# 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 os from .proto import server_configure_pb2 as server_sdk from .proto import general_model_config_pb2 as m_config import google.protobuf.text_format import tarfile import socket import paddle_serving_server_gpu as paddle_serving_server import time from .version import serving_server_version from contextlib import closing import argparse import collections import fcntl import shutil import numpy as np import grpc from .proto import multi_lang_general_model_service_pb2 import sys sys.path.append( os.path.join(os.path.abspath(os.path.dirname(__file__)), 'proto')) from .proto import multi_lang_general_model_service_pb2_grpc from multiprocessing import Pool, Process from concurrent import futures def serve_args(): parser = argparse.ArgumentParser("serve") parser.add_argument( "--thread", type=int, default=10, help="Concurrency of server") parser.add_argument( "--model", type=str, default="", help="Model for serving") parser.add_argument( "--port", type=int, default=9292, help="Port of the starting gpu") parser.add_argument( "--workdir", type=str, default="workdir", help="Working dir of current service") parser.add_argument( "--device", type=str, default="gpu", help="Type of device") parser.add_argument("--gpu_ids", type=str, default="", help="gpu ids") parser.add_argument( "--name", type=str, default="None", help="Default service name") parser.add_argument( "--mem_optim", default=False, action="store_true", help="Memory optimize") parser.add_argument( "--ir_optim", default=False, action="store_true", help="Graph optimize") parser.add_argument( "--max_body_size", type=int, default=512 * 1024 * 1024, help="Limit sizes of messages") parser.add_argument( "--use_multilang", default=False, action="store_true", help="Use Multi-language-service") return parser.parse_args() class OpMaker(object): def __init__(self): self.op_dict = { "general_infer": "GeneralInferOp", "general_reader": "GeneralReaderOp", "general_response": "GeneralResponseOp", "general_text_reader": "GeneralTextReaderOp", "general_text_response": "GeneralTextResponseOp", "general_single_kv": "GeneralSingleKVOp", "general_dist_kv_infer": "GeneralDistKVInferOp", "general_dist_kv": "GeneralDistKVOp" } self.node_name_suffix_ = collections.defaultdict(int) def create(self, node_type, engine_name=None, inputs=[], outputs=[]): if node_type not in self.op_dict: raise Exception("Op type {} is not supported right now".format( node_type)) node = server_sdk.DAGNode() # node.name will be used as the infer engine name if engine_name: node.name = engine_name else: node.name = '{}_{}'.format(node_type, self.node_name_suffix_[node_type]) self.node_name_suffix_[node_type] += 1 node.type = self.op_dict[node_type] if inputs: for dep_node_str in inputs: dep_node = server_sdk.DAGNode() google.protobuf.text_format.Parse(dep_node_str, dep_node) dep = server_sdk.DAGNodeDependency() dep.name = dep_node.name dep.mode = "RO" node.dependencies.extend([dep]) # Because the return value will be used as the key value of the # dict, and the proto object is variable which cannot be hashed, # so it is processed into a string. This has little effect on # overall efficiency. return google.protobuf.text_format.MessageToString(node) class OpSeqMaker(object): def __init__(self): self.workflow = server_sdk.Workflow() self.workflow.name = "workflow1" self.workflow.workflow_type = "Sequence" def add_op(self, node_str): node = server_sdk.DAGNode() google.protobuf.text_format.Parse(node_str, node) if len(node.dependencies) > 1: raise Exception( 'Set more than one predecessor for op in OpSeqMaker is not allowed.' ) if len(self.workflow.nodes) >= 1: if len(node.dependencies) == 0: dep = server_sdk.DAGNodeDependency() dep.name = self.workflow.nodes[-1].name dep.mode = "RO" node.dependencies.extend([dep]) elif len(node.dependencies) == 1: if node.dependencies[0].name != self.workflow.nodes[-1].name: raise Exception( 'You must add op in order in OpSeqMaker. The previous op is {}, but the current op is followed by {}.'. format(node.dependencies[0].name, self.workflow.nodes[ -1].name)) self.workflow.nodes.extend([node]) def get_op_sequence(self): workflow_conf = server_sdk.WorkflowConf() workflow_conf.workflows.extend([self.workflow]) return workflow_conf class OpGraphMaker(object): def __init__(self): self.workflow = server_sdk.Workflow() self.workflow.name = "workflow1" # Currently, SDK only supports "Sequence" self.workflow.workflow_type = "Sequence" def add_op(self, node_str): node = server_sdk.DAGNode() google.protobuf.text_format.Parse(node_str, node) self.workflow.nodes.extend([node]) def get_op_graph(self): workflow_conf = server_sdk.WorkflowConf() workflow_conf.workflows.extend([self.workflow]) return workflow_conf class Server(object): def __init__(self): self.server_handle_ = None self.infer_service_conf = None self.model_toolkit_conf = None self.resource_conf = None self.memory_optimization = False self.ir_optimization = False self.model_conf = None self.workflow_fn = "workflow.prototxt" self.resource_fn = "resource.prototxt" self.infer_service_fn = "infer_service.prototxt" self.model_toolkit_fn = "model_toolkit.prototxt" self.general_model_config_fn = "general_model.prototxt" self.cube_config_fn = "cube.conf" self.workdir = "" self.max_concurrency = 0 self.num_threads = 4 self.port = 8080 self.reload_interval_s = 10 self.max_body_size = 64 * 1024 * 1024 self.module_path = os.path.dirname(paddle_serving_server.__file__) self.cur_path = os.getcwd() self.use_local_bin = False self.gpuid = 0 self.model_config_paths = None # for multi-model in a workflow def set_max_concurrency(self, concurrency): self.max_concurrency = concurrency def set_num_threads(self, threads): self.num_threads = threads def set_max_body_size(self, body_size): if body_size >= self.max_body_size: self.max_body_size = body_size else: print( "max_body_size is less than default value, will use default value in service." ) def set_port(self, port): self.port = port def set_reload_interval(self, interval): self.reload_interval_s = interval def set_op_sequence(self, op_seq): self.workflow_conf = op_seq def set_op_graph(self, op_graph): self.workflow_conf = op_graph def set_memory_optimize(self, flag=False): self.memory_optimization = flag def set_ir_optimize(self, flag=False): self.ir_optimization = flag def check_local_bin(self): if "SERVING_BIN" in os.environ: self.use_local_bin = True self.bin_path = os.environ["SERVING_BIN"] def check_cuda(self): cuda_flag = False r = os.popen("ldd {} | grep cudart".format(self.bin_path)) r = r.read().split("=") if len(r) >= 2 and "cudart" in r[1] and os.system( "ls /dev/ | grep nvidia > /dev/null") == 0: cuda_flag = True if not cuda_flag: raise SystemExit( "CUDA not found, please check your environment or use cpu version by \"pip install paddle_serving_server\"" ) def set_gpuid(self, gpuid=0): self.gpuid = gpuid def _prepare_engine(self, model_config_paths, device): if self.model_toolkit_conf == None: self.model_toolkit_conf = server_sdk.ModelToolkitConf() for engine_name, model_config_path in model_config_paths.items(): engine = server_sdk.EngineDesc() engine.name = engine_name # engine.reloadable_meta = model_config_path + "/fluid_time_file" engine.reloadable_meta = self.workdir + "/fluid_time_file" os.system("touch {}".format(engine.reloadable_meta)) engine.reloadable_type = "timestamp_ne" engine.runtime_thread_num = 0 engine.batch_infer_size = 0 engine.enable_batch_align = 0 engine.model_data_path = model_config_path engine.enable_memory_optimization = self.memory_optimization engine.enable_ir_optimization = self.ir_optimization engine.static_optimization = False engine.force_update_static_cache = False if device == "cpu": engine.type = "FLUID_CPU_ANALYSIS_DIR" elif device == "gpu": engine.type = "FLUID_GPU_ANALYSIS_DIR" self.model_toolkit_conf.engines.extend([engine]) def _prepare_infer_service(self, port): if self.infer_service_conf == None: self.infer_service_conf = server_sdk.InferServiceConf() self.infer_service_conf.port = port infer_service = server_sdk.InferService() infer_service.name = "GeneralModelService" infer_service.workflows.extend(["workflow1"]) self.infer_service_conf.services.extend([infer_service]) def _prepare_resource(self, workdir, cube_conf): self.workdir = workdir if self.resource_conf == None: with open("{}/{}".format(workdir, self.general_model_config_fn), "w") as fout: fout.write(str(self.model_conf)) self.resource_conf = server_sdk.ResourceConf() for workflow in self.workflow_conf.workflows: for node in workflow.nodes: if "dist_kv" in node.name: self.resource_conf.cube_config_path = workdir self.resource_conf.cube_config_file = self.cube_config_fn if cube_conf == None: raise ValueError( "Please set the path of cube.conf while use dist_kv op." ) shutil.copy(cube_conf, workdir) self.resource_conf.model_toolkit_path = workdir self.resource_conf.model_toolkit_file = self.model_toolkit_fn self.resource_conf.general_model_path = workdir self.resource_conf.general_model_file = self.general_model_config_fn def _write_pb_str(self, filepath, pb_obj): with open(filepath, "w") as fout: fout.write(str(pb_obj)) def load_model_config(self, model_config_paths): # At present, Serving needs to configure the model path in # the resource.prototxt file to determine the input and output # format of the workflow. To ensure that the input and output # of multiple models are the same. workflow_oi_config_path = None if isinstance(model_config_paths, str): # If there is only one model path, use the default infer_op. # Because there are several infer_op type, we need to find # it from workflow_conf. default_engine_names = [ 'general_infer_0', 'general_dist_kv_infer_0', 'general_dist_kv_quant_infer_0' ] engine_name = None for node in self.workflow_conf.workflows[0].nodes: if node.name in default_engine_names: engine_name = node.name break if engine_name is None: raise Exception( "You have set the engine_name of Op. Please use the form {op: model_path} to configure model path" ) self.model_config_paths = {engine_name: model_config_paths} workflow_oi_config_path = self.model_config_paths[engine_name] elif isinstance(model_config_paths, dict): self.model_config_paths = {} for node_str, path in model_config_paths.items(): node = server_sdk.DAGNode() google.protobuf.text_format.Parse(node_str, node) self.model_config_paths[node.name] = path print("You have specified multiple model paths, please ensure " "that the input and output of multiple models are the same.") workflow_oi_config_path = list(self.model_config_paths.items())[0][ 1] else: raise Exception("The type of model_config_paths must be str or " "dict({op: model_path}), not {}.".format( type(model_config_paths))) self.model_conf = m_config.GeneralModelConfig() f = open( "{}/serving_server_conf.prototxt".format(workflow_oi_config_path), 'r') self.model_conf = google.protobuf.text_format.Merge( str(f.read()), self.model_conf) # check config here # print config here def download_bin(self): os.chdir(self.module_path) need_download = False device_version = "serving-gpu-" folder_name = device_version + serving_server_version tar_name = folder_name + ".tar.gz" bin_url = "https://paddle-serving.bj.bcebos.com/bin/" + tar_name self.server_path = os.path.join(self.module_path, folder_name) download_flag = "{}/{}.is_download".format(self.module_path, folder_name) #acquire lock version_file = open("{}/version.py".format(self.module_path), "r") fcntl.flock(version_file, fcntl.LOCK_EX) if os.path.exists(download_flag): os.chdir(self.cur_path) self.bin_path = self.server_path + "/serving" return if not os.path.exists(self.server_path): os.system("touch {}/{}.is_download".format(self.module_path, folder_name)) print('Frist time run, downloading PaddleServing components ...') r = os.system('wget ' + bin_url + ' --no-check-certificate') if r != 0: if os.path.exists(tar_name): os.remove(tar_name) raise SystemExit( 'Download failed, please check your network or permission of {}.'. format(self.module_path)) else: try: print('Decompressing files ..') tar = tarfile.open(tar_name) tar.extractall() tar.close() except: if os.path.exists(exe_path): os.remove(exe_path) raise SystemExit( 'Decompressing failed, please check your permission of {} or disk space left.'. format(self.module_path)) finally: os.remove(tar_name) #release lock version_file.close() os.chdir(self.cur_path) self.bin_path = self.server_path + "/serving" def prepare_server(self, workdir=None, port=9292, device="cpu", cube_conf=None): if workdir == None: workdir = "./tmp" os.system("mkdir {}".format(workdir)) else: os.system("mkdir {}".format(workdir)) os.system("touch {}/fluid_time_file".format(workdir)) if not self.port_is_available(port): raise SystemExit("Port {} is already used".format(port)) self.set_port(port) self._prepare_resource(workdir, cube_conf) self._prepare_engine(self.model_config_paths, device) self._prepare_infer_service(port) self.workdir = workdir infer_service_fn = "{}/{}".format(workdir, self.infer_service_fn) workflow_fn = "{}/{}".format(workdir, self.workflow_fn) resource_fn = "{}/{}".format(workdir, self.resource_fn) model_toolkit_fn = "{}/{}".format(workdir, self.model_toolkit_fn) self._write_pb_str(infer_service_fn, self.infer_service_conf) self._write_pb_str(workflow_fn, self.workflow_conf) self._write_pb_str(resource_fn, self.resource_conf) self._write_pb_str(model_toolkit_fn, self.model_toolkit_conf) def port_is_available(self, port): with closing(socket.socket(socket.AF_INET, socket.SOCK_STREAM)) as sock: sock.settimeout(2) result = sock.connect_ex(('0.0.0.0', port)) if result != 0: return True else: return False def run_server(self): # just run server with system command # currently we do not load cube self.check_local_bin() if not self.use_local_bin: self.download_bin() # wait for other process to download server bin while not os.path.exists(self.server_path): time.sleep(1) else: print("Use local bin : {}".format(self.bin_path)) self.check_cuda() command = "{} " \ "-enable_model_toolkit " \ "-inferservice_path {} " \ "-inferservice_file {} " \ "-max_concurrency {} " \ "-num_threads {} " \ "-port {} " \ "-reload_interval_s {} " \ "-resource_path {} " \ "-resource_file {} " \ "-workflow_path {} " \ "-workflow_file {} " \ "-bthread_concurrency {} " \ "-gpuid {} " \ "-max_body_size {} ".format( self.bin_path, self.workdir, self.infer_service_fn, self.max_concurrency, self.num_threads, self.port, self.reload_interval_s, self.workdir, self.resource_fn, self.workdir, self.workflow_fn, self.num_threads, self.gpuid, self.max_body_size) print("Going to Run Comand") print(command) os.system(command) class MultiLangServerServiceServicer(multi_lang_general_model_service_pb2_grpc. MultiLangGeneralModelServiceServicer): def __init__(self, model_config_path, is_multi_model, endpoints): self.is_multi_model_ = is_multi_model self.model_config_path_ = model_config_path self.endpoints_ = endpoints with open(self.model_config_path_) as f: self.model_config_str_ = str(f.read()) self._parse_model_config(self.model_config_str_) self._init_bclient(self.model_config_path_, self.endpoints_) def _init_bclient(self, model_config_path, endpoints, timeout_ms=None): from paddle_serving_client import Client self.bclient_ = Client() if timeout_ms is not None: self.bclient_.set_rpc_timeout_ms(timeout_ms) self.bclient_.load_client_config(model_config_path) self.bclient_.connect(endpoints) 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) 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 _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 _unpack_inference_request(self, request): feed_names = list(request.feed_var_names) fetch_names = list(request.fetch_var_names) is_python = request.is_python feed_batch = [] for feed_inst in request.insts: feed_dict = {} for idx, name in enumerate(feed_names): var = feed_inst.tensor_array[idx] v_type = self.feed_types_[name] data = None if is_python: if v_type == 0: data = np.frombuffer(var.data, dtype="int64") elif v_type == 1: data = np.frombuffer(var.data, dtype="float32") elif v_type == 2: data = np.frombuffer(var.data, dtype="int32") else: raise Exception("error type.") else: if v_type == 0: # int64 data = np.array(list(var.int64_data), dtype="int64") elif v_type == 1: # float32 data = np.array(list(var.float_data), dtype="float32") elif v_type == 2: data = np.array(list(var.int32_data), dtype="int32") else: raise Exception("error type.") data.shape = list(feed_inst.tensor_array[idx].shape) feed_dict[name] = data feed_batch.append(feed_dict) return feed_batch, fetch_names, is_python def _pack_inference_response(self, ret, fetch_names, is_python): resp = multi_lang_general_model_service_pb2.InferenceResponse() if ret is None: resp.err_code = 1 return resp results, tag = ret resp.tag = tag resp.err_code = 0 if not self.is_multi_model_: results = {'general_infer_0': results} for model_name, model_result in results.items(): model_output = multi_lang_general_model_service_pb2.ModelOutput() inst = multi_lang_general_model_service_pb2.FetchInst() for idx, name in enumerate(fetch_names): tensor = multi_lang_general_model_service_pb2.Tensor() v_type = self.fetch_types_[name] if is_python: tensor.data = model_result[name].tobytes() else: if v_type == 0: # int64 tensor.int64_data.extend(model_result[name].reshape(-1) .tolist()) elif v_type == 1: # float32 tensor.float_data.extend(model_result[name].reshape(-1) .tolist()) elif v_type == 2: # int32 tensor.int32_data.extend(model_result[name].reshape(-1) .tolist()) else: raise Exception("error type.") tensor.shape.extend(list(model_result[name].shape)) if name in self.lod_tensor_set_: tensor.lod.extend(model_result["{}.lod".format(name)] .tolist()) inst.tensor_array.append(tensor) model_output.insts.append(inst) model_output.engine_name = model_name resp.outputs.append(model_output) return resp def SetTimeout(self, request, context): # This porcess and Inference process cannot be operate at the same time. # For performance reasons, do not add thread lock temporarily. timeout_ms = request.timeout_ms self._init_bclient(self.model_config_path_, self.endpoints_, timeout_ms) resp = multi_lang_general_model_service_pb2.SimpleResponse() resp.err_code = 0 return resp def Inference(self, request, context): feed_dict, fetch_names, is_python = self._unpack_inference_request( request) ret = self.bclient_.predict( feed=feed_dict, fetch=fetch_names, need_variant_tag=True) return self._pack_inference_response(ret, fetch_names, is_python) def GetClientConfig(self, request, context): resp = multi_lang_general_model_service_pb2.GetClientConfigResponse() resp.client_config_str = self.model_config_str_ return resp class MultiLangServer(object): def __init__(self): self.bserver_ = Server() self.worker_num_ = 4 self.body_size_ = 64 * 1024 * 1024 self.concurrency_ = 100000 self.is_multi_model_ = False # for model ensemble def set_max_concurrency(self, concurrency): self.concurrency_ = concurrency self.bserver_.set_max_concurrency(concurrency) def set_num_threads(self, threads): self.worker_num_ = threads self.bserver_.set_num_threads(threads) def set_max_body_size(self, body_size): self.bserver_.set_max_body_size(body_size) if body_size >= self.body_size_: self.body_size_ = body_size else: print( "max_body_size is less than default value, will use default value in service." ) def set_port(self, port): self.gport_ = port def set_reload_interval(self, interval): self.bserver_.set_reload_interval(interval) def set_op_sequence(self, op_seq): self.bserver_.set_op_sequence(op_seq) def set_op_graph(self, op_graph): self.bserver_.set_op_graph(op_graph) def set_memory_optimize(self, flag=False): self.bserver_.set_memory_optimize(flag) def set_ir_optimize(self, flag=False): self.bserver_.set_ir_optimize(flag) def set_gpuid(self, gpuid=0): self.bserver_.set_gpuid(gpuid) def load_model_config(self, server_config_paths, client_config_path=None): self.bserver_.load_model_config(server_config_paths) if client_config_path is None: if isinstance(server_config_paths, dict): self.is_multi_model_ = True client_config_path = '{}/serving_server_conf.prototxt'.format( list(server_config_paths.items())[0][1]) else: client_config_path = '{}/serving_server_conf.prototxt'.format( server_config_paths) self.bclient_config_path_ = client_config_path def prepare_server(self, workdir=None, port=9292, device="cpu", cube_conf=None): if not self._port_is_available(port): raise SystemExit("Prot {} is already used".format(port)) default_port = 12000 self.port_list_ = [] for i in range(1000): if default_port + i != port and self._port_is_available(default_port + i): self.port_list_.append(default_port + i) break self.bserver_.prepare_server( workdir=workdir, port=self.port_list_[0], device=device, cube_conf=cube_conf) self.set_port(port) def _launch_brpc_service(self, bserver): bserver.run_server() def _port_is_available(self, port): with closing(socket.socket(socket.AF_INET, socket.SOCK_STREAM)) as sock: sock.settimeout(2) result = sock.connect_ex(('0.0.0.0', port)) return result != 0 def run_server(self): p_bserver = Process( target=self._launch_brpc_service, args=(self.bserver_, )) p_bserver.start() options = [('grpc.max_send_message_length', self.body_size_), ('grpc.max_receive_message_length', self.body_size_)] server = grpc.server( futures.ThreadPoolExecutor(max_workers=self.worker_num_), options=options, maximum_concurrent_rpcs=self.concurrency_) multi_lang_general_model_service_pb2_grpc.add_MultiLangGeneralModelServiceServicer_to_server( MultiLangServerServiceServicer( self.bclient_config_path_, self.is_multi_model_, ["0.0.0.0:{}".format(self.port_list_[0])]), server) server.add_insecure_port('[::]:{}'.format(self.gport_)) server.start() p_bserver.join() server.wait_for_termination()