1)
self.assertEqual(rsp.code, 0)
try:
rsp = client_f.FinishJoinPartition(rdreq2)
except Exception as e:
self.assertTrue(True)
else:
self.assertTrue(False)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
partition_id=0,
finish_raw_data=empty_pb2.Empty())
rsp = client_f.FinishRawData(rdreq)
self.assertEqual(rsp.code, 0)
#check idempotent
rsp = client_f.FinishRawData(rdreq)
self.assertEqual(rsp.code, 0)
manifest_f = client_f.QueryRawDataManifest(rdreq)
self.assertTrue(manifest_f is not None)
self.assertTrue(manifest_f.finished)
self.assertEqual(manifest_f.next_process_index, 2)
rdreq = dj_pb.RawDataRequest(
data_source_meta=data_source_f.data_source_meta,
rank_id=0,
partition_id=0,
added_raw_data_metas=dj_pb.AddedRawDataMetas(
dedup=True,
raw_data_metas=[
dj_pb.RawDataMeta(
file_path='x',
timestamp=timestamp_pb2.Timestamp(seconds=3))
]))
try:
rsp = client_f.AddRawData(rdreq)
except Exception as e:
self.assertTrue(True)
else:
self.assertTrue(False)
rsp = client_f.FinishJoinPartition(rdreq2)
self.assertEqual(rsp.code, 0)
#check idempotent
rsp = client_f.FinishJoinPartition(rdreq2)
self.assertEqual(rsp.code, 0)
rsp = client_l.FinishJoinPartition(rdreq2)
self.assertEqual(rsp.code, 0)
#check idempotent
rsp = client_l.FinishJoinPartition(rdreq2)
self.assertEqual(rsp.code, 0)
while True:
req_l = dj_pb.DataSourceRequest(
data_source_meta=data_source_l.data_source_meta)
req_f = dj_pb.DataSourceRequest(
data_source_meta=data_source_f.data_source_meta)
dss_l = client_l.GetDataSourceStatus(req_l)
dss_f = client_f.GetDataSourceStatus(req_f)
self.assertEqual(dss_l.role, common_pb.FLRole.Leader)
self.assertEqual(dss_f.role, common_pb.FLRole.Follower)
if dss_l.state == common_pb.DataSourceState.Finished and \
dss_f.state == common_pb.DataSourceState.Finished:
break
else:
time.sleep(2)
master_l.stop()
master_f.stop()
def setUp(self):
etcd_name = 'test_etcd'
etcd_addrs = 'localhost:2379'
etcd_base_dir_l = 'byefl_l'
etcd_base_dir_f = 'byefl_f'
data_source_name = 'test_data_source'
etcd_l = EtcdClient(etcd_name, etcd_addrs, etcd_base_dir_l, True)
etcd_f = EtcdClient(etcd_name, etcd_addrs, etcd_base_dir_f, True)
etcd_l.delete_prefix(data_source_name)
etcd_f.delete_prefix(data_source_name)
data_source_l = common_pb.DataSource()
data_source_l.role = common_pb.FLRole.Leader
data_source_l.state = common_pb.DataSourceState.Init
data_source_l.data_block_dir = "./data_block_l"
data_source_l.raw_data_dir = "./raw_data_l"
data_source_l.example_dumped_dir = "./example_dumped_l"
data_source_f = common_pb.DataSource()
data_source_f.role = common_pb.FLRole.Follower
data_source_f.state = common_pb.DataSourceState.Init
data_source_f.data_block_dir = "./data_block_f"
data_source_f.raw_data_dir = "./raw_data_f"
data_source_f.example_dumped_dir = "./example_dumped_f"
data_source_meta = common_pb.DataSourceMeta()
data_source_meta.name = data_source_name
data_source_meta.partition_num = 2
data_source_meta.start_time = 0
data_source_meta.end_time = 100000000
data_source_meta.min_matching_window = 64
data_source_meta.max_matching_window = 128
data_source_meta.data_source_type = common_pb.DataSourceType.Sequential
data_source_meta.max_example_in_data_block = 1000
data_source_l.data_source_meta.MergeFrom(data_source_meta)
etcd_l.set_data(os.path.join(data_source_name, 'master'),
text_format.MessageToString(data_source_l))
data_source_f.data_source_meta.MergeFrom(data_source_meta)
etcd_f.set_data(os.path.join(data_source_name, 'master'),
text_format.MessageToString(data_source_f))
customized_options.set_use_mock_etcd()
master_addr_l = 'localhost:4061'
master_addr_f = 'localhost:4062'
master_l = data_join_master.DataJoinMasterService(
int(master_addr_l.split(':')[1]), master_addr_f, data_source_name,
etcd_name, etcd_base_dir_l, etcd_addrs)
master_l.start()
master_f = data_join_master.DataJoinMasterService(
int(master_addr_f.split(':')[1]), master_addr_l, data_source_name,
etcd_name, etcd_base_dir_f, etcd_addrs)
master_f.start()
channel_l = make_insecure_channel(master_addr_l, ChannelType.INTERNAL)
master_client_l = dj_grpc.DataJoinMasterServiceStub(channel_l)
channel_f = make_insecure_channel(master_addr_f, ChannelType.INTERNAL)
master_client_f = dj_grpc.DataJoinMasterServiceStub(channel_f)
while True:
rsp_l = master_client_l.GetDataSourceState(
data_source_l.data_source_meta)
rsp_f = master_client_f.GetDataSourceState(
data_source_f.data_source_meta)
self.assertEqual(rsp_l.status.code, 0)
self.assertEqual(rsp_l.role, common_pb.FLRole.Leader)
self.assertEqual(rsp_l.data_source_type,
common_pb.DataSourceType.Sequential)
self.assertEqual(rsp_f.status.code, 0)
self.assertEqual(rsp_f.role, common_pb.FLRole.Follower)
self.assertEqual(rsp_f.data_source_type,
common_pb.DataSourceType.Sequential)
if (rsp_l.state == common_pb.DataSourceState.Processing
and rsp_f.state == common_pb.DataSourceState.Processing):
break
else:
time.sleep(2)
self.master_client_l = master_client_l
self.master_client_f = master_client_f
self.master_addr_l = master_addr_l
self.master_addr_f = master_addr_f
self.etcd_l = etcd_l
self.etcd_f = etcd_f
self.data_source_l = data_source_l
self.data_source_f = data_source_f
self.master_l = master_l
self.master_f = master_f
self.data_source_name = data_source_name,
self.etcd_name = etcd_name
self.etcd_addrs = etcd_addrs
self.etcd_base_dir_l = etcd_base_dir_l
self.etcd_base_dir_f = etcd_base_dir_f
if gfile.Exists(data_source_l.data_block_dir):
gfile.DeleteRecursively(data_source_l.data_block_dir)
if gfile.Exists(data_source_l.example_dumped_dir):
gfile.DeleteRecursively(data_source_l.example_dumped_dir)
if gfile.Exists(data_source_l.raw_data_dir):
gfile.DeleteRecursively(data_source_l.raw_data_dir)
if gfile.Exists(data_source_f.data_block_dir):
gfile.DeleteRecursively(data_source_f.data_block_dir)
if gfile.Exists(data_source_f.example_dumped_dir):
gfile.DeleteRecursively(data_source_f.example_dumped_dir)
if gfile.Exists(data_source_f.raw_data_dir):
gfile.DeleteRecursively(data_source_f.raw_data_dir)
self.total_index = 1 << 13