# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # pylint: disable=doc-string-missing import paddle_serving_client import os from .proto import sdk_configure_pb2 as sdk from .proto import general_model_config_pb2 as m_config import google.protobuf.text_format import numpy as np import time import sys import grpc from .proto import multi_lang_general_model_service_pb2 sys.path.append( os.path.join(os.path.abspath(os.path.dirname(__file__)), 'proto')) from .proto import multi_lang_general_model_service_pb2_grpc int64_type = 0 float32_type = 1 int32_type = 2 int_type = set([int64_type, int32_type]) float_type = set([float32_type]) class _NOPProfiler(object): def record(self, name): pass def print_profile(self): pass class _TimeProfiler(object): def __init__(self): self.pid = os.getpid() self.print_head = 'PROFILE\tpid:{}\t'.format(self.pid) self.time_record = [self.print_head] def record(self, name): self.time_record.append('{}:{} '.format( name, int(round(time.time() * 1000000)))) def print_profile(self): self.time_record.append('\n') sys.stderr.write(''.join(self.time_record)) self.time_record = [self.print_head] _is_profile = int(os.environ.get('FLAGS_profile_client', 0)) _Profiler = _TimeProfiler if _is_profile else _NOPProfiler class SDKConfig(object): def __init__(self): self.sdk_desc = sdk.SDKConf() self.tag_list = [] self.cluster_list = [] self.variant_weight_list = [] self.rpc_timeout_ms = 20000 self.load_balance_strategy = "la" def add_server_variant(self, tag, cluster, variant_weight): self.tag_list.append(tag) self.cluster_list.append(cluster) self.variant_weight_list.append(variant_weight) def set_load_banlance_strategy(self, strategy): self.load_balance_strategy = strategy def gen_desc(self, rpc_timeout_ms): predictor_desc = sdk.Predictor() predictor_desc.name = "general_model" predictor_desc.service_name = \ "baidu.paddle_serving.predictor.general_model.GeneralModelService" predictor_desc.endpoint_router = "WeightedRandomRender" predictor_desc.weighted_random_render_conf.variant_weight_list = "|".join( self.variant_weight_list) for idx, tag in enumerate(self.tag_list): variant_desc = sdk.VariantConf() variant_desc.tag = tag variant_desc.naming_conf.cluster = "list://{}".format(",".join( self.cluster_list[idx])) predictor_desc.variants.extend([variant_desc]) self.sdk_desc.predictors.extend([predictor_desc]) self.sdk_desc.default_variant_conf.tag = "default" self.sdk_desc.default_variant_conf.connection_conf.connect_timeout_ms = 2000 self.sdk_desc.default_variant_conf.connection_conf.rpc_timeout_ms = rpc_timeout_ms self.sdk_desc.default_variant_conf.connection_conf.connect_retry_count = 2 self.sdk_desc.default_variant_conf.connection_conf.max_connection_per_host = 100 self.sdk_desc.default_variant_conf.connection_conf.hedge_request_timeout_ms = -1 self.sdk_desc.default_variant_conf.connection_conf.hedge_fetch_retry_count = 2 self.sdk_desc.default_variant_conf.connection_conf.connection_type = "pooled" self.sdk_desc.default_variant_conf.naming_conf.cluster_filter_strategy = "Default" self.sdk_desc.default_variant_conf.naming_conf.load_balance_strategy = "la" self.sdk_desc.default_variant_conf.rpc_parameter.compress_type = 0 self.sdk_desc.default_variant_conf.rpc_parameter.package_size = 20 self.sdk_desc.default_variant_conf.rpc_parameter.protocol = "baidu_std" self.sdk_desc.default_variant_conf.rpc_parameter.max_channel_per_request = 3 return self.sdk_desc class Client(object): def __init__(self): self.feed_names_ = [] self.fetch_names_ = [] self.client_handle_ = None self.feed_shapes_ = {} self.feed_types_ = {} self.feed_names_to_idx_ = {} self.pid = os.getpid() self.predictor_sdk_ = None self.producers = [] self.consumer = None self.profile_ = _Profiler() self.all_numpy_input = True self.has_numpy_input = False self.rpc_timeout_ms = 20000 from .serving_client import PredictorRes self.predictorres_constructor = PredictorRes self.write_profile_into_fetch_map_ = False # only for grpc impl def load_client_config(self, path): from .serving_client import PredictorClient model_conf = m_config.GeneralModelConfig() f = open(path, 'r') model_conf = google.protobuf.text_format.Merge( str(f.read()), model_conf) # load configuraion here # get feed vars, fetch vars # get feed shapes, feed types # map feed names to index self.client_handle_ = PredictorClient() self.client_handle_.init(path) if "FLAGS_max_body_size" not in os.environ: os.environ["FLAGS_max_body_size"] = str(512 * 1024 * 1024) read_env_flags = ["profile_client", "profile_server", "max_body_size"] self.client_handle_.init_gflags([sys.argv[ 0]] + ["--tryfromenv=" + ",".join(read_env_flags)]) self.feed_names_ = [var.alias_name for var in model_conf.feed_var] self.fetch_names_ = [var.alias_name for var in model_conf.fetch_var] self.feed_names_to_idx_ = {} self.fetch_names_to_type_ = {} self.fetch_names_to_idx_ = {} self.lod_tensor_set = set() self.feed_tensor_len = {} for i, var in enumerate(model_conf.feed_var): self.feed_names_to_idx_[var.alias_name] = i self.feed_types_[var.alias_name] = var.feed_type self.feed_shapes_[var.alias_name] = var.shape if var.is_lod_tensor: self.lod_tensor_set.add(var.alias_name) else: counter = 1 for dim in self.feed_shapes_[var.alias_name]: counter *= dim self.feed_tensor_len[var.alias_name] = counter for i, var in enumerate(model_conf.fetch_var): self.fetch_names_to_idx_[var.alias_name] = i self.fetch_names_to_type_[var.alias_name] = var.fetch_type if var.is_lod_tensor: self.lod_tensor_set.add(var.alias_name) return def add_variant(self, tag, cluster, variant_weight): if self.predictor_sdk_ is None: self.predictor_sdk_ = SDKConfig() self.predictor_sdk_.add_server_variant(tag, cluster, str(variant_weight)) def set_rpc_timeout_ms(self, rpc_timeout): if not isinstance(rpc_timeout, int): raise ValueError("rpc_timeout must be int type.") else: self.rpc_timeout_ms = rpc_timeout def connect(self, endpoints=None): # check whether current endpoint is available # init from client config # create predictor here if endpoints is None: if self.predictor_sdk_ is None: raise ValueError( "You must set the endpoints parameter or use add_variant function to create a variant." ) else: if self.predictor_sdk_ is None: self.add_variant('default_tag_{}'.format(id(self)), endpoints, 100) else: print( "parameter endpoints({}) will not take effect, because you use the add_variant function.". format(endpoints)) sdk_desc = self.predictor_sdk_.gen_desc(self.rpc_timeout_ms) self.client_handle_.create_predictor_by_desc(sdk_desc.SerializeToString( )) def get_feed_names(self): return self.feed_names_ def get_fetch_names(self): return self.fetch_names_ def shape_check(self, feed, key): if key in self.lod_tensor_set: return if isinstance(feed[key], list) and len(feed[key]) != self.feed_tensor_len[key]: raise ValueError("The shape of feed tensor {} not match.".format( key)) if type(feed[key]).__module__ == np.__name__ and np.size(feed[ key]) != self.feed_tensor_len[key]: #raise SystemExit("The shape of feed tensor {} not match.".format( # key)) pass def predict(self, feed=None, fetch=None, need_variant_tag=False): self.profile_.record('py_prepro_0') if feed is None or fetch is None: raise ValueError("You should specify feed and fetch for prediction") fetch_list = [] if isinstance(fetch, str): fetch_list = [fetch] elif isinstance(fetch, list): fetch_list = fetch else: raise ValueError("Fetch only accepts string and list of string") feed_batch = [] if isinstance(feed, dict): feed_batch.append(feed) elif isinstance(feed, list): feed_batch = feed else: raise ValueError("Feed only accepts dict and list of dict") int_slot_batch = [] float_slot_batch = [] int_feed_names = [] float_feed_names = [] int_shape = [] float_shape = [] fetch_names = [] counter = 0 batch_size = len(feed_batch) for key in fetch_list: if key in self.fetch_names_: fetch_names.append(key) if len(fetch_names) == 0: raise ValueError( "Fetch names should not be empty or out of saved fetch list.") return {} for i, feed_i in enumerate(feed_batch): int_slot = [] float_slot = [] for key in feed_i: if key not in self.feed_names_: raise ValueError("Wrong feed name: {}.".format(key)) #if not isinstance(feed_i[key], np.ndarray): self.shape_check(feed_i, key) if self.feed_types_[key] in int_type: if i == 0: int_feed_names.append(key) if isinstance(feed_i[key], np.ndarray): int_shape.append(list(feed_i[key].shape)) else: int_shape.append(self.feed_shapes_[key]) if isinstance(feed_i[key], np.ndarray): int_slot.append(feed_i[key]) self.has_numpy_input = True else: int_slot.append(feed_i[key]) self.all_numpy_input = False elif self.feed_types_[key] in float_type: if i == 0: float_feed_names.append(key) if isinstance(feed_i[key], np.ndarray): float_shape.append(list(feed_i[key].shape)) else: float_shape.append(self.feed_shapes_[key]) if isinstance(feed_i[key], np.ndarray): float_slot.append(feed_i[key]) self.has_numpy_input = True else: float_slot.append(feed_i[key]) self.all_numpy_input = False int_slot_batch.append(int_slot) float_slot_batch.append(float_slot) self.profile_.record('py_prepro_1') self.profile_.record('py_client_infer_0') result_batch_handle = self.predictorres_constructor() if self.all_numpy_input: res = self.client_handle_.numpy_predict( float_slot_batch, float_feed_names, float_shape, int_slot_batch, int_feed_names, int_shape, fetch_names, result_batch_handle, self.pid) elif self.has_numpy_input == False: res = self.client_handle_.batch_predict( float_slot_batch, float_feed_names, float_shape, int_slot_batch, int_feed_names, int_shape, fetch_names, result_batch_handle, self.pid) else: raise ValueError( "Please make sure the inputs are all in list type or all in numpy.array type" ) self.profile_.record('py_client_infer_1') self.profile_.record('py_postpro_0') if res == -1: return None multi_result_map = [] model_engine_names = result_batch_handle.get_engine_names() for mi, engine_name in enumerate(model_engine_names): result_map = {} # result map needs to be a numpy array for i, name in enumerate(fetch_names): if self.fetch_names_to_type_[name] == int64_type: # result_map[name] will be py::array(numpy array) result_map[name] = result_batch_handle.get_int64_by_name( mi, name) shape = result_batch_handle.get_shape(mi, name) result_map[name].shape = shape if name in self.lod_tensor_set: result_map["{}.lod".format( name)] = result_batch_handle.get_lod(mi, name) elif self.fetch_names_to_type_[name] == float32_type: result_map[name] = result_batch_handle.get_float_by_name( mi, name) shape = result_batch_handle.get_shape(mi, name) result_map[name].shape = shape if name in self.lod_tensor_set: result_map["{}.lod".format( name)] = result_batch_handle.get_lod(mi, name) elif self.fetch_names_to_type_[name] == int32_type: # result_map[name] will be py::array(numpy array) result_map[name] = result_batch_handle.get_int32_by_name( mi, name) shape = result_batch_handle.get_shape(mi, name) result_map[name].shape = shape if name in self.lod_tensor_set: result_map["{}.lod".format( name)] = result_batch_handle.get_lod(mi, name) multi_result_map.append(result_map) ret = None if len(model_engine_names) == 1: # If only one model result is returned, the format of ret is result_map ret = multi_result_map[0] else: # If multiple model results are returned, the format of ret is {name: result_map} ret = { engine_name: multi_result_map[mi] for mi, engine_name in enumerate(model_engine_names) } self.profile_.record('py_postpro_1') self.profile_.print_profile() # When using the A/B test, the tag of variant needs to be returned return ret if not need_variant_tag else [ ret, result_batch_handle.variant_tag() ] def release(self): self.client_handle_.destroy_predictor() self.client_handle_ = None class MultiLangClient(object): def __init__(self): self.channel_ = None self.stub_ = None self.rpc_timeout_s_ = 2 self.profile_ = _Profiler() def add_variant(self, tag, cluster, variant_weight): # TODO raise Exception("cannot support ABtest yet") def set_rpc_timeout_ms(self, rpc_timeout): if self.stub_ is None: raise Exception("set timeout must be set after connect.") if not isinstance(rpc_timeout, int): # for bclient raise ValueError("rpc_timeout must be int type.") self.rpc_timeout_s_ = rpc_timeout / 1000.0 timeout_req = multi_lang_general_model_service_pb2.SetTimeoutRequest() timeout_req.timeout_ms = rpc_timeout resp = self.stub_.SetTimeout(timeout_req) return resp.err_code == 0 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.lb_policy_name', 'round_robin')] # TODO: weight round robin 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_) # get client model config get_client_config_req = multi_lang_general_model_service_pb2.GetClientConfigRequest( ) resp = self.stub_.GetClientConfig(get_client_config_req) model_config_str = resp.client_config_str self._parse_model_config(model_config_str) def _flatten_list(self, nested_list): for item in nested_list: if isinstance(item, (list, tuple)): for sub_item in self._flatten_list(item): yield sub_item else: yield item def _parse_model_config(self, model_config_str): model_conf = m_config.GeneralModelConfig() model_conf = google.protobuf.text_format.Merge(model_config_str, model_conf) self.feed_names_ = [var.alias_name for var in model_conf.feed_var] self.feed_types_ = {} self.feed_shapes_ = {} self.fetch_names_ = [var.alias_name for var in model_conf.fetch_var] self.fetch_types_ = {} self.lod_tensor_set_ = set() for i, var in enumerate(model_conf.feed_var): self.feed_types_[var.alias_name] = var.feed_type self.feed_shapes_[var.alias_name] = var.shape if var.is_lod_tensor: self.lod_tensor_set_.add(var.alias_name) else: counter = 1 for dim in self.feed_shapes_[var.alias_name]: counter *= dim for i, var in enumerate(model_conf.fetch_var): self.fetch_types_[var.alias_name] = var.fetch_type if var.is_lod_tensor: self.lod_tensor_set_.add(var.alias_name) def _pack_inference_request(self, feed, fetch, is_python): req = multi_lang_general_model_service_pb2.InferenceRequest() req.fetch_var_names.extend(fetch) req.is_python = is_python feed_batch = None if isinstance(feed, dict): feed_batch = [feed] elif isinstance(feed, list): feed_batch = feed else: raise Exception("{} not support".format(type(feed))) req.feed_var_names.extend(feed_batch[0].keys()) init_feed_names = False for feed_data in feed_batch: inst = multi_lang_general_model_service_pb2.FeedInst() for name in req.feed_var_names: tensor = multi_lang_general_model_service_pb2.Tensor() var = feed_data[name] v_type = self.feed_types_[name] if is_python: data = None if isinstance(var, list): if v_type == 0: # int64 data = np.array(var, dtype="int64") elif v_type == 1: # float32 data = np.array(var, dtype="float32") elif v_type == 2: # int32 data = np.array(var, dtype="int32") else: raise Exception("error tensor value type.") elif isinstance(var, np.ndarray): data = var if v_type == 0: if data.dtype != 'int64': data = data.astype("int64") elif v_type == 1: if data.dtype != 'float32': data = data.astype("float32") elif v_type == 2: if data.dtype != 'int32': data = data.astype("int32") else: raise Exception("error tensor value type.") else: raise Exception("var must be list or ndarray.") tensor.data = data.tobytes() else: if isinstance(var, np.ndarray): if v_type == 0: # int64 tensor.int64_data.extend( var.reshape(-1).astype("int64").tolist()) elif v_type == 1: tensor.float_data.extend( var.reshape(-1).astype('float32').tolist()) elif v_type == 2: tensor.int_data.extend( var.reshape(-1).astype('int32').tolist()) else: raise Exception("error tensor value type.") elif isinstance(var, list): if v_type == 0: tensor.int64_data.extend(self._flatten_list(var)) elif v_type == 1: tensor.float_data.extend(self._flatten_list(var)) elif v_type == 2: tensor.int_data.extend(self._flatten_list(var)) else: raise Exception("error tensor value type.") else: raise Exception("var must be list or ndarray.") if isinstance(var, np.ndarray): tensor.shape.extend(list(var.shape)) else: tensor.shape.extend(self.feed_shapes_[name]) inst.tensor_array.append(tensor) req.insts.append(inst) return req def _unpack_inference_response(self, resp, fetch, is_python, need_variant_tag): if resp.err_code != 0: return None tag = resp.tag multi_result_map = {} for model_result in resp.outputs: inst = model_result.insts[0] result_map = {} for i, name in enumerate(fetch): var = inst.tensor_array[i] v_type = self.fetch_types_[name] if is_python: if v_type == 0: # int64 result_map[name] = np.frombuffer( var.data, dtype="int64") elif v_type == 1: # float32 result_map[name] = np.frombuffer( var.data, dtype="float32") else: raise Exception("error type.") else: if v_type == 0: # int64 result_map[name] = np.array( list(var.int64_data), dtype="int64") elif v_type == 1: # float32 result_map[name] = np.array( list(var.float_data), dtype="float32") else: raise Exception("error type.") result_map[name].shape = list(var.shape) if name in self.lod_tensor_set_: result_map["{}.lod".format(name)] = np.array(list(var.lod)) multi_result_map[model_result.engine_name] = result_map ret = None if len(resp.outputs) == 1: ret = list(multi_result_map.values())[0] else: ret = multi_result_map ret["serving_status_code"] = 0 return ret if not need_variant_tag else [ret, tag] def _done_callback_func(self, fetch, is_python, need_variant_tag): def unpack_resp(resp): return self._unpack_inference_response(resp, fetch, is_python, need_variant_tag) return unpack_resp def get_feed_names(self): return self.feed_names_ def predict(self, feed, fetch, need_variant_tag=False, asyn=False, is_python=True): if not asyn: try: self.profile_.record('py_prepro_0') req = self._pack_inference_request( feed, fetch, is_python=is_python) self.profile_.record('py_prepro_1') self.profile_.record('py_client_infer_0') resp = self.stub_.Inference(req, timeout=self.rpc_timeout_s_) self.profile_.record('py_client_infer_1') self.profile_.record('py_postpro_0') ret = self._unpack_inference_response( resp, fetch, is_python=is_python, need_variant_tag=need_variant_tag) self.profile_.record('py_postpro_1') self.profile_.print_profile() return ret except grpc.RpcError as e: return {"serving_status_code": e.code()} else: req = self._pack_inference_request(feed, fetch, is_python=is_python) call_future = self.stub_.Inference.future( req, timeout=self.rpc_timeout_s_) return MultiLangPredictFuture( call_future, self._done_callback_func( fetch, is_python=is_python, need_variant_tag=need_variant_tag)) class MultiLangPredictFuture(object): def __init__(self, call_future, callback_func): self.call_future_ = call_future self.callback_func_ = callback_func def result(self): try: resp = self.call_future_.result() except grpc.RpcError as e: return {"serving_status_code": e.code()} return self.callback_func_(resp) def add_done_callback(self, fn): def __fn__(call_future): assert call_future == self.call_future_ fn(self) self.call_future_.add_done_callback(__fn__)