add name resolver config and deflaut load balancing config

python/paddle_serving_client/__init__.py

@@ -401,12 +401,10 @@ class MultiLangClient(object):
     def connect(self, endpoints):
         # https://github.com/tensorflow/serving/issues/1382
         options = [('grpc.max_receive_message_length', 512 * 1024 * 1024),
-                   ('grpc.max_send_message_length', 512 * 1024 * 1024)]
+                   ('grpc.max_send_message_length', 512 * 1024 * 1024),
+                   ('grpc.lb_policy_name', 'round_robin')]
 
-        # TODO
-        if len(endpoints) > 1:
-            print("Warn: grpc can only set a endpoint")
-        g_endpoint = endpoints[0]
+        g_endpoint = 'ipv4:{}'.format(','.join(endpoints))
         self.channel_ = grpc.insecure_channel(g_endpoint, options=options)
         self.stub_ = multi_lang_general_model_service_pb2_grpc.MultiLangGeneralModelServiceStub(
             self.channel_)

python/paddle_serving_server/__init__.py

@@ -440,13 +440,12 @@ class Server(object):
         os.system(command)
 
 
-class MultiLangServerService(multi_lang_general_model_service_pb2_grpc.
-                             MultiLangGeneralModelServiceServier):
-    def __init__(self, model_config_path, endpoints, rpc_timeout_ms):
+class MultiLangServerService(
+        multi_lang_general_model_service_pb2_grpc.MultiLangGeneralModelService):
+    def __init__(self, model_config_path, endpoints):
         from paddle_serving_client import Client
         self._parse_model_config(model_config_path)
         self.bclient_ = Client()
-        self.bclient_.set_rpc_timeout_ms(rpc_timeout_ms)
         self.bclient_.load_client_config(
             "{}/serving_server_conf.prototxt".format(model_config_path))
         self.bclient_.connect(endpoints)
@@ -538,6 +537,7 @@ class MultiLangServerService(multi_lang_general_model_service_pb2_grpc.
         return resp
 
     def inference(self, request, context):
+        print("get inference")
         feed_dict, fetch_names, is_python = self._unpack_request(request)
         data, tag = self.bclient_.predict(
             feed=feed_dict, fetch=fetch_names, need_variant_tag=True)
@@ -548,12 +548,8 @@ class MultiLangServer(object):
     def __init__(self):
         self.bserver_ = Server()
         self.worker_num_ = 4
-        self.body_size_ = 512 * 1024 * 1024
-        self.concurrency_ = 100
-        self.rpc_timeout_ms_ = 20000
-
-    def set_rpc_timeout_ms(self, rpc_timeout_ms):
-        self.rpc_timeout_ms_ = rpc_timeout_ms
+        self.body_size_ = 64 * 1024 * 1024
+        self.concurrency_ = 100000
 
     def set_max_concurrency(self, concurrency):
         self.concurrency_ = concurrency
@@ -636,8 +632,8 @@ class MultiLangServer(object):
             maximum_concurrent_rpcs=self.concurrency_)
         multi_lang_general_model_service_pb2_grpc.add_MultiLangGeneralModelServiceServicer_to_server(
             MultiLangServerService(self.model_config_path_,
-                                   ["0.0.0.0:{}".format(self.port_list_[0])],
-                                   self.rpc_timeout_ms_), server)
+                                   ["0.0.0.0:{}".format(self.port_list_[0])]),
+            server)
         server.add_insecure_port('[::]:{}'.format(self.gport_))
         server.start()
         p_bserver.join()