Merge remote-tracking branch 'upstream/develop' into develop

README.md

@@ -45,9 +45,10 @@ nvidia-docker exec -it test bash
 ```
 
 ```shell
-pip install paddle-serving-client 
-pip install paddle-serving-server # CPU
-pip install paddle-serving-server-gpu # GPU
+pip install paddle-serving-client==0.3.2 
+pip install paddle-serving-server==0.3.2 # CPU
+pip install paddle-serving-server-gpu==0.3.2.post9 # GPU with CUDA9.0
+pip install paddle-serving-server-gpu==0.3.2.post10 # GPU with CUDA9.0
 ```
 
 You may need to use a domestic mirror source (in China, you can use the Tsinghua mirror source, add `-i https://pypi.tuna.tsinghua.edu.cn/simple` to pip command) to speed up the download.
@@ -172,6 +173,11 @@ Here, `client.predict` function has two arguments. `feed` is a `python dict` wit
 - [An End-to-end tutorial from training to inference service deployment](doc/TRAIN_TO_SERVICE.md)
 - [Write Bert-as-Service in 10 minutes](doc/BERT_10_MINS.md)
 
+### Tutorial at AIStudio
+- [Introduction to PaddleServing](https://aistudio.baidu.com/aistudio/projectdetail/605819)
+- [Image Segmentation on Paddle Serving](https://aistudio.baidu.com/aistudio/projectdetail/457715)
+- [Sentimental Analysis](https://aistudio.baidu.com/aistudio/projectdetail/509014)
+
 ### Developers
 - [How to config Serving native operators on server side?](doc/SERVER_DAG.md)
 - [How to develop a new Serving operator?](doc/NEW_OPERATOR.md)

README_CN.md

@@ -47,9 +47,10 @@ nvidia-docker run -p 9292:9292 --name test -dit hub.baidubce.com/paddlepaddle/se
 nvidia-docker exec -it test bash
 ```
 ```shell
-pip install paddle-serving-client
-pip install paddle-serving-server # CPU
-pip install paddle-serving-server-gpu # GPU
+pip install paddle-serving-client==0.3.2
+pip install paddle-serving-server==0.3.2 # CPU
+pip install paddle-serving-server-gpu==0.3.2.post9 # GPU with CUDA9.0
+pip install paddle-serving-server-gpu==0.3.2.post10 # GPU with CUDA10.0
 ```
 
 您可能需要使用国内镜像源（例如清华源, 在pip命令中添加`-i https://pypi.tuna.tsinghua.edu.cn/simple`）来加速下载。
@@ -169,6 +170,11 @@ print(fetch_map)
 - [端到端完成从训练到部署全流程](doc/TRAIN_TO_SERVICE_CN.md)
 - [十分钟构建Bert-As-Service](doc/BERT_10_MINS_CN.md)
 
+### AIStudio教程
+- [PaddleServing作业](https://aistudio.baidu.com/aistudio/projectdetail/605819)
+- [PaddleServing图像分割](https://aistudio.baidu.com/aistudio/projectdetail/457715)
+- [PaddleServing情感分析](https://aistudio.baidu.com/aistudio/projectdetail/509014)
+
 ### 开发者教程
 - [如何配置Server端的计算图?](doc/SERVER_DAG_CN.md)
 - [如何开发一个新的General Op?](doc/NEW_OPERATOR_CN.md)

cmake/paddlepaddle.cmake

@@ -31,7 +31,7 @@ message( "WITH_GPU = ${WITH_GPU}")
 # Paddle Version should be one of:
 # latest: latest develop build
 # version number like 1.5.2
-SET(PADDLE_VERSION "1.7.2")
+SET(PADDLE_VERSION "1.8.4")
 
 if (WITH_GPU)
     SET(PADDLE_LIB_VERSION "${PADDLE_VERSION}-gpu-cuda${CUDA_VERSION_MAJOR}-cudnn7-avx-mkl")

core/configure/proto/multi_lang_general_model_service.proto

@@ -14,6 +14,8 @@
 
 syntax = "proto2";
 
+package baidu.paddle_serving.multi_lang;
+
 option java_multiple_files = true;
 option java_package = "io.paddle.serving.grpc";
 option java_outer_classname = "ServingProto";
@@ -37,6 +39,7 @@ message InferenceRequest {
   repeated string feed_var_names = 2;
   repeated string fetch_var_names = 3;
   required bool is_python = 4 [ default = false ];
+  required uint64 log_id = 5 [ default = 0 ];
 };
 
 message InferenceResponse {

core/configure/proto/server_configure.proto

@@ -58,6 +58,8 @@ message ResourceConf {
   optional string cube_config_path = 5;
   optional string cube_config_file = 6;
   optional int32 cube_quant_bits = 7; // set 0 if no quant.
+  optional string auth_product_name = 8;
+  optional string auth_container_id = 9;
 };
 
 // DAG node depency info

core/cube/cube-api/include/meta.h

@@ -22,7 +22,8 @@
 #ifdef BCLOUD
 #include "baidu/rpc/channel.h"
 #include "baidu/rpc/parallel_channel.h"
-#include "rapidjson/document.h"
+#include "rapidjson_1.0/document.h"
+#include "rapidjson_1.0/rapidjson.h"
 #else
 #include "brpc/channel.h"
 #include "brpc/parallel_channel.h"

core/general-client/include/general_model.h

@@ -227,7 +227,8 @@ class PredictorClient {
       const std::vector<std::vector<int>>& int_shape,
       const std::vector<std::string>& fetch_name,
       PredictorRes& predict_res_batch,  // NOLINT
-      const int& pid);
+      const int& pid,
+      const uint64_t log_id);
 
   int numpy_predict(
       const std::vector<std::vector<py::array_t<float>>>& float_feed_batch,
@@ -238,7 +239,8 @@ class PredictorClient {
       const std::vector<std::vector<int>>& int_shape,
       const std::vector<std::string>& fetch_name,
       PredictorRes& predict_res_batch,  // NOLINT
-      const int& pid);
+      const int& pid,
+      const uint64_t log_id);
 
  private:
   PredictorApi _api;

core/general-client/src/general_model.cpp

@@ -39,7 +39,9 @@ using configure::GeneralModelConfig;
 
 void PredictorClient::init_gflags(std::vector<std::string> argv) {
   std::call_once(gflags_init_flag, [&]() {
+#ifndef BCLOUD
     FLAGS_logtostderr = true;
+#endif
     argv.insert(argv.begin(), "dummy");
     int argc = argv.size();
     char **arr = new char *[argv.size()];
@@ -144,7 +146,8 @@ int PredictorClient::batch_predict(
     const std::vector<std::vector<int>> &int_shape,
     const std::vector<std::string> &fetch_name,
     PredictorRes &predict_res_batch,
-    const int &pid) {
+    const int &pid,
+    const uint64_t log_id) {
   int batch_size = std::max(float_feed_batch.size(), int_feed_batch.size());
 
   predict_res_batch.clear();
@@ -162,6 +165,7 @@ int PredictorClient::batch_predict(
   VLOG(2) << "int feed name size: " << int_feed_name.size();
   VLOG(2) << "max body size : " << brpc::fLU64::FLAGS_max_body_size;
   Request req;
+  req.set_log_id(log_id);
   for (auto &name : fetch_name) {
     req.add_fetch_var_names(name);
   }
@@ -356,7 +360,8 @@ int PredictorClient::numpy_predict(
     const std::vector<std::vector<int>> &int_shape,
     const std::vector<std::string> &fetch_name,
     PredictorRes &predict_res_batch,
-    const int &pid) {
+    const int &pid,
+    const uint64_t log_id) {
   int batch_size = std::max(float_feed_batch.size(), int_feed_batch.size());
   VLOG(2) << "batch size: " << batch_size;
   predict_res_batch.clear();
@@ -374,6 +379,7 @@ int PredictorClient::numpy_predict(
   VLOG(2) << "int feed name size: " << int_feed_name.size();
   VLOG(2) << "max body size : " << brpc::fLU64::FLAGS_max_body_size;
   Request req;
+  req.set_log_id(log_id);
   for (auto &name : fetch_name) {
     req.add_fetch_var_names(name);
   }

core/general-client/src/pybind_general_model.cpp

@@ -107,7 +107,8 @@ PYBIND11_MODULE(serving_client, m) {
               const std::vector<std::vector<int>> &int_shape,
               const std::vector<std::string> &fetch_name,
               PredictorRes &predict_res_batch,
-              const int &pid) {
+              const int &pid,
+              const uint64_t log_id) {
              return self.batch_predict(float_feed_batch,
                                        float_feed_name,
                                        float_shape,
@@ -116,7 +117,8 @@ PYBIND11_MODULE(serving_client, m) {
                                        int_shape,
                                        fetch_name,
                                        predict_res_batch,
-                                       pid);
+                                       pid,
+                                       log_id);
            },
            py::call_guard<py::gil_scoped_release>())
       .def("numpy_predict",
@@ -131,7 +133,8 @@ PYBIND11_MODULE(serving_client, m) {
               const std::vector<std::vector<int>> &int_shape,
               const std::vector<std::string> &fetch_name,
               PredictorRes &predict_res_batch,
-              const int &pid) {
+              const int &pid,
+              const uint64_t log_id) {
              return self.numpy_predict(float_feed_batch,
                                        float_feed_name,
                                        float_shape,
@@ -140,7 +143,8 @@ PYBIND11_MODULE(serving_client, m) {
                                        int_shape,
                                        fetch_name,
                                        predict_res_batch,
-                                       pid);
+                                       pid,
+                                       log_id);
            },
            py::call_guard<py::gil_scoped_release>());
 }

core/general-server/op/general_copy_op.cpp

@@ -45,36 +45,41 @@ int GeneralCopyOp::inference() {
   const std::string pre_name = pre_node_names[0];
 
   const GeneralBlob *input_blob = get_depend_argument<GeneralBlob>(pre_name);
-  VLOG(2) << "precedent name: " << pre_name;
+  uint64_t log_id = input_blob->GetLogId();
+
+  VLOG(2) << "(logid=" << log_id << ") precedent name: " << pre_name;
   const TensorVector *in = &input_blob->tensor_vector;
-  VLOG(2) << "input size: " << in->size();
+  VLOG(2) << "(logid=" << log_id << ") input size: " << in->size();
   int batch_size = input_blob->GetBatchSize();
   int input_var_num = 0;
 
   GeneralBlob *res = mutable_data<GeneralBlob>();
+  res->SetLogId(log_id);
   TensorVector *out = &res->tensor_vector;
 
-  VLOG(2) << "input batch size: " << batch_size;
+  VLOG(2) << "(logid=" << log_id << ") input batch size: " << batch_size;
   res->SetBatchSize(batch_size);
 
   if (!res) {
-    LOG(ERROR) << "Failed get op tls reader object output";
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed get op tls reader object output";
   }
 
   Timer timeline;
   int64_t start = timeline.TimeStampUS();
 
-  VLOG(2) << "Going to init lod tensor";
+  VLOG(2) << "(logid=" << log_id << ") Going to init lod tensor";
   for (int i = 0; i < in->size(); ++i) {
     paddle::PaddleTensor lod_tensor;
     CopyLod(&in->at(i), &lod_tensor);
     lod_tensor.dtype = in->at(i).dtype;
     lod_tensor.name = in->at(i).name;
-    VLOG(2) << "lod tensor [" << i << "].name = " << lod_tensor.name;
+    VLOG(2) << "(logid=" << log_id << ") lod tensor [" << i
+            << "].name = " << lod_tensor.name;
     out->push_back(lod_tensor);
   }
 
-  VLOG(2) << "pack done.";
+  VLOG(2) << "(logid=" << log_id << ") pack done.";
 
   for (int i = 0; i < out->size(); ++i) {
     int64_t *src_ptr = static_cast<int64_t *>(in->at(i).data.data());
@@ -86,7 +91,7 @@ int GeneralCopyOp::inference() {
     }
   }
 
-  VLOG(2) << "output done.";
+  VLOG(2) << "(logid=" << log_id << ") output done.";
 
   timeline.Pause();
   int64_t end = timeline.TimeStampUS();
@@ -94,7 +99,7 @@ int GeneralCopyOp::inference() {
   AddBlobInfo(res, start);
   AddBlobInfo(res, end);
 
-  VLOG(2) << "read data from client success";
+  VLOG(2) << "(logid=" << log_id << ") read data from client success";
   return 0;
 }
 

core/general-server/op/general_copy_op.h

@@ -13,20 +13,12 @@
 // limitations under the License.
 
 #pragma once
-#include <vector>
-#ifdef BCLOUD
-#ifdef WITH_GPU
-#include "paddle/paddle_inference_api.h"
-#else
-#include "paddle/fluid/inference/api/paddle_inference_api.h"
-#endif
-#else
-#include "paddle_inference_api.h"  // NOLINT
-#endif
 #include <string>
+#include <vector>
 #include "core/general-server/general_model_service.pb.h"
 #include "core/general-server/op/general_infer_helper.h"
 #include "core/predictor/framework/resource.h"
+#include "paddle_inference_api.h"  // NOLINT
 
 namespace baidu {
 namespace paddle_serving {

core/general-server/op/general_dist_kv_infer_op.cpp

@@ -50,18 +50,20 @@ int GeneralDistKVInferOp::inference() {
   const std::string pre_name = pre_node_names[0];
 
   const GeneralBlob *input_blob = get_depend_argument<GeneralBlob>(pre_name);
-  VLOG(2) << "Get precedent op name: " << pre_name;
+  uint64_t log_id = input_blob->GetLogId();
+  VLOG(2) << "(logid=" << log_id << ") Get precedent op name: " << pre_name;
   GeneralBlob *output_blob = mutable_data<GeneralBlob>();
 
   if (!input_blob) {
-    LOG(ERROR) << "Failed mutable depended argument, op:" << pre_name;
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed mutable depended argument, op:" << pre_name;
     return -1;
   }
 
   const TensorVector *in = &input_blob->tensor_vector;
   TensorVector *out = &output_blob->tensor_vector;
   int batch_size = input_blob->GetBatchSize();
-  VLOG(2) << "input batch size: " << batch_size;
+  VLOG(2) << "(logid=" << log_id << ") input batch size: " << batch_size;
   std::vector<uint64_t> keys;
   std::vector<rec::mcube::CubeValue> values;
   int sparse_count = 0;
@@ -96,13 +98,14 @@ int GeneralDistKVInferOp::inference() {
   rec::mcube::CubeAPI *cube = rec::mcube::CubeAPI::instance();
   std::vector<std::string> table_names = cube->get_table_names();
   if (table_names.size() == 0) {
-    LOG(ERROR) << "cube init error or cube config not given.";
+    LOG(ERROR) << "(logid=" << log_id
+               << ") cube init error or cube config not given.";
     return -1;
   }
   int ret = cube->seek(table_names[0], keys, &values);
   int64_t cube_end = timeline.TimeStampUS();
   if (values.size() != keys.size() || values[0].buff.size() == 0) {
-    LOG(ERROR) << "cube value return null";
+    LOG(ERROR) << "(logid=" << log_id << ") cube value return null";
   }
   size_t EMBEDDING_SIZE = values[0].buff.size() / sizeof(float);
   TensorVector sparse_out;
@@ -153,14 +156,16 @@ int GeneralDistKVInferOp::inference() {
   infer_in.insert(infer_in.end(), sparse_out.begin(), sparse_out.end());
 
   output_blob->SetBatchSize(batch_size);
+  output_blob->SetLogId(log_id);
 
-  VLOG(2) << "infer batch size: " << batch_size;
+  VLOG(2) << "(logid=" << log_id << ") infer batch size: " << batch_size;
 
   int64_t start = timeline.TimeStampUS();
 
   if (InferManager::instance().infer(
           engine_name().c_str(), &infer_in, out, batch_size)) {
-    LOG(ERROR) << "Failed do infer in fluid model: " << engine_name();
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed do infer in fluid model: " << engine_name();
     return -1;
   }
 

core/general-server/op/general_dist_kv_infer_op.h

@@ -15,17 +15,9 @@
 #pragma once
 #include <string>
 #include <vector>
-#ifdef BCLOUD
-#ifdef WITH_GPU
-#include "paddle/paddle_inference_api.h"
-#else
-#include "paddle/fluid/inference/api/paddle_inference_api.h"
-#endif
-#else
-#include "paddle_inference_api.h"  // NOLINT
-#endif
 #include "core/general-server/general_model_service.pb.h"
 #include "core/general-server/op/general_infer_helper.h"
+#include "paddle_inference_api.h"  // NOLINT
 
 namespace baidu {
 namespace paddle_serving {

core/general-server/op/general_dist_kv_quant_infer_op.cpp

@@ -59,10 +59,13 @@ int GeneralDistKVQuantInferOp::inference() {
     return -1;
   }
 
+  uint64_t log_id = input_blob->GetLogId();
+  output_blob->SetLogId(log_id);
+
   const TensorVector *in = &input_blob->tensor_vector;
   TensorVector *out = &output_blob->tensor_vector;
   int batch_size = input_blob->GetBatchSize();
-  VLOG(2) << "input batch size: " << batch_size;
+  VLOG(2) << "(logid=" << log_id << ") input batch size: " << batch_size;
   std::vector<uint64_t> keys;
   std::vector<rec::mcube::CubeValue> values;
   int sparse_count = 0;
@@ -94,13 +97,14 @@ int GeneralDistKVQuantInferOp::inference() {
   rec::mcube::CubeAPI *cube = rec::mcube::CubeAPI::instance();
   std::vector<std::string> table_names = cube->get_table_names();
   if (table_names.size() == 0) {
-    LOG(ERROR) << "cube init error or cube config not given.";
+    LOG(ERROR) << "(logid=" << log_id
+               << ") cube init error or cube config not given.";
     return -1;
   }
   int ret = cube->seek(table_names[0], keys, &values);
 
   if (values.size() != keys.size() || values[0].buff.size() == 0) {
-    LOG(ERROR) << "cube value return null";
+    LOG(ERROR) << "(logid=" << log_id << ") cube value return null";
   }
 
   TensorVector sparse_out;
@@ -182,7 +186,7 @@ int GeneralDistKVQuantInferOp::inference() {
 
   output_blob->SetBatchSize(batch_size);
 
-  VLOG(2) << "infer batch size: " << batch_size;
+  VLOG(2) << "(logid=" << log_id << ") infer batch size: " << batch_size;
 
   Timer timeline;
   int64_t start = timeline.TimeStampUS();
@@ -190,7 +194,8 @@ int GeneralDistKVQuantInferOp::inference() {
 
   if (InferManager::instance().infer(
           engine_name().c_str(), &infer_in, out, batch_size)) {
-    LOG(ERROR) << "Failed do infer in fluid model: " << engine_name();
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed do infer in fluid model: " << engine_name();
     return -1;
   }
 

core/general-server/op/general_dist_kv_quant_infer_op.h

@@ -15,17 +15,9 @@
 #pragma once
 #include <string>
 #include <vector>
-#ifdef BCLOUD
-#ifdef WITH_GPU
-#include "paddle/paddle_inference_api.h"
-#else
-#include "paddle/fluid/inference/api/paddle_inference_api.h"
-#endif
-#else
-#include "paddle_inference_api.h"  // NOLINT
-#endif
 #include "core/general-server/general_model_service.pb.h"
 #include "core/general-server/op/general_infer_helper.h"
+#include "paddle_inference_api.h"  // NOLINT
 
 namespace baidu {
 namespace paddle_serving {

core/general-server/op/general_infer_helper.h

@@ -15,17 +15,9 @@
 #pragma once
 
 #include <string.h>
+#include <string>
 #include <vector>
-#ifdef BCLOUD
-#ifdef WITH_GPU
-#include "paddle/paddle_inference_api.h"
-#else
-#include "paddle/fluid/inference/api/paddle_inference_api.h"
-#endif
-#else
 #include "paddle_inference_api.h"  // NOLINT
-#endif
-#include <string>
 
 namespace baidu {
 namespace paddle_serving {
@@ -35,6 +27,7 @@ struct GeneralBlob {
   std::vector<paddle::PaddleTensor> tensor_vector;
   int64_t time_stamp[20];
   int p_size = 0;
+  uint64_t _log_id = -1;  // for logging
 
   int _batch_size;
 
@@ -46,9 +39,11 @@ struct GeneralBlob {
     tensor_vector.clear();
   }
 
-  int SetBatchSize(int batch_size) { _batch_size = batch_size; }
+  void SetBatchSize(int batch_size) { _batch_size = batch_size; }
+  void SetLogId(uint64_t log_id) { _log_id = log_id; }
 
   int GetBatchSize() const { return _batch_size; }
+  uint64_t GetLogId() const { return _log_id; }
   std::string ShortDebugString() const { return "Not implemented!"; }
 };
 

core/general-server/op/general_infer_op.cpp

@@ -47,22 +47,26 @@ int GeneralInferOp::inference() {
   const std::string pre_name = pre_node_names[0];
 
   const GeneralBlob *input_blob = get_depend_argument<GeneralBlob>(pre_name);
-  VLOG(2) << "Get precedent op name: " << pre_name;
+  uint64_t log_id = input_blob->GetLogId();
+  VLOG(2) << "(logid=" << log_id << ") Get precedent op name: " << pre_name;
   GeneralBlob *output_blob = mutable_data<GeneralBlob>();
+  output_blob->SetLogId(log_id);
 
   if (!input_blob) {
-    LOG(ERROR) << "Failed mutable depended argument, op:" << pre_name;
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed mutable depended argument, op:" << pre_name;
     return -1;
   }
 
   const TensorVector *in = &input_blob->tensor_vector;
   TensorVector *out = &output_blob->tensor_vector;
-  int batch_size = input_blob->GetBatchSize();
-  VLOG(2) << "input batch size: " << batch_size;
 
-  output_blob->SetBatchSize(batch_size);
+  int batch_size = input_blob->_batch_size;
+  VLOG(2) << "(logid=" << log_id << ") input batch size: " << batch_size;
 
-  VLOG(2) << "infer batch size: " << batch_size;
+  output_blob->_batch_size = batch_size;
+
+  VLOG(2) << "(logid=" << log_id << ") infer batch size: " << batch_size;
 
   Timer timeline;
   int64_t start = timeline.TimeStampUS();
@@ -70,7 +74,8 @@ int GeneralInferOp::inference() {
 
   if (InferManager::instance().infer(
           engine_name().c_str(), in, out, batch_size)) {
-    LOG(ERROR) << "Failed do infer in fluid model: " << engine_name().c_str();
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed do infer in fluid model: " << engine_name().c_str();
     return -1;
   }
 

core/general-server/op/general_infer_op.h

@@ -15,17 +15,9 @@
 #pragma once
 #include <string>
 #include <vector>
-#ifdef BCLOUD
-#ifdef WITH_GPU
-#include "paddle/paddle_inference_api.h"
-#else
-#include "paddle/fluid/inference/api/paddle_inference_api.h"
-#endif
-#else
-#include "paddle_inference_api.h"  // NOLINT
-#endif
 #include "core/general-server/general_model_service.pb.h"
 #include "core/general-server/op/general_infer_helper.h"
+#include "paddle_inference_api.h"  // NOLINT
 
 namespace baidu {
 namespace paddle_serving {

core/general-server/op/general_reader_op.cpp

@@ -37,9 +37,9 @@ int conf_check(const Request *req,
                const std::shared_ptr<PaddleGeneralModelConfig> &model_config) {
   int var_num = req->insts(0).tensor_array_size();
   if (var_num != model_config->_feed_type.size()) {
-    VLOG(2) << "var num: " << var_num;
-    VLOG(2) << "model config var num: " << model_config->_feed_type.size();
-    LOG(ERROR) << "feed var number not match.";
+    LOG(ERROR) << "feed var number not match: model config["
+               << model_config->_feed_type.size() << "] vs. actual[" << var_num
+               << "]";
     return -1;
   }
 
@@ -72,6 +72,7 @@ int conf_check(const Request *req,
 int GeneralReaderOp::inference() {
   // reade request from client
   const Request *req = dynamic_cast<const Request *>(get_request_message());
+  uint64_t log_id = req->log_id();
 
   int batch_size = req->insts_size();
   int input_var_num = 0;
@@ -83,25 +84,29 @@ int GeneralReaderOp::inference() {
   TensorVector *out = &res->tensor_vector;
 
   res->SetBatchSize(batch_size);
+  res->SetLogId(log_id);
 
   if (!res) {
-    LOG(ERROR) << "Failed get op tls reader object output";
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed get op tls reader object output";
   }
 
   Timer timeline;
   int64_t start = timeline.TimeStampUS();
   int var_num = req->insts(0).tensor_array_size();
-  VLOG(2) << "var num: " << var_num;
+  VLOG(2) << "(logid=" << log_id << ") var num: " << var_num;
 
-  VLOG(2) << "start to call load general model_conf op";
+  VLOG(2) << "(logid=" << log_id
+          << ") start to call load general model_conf op";
   baidu::paddle_serving::predictor::Resource &resource =
       baidu::paddle_serving::predictor::Resource::instance();
 
-  VLOG(2) << "get resource pointer done.";
+  VLOG(2) << "(logid=" << log_id << ") get resource pointer done.";
+
   std::shared_ptr<PaddleGeneralModelConfig> model_config =
       resource.get_general_model_config();
 
-  VLOG(2) << "print general model config done.";
+  VLOG(2) << "(logid=" << log_id << ") print general model config done.";
 
   // TODO(guru4elephant): how to do conditional check?
   /*
@@ -122,7 +127,8 @@ int GeneralReaderOp::inference() {
   for (int i = 0; i < var_num; ++i) {
     paddle::PaddleTensor lod_tensor;
     elem_type[i] = req->insts(0).tensor_array(i).elem_type();
-    VLOG(2) << "var[" << i << "] has elem type: " << elem_type[i];
+    VLOG(2) << "(logid=" << log_id << ") var[" << i
+            << "] has elem type: " << elem_type[i];
     if (elem_type[i] == 0) {  // int64
       elem_size[i] = sizeof(int64_t);
       lod_tensor.dtype = paddle::PaddleDType::INT64;
@@ -137,17 +143,19 @@ int GeneralReaderOp::inference() {
     if (model_config->_is_lod_feed[i]) {
       lod_tensor.lod.resize(1);
       lod_tensor.lod[0].push_back(0);
-      VLOG(2) << "var[" << i << "] is lod_tensor";
+      VLOG(2) << "(logid=" << log_id << ") var[" << i << "] is lod_tensor";
     } else {
       lod_tensor.shape.push_back(batch_size);
       capacity[i] = 1;
       for (int k = 0; k < req->insts(0).tensor_array(i).shape_size(); ++k) {
         int dim = req->insts(0).tensor_array(i).shape(k);
-        VLOG(2) << "shape for var[" << i << "]: " << dim;
+        VLOG(2) << "(logid=" << log_id << ") shape for var[" << i
+                << "]: " << dim;
         capacity[i] *= dim;
         lod_tensor.shape.push_back(dim);
       }
-      VLOG(2) << "var[" << i << "] is tensor, capacity: " << capacity[i];
+      VLOG(2) << "(logid=" << log_id << ") var[" << i
+              << "] is tensor, capacity: " << capacity[i];
     }
     lod_tensor.name = model_config->_feed_name[i];
     out->push_back(lod_tensor);
@@ -167,11 +175,12 @@ int GeneralReaderOp::inference() {
         } else if (tensor.int_data_size() > 0) {
           data_len = tensor.int_data_size();
         }
-        VLOG(2) << "tensor size for var[" << i << "]: " << data_len;
+        VLOG(2) << "(logid=" << log_id << ") tensor size for var[" << i
+                << "]: " << data_len;
         tensor_size += data_len;
 
         int cur_len = out->at(i).lod[0].back();
-        VLOG(2) << "current len: " << cur_len;
+        VLOG(2) << "(logid=" << log_id << ") current len: " << cur_len;
 
         int sample_len = 0;
         if (tensor.shape_size() == 1) {
@@ -180,7 +189,7 @@ int GeneralReaderOp::inference() {
           sample_len = tensor.shape(0);
         }
         out->at(i).lod[0].push_back(cur_len + sample_len);
-        VLOG(2) << "new len: " << cur_len + sample_len;
+        VLOG(2) << "(logid=" << log_id << ") new len: " << cur_len + sample_len;
       }
       out->at(i).data.Resize(tensor_size * elem_size[i]);
       out->at(i).shape = {out->at(i).lod[0].back()};
@@ -190,11 +199,11 @@ int GeneralReaderOp::inference() {
       if (out->at(i).shape.size() == 1) {
         out->at(i).shape.push_back(1);
       }
-      VLOG(2) << "var[" << i
+      VLOG(2) << "(logid=" << log_id << ") var[" << i
               << "] is lod_tensor and len=" << out->at(i).lod[0].back();
     } else {
       out->at(i).data.Resize(batch_size * capacity[i] * elem_size[i]);
-      VLOG(2) << "var[" << i
+      VLOG(2) << "(logid=" << log_id << ") var[" << i
               << "] is tensor and capacity=" << batch_size * capacity[i];
     }
   }
@@ -203,8 +212,8 @@ int GeneralReaderOp::inference() {
   for (int i = 0; i < var_num; ++i) {
     if (elem_type[i] == 0) {
       int64_t *dst_ptr = static_cast<int64_t *>(out->at(i).data.data());
-      VLOG(2) << "first element data in var[" << i << "] is "
-              << req->insts(0).tensor_array(i).int64_data(0);
+      VLOG(2) << "(logid=" << log_id << ") first element data in var[" << i
+              << "] is " << req->insts(0).tensor_array(i).int64_data(0);
       int offset = 0;
       for (int j = 0; j < batch_size; ++j) {
         int elem_num = req->insts(j).tensor_array(i).int64_data_size();
@@ -219,8 +228,8 @@ int GeneralReaderOp::inference() {
       }
     } else if (elem_type[i] == 1) {
       float *dst_ptr = static_cast<float *>(out->at(i).data.data());
-      VLOG(2) << "first element data in var[" << i << "] is "
-              << req->insts(0).tensor_array(i).float_data(0);
+      VLOG(2) << "(logid=" << log_id << ") first element data in var[" << i
+              << "] is " << req->insts(0).tensor_array(i).float_data(0);
       int offset = 0;
       for (int j = 0; j < batch_size; ++j) {
         int elem_num = req->insts(j).tensor_array(i).float_data_size();
@@ -235,8 +244,8 @@ int GeneralReaderOp::inference() {
       }
     } else if (elem_type[i] == 2) {
       int32_t *dst_ptr = static_cast<int32_t *>(out->at(i).data.data());
-      VLOG(2) << "first element data in var[" << i << "] is "
-              << req->insts(0).tensor_array(i).int_data(0);
+      VLOG(2) << "(logid=" << log_id << ") first element data in var[" << i
+              << "] is " << req->insts(0).tensor_array(i).int_data(0);
       int offset = 0;
       for (int j = 0; j < batch_size; ++j) {
         int elem_num = req->insts(j).tensor_array(i).int_data_size();
@@ -252,15 +261,16 @@ int GeneralReaderOp::inference() {
     }
   }
 
-  VLOG(2) << "output size: " << out->size();
+  VLOG(2) << "(logid=" << log_id << ") output size: " << out->size();
 
   timeline.Pause();
   int64_t end = timeline.TimeStampUS();
   res->p_size = 0;
+  res->_batch_size = batch_size;
   AddBlobInfo(res, start);
   AddBlobInfo(res, end);
 
-  VLOG(2) << "read data from client success";
+  VLOG(2) << "(logid=" << log_id << ") read data from client success";
   return 0;
 }
 DEFINE_OP(GeneralReaderOp);

core/general-server/op/general_reader_op.h

@@ -13,21 +13,13 @@
 // limitations under the License.
 
 #pragma once
-#include <vector>
-#ifdef BCLOUD
-#ifdef WITH_GPU
-#include "paddle/paddle_inference_api.h"
-#else
-#include "paddle/fluid/inference/api/paddle_inference_api.h"
-#endif
-#else
-#include "paddle_inference_api.h"  // NOLINT
-#endif
 #include <string>
+#include <vector>
 #include "core/general-server/general_model_service.pb.h"
 #include "core/general-server/load_general_model_service.pb.h"
 #include "core/general-server/op/general_infer_helper.h"
 #include "core/predictor/framework/resource.h"
+#include "paddle_inference_api.h"  // NOLINT
 
 namespace baidu {
 namespace paddle_serving {

core/general-server/op/general_response_op.cpp

@@ -42,6 +42,9 @@ using baidu::paddle_serving::predictor::PaddleGeneralModelConfig;
 int GeneralResponseOp::inference() {
   const std::vector<std::string> pre_node_names = pre_names();
   VLOG(2) << "pre node names size: " << pre_node_names.size();
+  const GeneralBlob *input_blob;
+  uint64_t log_id =
+      get_depend_argument<GeneralBlob>(pre_node_names[0])->GetLogId();
 
   const Request *req = dynamic_cast<const Request *>(get_request_message());
   // response inst with only fetch_var_names
@@ -52,15 +55,17 @@ int GeneralResponseOp::inference() {
   // timeline.Start();
   int64_t start = timeline.TimeStampUS();
 
-  VLOG(2) << "start to call load general model_conf op";
+  VLOG(2) << "(logid=" << log_id
+          << ") start to call load general model_conf op";
   baidu::paddle_serving::predictor::Resource &resource =
       baidu::paddle_serving::predictor::Resource::instance();
 
-  VLOG(2) << "get resource pointer done.";
+  VLOG(2) << "(logid=" << log_id << ") get resource pointer done.";
   std::shared_ptr<PaddleGeneralModelConfig> model_config =
       resource.get_general_model_config();
 
-  VLOG(2) << "max body size : " << brpc::fLU64::FLAGS_max_body_size;
+  VLOG(2) << "(logid=" << log_id
+          << ") max body size : " << brpc::fLU64::FLAGS_max_body_size;
 
   std::vector<int> fetch_index;
   fetch_index.resize(req->fetch_var_names_size());
@@ -69,16 +74,16 @@ int GeneralResponseOp::inference() {
         model_config->_fetch_alias_name_to_index[req->fetch_var_names(i)];
   }
 
-  const GeneralBlob *input_blob;
   for (uint32_t pi = 0; pi < pre_node_names.size(); ++pi) {
     const std::string &pre_name = pre_node_names[pi];
-    VLOG(2) << "pre names[" << pi << "]: " << pre_name << " ("
-            << pre_node_names.size() << ")";
+    VLOG(2) << "(logid=" << log_id << ") pre names[" << pi << "]: " << pre_name
+            << " (" << pre_node_names.size() << ")";
     input_blob = get_depend_argument<GeneralBlob>(pre_name);
     // fprintf(stderr, "input(%s) blob address %x\n", pre_names.c_str(),
     // input_blob);
     if (!input_blob) {
-      LOG(ERROR) << "Failed mutable depended argument, op: " << pre_name;
+      LOG(ERROR) << "(logid=" << log_id
+                 << ") Failed mutable depended argument, op: " << pre_name;
       return -1;
     }
 
@@ -92,17 +97,19 @@ int GeneralResponseOp::inference() {
     for (auto &idx : fetch_index) {
       Tensor *tensor = fetch_inst->add_tensor_array();
       if (model_config->_is_lod_fetch[idx]) {
-        VLOG(2) << "out[" << idx << "] " << model_config->_fetch_name[idx]
-                << " is lod_tensor";
+        VLOG(2) << "(logid=" << log_id << ") out[" << idx << "] "
+                << model_config->_fetch_name[idx] << " is lod_tensor";
         for (int k = 0; k < in->at(idx).shape.size(); ++k) {
-          VLOG(2) << "shape[" << k << "]: " << in->at(idx).shape[k];
+          VLOG(2) << "(logid=" << log_id << ") shape[" << k
+                  << "]: " << in->at(idx).shape[k];
           tensor->add_shape(in->at(idx).shape[k]);
         }
       } else {
-        VLOG(2) << "out[" << idx << "] " << model_config->_fetch_name[idx]
-                << " is tensor";
+        VLOG(2) << "(logid=" << log_id << ") out[" << idx << "] "
+                << model_config->_fetch_name[idx] << " is tensor";
         for (int k = 0; k < in->at(idx).shape.size(); ++k) {
-          VLOG(2) << "shape[" << k << "]: " << in->at(idx).shape[k];
+          VLOG(2) << "(logid=" << log_id << ") shape[" << k
+                  << "]: " << in->at(idx).shape[k];
           tensor->add_shape(in->at(idx).shape[k]);
         }
       }
@@ -119,8 +126,8 @@ int GeneralResponseOp::inference() {
       auto dtype = in->at(idx).dtype;
 
       if (dtype == paddle::PaddleDType::INT64) {
-        VLOG(2) << "Prepare int64 var [" << model_config->_fetch_name[idx]
-                << "].";
+        VLOG(2) << "(logid=" << log_id << ") Prepare int64 var ["
+                << model_config->_fetch_name[idx] << "].";
         int64_t *data_ptr = static_cast<int64_t *>(in->at(idx).data.data());
         // from
         // https://stackoverflow.com/questions/15499641/copy-a-stdvector-to-a-repeated-field-from-protobuf-with-memcpy
@@ -130,16 +137,16 @@ int GeneralResponseOp::inference() {
         fetch_p->mutable_tensor_array(var_idx)->mutable_int64_data()->Swap(
             &tmp_data);
       } else if (dtype == paddle::PaddleDType::FLOAT32) {
-        VLOG(2) << "Prepare float var [" << model_config->_fetch_name[idx]
-                << "].";
+        VLOG(2) << "(logid=" << log_id << ") Prepare float var ["
+                << model_config->_fetch_name[idx] << "].";
         float *data_ptr = static_cast<float *>(in->at(idx).data.data());
         google::protobuf::RepeatedField<float> tmp_data(data_ptr,
                                                         data_ptr + cap);
         fetch_p->mutable_tensor_array(var_idx)->mutable_float_data()->Swap(
             &tmp_data);
       } else if (dtype == paddle::PaddleDType::INT32) {
-        VLOG(2) << "Prepare int32 var [" << model_config->_fetch_name[idx]
-                << "].";
+        VLOG(2) << "(logid=" << log_id << ")Prepare int32 var ["
+                << model_config->_fetch_name[idx] << "].";
         int32_t *data_ptr = static_cast<int32_t *>(in->at(idx).data.data());
         google::protobuf::RepeatedField<int32_t> tmp_data(data_ptr,
                                                           data_ptr + cap);
@@ -154,7 +161,8 @@ int GeneralResponseOp::inference() {
         }
       }
 
-      VLOG(2) << "fetch var [" << model_config->_fetch_name[idx] << "] ready";
+      VLOG(2) << "(logid=" << log_id << ") fetch var ["
+              << model_config->_fetch_name[idx] << "] ready";
       var_idx++;
     }
   }
@@ -167,7 +175,8 @@ int GeneralResponseOp::inference() {
     // a more elegant way.
     for (uint32_t pi = 0; pi < pre_node_names.size(); ++pi) {
       input_blob = get_depend_argument<GeneralBlob>(pre_node_names[pi]);
-      VLOG(2) << "p size for input blob: " << input_blob->p_size;
+      VLOG(2) << "(logid=" << log_id
+              << ") p size for input blob: " << input_blob->p_size;
       int profile_time_idx = -1;
       if (pi == 0) {
         profile_time_idx = 0;

core/general-server/op/general_response_op.h

@@ -15,16 +15,8 @@
 #pragma once
 #include <string>
 #include <vector>
-#ifdef BCLOUD
-#ifdef WITH_GPU
-#include "paddle/paddle_inference_api.h"
-#else
-#include "paddle/fluid/inference/api/paddle_inference_api.h"
-#endif
-#else
-#include "paddle_inference_api.h"  // NOLINT
-#endif
 #include "core/general-server/general_model_service.pb.h"
+#include "paddle_inference_api.h"  // NOLINT
 
 namespace baidu {
 namespace paddle_serving {

core/general-server/op/general_text_reader_op.cpp

@@ -35,6 +35,7 @@ using baidu::paddle_serving::predictor::PaddleGeneralModelConfig;
 int GeneralTextReaderOp::inference() {
   // reade request from client
   const Request *req = dynamic_cast<const Request *>(get_request_message());
+  uint64_t log_id = req->log_id();
 
   int batch_size = req->insts_size();
   int input_var_num = 0;
@@ -44,16 +45,18 @@ int GeneralTextReaderOp::inference() {
   std::vector<int64_t> capacity;
 
   GeneralBlob *res = mutable_data<GeneralBlob>();
-  TensorVector *out = &res->tensor_vector;
-
-  res->SetBatchSize(batch_size);
 
   if (!res) {
-    LOG(ERROR) << "Failed get op tls reader object output";
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed get op tls reader object output";
   }
 
+  TensorVector *out = &res->tensor_vector;
+  res->SetBatchSize(batch_size);
+  res->SetLogId(log_id);
+
   if (batch_size <= 0) {
-    LOG(ERROR) << "Batch size < 0";
+    LOG(ERROR) << "(logid=" << log_id << ") Batch size < 0";
     return -1;
   }
 
@@ -61,17 +64,18 @@ int GeneralTextReaderOp::inference() {
   int64_t start = timeline.TimeStampUS();
 
   int var_num = req->insts(0).tensor_array_size();
-  VLOG(2) << "var num: " << var_num;
+  VLOG(2) << "(logid=" << log_id << ") var num: " << var_num;
 
-  VLOG(2) << "start to call load general model_conf op";
+  VLOG(2) << "(logid=" << log_id
+          << ") start to call load general model_conf op";
   baidu::paddle_serving::predictor::Resource &resource =
       baidu::paddle_serving::predictor::Resource::instance();
 
-  VLOG(2) << "get resource pointer done.";
+  VLOG(2) << "(logid=" << log_id << ") get resource pointer done.";
   std::shared_ptr<PaddleGeneralModelConfig> model_config =
       resource.get_general_model_config();
 
-  VLOG(2) << "print general model config done.";
+  VLOG(2) << "(logid=" << log_id << ") print general model config done.";
 
   elem_type.resize(var_num);
   elem_size.resize(var_num);
@@ -79,7 +83,8 @@ int GeneralTextReaderOp::inference() {
   for (int i = 0; i < var_num; ++i) {
     paddle::PaddleTensor lod_tensor;
     elem_type[i] = req->insts(0).tensor_array(i).elem_type();
-    VLOG(2) << "var[" << i << "] has elem type: " << elem_type[i];
+    VLOG(2) << "(logid=" << log_id << ") var[" << i
+            << "] has elem type: " << elem_type[i];
     if (elem_type[i] == 0) {  // int64
       elem_size[i] = sizeof(int64_t);
       lod_tensor.dtype = paddle::PaddleDType::INT64;
@@ -91,17 +96,19 @@ int GeneralTextReaderOp::inference() {
     if (req->insts(0).tensor_array(i).shape(0) == -1) {
       lod_tensor.lod.resize(1);
       lod_tensor.lod[0].push_back(0);
-      VLOG(2) << "var[" << i << "] is lod_tensor";
+      VLOG(2) << "(logid=" << log_id << ") var[" << i << "] is lod_tensor";
     } else {
       lod_tensor.shape.push_back(batch_size);
       capacity[i] = 1;
       for (int k = 0; k < req->insts(0).tensor_array(i).shape_size(); ++k) {
         int dim = req->insts(0).tensor_array(i).shape(k);
-        VLOG(2) << "shape for var[" << i << "]: " << dim;
+        VLOG(2) << "(logid=" << log_id << ") shape for var[" << i
+                << "]: " << dim;
         capacity[i] *= dim;
         lod_tensor.shape.push_back(dim);
       }
-      VLOG(2) << "var[" << i << "] is tensor, capacity: " << capacity[i];
+      VLOG(2) << "(logid=" << log_id << ") var[" << i
+              << "] is tensor, capacity: " << capacity[i];
     }
     lod_tensor.name = model_config->_feed_name[i];
     out->push_back(lod_tensor);
@@ -117,11 +124,11 @@ int GeneralTextReaderOp::inference() {
       }
       out->at(i).data.Resize(out->at(i).lod[0].back() * elem_size[i]);
       out->at(i).shape = {out->at(i).lod[0].back(), 1};
-      VLOG(2) << "var[" << i
+      VLOG(2) << "(logid=" << log_id << ") var[" << i
               << "] is lod_tensor and len=" << out->at(i).lod[0].back();
     } else {
       out->at(i).data.Resize(batch_size * capacity[i] * elem_size[i]);
-      VLOG(2) << "var[" << i
+      VLOG(2) << "(logid=" << log_id << ") var[" << i
               << "] is tensor and capacity=" << batch_size * capacity[i];
     }
   }
@@ -163,7 +170,7 @@ int GeneralTextReaderOp::inference() {
   AddBlobInfo(res, start);
   AddBlobInfo(res, end);
 
-  VLOG(2) << "read data from client success";
+  VLOG(2) << "(logid=" << log_id << ") read data from client success";
   return 0;
 }
 DEFINE_OP(GeneralTextReaderOp);

core/general-server/op/general_text_reader_op.h

@@ -13,21 +13,13 @@
 // limitations under the License.
 
 #pragma once
-#include <vector>
-#ifdef BCLOUD
-#ifdef WITH_GPU
-#include "paddle/paddle_inference_api.h"
-#else
-#include "paddle/fluid/inference/api/paddle_inference_api.h"
-#endif
-#else
-#include "paddle_inference_api.h"  // NOLINT
-#endif
 #include <string>
+#include <vector>
 #include "core/general-server/general_model_service.pb.h"
 #include "core/general-server/load_general_model_service.pb.h"
 #include "core/general-server/op/general_infer_helper.h"
 #include "core/predictor/framework/resource.h"
+#include "paddle_inference_api.h"  // NOLINT
 
 namespace baidu {
 namespace paddle_serving {

core/general-server/op/general_text_response_op.cpp

@@ -40,6 +40,9 @@ int GeneralTextResponseOp::inference() {
   VLOG(2) << "Going to run inference";
   const std::vector<std::string> pre_node_names = pre_names();
   VLOG(2) << "pre node names size: " << pre_node_names.size();
+  const GeneralBlob *input_blob;
+  uint64_t log_id =
+      get_depend_argument<GeneralBlob>(pre_node_names[0])->GetLogId();
 
   const Request *req = dynamic_cast<const Request *>(get_request_message());
   // response inst with only fetch_var_names
@@ -48,11 +51,12 @@ int GeneralTextResponseOp::inference() {
   Timer timeline;
   int64_t start = timeline.TimeStampUS();
 
-  VLOG(2) << "start to call load general model_conf op";
+  VLOG(2) << "(logid=" << log_id
+          << ") start to call load general model_conf op";
   baidu::paddle_serving::predictor::Resource &resource =
       baidu::paddle_serving::predictor::Resource::instance();
 
-  VLOG(2) << "get resource pointer done.";
+  VLOG(2) << "(logid=" << log_id << ") get resource pointer done.";
   std::shared_ptr<PaddleGeneralModelConfig> model_config =
       resource.get_general_model_config();
 
@@ -63,20 +67,20 @@ int GeneralTextResponseOp::inference() {
         model_config->_fetch_alias_name_to_index[req->fetch_var_names(i)];
   }
 
-  const GeneralBlob *input_blob;
   for (uint32_t pi = 0; pi < pre_node_names.size(); ++pi) {
     const std::string &pre_name = pre_node_names[pi];
-    VLOG(2) << "pre names[" << pi << "]: " << pre_name << " ("
-            << pre_node_names.size() << ")";
+    VLOG(2) << "(logid=" << log_id << ") pre names[" << pi << "]: " << pre_name
+            << " (" << pre_node_names.size() << ")";
     input_blob = get_depend_argument<GeneralBlob>(pre_name);
     if (!input_blob) {
-      LOG(ERROR) << "Failed mutable depended argument, op: " << pre_name;
+      LOG(ERROR) << "(logid=" << log_id
+                 << ") Failed mutable depended argument, op: " << pre_name;
       return -1;
     }
 
     const TensorVector *in = &input_blob->tensor_vector;
     int batch_size = input_blob->GetBatchSize();
-    VLOG(2) << "input batch size: " << batch_size;
+    VLOG(2) << "(logid=" << log_id << ") input batch size: " << batch_size;
 
     ModelOutput *output = res->add_outputs();
     output->set_engine_name(
@@ -88,12 +92,13 @@ int GeneralTextResponseOp::inference() {
         // currently only response float tensor or lod_tensor
         tensor->set_elem_type(1);
         if (model_config->_is_lod_fetch[idx]) {
-          VLOG(2) << "out[" << idx << " is lod_tensor";
+          VLOG(2) << "(logid=" << log_id << ") out[" << idx << " is lod_tensor";
           tensor->add_shape(-1);
         } else {
-          VLOG(2) << "out[" << idx << "] is tensor";
+          VLOG(2) << "(logid=" << log_id << ") out[" << idx << "] is tensor";
           for (int k = 1; k < in->at(idx).shape.size(); ++k) {
-            VLOG(2) << "shape[" << k - 1 << "]: " << in->at(idx).shape[k];
+            VLOG(2) << "(logid=" << log_id << ") shape[" << k - 1
+                    << "]: " << in->at(idx).shape[k];
             tensor->add_shape(in->at(idx).shape[k]);
           }
         }
@@ -137,7 +142,8 @@ int GeneralTextResponseOp::inference() {
     // a more elegant way.
     for (uint32_t pi = 0; pi < pre_node_names.size(); ++pi) {
       input_blob = get_depend_argument<GeneralBlob>(pre_node_names[pi]);
-      VLOG(2) << "p size for input blob: " << input_blob->p_size;
+      VLOG(2) << "(logid=" << log_id
+              << ") p size for input blob: " << input_blob->p_size;
       int profile_time_idx = -1;
       if (pi == 0) {
         profile_time_idx = 0;

core/general-server/op/general_text_response_op.h

@@ -15,17 +15,9 @@
 #pragma once
 #include <string>
 #include <vector>
-#ifdef BCLOUD
-#ifdef WITH_GPU
-#include "paddle/paddle_inference_api.h"
-#else
-#include "paddle/fluid/inference/api/paddle_inference_api.h"
-#endif
-#else
-#include "paddle_inference_api.h"  // NOLINT
-#endif
 #include "core/general-server/general_model_service.pb.h"
 #include "core/general-server/op/general_infer_helper.h"
+#include "paddle_inference_api.h"  // NOLINT
 
 namespace baidu {
 namespace paddle_serving {

core/general-server/proto/general_model_service.proto

@@ -37,6 +37,7 @@ message Request {
   repeated FeedInst insts = 1;
   repeated string fetch_var_names = 2;
   optional bool profile_server = 3 [ default = false ];
+  required uint64 log_id = 4 [ default = 0 ];
 };
 
 message Response {

core/general-server/proto/load_general_model_service.proto

@@ -21,6 +21,7 @@ option cc_generic_services = true;
 message RequestAndResponse {
   required int32 a = 1;
   required float b = 2;
+  required uint64 log_id = 3 [ default = 0 ];
 };
 
 service LoadGeneralModelService {

core/pdcodegen/src/pdcodegen.cpp

@@ -280,25 +280,29 @@ class PdsCodeGenerator : public CodeGenerator {
             "  baidu::rpc::ClosureGuard done_guard(done);\n"
             "  baidu::rpc::Controller* cntl = \n"
             "        static_cast<baidu::rpc::Controller*>(cntl_base);\n"
+            "  uint64_t log_id = request->log_id();\n"
+            "  cntl->set_log_id(log_id);\n"
             "  ::baidu::paddle_serving::predictor::InferService* svr = \n"
             "       "
             "::baidu::paddle_serving::predictor::InferServiceManager::instance("
             ").item(\"$service$\");\n"
             "  if (svr == NULL) {\n"
-            "    LOG(ERROR) << \"Not found service: $service$\";\n"
+            "    LOG(ERROR) << \"(logid=\" << log_id << \") Not found service: "
+            "$service$\";\n"
             "    cntl->SetFailed(404, \"Not found service: $service$\");\n"
             "    return ;\n"
             "  }\n"
-            "  LOG(INFO) << \" remote_side=\[\" << cntl->remote_side() << "  // NOLINT
-            "\"\]\";\n"
-            "  LOG(INFO) << \" local_side=\[\" << cntl->local_side() << "  // NOLINT
-            "\"\]\";\n"
-            "  LOG(INFO) << \" service_name=\[\" << \"$name$\" << \"\]\";\n"  // NOLINT
-            "  LOG(INFO) << \" log_id=\[\" << cntl->log_id() << \"\]\";\n"  // NOLINT
-            "  int err_code = svr->inference(request, response);\n"
+            "  LOG(INFO) << \"(logid=\" << log_id << \") remote_side=\[\" "  // NOLINT
+            "<< cntl->remote_side() << \"\]\";\n"
+            "  LOG(INFO) << \"(logid=\" << log_id << \") local_side=\[\" "  // NOLINT
+            "<< cntl->local_side() << \"\]\";\n"
+            "  LOG(INFO) << \"(logid=\" << log_id << \") service_name=\[\" "  // NOLINT
+            "<< \"$name$\" << \"\]\";\n"
+            "  int err_code = svr->inference(request, response, log_id);\n"
             "  if (err_code != 0) {\n"
             "    LOG(WARNING)\n"
-            "        << \"Failed call inferservice[$name$], name[$service$]\"\n"
+            "        << \"(logid=\" << log_id << \") Failed call "
+            "inferservice[$name$], name[$service$]\"\n"
             "        << \", error_code: \" << err_code;\n"
             "    cntl->SetFailed(err_code, \"InferService inference "
             "failed!\");\n"
@@ -306,7 +310,8 @@ class PdsCodeGenerator : public CodeGenerator {
             "  gettimeofday(&tv, NULL);\n"
             "  long end = tv.tv_sec * 1000000 + tv.tv_usec;\n"
             "  // flush notice log\n"
-            "  LOG(INFO) << \" tc=\[\" << (end - start) << \"\]\";\n",  // NOLINT
+            "  LOG(INFO) << \"(logid=\" << log_id << \") tc=\[\" << (end - "  // NOLINT
+            "start) << \"\]\";\n",  // NOLINT
             "name",
             class_name,
             "service",
@@ -317,26 +322,31 @@ class PdsCodeGenerator : public CodeGenerator {
             "  baidu::rpc::ClosureGuard done_guard(done);\n"
             "  baidu::rpc::Controller* cntl = \n"
             "        static_cast<baidu::rpc::Controller*>(cntl_base);\n"
+            "  uint64_t log_id = equest->log_id();\n"
+            "  cntl->set_log_id(log_id);\n"
             "  ::baidu::paddle_serving::predictor::InferService* svr = \n"
             "       "
             "::baidu::paddle_serving::predictor::InferServiceManager::instance("
             ").item(\"$service$\");\n"
             "  if (svr == NULL) {\n"
-            "    LOG(ERROR) << \"Not found service: $service$\";\n"
+            "    LOG(ERROR) << \"(logid=\" << log_id << \") Not found service: "
+            "$service$\";\n"
             "    cntl->SetFailed(404, \"Not found service: $service$\");\n"
             "    return ;\n"
             "  }\n"
-            "  LOG(INFO) << \" remote_side=\[\" << cntl->remote_side() << "  // NOLINT
-            "\"\]\";\n"
-            "  LOG(INFO) << \" local_side=\[\" << cntl->local_side() << "  // NOLINT
-            "\"\]\";\n"
-            "  LOG(INFO) << \" service_name=\[\" << \"$name$\" << \"\]\";\n"  // NOLINT
-            "  LOG(INFO) << \" log_id=\[\" << cntl->log_id() << \"\]\";\n"  // NOLINT
+            "  LOG(INFO) << \"(logid=\" << log_id << \") remote_side=\[\" "  // NOLINT
+            "<< cntl->remote_side() << \"\]\";\n"
+            "  LOG(INFO) << \"(logid=\" << log_id << \") local_side=\[\" "  // NOLINT
+            "<< cntl->local_side() << \"\]\";\n"
+            "  LOG(INFO) << \"(logid=\" << log_id << \") service_name=\[\" "  // NOLINT
+            "<< \"$name$\" << \"\]\";\n"
             "  butil::IOBufBuilder debug_os;\n"
-            "  int err_code = svr->inference(request, response, &debug_os);\n"
+            "  int err_code = svr->inference(request, response, log_id, "
+            "&debug_os);\n"
             "  if (err_code != 0) {\n"
             "    LOG(WARNING)\n"
-            "        << \"Failed call inferservice[$name$], name[$service$]\"\n"
+            "        << \"(logid=\" << log_id << \") Failed call "
+            "inferservice[$name$], name[$service$]\"\n"
             "        << \", error_code: \" << err_code;\n"
             "    cntl->SetFailed(err_code, \"InferService inference "
             "failed!\");\n"
@@ -345,9 +355,11 @@ class PdsCodeGenerator : public CodeGenerator {
             "  gettimeofday(&tv, NULL);\n"
             "  long end = tv.tv_sec * 1000000 + tv.tv_usec;\n"
             "  // flush notice log\n"
-            "  LOG(INFO) << \" tc=\[\" << (end - start) << \"\]\";\n"  // NOLINT
+            "  LOG(INFO) << \"(logid=\" << log_id << \") tc=\[\" << (end - "  // NOLINT
+            "start) << \"\]\";\n"
             "  LOG(INFO)\n"
-            "      << \"TC=[\" << (end - start) << \"] Received debug "
+            "      << \"(logid=\" << log_id << \") TC=[\" << (end - start) << "
+            "\"] Received debug "
             "request[log_id=\" << cntl->log_id()\n"
             "      << \"] from \" << cntl->remote_side()\n"
             "      << \" to \" << cntl->local_side();\n",
@@ -1011,25 +1023,31 @@ class PdsCodeGenerator : public CodeGenerator {
             "  brpc::ClosureGuard done_guard(done);\n"
             "  brpc::Controller* cntl = \n"
             "        static_cast<brpc::Controller*>(cntl_base);\n"
+            "  uint64_t log_id = request->log_id();\n"
+            "  cntl->set_log_id(log_id);\n"
             "  ::baidu::paddle_serving::predictor::InferService* svr = \n"
             "       "
             "::baidu::paddle_serving::predictor::InferServiceManager::instance("
             ").item(\"$service$\");\n"
             "  if (svr == NULL) {\n"
-            "    LOG(ERROR) << \"Not found service: $service$\";\n"
+            "    LOG(ERROR) << \"(logid=\" << log_id << \") Not found service: "
+            "$service$\";\n"
             "    cntl->SetFailed(404, \"Not found service: $service$\");\n"
             "    return ;\n"
             "  }\n"
-            "  LOG(INFO) << \" remote_side=\[\" << cntl->remote_side() << "  // NOLINT
+            "  LOG(INFO) << \"(logid=\" << log_id << \") "
+            "remote_side=\[\" << cntl->remote_side() << "  // NOLINT
             "\"\]\";\n"
-            "  LOG(INFO) << \" local_side=\[\" << cntl->local_side() << "  // NOLINT
+            "  LOG(INFO) << \"(logid=\" << log_id << \") "
+            "local_side=\[\" << cntl->local_side() << "  // NOLINT
             "\"\]\";\n"
-            "  LOG(INFO) << \" service_name=\[\" << \"$name$\" << \"\]\";\n"  // NOLINT
-            "  LOG(INFO) << \" log_id=\[\" << cntl->log_id() << \"\]\";\n"  // NOLINT
-            "  int err_code = svr->inference(request, response);\n"
+            "  LOG(INFO) << \"(logid=\" << log_id << \") "
+            "service_name=\[\" << \"$name$\" << \"\]\";\n"  // NOLINT
+            "  int err_code = svr->inference(request, response, log_id);\n"
             "  if (err_code != 0) {\n"
             "    LOG(WARNING)\n"
-            "        << \"Failed call inferservice[$name$], name[$service$]\"\n"
+            "        << \"(logid=\" << log_id << \") Failed call "
+            "inferservice[$name$], name[$service$]\"\n"
             "        << \", error_code: \" << err_code;\n"
             "    cntl->SetFailed(err_code, \"InferService inference "
             "failed!\");\n"
@@ -1037,7 +1055,8 @@ class PdsCodeGenerator : public CodeGenerator {
             "  gettimeofday(&tv, NULL);\n"
             "  long end = tv.tv_sec * 1000000 + tv.tv_usec;\n"
             "  // flush notice log\n"
-            "  LOG(INFO) << \" tc=\[\" << (end - start) << \"\]\";\n",  // NOLINT
+            "  LOG(INFO) << \"(logid=\" << log_id << \") tc=\[\" << (end - "  // NOLINT
+            "start) << \"\]\";\n",  // NOLINT
             "name",
             class_name,
             "service",
@@ -1048,26 +1067,31 @@ class PdsCodeGenerator : public CodeGenerator {
             "  brpc::ClosureGuard done_guard(done);\n"
             "  brpc::Controller* cntl = \n"
             "        static_cast<brpc::Controller*>(cntl_base);\n"
+            "  uint64_t log_id = request->log_id();\n"
+            "  cntl->set_log_id(log_id);\n"
             "  ::baidu::paddle_serving::predictor::InferService* svr = \n"
             "       "
             "::baidu::paddle_serving::predictor::InferServiceManager::instance("
             ").item(\"$service$\");\n"
             "  if (svr == NULL) {\n"
-            "    LOG(ERROR) << \"Not found service: $service$\";\n"
+            "    LOG(ERROR) << \"(logid=\" << log_id << \") Not found service: "
+            "$service$\";\n"
             "    cntl->SetFailed(404, \"Not found service: $service$\");\n"
             "    return ;\n"
             "  }\n"
-            "  LOG(INFO) << \" remote_side=\[\" << cntl->remote_side() << "  // NOLINT
-            "\"\]\";\n"
-            "  LOG(INFO) << \" local_side=\[\" << cntl->local_side() << "  // NOLINT
-            "\"\]\";\n"
-            "  LOG(INFO) << \" service_name=\[\" << \"$name$\" << \"\]\";\n"  // NOLINT
-            "  LOG(INFO) << \" log_id=\[\" << cntl->log_id() << \"\]\";\n"  // NOLINT
+            "  LOG(INFO) << \"(logid=\" << log_id << \") remote_side=\[\" "  // NOLINT
+            " << cntl->remote_side() << \"\]\";\n"
+            "  LOG(INFO) << \"(logid=\" << log_id << \") local_side=\[\" "  // NOLINT
+            "<< cntl->local_side() << \"\]\";\n"
+            "  LOG(INFO) << \"(logid=\" << log_id << \") service_name=\[\" "  // NOLINT
+            "<< \"$name$\" << \"\]\";\n"
             "  butil::IOBufBuilder debug_os;\n"
-            "  int err_code = svr->inference(request, response, &debug_os);\n"
+            "  int err_code = svr->inference(request, response, log_id, "
+            "&debug_os);\n"
             "  if (err_code != 0) {\n"
             "    LOG(WARNING)\n"
-            "        << \"Failed call inferservice[$name$], name[$service$]\"\n"
+            "        << \"(logid=\" << log_id << \") Failed call "
+            "inferservice[$name$], name[$service$]\"\n"
             "        << \", error_code: \" << err_code;\n"
             "    cntl->SetFailed(err_code, \"InferService inference "
             "failed!\");\n"
@@ -1076,9 +1100,11 @@ class PdsCodeGenerator : public CodeGenerator {
             "  gettimeofday(&tv, NULL);\n"
             "  long end = tv.tv_sec * 1000000 + tv.tv_usec;\n"
             "  // flush notice log\n"
-            "  LOG(INFO) << \" tc=\[\" << (end - start) << \"\]\";\n"  // NOLINT
+            "  LOG(INFO) << \"(logid=\" << log_id << \") tc=\[\" << (end - "  // NOLINT
+            "start) << \"\]\";\n"  // NOLINT
             "  LOG(INFO)\n"
-            "      << \"TC=[\" << (end - start) << \"] Received debug "
+            "      << \"(logid=\" << log_id << \") TC=[\" << (end - start) << "
+            "\"] Received debug "
             "request[log_id=\" << cntl->log_id()\n"
             "      << \"] from \" << cntl->remote_side()\n"
             "      << \" to \" << cntl->local_side();\n",

core/predictor/CMakeLists.txt

@@ -6,7 +6,7 @@ include(framework/CMakeLists.txt)
 include(tools/CMakeLists.txt)
 include(src/CMakeLists.txt)
 
-
+add_definitions(-D__STDC_FORMAT_MACROS)
 add_library(pdserving ${pdserving_srcs})
 set_source_files_properties(
         ${pdserving_srcs}

core/predictor/common/inner_common.h

@@ -50,7 +50,7 @@
 #include "butil/time.h"
 #endif
 
-#include "glog/raw_logging.h"
+#define ERROR_STRING_LEN 10240
 
 #include "core/configure/general_model_config.pb.h"
 #include "core/configure/include/configure_parser.h"

core/predictor/framework/channel.h

@@ -72,9 +72,10 @@ class Channel {
 
   const std::string& op() { return _op; }
 
-  int share_to_bus(Bus* bus) {
+  int share_to_bus(Bus* bus, const uint64_t log_id) {
     if (bus->regist(_op, this) != 0) {
-      LOG(ERROR) << "Failed regist channel[" << _op << "] to bus!";
+      LOG(ERROR) << "(logid=" << log_id << ") Failed regist channel[" << _op
+                 << "] to bus!";
       return -1;
     }
 

core/predictor/framework/dag.cpp

@@ -155,13 +155,11 @@ int Dag::init(const configure::Workflow& conf, const std::string& name) {
   }
 
   if (FLAGS_el_log_level == 16) {
-    LOG(INFO) << "DAG: " << _dag_name;
-    LOG(INFO) << ", Op Num: " << _index_nodes.size();
+    LOG(INFO) << "DAG: " << _dag_name << ", Op Num: " << _index_nodes.size();
     for (uint32_t nid = 0; nid < _index_nodes.size(); nid++) {
       DagNode* node = _index_nodes[nid];
-      LOG(INFO) << ", OP-" << node->id << "-" << node->name << "-"
-                << node->type;
-      LOG(INFO) << " depends: " << node->depends.size();
+      LOG(INFO) << "OP-" << node->id << "-" << node->name << "-" << node->type
+                << " depends: " << node->depends.size();
 
       boost::unordered_map<std::string, EdgeMode>::iterator it;
       for (it = node->depends.begin(); it != node->depends.end(); it++) {
@@ -214,7 +212,7 @@ int Dag::topo_sort() {
     }
   }
   for (int i = 0; i < in_degree.size(); ++i) {
-    LOG(INFO) << "(" << _index_nodes[i]->name << ") in_degree[" << i
+    VLOG(2) << "(" << _index_nodes[i]->name << ") in_degree[" << i
             << "]: " << in_degree[i];
   }
   int sorted_num = 0;

core/predictor/framework/dag_view.cpp

@@ -26,7 +26,9 @@ namespace baidu {
 namespace paddle_serving {
 namespace predictor {
 
-int DagView::init(Dag* dag, const std::string& service_name) {
+int DagView::init(Dag* dag,
+                  const std::string& service_name,
+                  const uint64_t log_id) {
   _name = dag->name();
   _full_name = service_name + NAME_DELIMITER + dag->name();
   _bus = butil::get_object<Bus>();
@@ -36,17 +38,20 @@ int DagView::init(Dag* dag, const std::string& service_name) {
   for (uint32_t si = 0; si < stage_size; si++) {
     const DagStage* stage = dag->stage_by_index(si);
     if (stage == NULL) {
-      LOG(ERROR) << "Failed get stage by index:" << si;
+      LOG(ERROR) << "(logid=" << log_id << ") Failed get stage by index:" << si;
       return ERR_INTERNAL_FAILURE;
     }
     ViewStage* vstage = butil::get_object<ViewStage>();
     if (vstage == NULL) {
-      LOG(ERROR) << "Failed get vstage from object pool"
+      LOG(ERROR) << "(logid=" << log_id
+                 << ") Failed get vstage from object pool"
                  << "at:" << si;
       return ERR_MEM_ALLOC_FAILURE;
     }
-    VLOG(2) << "stage[" << si << "] name: " << stage->full_name;
-    VLOG(2) << "stage[" << si << "] node size: " << stage->nodes.size();
+    VLOG(2) << "(logid=" << log_id << ") stage[" << si
+            << "] name: " << stage->full_name;
+    VLOG(2) << "(logid=" << log_id << ") stage[" << si
+            << "] node size: " << stage->nodes.size();
     vstage->full_name = service_name + NAME_DELIMITER + stage->full_name;
     uint32_t node_size = stage->nodes.size();
     // create tls view node
@@ -54,31 +59,39 @@ int DagView::init(Dag* dag, const std::string& service_name) {
       DagNode* node = stage->nodes[ni];
       ViewNode* vnode = butil::get_object<ViewNode>();
       if (vnode == NULL) {
-        LOG(ERROR) << "Failed get vnode at:" << ni;
+        LOG(ERROR) << "(logid=" << log_id << ") Failed get vnode at:" << ni;
         return ERR_MEM_ALLOC_FAILURE;
       }
       // factory type
       Op* op = OpRepository::instance().get_op(node->type);
       if (op == NULL) {
-        LOG(ERROR) << "Failed get op with type:" << node->type;
+        LOG(ERROR) << "(logid=" << log_id
+                   << ") Failed get op with type:" << node->type;
         return ERR_INTERNAL_FAILURE;
       }
 
       // initialize a TLS op object
-      VLOG(2) << "dag view initialized: \n"
+      VLOG(2) << "(logid=" << log_id << ") dag view initialized: \n"
               << "node id: " << node->id << "\n"
               << "node name: " << node->name << "\n"
               << "node type: " << node->type;
-      if (op->init(_bus, dag, node->id, node->name, node->type, node->conf) !=
-          0) {
-        LOG(WARNING) << "Failed init op, type:" << node->type;
+      if (op->init(_bus,
+                   dag,
+                   node->id,
+                   node->name,
+                   node->type,
+                   node->conf,
+                   log_id) != 0) {
+        LOG(WARNING) << "(logid=" << log_id
+                     << ") Failed init op, type:" << node->type;
         return ERR_INTERNAL_FAILURE;
       }
 
       op->set_full_name(service_name + NAME_DELIMITER + node->full_name);
 
       // Set the name of the Op as the key of the matching engine.
-      VLOG(2) << "op->set_engine_name(" << node->name.c_str() << ")";
+      VLOG(2) << "(logid=" << log_id << ") op->set_engine_name("
+              << node->name.c_str() << ")";
       op->set_engine_name(node->name);
 
       vnode->conf = node;
@@ -88,7 +101,7 @@ int DagView::init(Dag* dag, const std::string& service_name) {
            it != vnode->conf->depends.end();
            ++it) {
         std::string pre_node_name = it->first;
-        VLOG(2) << "add op pre name: \n"
+        VLOG(2) << "(logid=" << log_id << ") add op pre name: \n"
                 << "current op name: " << vnode->op->op_name()
                 << ", previous op name: " << pre_node_name;
         vnode->op->add_pre_node_name(pre_node_name);
@@ -102,7 +115,7 @@ int DagView::init(Dag* dag, const std::string& service_name) {
     //<< " previous op name: "
     //<< _view[si - 1]->nodes.back()->op->op_name();
     // vstage->nodes.back()->op->set_pre_node_name(
-    //_view[si - 1]->nodes.back()->op->op_name());
+    // _view[si - 1]->nodes.back()->op->op_name());
     /*}*/
     _view.push_back(vstage);
   }
@@ -133,14 +146,15 @@ int DagView::deinit() {
   return ERR_OK;
 }
 
-int DagView::execute(butil::IOBufBuilder* debug_os) {
+int DagView::execute(const uint64_t log_id, butil::IOBufBuilder* debug_os) {
   uint32_t stage_size = _view.size();
   for (uint32_t si = 0; si < stage_size; si++) {
-    TRACEPRINTF("start to execute stage[%u]", si);
-    int errcode = execute_one_stage(_view[si], debug_os);
-    TRACEPRINTF("finish to execute stage[%u]", si);
+    TRACEPRINTF("(logid=%" PRIu64 ") start to execute stage[%u]", log_id, si);
+    int errcode = execute_one_stage(_view[si], log_id, debug_os);
+    TRACEPRINTF("(logid=%" PRIu64 ") finish to execute stage[%u]", log_id, si);
     if (errcode < 0) {
-      LOG(ERROR) << "failed execute stage[" << _view[si]->debug();
+      LOG(ERROR) << "(logid=" << log_id << ") Failed execute stage["
+                 << _view[si]->debug();
       return errcode;
     }
   }
@@ -151,29 +165,34 @@ int DagView::execute(butil::IOBufBuilder* debug_os) {
 // You can derive a subclass to implement this func.
 // ParallelDagView maybe the one you want.
 int DagView::execute_one_stage(ViewStage* vstage,
+                               const uint64_t log_id,
                                butil::IOBufBuilder* debug_os) {
   butil::Timer stage_time(butil::Timer::STARTED);
   uint32_t node_size = vstage->nodes.size();
-  VLOG(2) << "vstage->nodes.size(): " << node_size;
+  VLOG(2) << "(logid=" << log_id << ") vstage->nodes.size(): " << node_size;
   for (uint32_t ni = 0; ni < node_size; ni++) {
     ViewNode* vnode = vstage->nodes[ni];
     DagNode* conf = vnode->conf;
     Op* op = vnode->op;
-    TRACEPRINTF("start to execute op[%s]", op->name());
-    int errcode = op->process(debug_os != NULL);
-    TRACEPRINTF("finish to execute op[%s]", op->name());
+    TRACEPRINTF(
+        "(logid=%" PRIu64 ") start to execute op[%s]", log_id, op->name());
+    int errcode = op->process(log_id, debug_os != NULL);
+    TRACEPRINTF(
+        "(logid=%" PRIu64 ") finish to execute op[%s]", log_id, op->name());
     if (errcode < 0) {
-      LOG(ERROR) << "Execute failed, Op:" << op->debug_string();
+      LOG(ERROR) << "(logid=" << log_id
+                 << ") Execute failed, Op:" << op->debug_string();
       return errcode;
     }
 
     if (errcode > 0) {
-      LOG(INFO) << "Execute ignore, Op:" << op->debug_string();
+      LOG(INFO) << "(logid=" << log_id
+                << ") Execute ignore, Op:" << op->debug_string();
       continue;
     }
 
     if (debug_os) {
-      (*debug_os) << "{\"op_name\": \"" << op->name()
+      (*debug_os) << "(logid=" << log_id << ") {\"op_name\": \"" << op->name()
                   << "\", \"debug_str:\": \"" << op->debug_string()
                   << "\", \"time_info\": \"" << op->time_info() << "\"}";
     }
@@ -186,34 +205,34 @@ int DagView::execute_one_stage(ViewStage* vstage,
   return ERR_OK;
 }
 
-int DagView::set_request_channel(Channel& request) {
+int DagView::set_request_channel(Channel& request, const uint64_t log_id) {
   // Each workflow should get the very beginning
   // request (channel), and commit it to bus, for
   // the first stage ops consuming.
 
-  request.share_to_bus(_bus);
+  request.share_to_bus(_bus, log_id);
 
   return ERR_OK;
 }
 
-const Channel* DagView::get_response_channel() const {
+const Channel* DagView::get_response_channel(const uint64_t log_id) const {
   // Caller obtains response channel from bus, and
   // writes it to rpc response(protbuf/json)
   if (_view.size() < 1) {
-    LOG(ERROR) << "invalid empty view stage!";
+    LOG(ERROR) << "(logid=" << log_id << ") invalid empty view stage!";
     return NULL;
   }
 
   ViewStage* last_stage = _view[_view.size() - 1];
   if (last_stage->nodes.size() != 1 || last_stage->nodes[0] == NULL) {
-    LOG(ERROR) << "Invalid last stage, size[" << last_stage->nodes.size()
-               << "] != 1";
+    LOG(ERROR) << "(logid=" << log_id << ") Invalid last stage, size["
+               << last_stage->nodes.size() << "] != 1";
     return NULL;
   }
 
   Op* last_op = last_stage->nodes[0]->op;
   if (last_op == NULL) {
-    LOG(ERROR) << "Last op is NULL";
+    LOG(ERROR) << "(logid=" << log_id << ") Last op is NULL";
     return NULL;
   }
   return last_op->mutable_channel();

core/predictor/framework/dag_view.h

@@ -47,21 +47,22 @@ class DagView {
 
   ~DagView() {}
 
-  int init(Dag* dag, const std::string& service_name);
+  int init(Dag* dag, const std::string& service_name, const uint64_t log_id);
 
   int deinit();
 
-  int execute(butil::IOBufBuilder* debug_os);
+  int execute(const uint64_t log_id, butil::IOBufBuilder* debug_os);
 
   // The default execution strategy is in sequencing
   // You can derive a subclass to implement this func.
   // ParallelDagView maybe the one you want.
   virtual int execute_one_stage(ViewStage* vstage,
+                                const uint64_t log_id,
                                 butil::IOBufBuilder* debug_os);
 
-  int set_request_channel(Channel& request);  // NOLINT
+  int set_request_channel(Channel& request, const uint64_t log_id);  // NOLINT
 
-  const Channel* get_response_channel() const;
+  const Channel* get_response_channel(const uint64_t log_id) const;
 
   const std::string& name() const { return _name; }
 

core/predictor/framework/factory.h

@@ -17,7 +17,7 @@
 #include <string>
 #include <utility>
 #include "core/predictor/common/inner_common.h"
-#include "glog/raw_logging.h"
+
 namespace baidu {
 namespace paddle_serving {
 namespace predictor {
@@ -28,7 +28,12 @@ namespace predictor {
     FactoryDerive<D, B>* factory = new (std::nothrow) FactoryDerive<D, B>(); \
     if (factory == NULL ||                                                   \
         FactoryPool<B>::instance().register_factory(tag, factory) != 0) {    \
-      RAW_LOG_FATAL("Failed regist factory: %s in macro!", #D);              \
+      char err_str[ERROR_STRING_LEN];                                        \
+      snprintf(err_str,                                                      \
+               ERROR_STRING_LEN - 1,                                         \
+               "Failed regist factory: %s in macro!",                        \
+               #D);                                                          \
+      RAW_LOG(FATAL, err_str);                                               \
       return -1;                                                             \
     }                                                                        \
     return 0;                                                                \
@@ -54,7 +59,13 @@ namespace predictor {
     if (factory == NULL ||                                                     \
         ::baidu::paddle_serving::predictor::FactoryPool<B>::instance()         \
                 .register_factory(#D, factory) != 0) {                         \
-      RAW_LOG_FATAL("Failed regist factory: %s->%s in macro!", #D, #B);        \
+      char err_str[ERROR_STRING_LEN];                                          \
+      snprintf(err_str,                                                        \
+               ERROR_STRING_LEN - 1,                                           \
+               "Failed regist factory: %s->%s in macro!",                      \
+               #D,                                                             \
+               #B);                                                            \
+      RAW_LOG(FATAL, err_str);                                                 \
       return;                                                                  \
     }                                                                          \
     return;                                                                    \
@@ -66,15 +77,26 @@ namespace predictor {
     ::baidu::paddle_serving::predictor::FactoryDerive<D, B>* factory = new (   \
         ::std::nothrow)::baidu::paddle_serving::predictor::FactoryDerive<D,    \
                                                                          B>(); \
+    char err_str[ERROR_STRING_LEN];                                            \
     if (factory == NULL ||                                                     \
         ::baidu::paddle_serving::predictor::FactoryPool<B>::instance()         \
                 .register_factory(N, factory) != 0) {                          \
-      RAW_LOG_FATAL(                                                           \
-          "Failed regist factory: %s->%s, tag: %s in macro!", #D, #B, N);      \
+      snprintf(err_str,                                                        \
+               ERROR_STRING_LEN - 1,                                           \
+               "Failed regist factory: %s->%s, tag: %s in macro!",             \
+               #D,                                                             \
+               #B,                                                             \
+               N);                                                             \
+      RAW_LOG(FATAL, err_str);                                                 \
       return;                                                                  \
     }                                                                          \
-    RAW_LOG_WARNING(                                                           \
-        "Succ regist factory: %s->%s, tag: %s in macro!", #D, #B, N);          \
+    snprintf(err_str,                                                          \
+             ERROR_STRING_LEN - 1,                                             \
+             "Succ regist factory: %s->%s, tag: %s in macro!",                 \
+             #D,                                                               \
+             #B,                                                               \
+             N);                                                               \
+    RAW_LOG(WARNING, err_str);                                                 \
     return;                                                                    \
   }
 
@@ -102,24 +124,35 @@ class FactoryPool {
   }
 
   int register_factory(const std::string& tag, FactoryBase<B>* factory) {
+    char err_str[ERROR_STRING_LEN];
     typename std::map<std::string, FactoryBase<B>*>::iterator it =
         _pool.find(tag);
     if (it != _pool.end()) {
-      RAW_LOG_FATAL("Insert duplicate with tag: %s", tag.c_str());
+      snprintf(err_str,
+               ERROR_STRING_LEN - 1,
+               "Insert duplicate with tag: %s",
+               tag.c_str());
+      RAW_LOG(FATAL, err_str);
       return -1;
     }
 
     std::pair<typename std::map<std::string, FactoryBase<B>*>::iterator, bool>
         r = _pool.insert(std::make_pair(tag, factory));
     if (!r.second) {
-      RAW_LOG_FATAL("Failed insert new factory with: %s", tag.c_str());
+      snprintf(err_str,
+               ERROR_STRING_LEN - 1,
+               "Failed insert new factory with: %s",
+               tag.c_str());
+      RAW_LOG(FATAL, err_str);
       return -1;
     }
 
-    RAW_LOG_INFO("Succ insert one factory, tag: %s, base type %s",
+    snprintf(err_str,
+             ERROR_STRING_LEN - 1,
+             "Succ insert one factory, tag: %s, base type %s",
              tag.c_str(),
              typeid(B).name());
-
+    RAW_LOG(INFO, err_str);
     return 0;
   }
 
@@ -127,9 +160,13 @@ class FactoryPool {
     typename std::map<std::string, FactoryBase<B>*>::iterator it =
         _pool.find(tag);
     if (it == _pool.end() || it->second == NULL) {
-      RAW_LOG_FATAL("Not found factory pool, tag: %s, pool size %u",
+      char err_str[ERROR_STRING_LEN];
+      snprintf(err_str,
+               ERROR_STRING_LEN - 1,
+               "Not found factory pool, tag: %s, pool size %u",
                tag.c_str(),
                _pool.size());
+      RAW_LOG(FATAL, err_str);
       return NULL;
     }
 

core/predictor/framework/infer.h

@@ -603,6 +603,7 @@ class VersionedInferEngine : public InferEngine {
       LOG(ERROR) << "Failed generate engine with type:" << engine_type;
       return -1;
     }
+#ifndef BCLOUD
     VLOG(2) << "FLAGS_logtostderr " << FLAGS_logtostderr;
     int tmp = FLAGS_logtostderr;
     if (engine->proc_initialize(conf, version) != 0) {
@@ -611,6 +612,12 @@ class VersionedInferEngine : public InferEngine {
     }
     VLOG(2) << "FLAGS_logtostderr " << FLAGS_logtostderr;
     FLAGS_logtostderr = tmp;
+#else
+    if (engine->proc_initialize(conf, version) != 0) {
+      LOG(ERROR) << "Failed initialize engine, type:" << engine_type;
+      return -1;
+    }
+#endif
     auto r = _versions.insert(std::make_pair(engine->version(), engine));
     if (!r.second) {
       LOG(ERROR) << "Failed insert item: " << engine->version()

core/predictor/framework/op_repository.h

@@ -62,7 +62,10 @@ class OpRepository {
   template <typename OP_TYPE>
   void regist_op(std::string op_type) {
     _repository[op_type] = &OpFactory<OP_TYPE>::instance();
-    RAW_LOG_INFO("Succ regist op: %s", op_type.c_str());
+    char err_str[ERROR_STRING_LEN];
+    snprintf(
+        err_str, ERROR_STRING_LEN - 1, "Succ regist op: %s", op_type.c_str());
+    RAW_LOG(INFO, err_str);
   }
 
   Op* get_op(std::string op_type);

core/predictor/framework/resource.cpp

@@ -17,6 +17,9 @@
 #include <string>
 #include "core/predictor/common/inner_common.h"
 #include "core/predictor/framework/kv_manager.h"
+#ifdef BCLOUD
+#include "aipe_sec_client.h"  // NOLINT
+#endif
 namespace baidu {
 namespace paddle_serving {
 namespace predictor {
@@ -109,6 +112,42 @@ int Resource::initialize(const std::string& path, const std::string& file) {
   }
   LOG(WARNING) << "Successfully proc initialized mempool wrapper";
 
+#ifdef WITH_AUTH
+  std::string product_name_str = resource_conf.auth_product_name();
+  std::string container_id_str = resource_conf.auth_container_id();
+
+  char* product_name = new char[product_name_str.size() + 1];
+  snprintf(product_name,
+           product_name_str.size() + 1,
+           "%s",
+           product_name_str.c_str());
+  char* container_id = new char[container_id_str.size() + 1];
+  snprintf(container_id,
+           container_id_str.size() + 1,
+           "%s",
+           container_id_str.c_str());
+
+  aipe_auth_request request;
+  request.product_name = product_name;
+  request.container_id = container_id;
+  request.request_ts = (int64_t)time(NULL);
+
+  LOG(INFO) << "\nEasypack info"
+            << "\nproduct name: " << request.product_name
+            << "\ncontainer_id: " << request.container_id
+            << "\nrequest time stamp: " << request.request_ts;
+
+  aipe_auth_response response;
+  response = check_auth(request);
+
+  if (response.result == 0) {
+    LOG(INFO) << "Authentication succeed.";
+  } else {
+    LOG(ERROR) << "Authentication failed. Error code: " << response.result;
+    return -1;
+  }
+#endif
+
   if (FLAGS_enable_model_toolkit) {
     int err = 0;
     std::string model_toolkit_path = resource_conf.model_toolkit_path();

core/predictor/framework/service.cpp

@@ -19,6 +19,7 @@
 #include <butil/time.h>  // butil::Timer
 #endif
 
+#include <inttypes.h>
 #include <list>
 #include <string>
 #include <vector>
@@ -135,50 +136,63 @@ const std::string& InferService::name() const { return _infer_service_format; }
 // ´®ÐÐÖ´ÐÐÿ¸öworkflow
 int InferService::inference(const google::protobuf::Message* request,
                             google::protobuf::Message* response,
+                            const uint64_t log_id,
                             butil::IOBufBuilder* debug_os) {
-  TRACEPRINTF("start to inference");
+  TRACEPRINTF("(logid=%" PRIu64 ") start to inference", log_id);
   // when funtion call begins, framework will reset
   // thread local variables&resources automatically.
   if (Resource::instance().thread_clear() != 0) {
-    LOG(ERROR) << "Failed thread clear whole resource";
+    LOG(ERROR) << "(logid=" << log_id << ") Failed thread clear whole resource";
     return ERR_INTERNAL_FAILURE;
   }
 
-  TRACEPRINTF("finish to thread clear");
+  TRACEPRINTF("(logid=%" PRIu64 ") finish to thread clear", log_id);
 
   if (_enable_map_request_to_workflow) {
-    LOG(INFO) << "enable map request == True";
-    std::vector<Workflow*>* workflows = _map_request_to_workflow(request);
+    VLOG(2) << "(logid=" << log_id << ") enable map request == True";
+    std::vector<Workflow*>* workflows =
+        _map_request_to_workflow(request, log_id);
     if (!workflows || workflows->size() == 0) {
-      LOG(ERROR) << "Failed to map request to workflow";
+      LOG(ERROR) << "(logid=" << log_id
+                 << ") Failed to map request to workflow";
       return ERR_INTERNAL_FAILURE;
     }
     size_t fsize = workflows->size();
     for (size_t fi = 0; fi < fsize; ++fi) {
       Workflow* workflow = (*workflows)[fi];
       if (workflow == NULL) {
-        LOG(ERROR) << "Failed to get valid workflow at: " << fi;
+        LOG(ERROR) << "(logid=" << log_id
+                   << ") Failed to get valid workflow at: " << fi;
         return ERR_INTERNAL_FAILURE;
       }
-      TRACEPRINTF("start to execute workflow[%s]", workflow->name().c_str());
-      int errcode = _execute_workflow(workflow, request, response, debug_os);
-      TRACEPRINTF("finish to execute workflow[%s]", workflow->name().c_str());
+      TRACEPRINTF("(logid=%" PRIu64 ") start to execute workflow[%s]",
+                  log_id,
+                  workflow->name().c_str());
+      int errcode =
+          _execute_workflow(workflow, request, response, log_id, debug_os);
+      TRACEPRINTF("(logid=%" PRIu64 ") finish to execute workflow[%s]",
+                  log_id,
+                  workflow->name().c_str());
       if (errcode < 0) {
-        LOG(ERROR) << "Failed execute workflow[" << workflow->name()
-                   << "] in:" << name();
+        LOG(ERROR) << "(logid=" << log_id << ") Failed execute workflow["
+                   << workflow->name() << "] in:" << name();
         return errcode;
       }
     }
   } else {
-    LOG(INFO) << "enable map request == False";
-    TRACEPRINTF("start to execute one workflow");
+    VLOG(2) << "(logid=" << log_id << ") enable map request == False";
+    TRACEPRINTF("(logid=%" PRIu64 ") start to execute one workflow", log_id);
     size_t fsize = _flows.size();
     for (size_t fi = 0; fi < fsize; ++fi) {
-      TRACEPRINTF("start to execute one workflow-%lu", fi);
-      int errcode = execute_one_workflow(fi, request, response, debug_os);
-      TRACEPRINTF("finish to execute one workflow-%lu", fi);
+      TRACEPRINTF(
+          "(logid=%" PRIu64 ") start to execute one workflow-%lu", log_id, fi);
+      int errcode =
+          execute_one_workflow(fi, request, response, log_id, debug_os);
+      TRACEPRINTF(
+          "(logid=%" PRIu64 ") finish to execute one workflow-%lu", log_id, fi);
       if (errcode < 0) {
-        LOG(ERROR) << "Failed execute 0-th workflow in:" << name();
+        LOG(ERROR) << "(logid=" << log_id
+                   << ") Failed execute 0-th workflow in:" << name();
         return errcode;
       }
     }
@@ -188,26 +202,30 @@ int InferService::inference(const google::protobuf::Message* request,
 
 int InferService::debug(const google::protobuf::Message* request,
                         google::protobuf::Message* response,
+                        const uint64_t log_id,
                         butil::IOBufBuilder* debug_os) {
-  return inference(request, response, debug_os);
+  return inference(request, response, log_id, debug_os);
 }
 
 int InferService::execute_one_workflow(uint32_t index,
                                        const google::protobuf::Message* request,
                                        google::protobuf::Message* response,
+                                       const uint64_t log_id,
                                        butil::IOBufBuilder* debug_os) {
   if (index >= _flows.size()) {
-    LOG(ERROR) << "Faield execute workflow, index: " << index
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Faield execute workflow, index: " << index
                << " >= max:" << _flows.size();
     return ERR_OVERFLOW_FAILURE;
   }
   Workflow* workflow = _flows[index];
-  return _execute_workflow(workflow, request, response, debug_os);
+  return _execute_workflow(workflow, request, response, log_id, debug_os);
 }
 
 int InferService::_execute_workflow(Workflow* workflow,
                                     const google::protobuf::Message* request,
                                     google::protobuf::Message* response,
+                                    const uint64_t log_id,
                                     butil::IOBufBuilder* debug_os) {
   butil::Timer workflow_time(butil::Timer::STARTED);
   // create and submit beginer channel
@@ -215,54 +233,62 @@ int InferService::_execute_workflow(Workflow* workflow,
   req_channel.init(0, START_OP_NAME);
   req_channel = request;
 
-  DagView* dv = workflow->fetch_dag_view(full_name());
-  dv->set_request_channel(req_channel);
+  DagView* dv = workflow->fetch_dag_view(full_name(), log_id);
+  dv->set_request_channel(req_channel, log_id);
 
   // call actual inference interface
-  int errcode = dv->execute(debug_os);
+  int errcode = dv->execute(log_id, debug_os);
   if (errcode < 0) {
-    LOG(ERROR) << "Failed execute dag for workflow:" << workflow->name();
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed execute dag for workflow:" << workflow->name();
     return errcode;
   }
 
-  TRACEPRINTF("finish to dv execute");
+  TRACEPRINTF("(logid=%" PRIu64 ") finish to dv execute", log_id);
   // create ender channel and copy
-  const Channel* res_channel = dv->get_response_channel();
+  const Channel* res_channel = dv->get_response_channel(log_id);
+  if (res_channel == NULL) {
+    LOG(ERROR) << "(logid=" << log_id << ") Failed get response channel";
+    return ERR_INTERNAL_FAILURE;
+  }
+
   if (!_merger || !_merger->merge(res_channel->message(), response)) {
-    LOG(ERROR) << "Failed merge channel res to response";
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed merge channel res to response";
     return ERR_INTERNAL_FAILURE;
   }
-  TRACEPRINTF("finish to copy from");
+  TRACEPRINTF("(logid=%" PRIu64 ") finish to copy from", log_id);
 
   workflow_time.stop();
-  LOG(INFO) << "workflow total time: " << workflow_time.u_elapsed();
+  LOG(INFO) << "(logid=" << log_id
+            << ") workflow total time: " << workflow_time.u_elapsed();
   PredictorMetric::GetInstance()->update_latency_metric(
       WORKFLOW_METRIC_PREFIX + dv->full_name(), workflow_time.u_elapsed());
 
   // return tls data to object pool
   workflow->return_dag_view(dv);
-  TRACEPRINTF("finish to return dag view");
+  TRACEPRINTF("(logid=%" PRIu64 ") finish to return dag view", log_id);
   return ERR_OK;
 }
 
 std::vector<Workflow*>* InferService::_map_request_to_workflow(
-    const google::protobuf::Message* request) {
+    const google::protobuf::Message* request, const uint64_t log_id) {
   const google::protobuf::Descriptor* desc = request->GetDescriptor();
   const google::protobuf::FieldDescriptor* field =
       desc->FindFieldByName(_request_field_key);
   if (field == NULL) {
-    LOG(ERROR) << "No field[" << _request_field_key << "] in ["
-               << desc->full_name() << "].";
+    LOG(ERROR) << "(logid=" << log_id << ") No field[" << _request_field_key
+               << "] in [" << desc->full_name() << "].";
     return NULL;
   }
   if (field->is_repeated()) {
-    LOG(ERROR) << "field[" << desc->full_name() << "." << _request_field_key
-               << "] is repeated.";
+    LOG(ERROR) << "(logid=" << log_id << ") field[" << desc->full_name() << "."
+               << _request_field_key << "] is repeated.";
     return NULL;
   }
   if (field->cpp_type() != google::protobuf::FieldDescriptor::CPPTYPE_STRING) {
-    LOG(ERROR) << "field[" << desc->full_name() << "." << _request_field_key
-               << "] should be string";
+    LOG(ERROR) << "(logid=" << log_id << ") field[" << desc->full_name() << "."
+               << _request_field_key << "] should be string";
     return NULL;
   }
   const std::string& field_value =
@@ -270,7 +296,7 @@ std::vector<Workflow*>* InferService::_map_request_to_workflow(
   std::vector<Workflow*>* p_workflow =
       _request_to_workflow_map.seek(field_value);
   if (p_workflow == NULL) {
-    LOG(ERROR) << "cannot find key[" << field_value
+    LOG(ERROR) << "(logid=" << log_id << ") cannot find key[" << field_value
                << "] in _request_to_workflow_map";
     return NULL;
   }

core/predictor/framework/service.h

@@ -52,25 +52,29 @@ class InferService {
   // Execute each workflow serially
   virtual int inference(const google::protobuf::Message* request,
                         google::protobuf::Message* response,
+                        const uint64_t log_id,
                         butil::IOBufBuilder* debug_os = NULL);
 
   int debug(const google::protobuf::Message* request,
             google::protobuf::Message* response,
+            const uint64_t log_id,
             butil::IOBufBuilder* debug_os);
 
   int execute_one_workflow(uint32_t index,
                            const google::protobuf::Message* request,
                            google::protobuf::Message* response,
+                           const uint64_t log_id,
                            butil::IOBufBuilder* debug_os);
 
  private:
   int _execute_workflow(Workflow* workflow,
                         const google::protobuf::Message* request,
                         google::protobuf::Message* response,
+                        const uint64_t log_id,
                         butil::IOBufBuilder* debug_os);
 
   std::vector<Workflow*>* _map_request_to_workflow(
-      const google::protobuf::Message* request);
+      const google::protobuf::Message* request, const uint64_t log_id);
 
  private:
   std::vector<Workflow*> _flows;
@@ -88,6 +92,7 @@ class ParallelInferService : public InferService {
   // Execute workflows in parallel
   int inference(const google::protobuf::Message* request,
                 google::protobuf::Message* response,
+                const uint64_t log_id,
                 butil::IOBufBuilder* debug_os) {
     return 0;
   }

core/predictor/framework/service_manager.h

@@ -23,17 +23,24 @@ namespace predictor {
 
 #define REGIST_FORMAT_SERVICE(svr_name, svr)                                 \
   do {                                                                       \
+    char err_str[ERROR_STRING_LEN];                                          \
     int ret =                                                                \
         ::baidu::paddle_serving::predictor::FormatServiceManager::instance() \
             .regist_service(svr_name, svr);                                  \
     if (ret != 0) {                                                          \
-      RAW_LOG_ERROR("Failed regist service[%s][%s]",                         \
+      snprintf(err_str,                                                      \
+               ERROR_STRING_LEN - 1,                                         \
+               "Failed regist service[%s][%s]",                              \
                svr_name.c_str(),                                             \
                typeid(svr).name());                                          \
+      RAW_LOG(ERROR, err_str);                                               \
     } else {                                                                 \
-      RAW_LOG_INFO("Success regist service[%s][%s]",                         \
+      snprintf(err_str,                                                      \
+               ERROR_STRING_LEN - 1,                                         \
+               "Success regist service[%s][%s]",                             \
                svr_name.c_str(),                                             \
                typeid(svr).name());                                          \
+      RAW_LOG(INFO, err_str);                                                \
     }                                                                        \
   } while (0)
 
@@ -42,31 +49,46 @@ class FormatServiceManager {
   typedef google::protobuf::Service Service;
 
   int regist_service(const std::string& svr_name, Service* svr) {
+    char err_str[ERROR_STRING_LEN];
     if (_service_map.find(svr_name) != _service_map.end()) {
-      RAW_LOG_ERROR("Service[%s][%s] already exist!",
+      snprintf(err_str,
+               ERROR_STRING_LEN - 1,
+               "Service[%s][%s] already exist!",
                svr_name.c_str(),
                typeid(svr).name());
+      RAW_LOG(ERROR, err_str);
       return -1;
     }
 
     std::pair<boost::unordered_map<std::string, Service*>::iterator, bool> ret;
     ret = _service_map.insert(std::make_pair(svr_name, svr));
     if (ret.second == false) {
-      RAW_LOG_ERROR("Service[%s][%s] insert failed!",
+      snprintf(err_str,
+               ERROR_STRING_LEN - 1,
+               "Service[%s][%s] insert failed!",
                svr_name.c_str(),
                typeid(svr).name());
+      RAW_LOG(ERROR, err_str);
       return -1;
     }
 
-    RAW_LOG_INFO("Service[%s] insert successfully!", svr_name.c_str());
+    snprintf(err_str,
+             ERROR_STRING_LEN - 1,
+             "Service[%s] insert successfully!",
+             svr_name.c_str());
+    RAW_LOG(INFO, err_str);
     return 0;
   }
 
   Service* get_service(const std::string& svr_name) {
+    char err_str[ERROR_STRING_LEN];
     boost::unordered_map<std::string, Service*>::iterator res;
     if ((res = _service_map.find(svr_name)) == _service_map.end()) {
-      RAW_LOG_WARNING("Service[%s] not found in service manager!",
+      snprintf(err_str,
+               ERROR_STRING_LEN - 1,
+               "Service[%s] not found in service manager!",
                svr_name.c_str());
+      RAW_LOG(WARNING, err_str);
       return NULL;
     }
     return (*res).second;

core/predictor/framework/workflow.cpp

@@ -32,21 +32,22 @@ int Workflow::init(const configure::Workflow& conf) {
   return 0;
 }
 
-DagView* Workflow::fetch_dag_view(const std::string& service_name) {
+DagView* Workflow::fetch_dag_view(const std::string& service_name,
+                                  const uint64_t log_id) {
   DagView* view = NULL;
   if (_type == "Sequence") {
     view = butil::get_object<DagView>();
   } else if (_type == "Parallel") {
     view = butil::get_object<ParallelDagView>();
   } else {
-    LOG(ERROR) << "Unknown dag type:" << _type << "!";
+    LOG(ERROR) << "(logid=" << log_id << ") Unknown dag type:" << _type << "!";
     return NULL;
   }
   if (view == NULL) {
-    LOG(ERROR) << "create dag view from pool failed!";
+    LOG(ERROR) << "(logid=" << log_id << ") create dag view from pool failed!";
     return NULL;
   }
-  view->init(&_dag, service_name);
+  view->init(&_dag, service_name, log_id);
   return view;
 }
 

core/predictor/framework/workflow.h

@@ -36,7 +36,8 @@ class Workflow {
   // different apps.
   int init(const configure::Workflow& conf);
 
-  DagView* fetch_dag_view(const std::string& service_name);
+  DagView* fetch_dag_view(const std::string& service_name,
+                          const uint64_t log_id);
 
   int deinit() { return 0; }
 

core/predictor/op/op.cpp

@@ -35,7 +35,8 @@ int Op::init(Bus* bus,
              uint32_t id,
              const std::string& name,
              const std::string& type,
-             void* conf) {
+             void* conf,
+             const uint64_t log_id) {
   _bus = bus;
   _dag = dag;
   _id = id;
@@ -45,7 +46,8 @@ int Op::init(Bus* bus,
 
   _timer = butil::get_object<TimerFlow>();
   if (!_timer) {
-    LOG(ERROR) << "Invalid timerflow in op:" << this->name();
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Invalid timerflow in op:" << this->name();
     return -1;
   }
 
@@ -55,7 +57,8 @@ int Op::init(Bus* bus,
 
   Channel* channel = mutable_channel();
   if (channel == NULL) {
-    LOG(ERROR) << "Failed mutable channel in op: " << this->id() << ", "
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Failed mutable channel in op: " << this->id() << ", "
                << this->name() << "!";
     return -1;
   }
@@ -96,18 +99,20 @@ int Op::check_time(const char* tag) {
   return 0;
 }
 
-int Op::process(bool debug) {
+int Op::process(const uint64_t log_id, bool debug) {
   butil::Timer op_time(butil::Timer::STARTED);
   if (debug && _timer) {
     _timer->start();
   }
   if (!_has_init) {
-    LOG(ERROR) << "Make sure op has been init before inference";
+    LOG(ERROR) << "(logid=" << log_id
+               << ") Make sure op has been init before inference";
     return ERR_INTERNAL_FAILURE;
   }
 
   if (_has_calc) {
-    LOG(INFO) << "Op: " << _name << " already processed before";
+    LOG(INFO) << "(logid=" << log_id << ") Op: " << _name
+              << " already processed before";
     return ERR_OK;
   }
 
@@ -143,7 +148,7 @@ int Op::process(bool debug) {
 
   // 3. share output to bus
   Channel* channel = mutable_channel();
-  channel->share_to_bus(_bus);
+  channel->share_to_bus(_bus, log_id);
 
   // 4. mark has calculated
   _has_calc = true;
@@ -156,7 +161,8 @@ int Op::process(bool debug) {
   op_time.stop();
   PredictorMetric::GetInstance()->update_latency_metric(
       OP_METRIC_PREFIX + full_name(), op_time.u_elapsed());
-  LOG(INFO) << " " << name() << "_time=[" << op_time.u_elapsed() << "]";
+  LOG(INFO) << "(logid=" << log_id << ") " << name() << "_time=["
+            << op_time.u_elapsed() << "]";
   return ERR_OK;
 }
 

core/predictor/op/op.h

@@ -113,13 +113,14 @@ class Op {
            uint32_t id,
            const std::string& name,
            const std::string& type,
-           void* conf);
+           void* conf,
+           const uint64_t log_id);
 
   int deinit();
 
   int check_time(const char* tag);
 
-  int process(bool debug);
+  int process(const uint64_t log_id, bool debug);
 
   std::string time_info();
 

core/predictor/src/pdserving.cpp

@@ -202,8 +202,6 @@ int main(int argc, char** argv) {
   }
   VLOG(2) << "Succ call pthread worker start function";
 
-#ifndef BCLOUD
-
   if (Resource::instance().general_model_initialize(FLAGS_resource_path,
                                                     FLAGS_resource_file) != 0) {
     LOG(ERROR) << "Failed to initialize general model conf: "
@@ -213,6 +211,7 @@ int main(int argc, char** argv) {
 
   VLOG(2) << "Succ initialize general model";
 
+#ifndef BCLOUD
   // FATAL messages are output to stderr
   FLAGS_stderrthreshold = 3;
 #endif

core/sdk-cpp/include/abtest.h

@@ -50,9 +50,9 @@ class WeightedRandomRender : public EndpointRouterBase {
     Factory<WeightedRandomRender, EndpointRouterBase>* factory =
         new (std::nothrow) Factory<WeightedRandomRender, EndpointRouterBase>();
     if (factory == NULL) {
-      RAW_LOG_ERROR(
-          "Failed regist factory: WeightedRandomRender->EndpointRouterBase in "
-          "macro!");
+      RAW_LOG(ERROR,
+              "Failed regist factory: WeightedRandomRender->EndpointRouterBase \
+          in macro!");
       return -1;
     }
 
@@ -62,9 +62,9 @@ class WeightedRandomRender : public EndpointRouterBase {
     // together.
     if (FactoryPool<EndpointRouterBase>::instance().register_factory(
             "WeightedRandomRender", factory) != 0) {
-      RAW_LOG_INFO(
-          "Factory has been registed: "
-          "WeightedRandomRender->EndpointRouterBase.");
+      RAW_LOG(INFO,
+              "Factory has been registed: \
+              WeightedRandomRender->EndpointRouterBase.");
     }
 
     return 0;

core/sdk-cpp/include/factory.h

@@ -18,7 +18,6 @@
 #include <utility>
 #include "core/sdk-cpp/include/common.h"
 #include "core/sdk-cpp/include/stub_impl.h"
-#include "glog/raw_logging.h"
 
 namespace baidu {
 namespace paddle_serving {
@@ -28,12 +27,20 @@ namespace sdk_cpp {
 namespace brpc = baidu::rpc;
 #endif
 
+#define ERROR_STRING_LEN 10240
+
 #define INLINE_REGIST_OBJECT(D, B, E)                                    \
   do {                                                                   \
     Factory<D, B>* factory = new (std::nothrow) Factory<D, B>();         \
     if (factory == NULL ||                                               \
         FactoryPool<B>::instance().register_factory(#D, factory) != 0) { \
-      RAW_LOG_ERROR("Failed regist factory: %s->%s in macro!", #D, #B);  \
+      char err_str[ERROR_STRING_LEN];                                    \
+      snprintf(err_str,                                                  \
+               ERROR_STRING_LEN - 1,                                     \
+               "Failed regist factory: %s->%s in macro!",                \
+               #D,                                                       \
+               #B);                                                      \
+      RAW_LOG(ERROR, err_str);                                           \
       return E;                                                          \
     }                                                                    \
   } while (0)
@@ -43,7 +50,12 @@ namespace brpc = baidu::rpc;
     Factory<D, B>* factory = new (std::nothrow) Factory<D, B>();          \
     if (factory == NULL ||                                                \
         FactoryPool<B>::instance().register_factory(tag, factory) != 0) { \
-      RAW_LOG_ERROR("Failed regist factory: %s in macro!", #D);           \
+      char err_str[ERROR_STRING_LEN];                                     \
+      snprintf(err_str,                                                   \
+               ERROR_STRING_LEN - 1,                                      \
+               "Failed regist factory: %s in macro!",                     \
+               #D);                                                       \
+      RAW_LOG(ERROR, err_str);                                            \
       return -1;                                                          \
     }                                                                     \
     return 0;                                                             \
@@ -66,7 +78,13 @@ namespace brpc = baidu::rpc;
     if (factory == NULL ||                                                     \
         ::baidu::paddle_serving::sdk_cpp::FactoryPool<B>::instance()           \
                 .register_factory(#D, factory) != 0) {                         \
-      RAW_LOG_ERROR("Failed regist factory: %s->%s in macro!", #D, #B);        \
+      char err_str[ERROR_STRING_LEN];                                          \
+      snprintf(err_str,                                                        \
+               ERROR_STRING_LEN - 1,                                           \
+               "Failed regist factory: %s->%s in macro!",                      \
+               #D,                                                             \
+               #B);                                                            \
+      RAW_LOG(ERROR, err_str);                                                 \
       return;                                                                  \
     }                                                                          \
     return;                                                                    \
@@ -80,8 +98,14 @@ namespace brpc = baidu::rpc;
     if (factory == NULL ||                                                     \
         ::baidu::paddle_serving::sdk_cpp::FactoryPool<B>::instance()           \
                 .register_factory(T, factory) != 0) {                          \
-      RAW_LOG_ERROR(                                                           \
-          "Failed regist factory: %s->%s, tag %s in macro!", #D, #B, T);       \
+      char err_str[ERROR_STRING_LEN];                                          \
+      snprintf(err_str,                                                        \
+               ERROR_STRING_LEN - 1,                                           \
+               "Failed regist factory: %s->%s, tag %s in macro!",              \
+               #D,                                                             \
+               #B,                                                             \
+               T);                                                             \
+      RAW_LOG(ERROR, err_str);                                                 \
       return;                                                                  \
     }                                                                          \
     return;                                                                    \
@@ -108,8 +132,13 @@ namespace brpc = baidu::rpc;
         ::baidu::paddle_serving::sdk_cpp::FactoryPool<                     \
             ::baidu::paddle_serving::sdk_cpp::Stub>::instance()            \
                 .register_factory(T, factory) != 0) {                      \
-      RAW_LOG_ERROR(                                                       \
-          "Failed regist factory: %s->Stub, tag: %s in macro!", #D, T);    \
+      char err_str[ERROR_STRING_LEN];                                      \
+      snprintf(err_str,                                                    \
+               ERROR_STRING_LEN - 1,                                       \
+               "Failed regist factory: %s->Stub, tag: %s in macro!",       \
+               #D,                                                         \
+               T);                                                         \
+      RAW_LOG(ERROR, err_str);                                             \
       return;                                                              \
     }                                                                      \
     return;                                                                \
@@ -146,14 +175,24 @@ class FactoryPool {
     typename std::map<std::string, FactoryBase<B>*>::iterator it =
         _pool.find(tag);
     if (it != _pool.end()) {
-      RAW_LOG_ERROR("Insert duplicate with tag: %s", tag.c_str());
+      char err_str[ERROR_STRING_LEN];
+      snprintf(err_str,
+               ERROR_STRING_LEN - 1,
+               "Insert duplicate with tag: %s",
+               tag.c_str());
+      RAW_LOG(ERROR, err_str);
       return -1;
     }
 
     std::pair<typename std::map<std::string, FactoryBase<B>*>::iterator, bool>
         r = _pool.insert(std::make_pair(tag, factory));
     if (!r.second) {
-      RAW_LOG_ERROR("Failed insert new factory with: %s", tag.c_str());
+      char err_str[ERROR_STRING_LEN];
+      snprintf(err_str,
+               ERROR_STRING_LEN - 1,
+               "Failed insert new factory with: %s",
+               tag.c_str());
+      RAW_LOG(ERROR, err_str);
       return -1;
     }
 
@@ -164,9 +203,13 @@ class FactoryPool {
     typename std::map<std::string, FactoryBase<B>*>::iterator it =
         _pool.find(tag);
     if (it == _pool.end() || it->second == NULL) {
-      RAW_LOG_ERROR("Not found factory pool, tag: %s, pool size: %u",
+      char err_str[ERROR_STRING_LEN];
+      snprintf(err_str,
+               ERROR_STRING_LEN - 1,
+               "Not found factory pool, tag: %s, pool size: %u",
                tag.c_str(),
                _pool.size());
+      RAW_LOG(ERROR, err_str);
       return NULL;
     }
 

core/sdk-cpp/proto/general_model_service.proto

@@ -37,6 +37,7 @@ message Request {
   repeated FeedInst insts = 1;
   repeated string fetch_var_names = 2;
   optional bool profile_server = 3 [ default = false ];
+  required uint64 log_id = 4 [ default = 0 ];
 };
 
 message Response {

doc/COMPILE.md

@@ -5,7 +5,7 @@
 ## Compilation environment requirements
 
 |            module            |              version              |
-| :--------------------------: | :----------------------------------------------------------: |
+| :--------------------------: | :-------------------------------: |
 |              OS              |             CentOS 7              |
 |             gcc              |          4.8.5 and later          |
 |           gcc-c++            |          4.8.5 and later          |
@@ -23,7 +23,6 @@
 |           libXext            |               1.3.3               |
 |            libSM             |               1.2.2               |
 |          libXrender          |              0.9.10               |
-|          python-whl          | numpy>=1.12, <=1.16.4<br/>google>=2.0.3<br/>protobuf>=3.12.2<br/>grpcio-tools>=1.28.1<br/>grpcio>=1.28.1<br/>func-timeout>=4.3.5<br/>pyyaml>=1.3.0<br/>sentencepiece==0.1.92<br>flask>=1.1.2<br>ujson>=2.0.3 |
 
 It is recommended to use Docker for compilation. We have prepared the Paddle Serving compilation environment for you, see [this document](DOCKER_IMAGES.md).
 
@@ -62,6 +61,25 @@ pip install -r python/requirements.txt
 
 If Python3 is used, replace `pip` with `pip3`.
 
+## GOPATH Setting
+
+
+## Compile Arguments
+
+The default GOPATH is `$HOME/go`, which you can set to other values.
+```shell
+export GOPATH=$HOME/go
+export PATH=$PATH:$GOPATH/bin
+```
+
+## Get go packages
+
+```shell
+go get -u github.com/grpc-ecosystem/grpc-gateway/protoc-gen-grpc-gateway
+go get -u github.com/grpc-ecosystem/grpc-gateway/protoc-gen-swagger
+go get -u github.com/golang/protobuf/protoc-gen-go
+go get -u google.golang.org/grpc
+```
 
 
 ## Compile Server
@@ -70,7 +88,10 @@ If Python3 is used, replace `pip` with `pip3`.
 
 ``` shell
 mkdir server-build-cpu && cd server-build-cpu
-cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ -DPYTHON_LIBRARIES=$PYTHONROOT/lib/libpython2.7.so -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python -DSERVER=ON ..
+cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ \
+      -DPYTHON_LIBRARIES=$PYTHONROOT/lib/libpython2.7.so \
+      -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python \
+      -DSERVER=ON ..
 make -j10
 ```
 
@@ -80,7 +101,11 @@ you can execute `make install` to put targets under directory `./output`, you ne
 
 ``` shell
 mkdir server-build-gpu && cd server-build-gpu
-cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ -DPYTHON_LIBRARIES=$PYTHONROOT/lib/libpython2.7.so -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python -DSERVER=ON -DWITH_GPU=ON ..
+cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ \
+      -DPYTHON_LIBRARIES=$PYTHONROOT/lib/libpython2.7.so \
+      -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python \
+      -DSERVER=ON \
+      -DWITH_GPU=ON ..
 make -j10
 ```
 
@@ -94,7 +119,10 @@ execute `make install` to put targets under directory `./output`
 
 ``` shell
 mkdir client-build && cd client-build
-cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ -DPYTHON_LIBRARIES=$PYTHONROOT/lib/libpython2.7.so -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python -DCLIENT=ON ..
+cmake -DPYTHON_INCLUDE_DIR=$PYTHONROOT/include/python2.7/ \
+      -DPYTHON_LIBRARIES=$PYTHONROOT/lib/libpython2.7.so \
+      -DPYTHON_EXECUTABLE=$PYTHONROOT/bin/python \
+      -DCLIENT=ON ..
 make -j10
 ```
 
@@ -114,7 +142,7 @@ make
 
 ## Install wheel package
 
-Regardless of the client, server or App part, after compiling, install the whl package under `python/dist/`.
+Regardless of the client, server or App part, after compiling, install the whl package in `python/dist/` in the temporary directory(`server-build-cpu`, `server-build-gpu`, `client-build`,`app-build`) of the compilation process.
 
 
 
@@ -124,6 +152,12 @@ When running the python server, it will check the `SERVING_BIN` environment vari
 
 
 
+## Verify
+
+Please use the example under `python/examples` to verify.
+
+
+
 ## CMake Option Description
 
 | Compile Options  |                    Description             | Default |

doc/COMPILE_CN.md

@@ -5,7 +5,7 @@
 ## 编译环境设置
 
 |             组件             |             版本要求              |
-| :--------------------------: | :----------------------------------------------------------: |
+| :--------------------------: | :-------------------------------: |
 |              OS              |             CentOS 7              |
 |             gcc              |          4.8.5 and later          |
 |           gcc-c++            |          4.8.5 and later          |
@@ -23,7 +23,6 @@
 |           libXext            |               1.3.3               |
 |            libSM             |               1.2.2               |
 |          libXrender          |              0.9.10               |
-|          python-whl          | numpy>=1.12, <=1.16.4<br/>google>=2.0.3<br/>protobuf>=3.12.2<br/>grpcio-tools>=1.28.1<br/>grpcio>=1.28.1<br/>func-timeout>=4.3.5<br/>pyyaml>=1.3.0<br/>sentencepiece==0.1.92<br/>flask>=1.1.2<br/>ujson>=2.0.3 |
 
 推荐使用Docker编译，我们已经为您准备好了Paddle Serving编译环境，详见[该文档](DOCKER_IMAGES_CN.md)。
 
@@ -62,6 +61,22 @@ pip install -r python/requirements.txt
 
 如果使用 Python3，请以 `pip3` 替换 `pip`。
 
+## GOPATH 设置
+
+默认 GOPATH 设置为 `$HOME/go`，您也可以设置为其他值。
+```shell
+export GOPATH=$HOME/go
+export PATH=$PATH:$GOPATH/bin
+```
+
+## 获取 Go packages
+
+```shell
+go get -u github.com/grpc-ecosystem/grpc-gateway/protoc-gen-grpc-gateway
+go get -u github.com/grpc-ecosystem/grpc-gateway/protoc-gen-swagger
+go get -u github.com/golang/protobuf/protoc-gen-go
+go get -u google.golang.org/grpc
+```
 
 
 ## 编译Server部分
@@ -114,7 +129,7 @@ make
 
 ## 安装wheel包
 
-无论是Client端，Server端还是App部分，编译完成后，安装`python/dist/`下的whl包即可。
+无论是Client端，Server端还是App部分，编译完成后，安装编译过程临时目录（`server-build-cpu`、`server-build-gpu`、`client-build`、`app-build`）下的`python/dist/` 中的whl包即可。
 
 
 
@@ -124,6 +139,12 @@ make
 
 
 
+## 如何验证
+
+请使用 `python/examples` 下的例子进行验证。
+
+
+
 ## CMake选项说明
 
 |     编译选项     |                    说明                    | 默认 |

doc/FAQ.md

 # FAQ
 
-- Q：如何调整RPC服务的等待时间，避免超时？ 
+- Q: 如何调整RPC服务的等待时间，避免超时？ 
 
-  A：使用set_rpc_timeout_ms设置更长的等待时间，单位为毫秒，默认时间为20秒。
+  A: 使用set_rpc_timeout_ms设置更长的等待时间，单位为毫秒，默认时间为20秒。
   
   示例：
   ```
@@ -15,4 +15,25 @@
    ```
 
 - Q: 如何使用自己编译的Paddle Serving进行预测？
-  A：通过pip命令安装自己编译出的whl包，并设置SERVING_BIN环境变量为编译出的serving二进制文件路径。
+
+  A: 通过pip命令安装自己编译出的whl包，并设置SERVING_BIN环境变量为编译出的serving二进制文件路径。
+
+- Q: 执行GPU预测时遇到InvalidArgumentError: Device id must be less than GPU count, but received id is: 0. GPU count is: 0.
+
+  A: 将显卡驱动对应的libcuda.so的目录添加到LD_LIBRARY_PATH环境变量中
+
+- Q: 执行GPU预测时遇到ExternalError: Cudnn error, CUDNN_STATUS_BAD_PARAM at (/home/scmbuild/workspaces_cluster.dev/baidu.lib.paddlepaddle/baidu/lib/paddlepaddle/Paddle/paddle/fluid/operators/batch_norm_op.cu:198)
+
+  A: 将cudnn的lib64路径添加到LD_LIBRARY_PATH，安装自pypi的Paddle Serving中post9版使用的是cudnn 7.3,post10使用的是cudnn 7.5。如果是使用自己编译的Paddle Serving，可以在log/serving.INFO日志文件中查看对应的cudnn版本。
+
+- Q: 执行GPU预测时遇到Error: Failed to find dynamic library: libcublas.so
+
+  A: 将cuda的lib64路径添加到LD_LIBRARY_PATH, post9版本的Paddle Serving使用的是cuda 9.0，post10版本使用的cuda 10.0。
+
+- Q: 部署和预测中的日志信息在哪里查看？
+
+- A: server端的日志分为两部分，一部分打印到标准输出，一部分打印到启动服务时的目录下的log/serving.INFO文件中。
+
+    client端的日志直接打印到标准输出。
+
+    通过在部署服务之前 'export  GLOG_v=3'可以输出更为详细的日志信息。

doc/INFERENCE_TO_SERVING.md

@@ -2,6 +2,20 @@
 
 ([简体中文](./INFERENCE_TO_SERVING_CN.md)|English)
 
+We should know something before converting to serving model
+
+**inference_model_dir**：the directory of Paddle inference model
+
+**serving_client_dir**: the directory of server side configuration
+
+**serving_client_dir**: the directory of client side configuration
+
+**model_filename**: this is model description file whose default value is `__model__`, if it's not default name, set `model_filename` explicitly
+
+**params_filename**: during `save_inference_model` every Variable will be save as a single file. If we have the inference model whose params are compressed into one file, please set `params_filename` explicitly
+
+
+
 ## Example
 
 ``` python
@@ -12,3 +26,11 @@ serving_server_dir = "serving_server_dir"
 feed_var_names, fetch_var_names = inference_model_to_serving(
 		inference_model_dir, serving_client_dir, serving_server_dir)
 ```
+
+if your model file and params file are both standalone, please use the following api.
+
+```
+feed_var_names, fetch_var_names = inference_model_to_serving(
+		inference_model_dir, serving_client_dir, serving_server_dir,
+		model_filename="model", params_filename="params")
+```

doc/INFERENCE_TO_SERVING_CN.md

@@ -4,6 +4,19 @@
 
 ## 示例
 
+在下列代码中，我们需要知道以下信息。
+
+**模型文件夹**：这个文件夹就是Paddle的inference_model所在的文件夹
+
+**serving_client_dir**: 这个文件夹是inference_model转换成Serving模型后，服务端配置的保存路径
+
+**serving_client_dir**: 这个文件夹是inference_model转换成Serving模型后，客户端配置的保存路径
+
+**模型描述文件**: 模型描述文件也就是`model_filename`默认值为`__model__`,是一个pb2文本文件，如果是别的文件名需要显式指定
+
+**模型参数文件**: 在`save_inference_model`阶段，默认方式是每一个Variable保存一个二进制文件，如果是这种情况就不需要做指定。如果所有参数用压缩成一个文件的形式保存，则需要显式指定`params_filename`
+
+
 ``` python
 from paddle_serving_client.io import inference_model_to_serving
 inference_model_dir = "your_inference_model"
@@ -12,3 +25,9 @@ serving_server_dir = "serving_server_dir"
 feed_var_names, fetch_var_names = inference_model_to_serving(
 		inference_model_dir, serving_client_dir, serving_server_dir)
 ```
+如果模型中有模型描述文件`model_filename` 和 模型参数文件`params_filename`，那么请用
+```
+feed_var_names, fetch_var_names = inference_model_to_serving(
+		inference_model_dir, serving_client_dir, serving_server_dir,
+		 model_filename="model", params_filename="params")
+```

doc/LATEST_PACKAGES.md

@@ -3,51 +3,51 @@
 ## CPU server
 ### Python 3
 ```
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server-0.3.2-py3-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server-0.0.0-py3-none-any.whl
 ```
 
 ### Python 2
 ```
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server-0.3.2-py2-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server-0.0.0-py2-none-any.whl
 ```
 
 ## GPU server
 ### Python 3
 ```
 #cuda 9.0
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.3.2.post9-py3-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.0.0.post9-py3-none-any.whl
 #cuda 10.0
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.3.2.post10-py3-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.0.0.post10-py3-none-any.whl
 ```
 ### Python 2
 ```
 #cuda 9.0
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.3.2.post9-py2-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.0.0.post9-py2-none-any.whl
 #cuda 10.0
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.3.2.post10-py2-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_server_gpu-0.0.0.post10-py2-none-any.whl
 ```
 
 ## Client
 ### Python 3.7
 ```
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_client-0.3.2-cp37-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_client-0.0.0-cp37-none-any.whl
 ```
 ### Python 3.6
 ```
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_client-0.3.2-cp36-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_client-0.0.0-cp36-none-any.whl
 ```
 ### Python 2.7
 ```
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_client-0.3.2-cp27-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_client-0.0.0-cp27-none-any.whl
 ```
 
 ## App
 ### Python 3
 ```
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_app-0.1.2-py3-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_app-0.0.0-py3-none-any.whl
 ```
 
 ### Python 2
 ```
-https://paddle-serving.bj.bcebos.com/whl/paddle_serving_app-0.1.2-py2-none-any.whl
+https://paddle-serving.bj.bcebos.com/whl/paddle_serving_app-0.0.0-py2-none-any.whl
 ```

doc/NEW_WEB_SERVICE.md

 # How to develop a new Web service?
 
+
 ([简体中文](NEW_WEB_SERVICE_CN.md)|English)
 
-This document will take the image classification service based on the Imagenet data set as an example to introduce how to develop a new web service. The complete code can be visited at [here](../python/examples/imagenet/resnet50_web_service.py).
+This document will take Uci service as an example to introduce how to develop a new Web Service. You can check out the complete code [here](../python/examples/pipeline/simple_web_service/web_service.py).
 
-## WebService base class
+## Op base class
 
-Paddle Serving implements the [WebService](https://github.com/PaddlePaddle/Serving/blob/develop/python/paddle_serving_server/web_service.py#L23) base class. You need to override its `preprocess` and `postprocess` method. The default implementation is as follows:
+In some services, a single model may not meet business needs, requiring multiple models to be concatenated or parallel to complete the entire service. We call a single model operation Op and provide a simple set of interfaces to implement the complex logic of Op concatenation or parallelism.
 
-```python
-class WebService(object):
+Data between Ops is passed as a dictionary, Op can be started as threads or process, and Op can be configured for the number of concurrencies, etc.
+
+Typically, you need to inherit the Op base class and override its `init_op`,  `preprocess` and `postprocess` methods, which are implemented by default as follows:
 
-    def preprocess(self, feed={}, fetch=[]):
-        return feed, fetch
-    def postprocess(self, feed={}, fetch=[], fetch_map=None):
-        return fetch_map
+```python
+class Op(object):
+  def init_op(self):
+    pass
+  def preprocess(self, input_dicts):
+    # multiple previous Op
+    if len(input_dicts) != 1:
+      _LOGGER.critical(
+        "Failed to run preprocess: this Op has multiple previous "
+        "inputs. Please override this func.")
+      os._exit(-1)
+    (_, input_dict), = input_dicts.items()
+    return input_dict
+  def postprocess(self, input_dicts, fetch_dict):
+    return fetch_dict
 ```
 
+### init_op
+
+This method is used to load user-defined resources such as dictionaries. A separator is loaded in the [UciOp](../python/examples/pipeline/simple_web_service/web_service.py).
+
+**Note**: If Op is launched in threaded mode, different threads of the same Op execute `init_op` only once and share `init_op` loaded resources when Op is multi-concurrent.
+
 ### preprocess
 
-The preprocess method has two input parameters, `feed` and `fetch`. For an HTTP request `request`:
+This method is used to preprocess the data before model prediction. It has an `input_dicts` parameter, `input_dicts` is a dictionary, key is the `name` of the previous Op, and value is the data transferred from the corresponding previous op (the data is also in dictionary format).
 
-- The value of `feed` is the feed part `request.json["feed"]` in the request data 
-- The value of `fetch` is the fetch part `request.json["fetch"]` in the request data
+The `preprocess` method needs to process the data into a ndarray dictionary (key is the feed variable name, and value is the corresponding ndarray value). Op will take the return value as the input of the model prediction and pass the output to the `postprocess` method.
 
-The return values are the feed and fetch values used in the prediction.
+**Note**: if Op does not have a model configuration file, the return value of `preprocess` will be directly passed to `postprocess`.
 
 ### postprocess
 
-The postprocess method has three input parameters, `feed`, `fetch` and `fetch_map`:
+This method is used for data post-processing after model prediction. It has two parameters, `input_dicts` and `fetch_dict`.
+
+Where the `input_dicts` parameter is consistent with the parameter in `preprocess` method, and `fetch_dict` is the output of the model prediction (key is the name of the fetch variable, and value is the corresponding ndarray value). Op will take the return value of `postprocess` as the input of subsequent Op `preprocess`.
+
+**Note**: if Op does not have a model configuration file, `fetch_dict` will be the return value of `preprocess`.
+
+
+
+Here is the op of the UCI example:
+
+```python
+class UciOp(Op):
+    def init_op(self):
+        self.separator = ","
+
+    def preprocess(self, input_dicts):
+        (_, input_dict), = input_dicts.items()
+        x_value = input_dict["x"]
+        if isinstance(x_value, (str, unicode)):
+            input_dict["x"] = np.array(
+                [float(x.strip()) for x in x_value.split(self.separator)])
+        return input_dict
+
+    def postprocess(self, input_dicts, fetch_dict):
+        fetch_dict["price"] = str(fetch_dict["price"][0][0])
+        return fetch_dict
+```
+
 
-- The value of `feed` is the feed part `request.json["feed"]` in the request data 
-- The value of `fetch` is the fetch part `request.json["fetch"]` in the request data
-- The value of `fetch_map` is the model output value.
 
-The return value will be processed as `{"reslut": fetch_map}` as the return of the HTTP request.
+## WebService base class
+
+Paddle Serving implements the [WebService](https://github.com/PaddlePaddle/Serving/blob/develop/python/paddle_serving_server/web_service.py#L23) base class. You need to override its `get_pipeline_response` method to define the topological relationship between Ops. The default implementation is as follows:
+
+```python
+class WebService(object):
+  def get_pipeline_response(self, read_op):
+    return None
+```
+
+Where `read_op` serves as the entry point of the topology map of the whole service (that is, the first op defined by the user is followed by `read_op`).
+
+For single Op service (single model), take Uci service as an example (there is only one Uci prediction model in the whole service):
+
+```python
+class UciService(WebService):
+  def get_pipeline_response(self, read_op):
+    uci_op = UciOp(name="uci", input_ops=[read_op])
+    return uci_op
+```
 
-## Develop ImageService class
+For multiple Op services (multiple models), take Ocr service as an example (the whole service is completed in series by Det model and Rec model):
 
 ```python
-class ImageService(WebService):
-
-    def preprocess(self, feed={}, fetch=[]):
-        reader = ImageReader()
-        feed_batch = []
-        for ins in feed:
-            if "image" not in ins:
-                raise ("feed data error!")
-            sample = base64.b64decode(ins["image"])
-            img = reader.process_image(sample)
-            feed_batch.append({"image": img})
-        return feed_batch, fetch
+class OcrService(WebService):
+  def get_pipeline_response(self, read_op):
+    det_op = DetOp(name="det", input_ops=[read_op])
+    rec_op = RecOp(name="rec", input_ops=[det_op])
+    return rec_op
+```
+
+
+
+WebService objects need to load a yaml configuration file through the `prepare_pipeline_config` to configure each Op and the entire service. The simplest configuration file is as follows (Uci example):
+
+```yaml
+http_port: 18080
+op:
+    uci:
+        local_service_conf:
+            model_config: uci_housing_model # path
+```
+
+All field names of yaml file are as follows:
+
+```yaml
+rpc_port: 18080  # gRPC port
+build_dag_each_worker: false  # Whether to use process server or not. The default is false
+worker_num: 1  # gRPC thread pool size (the number of processes in the process version servicer). The default is 1
+http_port: 0 # HTTP service port. Do not start HTTP service when the value is less or equals 0. The default value is 0.
+dag:
+    is_thread_op: true  # Whether to use the thread version of OP. The default is true
+    client_type: brpc  # Use brpc or grpc client. The default is brpc
+    retry: 1  # The number of times DAG executor retries after failure. The default value is 1, that is, no retrying
+    use_profile: false  # Whether to print the log on the server side. The default is false
+    tracer:
+        interval_s: -1 # Monitoring time interval of Tracer (in seconds). Do not start monitoring when the value is less than 1. The default value is -1
+op:
+    <op_name>: # op name, corresponding to the one defined in the program
+        concurrency: 1 # op concurrency number, the default is 1
+        timeout: -1 # predict timeout in milliseconds. The default value is -1, that is, no timeout
+        retry: 1 # timeout retransmissions. The default value is 1, that is, do not try again
+        batch_size: 1 # If this field is set, Op will merge multiple request outputs into a single batch
+        auto_batching_timeout: -1 # auto-batching timeout in milliseconds. The default value is -1, that is, no timeout
+        local_service_conf:
+            model_config: # the path of the corresponding model file. There is no default value(None). If this item is not configured, the model file will not be loaded.
+            workdir: "" # working directory of corresponding model
+            thread_num: 2 # the corresponding model is started with thread_num threads
+            devices: "" # on which device does the model launched. You can specify the GPU card number(such as "0,1,2"), which is CPU by default
+            mem_optim: true # mem optimization option, the default is true
+            ir_optim: false # ir optimization option, the default is false
 ```
 
-For the above `ImageService`, only the `preprocess` method is rewritten to process the image data in Base64 format into the data format required by prediction.
+All fields of Op can be defined when Op is created in the program (which will override yaml fields).

doc/NEW_WEB_SERVICE_CN.md

 # 如何开发一个新的Web Service？
 
+
 (简体中文|[English](NEW_WEB_SERVICE.md))
 
-本文档将以Imagenet图像分类服务为例，来介绍如何开发一个新的Web Service。您可以在[这里](../python/examples/imagenet/resnet50_web_service.py)查阅完整的代码。
+本文档将以 Uci 房价预测服务为例，来介绍如何开发一个新的Web Service。您可以在[这里](../python/examples/pipeline/simple_web_service/web_service.py)查阅完整的代码。
+
+## Op 基类
 
-## WebService基类
+在一些服务中，单个模型可能无法满足需求，需要多个模型串联或并联来完成整个服务。我们将单个模型操作称为 Op，并提供了一套简单的接口来实现 Op 串联或并联的复杂逻辑。
 
-Paddle Serving实现了[WebService](https://github.com/PaddlePaddle/Serving/blob/develop/python/paddle_serving_server/web_service.py#L23)基类，您需要重写它的`preprocess`方法和`postprocess`方法，默认实现如下：
+Op 间数据是以字典形式进行传递的，Op 可以以线程或进程方式启动，同时可以对 Op 的并发数等进行配置。
+
+通常情况下，您需要继承 Op 基类，重写它的 `init_op`、`preprocess` 和 `postprocess` 方法，默认实现如下：
 
 ```python
-class WebService(object):
+class Op(object):
+  def init_op(self):
+    pass
+  def preprocess(self, input_dicts):
+    # multiple previous Op
+    if len(input_dicts) != 1:
+      _LOGGER.critical(
+        "Failed to run preprocess: this Op has multiple previous "
+        "inputs. Please override this func.")
+      os._exit(-1)
+    (_, input_dict), = input_dicts.items()
+    return input_dict
+  def postprocess(self, input_dicts, fetch_dict):
+    return fetch_dict
+```
+
+### init_op 方法
+
+该方法用于加载用户自定义资源（如字典等），在 [UciOp](../python/examples/pipeline/simple_web_service/web_service.py) 中加载了一个分隔符。
+
+**注意**：如果 Op 是以线程模式加载的，那么在 Op 多并发时，同种 Op 的不同线程只执行一次 `init_op`，且共用 `init_op` 加载的资源。
+
+### preprocess 方法
+
+该方法用于模型预测前对数据的预处理，它有一个 `input_dicts` 参数，`input_dicts` 是一个字典，key 为前继 Op 的 `name`，value 为对应前继 Op 传递过来的数据（数据同样是字典格式）。
+
+`preprocess` 方法需要将数据处理成 ndarray 字典（key 为 feed 变量名，value 为对应的 ndarray 值），Op 会将该返回值作为模型预测的输入，并将输出传递给 `postprocess` 方法。
+
+**注意**：如果 Op 没有配置模型，则 `preprocess` 的返回值会直接传递给 `postprocess`。
+
+### postprocess 方法
+
+该方法用于模型预测后对数据的后处理，它有两个参数，`input_dicts` 和 `fetch_dict`。
+
+其中，`input_dicts` 与 `preprocess` 的参数相同，`fetch_dict` 为模型预测的输出（key 为 fetch 变量名，value 为对应的 ndarray 值）。Op 会将 `postprocess` 的返回值作为后继 Op `preprocess` 的输入。
+
+**注意**：如果 Op 没有配置模型，则 `fetch_dict` 将为 `preprocess` 的返回值。
 
-    def preprocess(self, feed={}, fetch=[]):
-        return feed, fetch
-    def postprocess(self, feed={}, fetch=[], fetch_map=None):
-        return fetch_map
+
+
+下面是 Uci 例子的 Op：
+
+```python
+class UciOp(Op):
+    def init_op(self):
+        self.separator = ","
+
+    def preprocess(self, input_dicts):
+        (_, input_dict), = input_dicts.items()
+        x_value = input_dict["x"]
+        if isinstance(x_value, (str, unicode)):
+            input_dict["x"] = np.array(
+                [float(x.strip()) for x in x_value.split(self.separator)])
+        return input_dict
+
+    def postprocess(self, input_dicts, fetch_dict):
+        fetch_dict["price"] = str(fetch_dict["price"][0][0])
+        return fetch_dict
 ```
 
-### preprocess方法
 
-preprocess方法有两个输入参数，`feed`和`fetch`。对于一个HTTP请求`request`：
 
-- `feed`的值为请求数据中的feed部分`request.json["feed"]`
-- `fetch`的值为请求数据中的fetch部分`request.json["fetch"]`
+## WebService 基类
 
-返回值分别是预测过程中用到的feed和fetch值。
+Paddle Serving 实现了 [WebService](https://github.com/PaddlePaddle/Serving/blob/develop/python/paddle_serving_server/web_service.py#L28) 基类，您需要重写它的 `get_pipeline_response` 方法来定义 Op 间的拓扑关系，并返回作为 Response 的 Op，默认实现如下：
 
-### postprocess方法
+```python
+class WebService(object):
+  def get_pipeline_response(self, read_op):
+    return None
+```
 
-postprocess方法有三个输入参数，`feed`、`fetch`和`fetch_map`：
+其中，`read_op` 作为整个服务拓扑图的入口（即用户自定义的第一个 Op 的前继为 `read_op`）。
 
-- `feed`的值为请求数据中的feed部分`request.json["feed"]`
-- `fetch`的值为请求数据中的fetch部分`request.json["fetch"]`
-- `fetch_map`的值为fetch到的模型输出值
+对于单 Op 服务（单模型），以 Uci 服务为例（整个服务中只有一个 Uci 房价预测模型）：
 
-返回值将会被处理成`{"reslut": fetch_map}`作为HTTP请求的返回。
+```python
+class UciService(WebService):
+  def get_pipeline_response(self, read_op):
+    uci_op = UciOp(name="uci", input_ops=[read_op])
+    return uci_op
+```
 
-## 开发ImageService类
+对于多 Op 服务（多模型），以 Ocr 服务为例（整个服务由 Det 模型和 Rec 模型串联完成）：
 
 ```python
-class ImageService(WebService):
-
-    def preprocess(self, feed={}, fetch=[]):
-        reader = ImageReader()
-        feed_batch = []
-        for ins in feed:
-            if "image" not in ins:
-                raise ("feed data error!")
-            sample = base64.b64decode(ins["image"])
-            img = reader.process_image(sample)
-            feed_batch.append({"image": img})
-        return feed_batch, fetch
+class OcrService(WebService):
+  def get_pipeline_response(self, read_op):
+    det_op = DetOp(name="det", input_ops=[read_op])
+    rec_op = RecOp(name="rec", input_ops=[det_op])
+    return rec_op
+```
+
+
+
+WebService 对象需要通过 `prepare_pipeline_config` 加载一个 yaml 配置文件，用来对各个 Op 以及整个服务进行配置，最简单的配置文件如下（Uci 例子）：
+
+```yaml
+http_port: 18080
+op:
+    uci:
+        local_service_conf:
+            model_config: uci_housing_model # 路径
+```
+
+yaml 文件的所有字段名详见下面：
+
+```yaml
+rpc_port: 18080  # gRPC端口号
+build_dag_each_worker: false  # 是否使用进程版 Servicer，默认为 false
+worker_num: 1  # gRPC线程池大小（进程版 Servicer 中为进程数），默认为 1
+http_port: 0 # HTTP 服务的端口号，若该值小于或等于 0 则不开启 HTTP 服务，默认为 0
+dag:
+    is_thread_op: true  # 是否使用线程版Op，默认为 true
+    client_type: brpc  # 使用 brpc 或 grpc client，默认为 brpc
+    retry: 1  # DAG Executor 在失败后重试次数，默认为 1，即不重试
+    use_profile: false  # 是否在 Server 端打印日志，默认为 false
+    tracer:
+        interval_s: -1 # Tracer 监控的时间间隔，单位为秒。当该值小于 1 时不启动监控，默认为 -1
+op:
+    <op_name>: # op 名，与程序中定义的相对应
+        concurrency: 1 # op 并发数，默认为 1
+        timeout: -1 # 预测超时时间，单位为毫秒。默认为 -1 即不超时
+        retry: 1 # 超时重发次数。默认为 1 即不重试
+        batch_size: 1 # auto-batching 中的 batch_size，若设置该字段则 Op 会将多个请求输出合并为一个 batch
+        auto_batching_timeout: -1 # auto-batching 超时时间，单位为毫秒。默认为 -1 即不超时
+        local_service_conf:
+            model_config: # 对应模型文件的路径，无默认值（None）。若不配置该项则不会加载模型文件。
+            workdir: "" # 对应模型的工作目录
+            thread_num: 2 # 对应模型用几个线程启动
+            devices: "" # 模型启动在哪个设备上，可以指定 gpu 卡号（如 "0,1,2"），默认为 cpu
+            mem_optim: true # mem 优化选项，默认为 true
+            ir_optim: false # ir 优化选项，默认为 false
 ```
 
-对于上述的`ImageService`，只重写了前处理方法，将base64格式的图片数据处理成模型预测需要的数据格式。
+其中，Op 的所有字段均可以在程序中创建 Op 时定义（会覆盖 yaml 的字段）。

doc/PIPELINE_SERVING.md

@@ -33,6 +33,7 @@ The graph execution engine consists of OPs and Channels, and the connected OPs s
 - The default function of a single OP is to access a single Paddle Serving Service based on the input Channel data and put the result into the output Channel.
 - OP supports user customization, including preprocess, process, postprocess functions that can be inherited and implemented by the user.
 - OP can set the number of concurrencies to increase the number of concurrencies processed.
+- OP can obtain data from multiple different RPC requests for Auto-Batching.
 - OP can be started by a thread or process.
 
 ### Channel Design
@@ -46,6 +47,7 @@ The graph execution engine consists of OPs and Channels, and the connected OPs s
 </center>
 
 
+
 ### Extreme Case Consideration
 
 - Request timeout
@@ -59,9 +61,9 @@ The graph execution engine consists of OPs and Channels, and the connected OPs s
 - Whether input buffers and output buffers in Channel will increase indefinitely
 
   - It will not increase indefinitely. The input to the entire graph execution engine is placed inside a Channel's internal queue, directly acting as a traffic control buffer queue for the entire service.
-  - For input buffer, adjust the number of concurrencies of OP1 and OP2 according to the amount of computation, so that the number of input buffers from each input OP is relatively balanced.
-  - For output buffer, you can use a similar process as input buffer, which adjusts the concurrency of OP3 and OP4 to control the buffer length of output buffer.
-  - Note: The length of the input buffer depends on the speed at which each item in the internal queue is ready, and the length of the output buffer depends on the speed at which downstream OPs obtain data from the output buffer.
+  - For input buffer, adjust the number of concurrencies of OP1 and OP2 according to the amount of computation, so that the number of input buffers from each input OP is relatively balanced. (The length of the input buffer depends on the speed at which each item in the internal queue is ready)
+  - For output buffer, you can use a similar process as input buffer, which adjusts the concurrency of OP3 and OP4 to control the buffer length of output buffer. (The length of the output buffer depends on the speed at which downstream OPs obtain data from the output buffer)
+  - The amount of data in the Channel will not exceed `worker_num` of gRPC, that is, it will not exceed the thread pool size.
 
 ## Detailed Design
 
@@ -79,31 +81,36 @@ def __init__(name=None,
              client_config=None,
              concurrency=1,
              timeout=-1,
-             retry=1)
+             retry=1,
+             batch_size=1,
+             auto_batching_timeout=None)
 ```
 
 The meaning of each parameter is as follows:
 
 |       Parameter       |                           Meaning                            |
-| :--------------: | :----------------------------------------------------------: |
+| :-------------------: | :----------------------------------------------------------: |
 |         name          | (str) String used to identify the OP type, which must be globally unique. |
 |       input_ops       |     (list) A list of all previous OPs of the current Op.     |
 |   server_endpoints    | (list) List of endpoints for remote Paddle Serving Service. If this parameter is not set, the OP will not access the remote Paddle Serving Service, that is, the process operation will not be performed. |
 |      fetch_list       | (list) List of fetch variable names for remote Paddle Serving Service. |
 |     client_config     | (str) The path of the client configuration file corresponding to the Paddle Serving Service. |
 |      concurrency      |             (int) The number of concurrent OPs.              |
-|     timeout      | (int) The timeout time of the process operation, in seconds. If the value is less than zero, no timeout is considered. |
+|        timeout        | (int) The timeout time of the process operation, in ms. If the value is less than zero, no timeout is considered. |
 |         retry         | (int) Timeout number of retries. When the value is 1, no retries are made. |
+|      batch_size       | (int) The expected batch_size of Auto-Batching, since building batches may time out, the actual batch_size may be less than the set value. |
+| auto_batching_timeout | (float) Timeout for building batches of Auto-Batching (the unit is ms). |
+
 
 #### 2. General OP Secondary Development Interface
 
 |              Interface or Variable               |                           Explain                            |
-| :--------------------------------------------: | :----------------------------------------------------------: |
-|       def preprocess(self, input_dicts)        | Process the data obtained from the channel, and the processed data will be used as the input of the **process** function. |
-|          def process(self, feed_dict)          | The RPC prediction process is based on the Paddle Serving Client, and the processed data will be used as the input of the **postprocess** function. |
-| def postprocess(self, input_dicts, fetch_dict) | After processing the prediction results, the processed data will be put into the subsequent Channel to be obtained by the subsequent OP. |
+| :----------------------------------------------: | :----------------------------------------------------------: |
+|        def preprocess(self, input_dicts)         | Process the data obtained from the channel, and the processed data will be used as the input of the **process** function. (This function handles a **sample**) |
+| def process(self, feed_dict_list, typical_logid) | The RPC prediction process is based on the Paddle Serving Client, and the processed data will be used as the input of the **postprocess** function. (This function handles a **batch**) |
+|  def postprocess(self, input_dicts, fetch_dict)  | After processing the prediction results, the processed data will be put into the subsequent Channel to be obtained by the subsequent OP. (This function handles a **sample**) |
 |                def init_op(self)                 |      Used to load resources (such as word dictionary).       |
-|              self.concurrency_idx              | Concurrency index of current thread / process (different kinds of OP are calculated separately). |
+|               self.concurrency_idx               | Concurrency index of current process(not thread) (different kinds of OP are calculated separately). |
 
 In a running cycle, OP will execute three operations: preprocess, process, and postprocess (when the `server_endpoints` parameter is not set, the process operation is not executed). Users can rewrite these three functions. The default implementation is as follows:
 
@@ -117,24 +124,28 @@ def preprocess(self, input_dicts):
   (_, input_dict), = input_dicts.items()
   return input_dict
 
-def process(self, feed_dict):
-  err, err_info = ChannelData.check_npdata(feed_dict)
+def process(self, feed_dict_list, typical_logid):
+  err, err_info = ChannelData.check_batch_npdata(feed_dict_list)
   if err != 0:
     raise NotImplementedError(
       "{} Please override preprocess func.".format(err_info))
   call_result = self.client.predict(
-    feed=feed_dict, fetch=self._fetch_names)
+    feed=feed_dict_list, fetch=self._fetch_names, log_id=typical_logid)
+  if isinstance(self.client, MultiLangClient):
+    if call_result is None or call_result["serving_status_code"] != 0:
+      return None
+    call_result.pop("serving_status_code")
   return call_result
 
 def postprocess(self, input_dicts, fetch_dict):
   return fetch_dict
 ```
 
-The parameter of **preprocess** is the data `input_dicts` in the previous Channel. This variable is a dictionary with the name of the previous OP as key and the output of the corresponding OP as value.
+The parameter of **preprocess** is the data `input_dicts` in the previous Channel. This variable (as a **sample**) is a dictionary with the name of the previous OP as key and the output of the corresponding OP as value.
 
-The parameter of **process** is the input variable `fetch_dict` (the return value of the preprocess function) of the Paddle Serving Client prediction interface. This variable is a dictionary with feed_name as the key and the data in the ndarray format as the value.
+The parameter of **process** is the input variable `fetch_dict_list` (a list of the return value of the preprocess function) of the Paddle Serving Client prediction interface. This variable (as a **batch**) is a list of dictionaries with feed_name as the key and the data in the ndarray format as the value. `typical_logid` is used as the logid that penetrates to PaddleServingService.
 
-The parameters of **postprocess** are `input_dicts` and `fetch_dict`. `input_dicts` is consistent with the parameter of preprocess, and `fetch_dict` is the return value of the process function (if process is not executed, this value is the return value of preprocess).
+The parameters of **postprocess** are `input_dicts` and `fetch_dict`. `input_dicts` is consistent with the parameter of preprocess, and `fetch_dict` (as a **sample**) is a sample of the return batch of the process function (if process is not executed, this value is the return value of preprocess).
 
 Users can also rewrite the **init_op** function to load some custom resources (such as word dictionary). The default implementation is as follows:
 
@@ -143,7 +154,7 @@ def init_op(self):
   pass
 ```
 
-It should be noted that in the threaded version of OP, each OP will only call this function once, so the loaded resources must be thread safe.
+It should be **noted** that in the threaded version of OP, each OP will only call this function once, so the loaded resources must be thread safe.
 
 #### 3. RequestOp Definition
 
@@ -240,14 +251,17 @@ server.run_server()
 Where `response_op` is the responseop mentioned above, PipelineServer will initialize Channels according to the topology relationship of each OP and build the calculation graph. `config_yml_path` is the configuration file of PipelineServer. The example file is as follows:
 
 ```yaml
-port: 18080  # gRPC port
+rpc_port: 18080  # gRPC port
 worker_num: 1  # gRPC thread pool size (the number of processes in the process version servicer). The default is 1
 build_dag_each_worker: false  # Whether to use process server or not. The default is false
+http_port: 0 # HTTP service port. Do not start HTTP service when the value is less or equals 0. The default value is 0.
 dag:
     is_thread_op: true  # Whether to use the thread version of OP. The default is true
     client_type: brpc  # Use brpc or grpc client. The default is brpc
     retry: 1  # The number of times DAG executor retries after failure. The default value is 1, that is, no retrying
     use_profile: false  # Whether to print the log on the server side. The default is false
+    tracer:
+        interval_s: 600 # Monitoring time interval of Tracer (in seconds). Do not start monitoring when the value is less than 1. The default value is -1
 ```
 
 
@@ -272,6 +286,8 @@ python -m paddle_serving_server.serve --model imdb_cnn_model --port 9292 &> cnn.
 python -m paddle_serving_server.serve --model imdb_bow_model --port 9393 &> bow.log &
 ```
 
+PipelineServing also supports local automatic startup of PaddleServingService. Please refer to the example `python/examples/pipeline/ocr`.
+
 ### Start PipelineServer
 
 Run the following code
@@ -282,14 +298,8 @@ from paddle_serving_server.pipeline import PipelineServer
 from paddle_serving_server.pipeline.proto import pipeline_service_pb2
 from paddle_serving_server.pipeline.channel import ChannelDataEcode
 import numpy as np
-import logging
 from paddle_serving_app.reader import IMDBDataset
 
-logging.basicConfig(level=logging.DEBUG)
-
-_LOGGER = logging.getLogger()
-
-
 class ImdbRequestOp(RequestOp):
     def init_op(self):
         self.imdb_dataset = IMDBDataset()
@@ -377,7 +387,7 @@ for f in futures:
 
 
 
-## How to optimize through the timeline tool
+## How to optimize with the timeline tool
 
 In order to better optimize the performance, PipelineServing provides a timeline tool to monitor the time of each stage of the whole service.
 
@@ -390,15 +400,23 @@ dag:
     use_profile: true
 ```
 
-After the function is enabled, the server will print the corresponding log information to the standard output in the process of prediction. In order to show the time consumption of each stage more intuitively, scripts are provided for further analysis and processing of log files.
+After the function is enabled, the server will print the corresponding log information to the standard output in the process of prediction. In order to show the time consumption of each stage more intuitively, Analyst module is provided for further analysis and processing of log files.
 
-The output of the server is first saved to a file. Taking profile as an example, the script converts the time monitoring information in the log into JSON format and saves it to the trace file. The trace file can be visualized through the tracing function of Chrome browser.
+The output of the server is first saved to a file. Taking `profile.txt` as an example, the script converts the time monitoring information in the log into JSON format and saves it to the `trace` file. The `trace` file can be visualized through the tracing function of Chrome browser.
 
 ```shell
-python timeline_trace.py profile trace
+from paddle_serving_server.pipeline import Analyst
+import json
+import sys
+
+if __name__ == "__main__":
+    log_filename = "profile.txt"
+    trace_filename = "trace"
+    analyst = Analyst(log_filename)
+    analyst.save_trace(trace_filename)
 ```
 
-Specific operation: open Chrome browser, input in the address bar `chrome://tracing/` , jump to the tracing page, click the load button, open the saved trace file, and then visualize the time information of each stage of the prediction service.
+Specific operation: open Chrome browser, input in the address bar `chrome://tracing/` , jump to the tracing page, click the load button, open the saved `trace` file, and then visualize the time information of each stage of the prediction service.
 
 ### Output profile information on client side
 

doc/PIPELINE_SERVING_CN.md

@@ -6,6 +6,7 @@ Paddle Serving 通常用于单模型的一键部署，但端到端的深度学
 
 Paddle Serving 提供了用户友好的多模型组合服务编程框架，Pipeline Serving，旨在降低编程门槛，提高资源使用率（尤其是GPU设备），提升整体的预估效率。
 
+
 ## 整体架构设计
 
 Server端基于 gRPC 和图执行引擎构建，两者的关系如下图所示。
@@ -30,9 +31,10 @@ Server端基于 gRPC 和图执行引擎构建，两者的关系如下图所示
 
 ### OP的设计
 
-- 单个OP默认的功能是根据输入的 Channel 数据，访问一个 Paddle Serving 的单模型服务，并将结果存在输出的 Channel
+- 单个 OP 默认的功能是根据输入的 Channel 数据，访问一个 Paddle Serving 的单模型服务，并将结果存在输出的 Channel
 - 单个 OP 可以支持用户自定义，包括 preprocess，process，postprocess 三个函数都可以由用户继承和实现
 - 单个 OP 可以控制并发数，从而增加处理并发数
+- 单个 OP 可以获取多个不同 RPC 请求的数据，以实现 Auto-Batching
 - OP 可以由线程或进程启动
 
 ### Channel的设计
@@ -59,11 +61,9 @@ Server端基于 gRPC 和图执行引擎构建，两者的关系如下图所示
 - Channel 设计中的 input buffer 和 output buffer 是否会无限增加
 
   - 不会。整个图执行引擎的输入会放到一个 Channel 的 internal queue 里面，直接作为整个服务的流量控制缓冲队列
-  - 对于 input buffer，根据计算量的情况调整 OP1 和 OP2 的并发数，使得 input buffer 来自各个输入 OP 的数量相对平衡
-  - 对于 output buffer，可以采用和 input buffer 类似的处理方法，即调整 OP3 和 OP4 的并发数，使得 output buffer 的缓冲长度得到控制
-  - 注：input buffer 的长度取决于 internal queue 中每个 item 完全 ready 的速度，output buffer 的长度取决于下游 OP 从 output buffer 获取数据的速度
-
-## 详细设计
+  - 对于 input buffer，根据计算量的情况调整 OP1 和 OP2 的并发数，使得 input buffer 来自各个输入 OP 的数量相对平衡（input buffer 的长度取决于 internal queue 中每个 item 完全 ready 的速度）
+  - 对于 output buffer，可以采用和 input buffer 类似的处理方法，即调整 OP3 和 OP4 的并发数，使得 output buffer 的缓冲长度得到控制（output buffer 的长度取决于下游 OP 从 output buffer 获取数据的速度）
+  - 同时 Channel 中数据量不会超过 gRPC 的 `worker_num`，即线程池大小
 
 ### 用户接口设计
 
@@ -79,31 +79,36 @@ def __init__(name=None,
              client_config=None,
              concurrency=1,
              timeout=-1,
-             retry=1)
+             retry=1,
+             batch_size=1,
+             auto_batching_timeout=None)
 ```
 
 各参数含义如下
 
 |        参数名         |                             含义                             |
-| :--------------: | :----------------------------------------------------------: |
+| :-------------------: | :----------------------------------------------------------: |
 |         name          |    （str）用于标识 OP 类型的字符串，该字段必须全局唯一。     |
 |       input_ops       |            （list）当前 OP 的所有前继 OP 的列表。            |
 |   server_endpoints    | （list）远程 Paddle Serving Service 的 endpoints 列表。如果不设置该参数，则不访问远程 Paddle Serving Service，即 不会执行 process 操作。 |
 |      fetch_list       |     （list）远程 Paddle Serving Service 的 fetch 列表。      |
 |     client_config     | （str）Paddle Serving Service 对应的 Client 端配置文件路径。 |
 |      concurrency      |                     （int）OP 的并发数。                     |
-|     timeout      | （int）process 操作的超时时间，单位为秒。若该值小于零，则视作不超时。 |
+|        timeout        | （int）process 操作的超时时间，单位为毫秒。若该值小于零，则视作不超时。 |
 |         retry         |       （int）超时重试次数。当该值为 1 时，不进行重试。       |
+|      batch_size       | （int）进行 Auto-Batching 的期望 batch_size 大小，由于构建 batch 可能超时，实际 batch_size 可能小于设定值。 |
+| auto_batching_timeout | （float）进行 Auto-Batching 构建 batch 的超时时间，单位为毫秒。 |
+
 
 #### 2. 普通 OP二次开发接口
 
 |                    变量或接口                    |                             说明                             |
-| :--------------------------------------------: | :----------------------------------------------------------: |
-|       def preprocess(self, input_dicts)        | 对从 Channel 中获取的数据进行处理，处理完的数据将作为 **process** 函数的输入。 |
-|          def process(self, feed_dict)          | 基于 Paddle Serving Client 进行 RPC 预测，处理完的数据将作为 **postprocess** 函数的输入。 |
-| def postprocess(self, input_dicts, fetch_dict) | 处理预测结果，处理完的数据将被放入后继 Channel 中，以被后继 OP 获取。 |
+| :----------------------------------------------: | :----------------------------------------------------------: |
+|        def preprocess(self, input_dicts)         | 对从 Channel 中获取的数据进行处理，处理完的数据将作为 **process** 函数的输入。（该函数对一个 **sample** 进行处理） |
+| def process(self, feed_dict_list, typical_logid) | 基于 Paddle Serving Client 进行 RPC 预测，处理完的数据将作为 **postprocess** 函数的输入。（该函数对一个 **batch** 进行处理） |
+|  def postprocess(self, input_dicts, fetch_dict)  | 处理预测结果，处理完的数据将被放入后继 Channel 中，以被后继 OP 获取。（该函数对一个 **sample** 进行处理） |
 |                def init_op(self)                 |                  用于加载资源（如字典等）。                  |
-|              self.concurrency_idx              |   当前线程（进程）的并发数索引（不同种类的 OP 单独计算）。   |
+|               self.concurrency_idx               |  当前进程（非线程）的并发数索引（不同种类的 OP 单独计算）。  |
 
 OP 在一个运行周期中会依次执行 preprocess，process，postprocess 三个操作（当不设置 `server_endpoints` 参数时，不执行 process 操作），用户可以对这三个函数进行重写，默认实现如下：
 
@@ -117,25 +122,28 @@ def preprocess(self, input_dicts):
   (_, input_dict), = input_dicts.items()
   return input_dict
 
-def process(self, feed_dict):
-  err, err_info = ChannelData.check_npdata(feed_dict)
+def process(self, feed_dict_list, typical_logid):
+  err, err_info = ChannelData.check_batch_npdata(feed_dict_list)
   if err != 0:
     raise NotImplementedError(
       "{} Please override preprocess func.".format(err_info))
   call_result = self.client.predict(
-    feed=feed_dict, fetch=self._fetch_names)
+    feed=feed_dict_list, fetch=self._fetch_names, log_id=typical_logid)
+  if isinstance(self.client, MultiLangClient):
+    if call_result is None or call_result["serving_status_code"] != 0:
+      return None
+    call_result.pop("serving_status_code")
   return call_result
 
 def postprocess(self, input_dicts, fetch_dict):
   return fetch_dict
 ```
 
+**preprocess** 的参数是前继 Channel 中的数据 `input_dicts`，该变量（作为一个 **sample**）是一个以前继 OP 的 name 为 Key，对应 OP 的输出为 Value 的字典。
 
-**preprocess** 的参数是前继 Channel 中的数据 `input_dicts`，该变量是一个以前继 OP 的 name 为 Key，对应 OP 的输出为 Value 的字典。
-
-**process** 的参数是 Paddle Serving Client 预测接口的输入变量 `fetch_dict`（preprocess 函数的返回值），该变量是一个以 feed_name 为 Key，对应 ndarray 格式的数据为 Value 的字典。
+**process** 的参数是 Paddle Serving Client 预测接口的输入变量 `fetch_dict_list`（preprocess 函数的返回值的列表），该变量（作为一个 **batch**）是一个列表，列表中的元素为以 feed_name 为 Key，对应 ndarray 格式的数据为 Value 的字典。`typical_logid` 作为向 PaddleServingService 穿透的 logid。
 
-**postprocess** 的参数是 `input_dicts` 和 `fetch_dict`，`input_dicts` 与 preprocess 的参数一致，`fetch_dict` 是 process 函数的返回值（如果没有执行 process ，则该值为 preprocess 的返回值）。
+**postprocess** 的参数是 `input_dicts` 和 `fetch_dict`，`input_dicts` 与 preprocess 的参数一致，`fetch_dict` （作为一个 **sample**）是 process 函数的返回 batch 中的一个 sample（如果没有执行 process ，则该值为 preprocess 的返回值）。
 
 用户还可以对 **init_op** 函数进行重写，已加载自定义的一些资源（比如字典等），默认实现如下：
 
@@ -144,7 +152,7 @@ def init_op(self):
   pass
 ```
 
-需要注意的是，在线程版 OP 中，每个 OP 只会调用一次该函数，故加载的资源必须要求是线程安全的。
+需要**注意**的是，在线程版 OP 中，每个 OP 只会调用一次该函数，故加载的资源必须要求是线程安全的。
 
 #### 3. RequestOp 定义
 
@@ -241,14 +249,17 @@ server.run_server()
 其中，`response_op` 为上面提到的 ResponseOp，PipelineServer 将会根据各个 OP 的拓扑关系初始化 Channel 并构建计算图。`config_yml_path` 为 PipelineServer 的配置文件，示例文件如下：
 
 ```yaml
-port: 18080  # gRPC端口号
+rpc_port: 18080  # gRPC端口号
 worker_num: 1  # gRPC线程池大小（进程版 Servicer 中为进程数），默认为 1
 build_dag_each_worker: false  # 是否使用进程版 Servicer，默认为 false
+http_port: 0 # HTTP 服务的端口号，若该值小于或等于 0 则不开启 HTTP 服务，默认为 0
 dag:
     is_thread_op: true  # 是否使用线程版Op，默认为 true
     client_type: brpc  # 使用 brpc 或 grpc client，默认为 brpc
     retry: 1  # DAG Executor 在失败后重试次数，默认为 1，即不重试
     use_profile: false  # 是否在 Server 端打印日志，默认为 false
+    tracer:
+        interval_s: 600 # Tracer 监控的时间间隔，单位为秒。当该值小于 1 时不启动监控，默认为 -1
 ```
 
 
@@ -273,6 +284,8 @@ python -m paddle_serving_server.serve --model imdb_cnn_model --port 9292 &> cnn.
 python -m paddle_serving_server.serve --model imdb_bow_model --port 9393 &> bow.log &
 ```
 
+PipelineServing 也支持本地自动启动 PaddleServingService，请参考 `python/examples/pipeline/ocr` 下的例子。
+
 ### 启动 PipelineServer
 
 运行下面代码
@@ -283,14 +296,8 @@ from paddle_serving_server.pipeline import PipelineServer
 from paddle_serving_server.pipeline.proto import pipeline_service_pb2
 from paddle_serving_server.pipeline.channel import ChannelDataEcode
 import numpy as np
-import logging
 from paddle_serving_app.reader import IMDBDataset
 
-logging.basicConfig(level=logging.DEBUG)
-
-_LOGGER = logging.getLogger()
-
-
 class ImdbRequestOp(RequestOp):
     def init_op(self):
         self.imdb_dataset = IMDBDataset()
@@ -311,7 +318,6 @@ class CombineOp(Op):
     def preprocess(self, input_data):
         combined_prediction = 0
         for op_name, data in input_data.items():
-            _LOGGER.info("{}: {}".format(op_name, data["prediction"]))
             combined_prediction += data["prediction"]
         data = {"prediction": combined_prediction / 2}
         return data
@@ -391,15 +397,23 @@ dag:
     use_profile: true
 ```
 
-开启该功能后，Server 端在预测的过程中会将对应的日志信息打印到标准输出，为了更直观地展现各阶段的耗时，提供脚本对日志文件做进一步的分析处理。
+开启该功能后，Server 端在预测的过程中会将对应的日志信息打印到标准输出，为了更直观地展现各阶段的耗时，提供 Analyst 模块对日志文件做进一步的分析处理。
 
-使用时先将 Server 的输出保存到文件，以 profile 为例，脚本将日志中的时间打点信息转换成 json 格式保存到trace 文件，trace 文件可以通过 chrome 浏览器的 tracing 功能进行可视化。
+使用时先将 Server 的输出保存到文件，以 `profile.txt` 为例，脚本将日志中的时间打点信息转换成 json 格式保存到 `trace` 文件，`trace` 文件可以通过 chrome 浏览器的 tracing 功能进行可视化。
 
-```shell
-python timeline_trace.py profile trace
+```python
+from paddle_serving_server.pipeline import Analyst
+import json
+import sys
+
+if __name__ == "__main__":
+    log_filename = "profile.txt"
+    trace_filename = "trace"
+    analyst = Analyst(log_filename)
+    analyst.save_trace(trace_filename)
 ```
 
-具体操作：打开 chrome 浏览器，在地址栏输入 chrome://tracing/ ，跳转至 tracing 页面，点击 load 按钮，打开保存的 trace 文件，即可将预测服务的各阶段时间信息可视化。
+具体操作：打开 chrome 浏览器，在地址栏输入 `chrome://tracing/` ，跳转至 tracing 页面，点击 load 按钮，打开保存的 `trace` 文件，即可将预测服务的各阶段时间信息可视化。
 
 ### 在 Client 端输出 Profile 信息
 

doc/SAVE.md

@@ -38,12 +38,15 @@ If you have saved model files using Paddle's `save_inference_model` API, you can
 import paddle_serving_client.io as serving_io
 serving_io.inference_model_to_serving(dirname, serving_server="serving_server", serving_client="serving_client", model_filename=None, params_filename=None )
 ```
-dirname (str) - Path of saved model files. Program file and parameter files are saved in this directory.
-
-serving_server (str, optional) - The path of model files and configuration files for server. Default: "serving_server".
-
-serving_client (str, optional) - The path of configuration files for client. Default: "serving_client".
-
-model_filename (str, optional) - The name of file to load the inference program. If it is None, the default filename `__model__` will be used. Default: None.
-
-paras_filename (str, optional) - The name of file to load all parameters. It is only used for the case that all parameters were saved in a single binary file. If parameters were saved in separate files, set it as None. Default: None.
+Or you can use a build-in python module called `paddle_serving_client.convert` to convert it.
+```python
+python -m paddle_serving_client.convert --dirname ./your_inference_model_dir
+```
+Arguments are the same as `inference_model_to_serving` API.
+| Argument | Type | Default | Description |
+|--------------|------|-----------|--------------------------------|
+| `dirname` | str | - | Path of saved model files. Program file and parameter files are saved in this directory. |
+| `serving_server` | str | `"serving_server"` | The path of model files and configuration files for server. |
+| `serving_client` | str | `"serving_client"` | The path of configuration files for client. |
+| `model_filename` | str | None | The name of file to load the inference program. If it is None, the default filename `__model__` will be used. |
+| `paras_filename` | str | None | The name of file to load all parameters. It is only used for the case that all parameters were saved in a single binary file. If parameters were saved in separate files, set it as None. |

doc/SAVE_CN.md

@@ -39,12 +39,15 @@ for line in sys.stdin:
 import paddle_serving_client.io as serving_io
 serving_io.inference_model_to_serving(dirname, serving_server="serving_server", serving_client="serving_client",  model_filename=None, params_filename=None)
 ```
-dirname (str) – 需要转换的模型文件存储路径，Program结构文件和参数文件均保存在此目录。
-
-serving_server (str, 可选) - 转换后的模型文件和配置文件的存储路径。默认值为serving_server。
-
-serving_client (str, 可选) - 转换后的客户端配置文件存储路径。默认值为serving_client。
-
-model_filename (str，可选) – 存储需要转换的模型Inference Program结构的文件名称。如果设置为None，则使用 `__model__` 作为默认的文件名。默认值为None。
-
-params_filename (str，可选) – 存储需要转换的模型所有参数的文件名称。当且仅当所有模型参数被保存在一个单独的二进制文件中，它才需要被指定。如果模型参数是存储在各自分离的文件中，设置它的值为None。默认值为None。
+或者你可以使用Paddle Serving提供的名为`paddle_serving_client.convert`的内置模块进行转换。
+```python
+python -m paddle_serving_client.convert --dirname ./your_inference_model_dir
+```
+模块参数与`inference_model_to_serving`接口参数相同。
+| 参数 | 类型 | 默认值 | 描述 |
+|--------------|------|-----------|--------------------------------|
+| `dirname` | str | - | 需要转换的模型文件存储路径，Program结构文件和参数文件均保存在此目录。|
+| `serving_server` | str | `"serving_server"` | 转换后的模型文件和配置文件的存储路径。默认值为serving_server |
+| `serving_client` | str | `"serving_client"` | 转换后的客户端配置文件存储路径。默认值为serving_client |
+| `model_filename` | str | None | 存储需要转换的模型Inference Program结构的文件名称。如果设置为None，则使用 `__model__` 作为默认的文件名 |
+| `paras_filename` | str | None | 存储需要转换的模型所有参数的文件名称。当且仅当所有模型参数被保存在一个单独的二进制文件中，它才需要被指定。如果模型参数是存储在各自分离的文件中，设置它的值为None |

doc/deprecated/NEW_WEB_SERVICE.md

+# How to develop a new Web service?
+
+([简体中文](NEW_WEB_SERVICE_CN.md)|English)
+
+This document will take the image classification service based on the Imagenet data set as an example to introduce how to develop a new web service. The complete code can be visited at [here](../python/examples/imagenet/resnet50_web_service.py).
+
+## WebService base class
+
+Paddle Serving implements the [WebService](https://github.com/PaddlePaddle/Serving/blob/develop/python/paddle_serving_server/web_service.py#L23) base class. You need to override its `preprocess` and `postprocess` method. The default implementation is as follows:
+
+```python
+class WebService(object):
+  
+    def preprocess(self, feed={}, fetch=[]):
+        return feed, fetch
+    def postprocess(self, feed={}, fetch=[], fetch_map=None):
+        return fetch_map
+```
+
+### preprocess
+
+The preprocess method has two input parameters, `feed` and `fetch`. For an HTTP request `request`:
+
+- The value of `feed` is the feed part `request.json["feed"]` in the request data 
+- The value of `fetch` is the fetch part `request.json["fetch"]` in the request data
+
+The return values are the feed and fetch values used in the prediction.
+
+### postprocess
+
+The postprocess method has three input parameters, `feed`, `fetch` and `fetch_map`:
+
+- The value of `feed` is the feed part `request.json["feed"]` in the request data 
+- The value of `fetch` is the fetch part `request.json["fetch"]` in the request data
+- The value of `fetch_map` is the model output value.
+
+The return value will be processed as `{"reslut": fetch_map}` as the return of the HTTP request.
+
+## Develop ImageService class
+
+```python
+class ImageService(WebService):
+
+    def preprocess(self, feed={}, fetch=[]):
+        reader = ImageReader()
+        feed_batch = []
+        for ins in feed:
+            if "image" not in ins:
+                raise ("feed data error!")
+            sample = base64.b64decode(ins["image"])
+            img = reader.process_image(sample)
+            feed_batch.append({"image": img})
+        return feed_batch, fetch
+```
+
+For the above `ImageService`, only the `preprocess` method is rewritten to process the image data in Base64 format into the data format required by prediction.

doc/deprecated/NEW_WEB_SERVICE_CN.md

+# 如何开发一个新的Web Service？
+
+(简体中文|[English](NEW_WEB_SERVICE.md))
+
+本文档将以Imagenet图像分类服务为例，来介绍如何开发一个新的Web Service。您可以在[这里](../python/examples/imagenet/resnet50_web_service.py)查阅完整的代码。
+
+## WebService基类
+
+Paddle Serving实现了[WebService](https://github.com/PaddlePaddle/Serving/blob/develop/python/paddle_serving_server/web_service.py#L23)基类，您需要重写它的`preprocess`方法和`postprocess`方法，默认实现如下：
+
+```python
+class WebService(object):
+  
+    def preprocess(self, feed={}, fetch=[]):
+        return feed, fetch
+    def postprocess(self, feed={}, fetch=[], fetch_map=None):
+        return fetch_map
+```
+
+### preprocess方法
+
+preprocess方法有两个输入参数，`feed`和`fetch`。对于一个HTTP请求`request`：
+
+- `feed`的值为请求数据中的feed部分`request.json["feed"]`
+- `fetch`的值为请求数据中的fetch部分`request.json["fetch"]`
+
+返回值分别是预测过程中用到的feed和fetch值。
+
+### postprocess方法
+
+postprocess方法有三个输入参数，`feed`、`fetch`和`fetch_map`：
+
+- `feed`的值为请求数据中的feed部分`request.json["feed"]`
+- `fetch`的值为请求数据中的fetch部分`request.json["fetch"]`
+- `fetch_map`的值为fetch到的模型输出值
+
+返回值将会被处理成`{"reslut": fetch_map}`作为HTTP请求的返回。
+
+## 开发ImageService类
+
+```python
+class ImageService(WebService):
+
+    def preprocess(self, feed={}, fetch=[]):
+        reader = ImageReader()
+        feed_batch = []
+        for ins in feed:
+            if "image" not in ins:
+                raise ("feed data error!")
+            sample = base64.b64decode(ins["image"])
+            img = reader.process_image(sample)
+            feed_batch.append({"image": img})
+        return feed_batch, fetch
+```
+
+对于上述的`ImageService`，只重写了前处理方法，将base64格式的图片数据处理成模型预测需要的数据格式。

java/src/main/java/io/paddle/serving/client/Client.java

@@ -192,14 +192,16 @@ public class Client {
 
     private InferenceRequest _packInferenceRequest(
             List<HashMap<String, INDArray>> feed_batch,
-            Iterable<String> fetch) throws IllegalArgumentException {
+            Iterable<String> fetch,
+            long log_id) throws IllegalArgumentException {
         List<String> feed_var_names = new ArrayList<String>();
         feed_var_names.addAll(feed_batch.get(0).keySet());
 
         InferenceRequest.Builder req_builder = InferenceRequest.newBuilder()
             .addAllFeedVarNames(feed_var_names)
             .addAllFetchVarNames(fetch)
-            .setIsPython(false);
+            .setIsPython(false)
+            .setLogId(log_id);
         for (HashMap<String, INDArray> feed_data: feed_batch) {
             FeedInst.Builder inst_builder = FeedInst.newBuilder();
             for (String name: feed_var_names) {
@@ -332,76 +334,151 @@ public class Client {
     public Map<String, INDArray> predict(
             HashMap<String, INDArray> feed,
             Iterable<String> fetch) {
-        return predict(feed, fetch, false);
+        return predict(feed, fetch, false, 0);
+    }
+
+    public Map<String, INDArray> predict(
+            HashMap<String, INDArray> feed,
+            Iterable<String> fetch,
+            long log_id) {
+        return predict(feed, fetch, false, log_id);
     }
 
     public Map<String, HashMap<String, INDArray>> ensemble_predict(
             HashMap<String, INDArray> feed,
             Iterable<String> fetch) {
-        return ensemble_predict(feed, fetch, false);
+        return ensemble_predict(feed, fetch, false, 0);
+    }
+
+    public Map<String, HashMap<String, INDArray>> ensemble_predict(
+            HashMap<String, INDArray> feed,
+            Iterable<String> fetch,
+            long log_id) {
+        return ensemble_predict(feed, fetch, false, log_id);
     }
 
     public PredictFuture asyn_predict(
             HashMap<String, INDArray> feed,
             Iterable<String> fetch) {
-        return asyn_predict(feed, fetch, false);
+        return asyn_predict(feed, fetch, false, 0);
+    }
+
+    public PredictFuture asyn_predict(
+            HashMap<String, INDArray> feed,
+            Iterable<String> fetch,
+            long log_id) {
+        return asyn_predict(feed, fetch, false, log_id);
     }
 
     public Map<String, INDArray> predict(
             HashMap<String, INDArray> feed,
             Iterable<String> fetch,
             Boolean need_variant_tag) {
+        return predict(feed, fetch, need_variant_tag, 0);
+    }
+
+    public Map<String, INDArray> predict(
+            HashMap<String, INDArray> feed,
+            Iterable<String> fetch,
+            Boolean need_variant_tag,
+            long log_id) {
         List<HashMap<String, INDArray>> feed_batch
             = new ArrayList<HashMap<String, INDArray>>();
         feed_batch.add(feed);
-        return predict(feed_batch, fetch, need_variant_tag);
+        return predict(feed_batch, fetch, need_variant_tag, log_id);
     }
 
     public Map<String, HashMap<String, INDArray>> ensemble_predict(
             HashMap<String, INDArray> feed,
             Iterable<String> fetch,
             Boolean need_variant_tag) {
+        return ensemble_predict(feed, fetch, need_variant_tag, 0);
+    }
+    
+    public Map<String, HashMap<String, INDArray>> ensemble_predict(
+            HashMap<String, INDArray> feed,
+            Iterable<String> fetch,
+            Boolean need_variant_tag,
+            long log_id) {
         List<HashMap<String, INDArray>> feed_batch
             = new ArrayList<HashMap<String, INDArray>>();
         feed_batch.add(feed);
-        return ensemble_predict(feed_batch, fetch, need_variant_tag);
+        return ensemble_predict(feed_batch, fetch, need_variant_tag, log_id);
     }
 
     public PredictFuture asyn_predict(
             HashMap<String, INDArray> feed,
             Iterable<String> fetch,
             Boolean need_variant_tag) {
+        return asyn_predict(feed, fetch, need_variant_tag, 0);
+    }
+
+    public PredictFuture asyn_predict(
+            HashMap<String, INDArray> feed,
+            Iterable<String> fetch,
+            Boolean need_variant_tag,
+            long log_id) {
         List<HashMap<String, INDArray>> feed_batch
             = new ArrayList<HashMap<String, INDArray>>();
         feed_batch.add(feed);
-        return asyn_predict(feed_batch, fetch, need_variant_tag);
+        return asyn_predict(feed_batch, fetch, need_variant_tag, log_id);
     }
 
     public Map<String, INDArray> predict(
             List<HashMap<String, INDArray>> feed_batch,
             Iterable<String> fetch) {
-        return predict(feed_batch, fetch, false);
+        return predict(feed_batch, fetch, false, 0);
+    }
+
+    public Map<String, INDArray> predict(
+            List<HashMap<String, INDArray>> feed_batch,
+            Iterable<String> fetch,
+            long log_id) {
+        return predict(feed_batch, fetch, false, log_id);
     }
     
     public Map<String, HashMap<String, INDArray>> ensemble_predict(
             List<HashMap<String, INDArray>> feed_batch,
             Iterable<String> fetch) {
-        return ensemble_predict(feed_batch, fetch, false);
+        return ensemble_predict(feed_batch, fetch, false, 0);
+    }
+
+    public Map<String, HashMap<String, INDArray>> ensemble_predict(
+            List<HashMap<String, INDArray>> feed_batch,
+            Iterable<String> fetch,
+            long log_id) {
+        return ensemble_predict(feed_batch, fetch, false, log_id);
     }
 
     public PredictFuture asyn_predict(
             List<HashMap<String, INDArray>> feed_batch,
             Iterable<String> fetch) {
-        return asyn_predict(feed_batch, fetch, false);
+        return asyn_predict(feed_batch, fetch, false, 0);
+    }
+
+    public PredictFuture asyn_predict(
+            List<HashMap<String, INDArray>> feed_batch,
+            Iterable<String> fetch,
+            long log_id) {
+        return asyn_predict(feed_batch, fetch, false, log_id);
     }
 
     public Map<String, INDArray> predict(
             List<HashMap<String, INDArray>> feed_batch,
             Iterable<String> fetch,
             Boolean need_variant_tag) {
+        return predict(feed_batch, fetch, need_variant_tag, 0);        
+    }
+
+    public Map<String, INDArray> predict(
+            List<HashMap<String, INDArray>> feed_batch,
+            Iterable<String> fetch,
+            Boolean need_variant_tag,
+            long log_id) {
         try {
             profiler_.record("java_prepro_0");
-            InferenceRequest req = _packInferenceRequest(feed_batch, fetch);
+            InferenceRequest req = _packInferenceRequest(
+                    feed_batch, fetch, log_id);
             profiler_.record("java_prepro_1");
             
             profiler_.record("java_client_infer_0");
@@ -415,7 +492,7 @@ public class Client {
                 = new ArrayList<Map.Entry<String, HashMap<String, INDArray>>>(
                     ensemble_result.entrySet());
             if (list.size() != 1) {
-                System.out.format("predict failed: please use ensemble_predict impl.\n");
+                System.out.format("Failed to predict: please use ensemble_predict impl.\n");
                 return null;
             }
             profiler_.record("java_postpro_1");
@@ -423,7 +500,7 @@ public class Client {
 
             return list.get(0).getValue();
         } catch (StatusRuntimeException e) {
-            System.out.format("predict failed: %s\n", e.toString());
+            System.out.format("Failed to predict: %s\n", e.toString());
             return null;
         }
     }
@@ -432,9 +509,18 @@ public class Client {
             List<HashMap<String, INDArray>> feed_batch,
             Iterable<String> fetch,
             Boolean need_variant_tag) {
+        return ensemble_predict(feed_batch, fetch, need_variant_tag, 0);        
+    }
+     
+    public Map<String, HashMap<String, INDArray>> ensemble_predict(
+            List<HashMap<String, INDArray>> feed_batch,
+            Iterable<String> fetch,
+            Boolean need_variant_tag,
+            long log_id) {
         try {
             profiler_.record("java_prepro_0");
-            InferenceRequest req = _packInferenceRequest(feed_batch, fetch);
+            InferenceRequest req = _packInferenceRequest(
+                    feed_batch, fetch, log_id);
             profiler_.record("java_prepro_1");
             
             profiler_.record("java_client_infer_0");
@@ -449,7 +535,7 @@ public class Client {
 
             return ensemble_result;
         } catch (StatusRuntimeException e) {
-            System.out.format("predict failed: %s\n", e.toString());
+            System.out.format("Failed to predict: %s\n", e.toString());
             return null;
         }
     }
@@ -458,7 +544,16 @@ public class Client {
             List<HashMap<String, INDArray>> feed_batch,
             Iterable<String> fetch,
             Boolean need_variant_tag) {
-        InferenceRequest req = _packInferenceRequest(feed_batch, fetch);
+        return asyn_predict(feed_batch, fetch, need_variant_tag, 0);
+    }
+
+    public PredictFuture asyn_predict(
+            List<HashMap<String, INDArray>> feed_batch,
+            Iterable<String> fetch,
+            Boolean need_variant_tag,
+            long log_id) {
+        InferenceRequest req = _packInferenceRequest(
+                feed_batch, fetch, log_id);
         ListenableFuture<InferenceResponse> future = futureStub_.inference(req);
         PredictFuture predict_future = new PredictFuture(future, 
             (InferenceResponse resp) -> {

java/src/main/proto/multi_lang_general_model_service.proto

@@ -14,6 +14,8 @@
 
 syntax = "proto2";
 
+package baidu.paddle_serving.multi_lang;
+
 option java_multiple_files = true;
 option java_package = "io.paddle.serving.grpc";
 option java_outer_classname = "ServingProto";
@@ -37,6 +39,7 @@ message InferenceRequest {
   repeated string feed_var_names = 2;
   repeated string fetch_var_names = 3;
   required bool is_python = 4 [ default = false ];
+  required uint64 log_id = 5 [ default = 0 ];
 };
 
 message InferenceResponse {

paddle_inference/inferencer-fluid-cpu/include/fluid_cpu_engine.h

@@ -23,6 +23,7 @@
 #include "core/configure/inferencer_configure.pb.h"
 #include "core/predictor/framework/infer.h"
 #include "paddle_inference_api.h"  // NOLINT
+//#include "predictor/framework/infer.h"
 
 namespace baidu {
 namespace paddle_serving {

python/CMakeLists.txt

 if (CLIENT)
     file(INSTALL pipeline DESTINATION paddle_serving_client)
-    execute_process(COMMAND ${PYTHON_EXECUTABLE} run_codegen.py
-        WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/paddle_serving_client/pipeline/proto)
     file(GLOB_RECURSE SERVING_CLIENT_PY_FILES paddle_serving_client/*.py)
     set(PY_FILES ${SERVING_CLIENT_PY_FILES})
     SET(PACKAGE_NAME "serving_client")
@@ -11,13 +9,9 @@ endif()
 if (SERVER)
     if (NOT WITH_GPU)
         file(INSTALL pipeline DESTINATION paddle_serving_server)
-        execute_process(COMMAND ${PYTHON_EXECUTABLE} run_codegen.py
-            WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/paddle_serving_server/pipeline/proto)
         file(GLOB_RECURSE SERVING_SERVER_PY_FILES paddle_serving_server/*.py)
     else()
         file(INSTALL pipeline DESTINATION paddle_serving_server_gpu)
-        execute_process(COMMAND ${PYTHON_EXECUTABLE} run_codegen.py
-            WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/paddle_serving_server_gpu/pipeline/proto)
         file(GLOB_RECURSE SERVING_SERVER_PY_FILES paddle_serving_server_gpu/*.py)
     endif()
         set(PY_FILES ${SERVING_SERVER_PY_FILES})
@@ -25,6 +19,8 @@ if (SERVER)
         set(SETUP_LOG_FILE "setup.py.server.log")
 endif()
 
+configure_file(${CMAKE_CURRENT_SOURCE_DIR}/util.py
+    ${CMAKE_CURRENT_BINARY_DIR}/util.py)
 if (CLIENT)
 configure_file(${CMAKE_CURRENT_SOURCE_DIR}/setup.py.client.in
     ${CMAKE_CURRENT_BINARY_DIR}/setup.py)
@@ -47,6 +43,9 @@ if (SERVER)
     endif()
 endif()
 
+configure_file(${CMAKE_CURRENT_SOURCE_DIR}/gen_version.py
+    ${CMAKE_CURRENT_BINARY_DIR}/gen_version.py)
+
 set (SERVING_CLIENT_CORE ${PADDLE_SERVING_BINARY_DIR}/core/general-client/*.so)
 message("python env: " ${py_env})
 
@@ -54,6 +53,7 @@ if (APP)
 add_custom_command(
         OUTPUT ${PADDLE_SERVING_BINARY_DIR}/.timestamp
         COMMAND cp -r ${CMAKE_CURRENT_SOURCE_DIR}/paddle_serving_app/ ${PADDLE_SERVING_BINARY_DIR}/python/
+        COMMAND env ${py_env} ${PYTHON_EXECUTABLE} gen_version.py "app"
         COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py bdist_wheel
         DEPENDS ${SERVING_APP_CORE} general_model_config_py_proto ${PY_FILES})
 add_custom_target(paddle_python ALL DEPENDS ${PADDLE_SERVING_BINARY_DIR}/.timestamp)
@@ -65,6 +65,7 @@ add_custom_command(
 	COMMAND cp -r ${CMAKE_CURRENT_SOURCE_DIR}/paddle_serving_client/ ${PADDLE_SERVING_BINARY_DIR}/python/
 	COMMAND ${CMAKE_COMMAND} -E copy ${SERVING_CLIENT_CORE} ${PADDLE_SERVING_BINARY_DIR}/python/paddle_serving_client/serving_client.so
     COMMAND env ${py_env} ${PYTHON_EXECUTABLE} python_tag.py
+    COMMAND env ${py_env} ${PYTHON_EXECUTABLE} gen_version.py "client"
 	COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py bdist_wheel
 	DEPENDS ${SERVING_CLIENT_CORE} sdk_configure_py_proto ${PY_FILES})
 add_custom_target(paddle_python ALL DEPENDS serving_client ${PADDLE_SERVING_BINARY_DIR}/.timestamp)
@@ -75,6 +76,7 @@ if (SERVER)
         add_custom_command(
             OUTPUT ${PADDLE_SERVING_BINARY_DIR}/.timestamp
             COMMAND cp -r ${CMAKE_CURRENT_SOURCE_DIR}/paddle_serving_server/ ${PADDLE_SERVING_BINARY_DIR}/python/
+            COMMAND env ${py_env} ${PYTHON_EXECUTABLE} gen_version.py "server"
             COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py bdist_wheel
             DEPENDS ${SERVING_SERVER_CORE} server_config_py_proto ${PY_FILES})
         add_custom_target(paddle_python ALL DEPENDS ${PADDLE_SERVING_BINARY_DIR}/.timestamp)
@@ -83,7 +85,8 @@ if (SERVER)
             OUTPUT ${PADDLE_SERVING_BINARY_DIR}/.timestamp
             COMMAND cp -r
             ${CMAKE_CURRENT_SOURCE_DIR}/paddle_serving_server_gpu/ ${PADDLE_SERVING_BINARY_DIR}/python/
-            COMMAND env ${py_env} ${PYTHON_EXECUTABLE} paddle_serving_server_gpu/gen_cuda_version.py ${CUDA_VERSION_MAJOR}
+            COMMAND env ${py_env} ${PYTHON_EXECUTABLE} gen_version.py
+            "server_gpu" ${CUDA_VERSION_MAJOR}
             COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py bdist_wheel
             DEPENDS ${SERVING_SERVER_CORE} server_config_py_proto ${PY_FILES})
         add_custom_target(paddle_python ALL DEPENDS ${PADDLE_SERVING_BINARY_DIR}/.timestamp)

python/examples/blazeface/README.md

@@ -3,7 +3,7 @@
 ## Get Model
 ```
 python -m paddle_serving_app.package --get_model blazeface
-tar -xzvf blazeface.tar.gz
+tar -xf blazeface.tar.gz
 ```
 
 ## RPC Service

python/examples/blazeface/test_client.py

@@ -16,6 +16,7 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import *
 import sys
 import numpy as np
+from paddle_serving_app.reader import BlazeFacePostprocess
 
 preprocess = Sequential([
     File2Image(),

python/examples/imagenet/benchmark.py

@@ -90,6 +90,7 @@ def single_func(idx, resource):
                 image = base64.b64encode(
                     open("./image_data/n01440764/" + file_list[i]).read())
             else:
+                image_path = "./image_data/n01440764/" + file_list[i]
                 image = base64.b64encode(open(image_path, "rb").read()).decode(
                     "utf-8")
             req = json.dumps({"feed": [{"image": image}], "fetch": ["score"]})

python/examples/imagenet/resnet50_web_service.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 import sys
 from paddle_serving_client import Client
-from paddle_serving_app.reader import Sequential, URL2Image, Resize, CenterCrop, RGB2BGR, Transpose, Div, Normalize
+from paddle_serving_app.reader import Sequential, URL2Image, Resize, CenterCrop, RGB2BGR, Transpose, Div, Normalize, Base64ToImage
 
 if len(sys.argv) != 4:
     print("python resnet50_web_service.py model device port")

python/examples/ocr/README.md

@@ -21,8 +21,13 @@ tar xf test_imgs.tar
 ### Start Service
 
 ```
+#choose one of cpu/gpu commands as following
+#for cpu user
+python -m paddle_serving_server.serve --model ocr_det_model --port 9293
+python ocr_web_server.py cpu
+#for gpu user
 python -m paddle_serving_server_gpu.serve --model ocr_det_model --port 9293 --gpu_id 0
-python ocr_web_server.py
+python ocr_web_server.py gpu
 ```
 
 ### Client Prediction
@@ -33,7 +38,11 @@ If you want a faster web service, please try Web Debugger Service
 
 ## Web Debugger Service
 ```
-python ocr_debugger_server.py
+#choose one of cpu/gpu commands as following
+#for cpu user
+python ocr_debugger_server.py cpu
+#for gpu user
+python ocr_debugger_server.py gpu 
 ```
 
 ## Web Debugger Client Prediction
@@ -54,15 +63,17 @@ Dataset: RCTW 500 sample images
 | Serving web service          | 8.69         | 13.41                      | 109.97           | 2.82               | 87.76            | 4.29               | 3.98               | 78.51            | 3.66               | 4.12                     | 181.02             | 136.49       | 317.51        |
 | Serving Debugger web service |  8.73        | 16.42                      | 115.27           | 2.93               | 20.63            | 3.97               | 4.48               | 13.84            | 3.60               | 6.91                     | 49.45              | 147.33       | 196.78        |
 
-## Appendix: Det or Rec only
+## Appendix: For Users who want to launch Det or Rec only
 if you are going to detect images not recognize it or directly recognize the words from images. We also provide Det and Rec server for you.
 
 ### Det Server 
 
 ```
-python det_web_server.py 
+python det_web_server.py cpu #for cpu user
+python det_web_server.py gpu #for gpu user
 #or
-python det_debugger_server.py
+python det_debugger_server.py cpu #for cpu user
+python det_debugger_server.py gpu #for gpu user
 ```
 
 ### Det Client
@@ -75,9 +86,11 @@ python ocr_web_client.py
 ### Rec Server
 
 ```
-python rec_web_server.py 
+python rec_web_server.py cpu #for cpu user
+python rec_web_server.py gpu #for gpu user
 #or
-python rec_debugger_server.py
+python rec_debugger_server.py cpu #for cpu user
+python rec_debugger_server.py gpu #for gpu user
 ```
 
 ### Rec Client

python/examples/ocr/README_CN.md

@@ -15,19 +15,18 @@ wget --no-check-certificate https://paddle-serving.bj.bcebos.com/ocr/test_imgs.t
 tar xf test_imgs.tar
 ```
 
-### 客户端预测
-
-```
-python ocr_rpc_client.py
-```
-
 ## Web Service服务
 
 ### 启动服务
 
 ```
+#根据CPU/GPU设备选择一种启动方式
+#for cpu user
+python -m paddle_serving_server.serve --model ocr_det_model --port 9293
+python ocr_web_server.py cpu
+#for gpu user
 python -m paddle_serving_server_gpu.serve --model ocr_det_model --port 9293 --gpu_id 0
-python ocr_web_server.py
+python ocr_web_server.py gpu
 ```
 
 ### 启动客户端
@@ -38,7 +37,11 @@ python ocr_web_client.py
 如果用户需要更快的执行速度，请尝试Debugger版Web服务
 ## 启动Debugger版Web服务
 ```
-python ocr_debugger_server.py
+#根据CPU/GPU设备选择一种启动方式
+#for cpu user
+python ocr_debugger_server.py cpu
+#for gpu user
+python ocr_debugger_server.py gpu
 ```
 
 ## 启动客户端
@@ -66,9 +69,11 @@ GPU: Nvidia Tesla V100单卡
 ### 启动检测服务
 
 ```
-python det_web_server.py 
+python det_web_server.py cpu #for cpu user
+python det_web_server.py gpu #for gpu user
 #or
-python det_debugger_server.py
+python det_debugger_server.py cpu #for cpu user
+python det_debugger_server.py gpu #for gpu user
 ```
 
 ### 检测服务客户端
@@ -81,9 +86,11 @@ python ocr_web_client.py
 ### 启动识别服务
 
 ```
-python rec_web_server.py 
+python rec_web_server.py cpu #for cpu user
+python rec_web_server.py gpu #for gpu user
 #or
-python rec_debugger_server.py
+python rec_debugger_server.py cpu #for cpu user
+python rec_debugger_server.py gpu #for gpu user
 ```
 
 ### 识别服务客户端

python/examples/ocr/det_debugger_server.py

@@ -21,7 +21,10 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import Sequential, ResizeByFactor
 from paddle_serving_app.reader import Div, Normalize, Transpose
 from paddle_serving_app.reader import DBPostProcess, FilterBoxes
-from paddle_serving_server_gpu.web_service import WebService
+if sys.argv[1] == 'gpu':
+    from paddle_serving_server_gpu.web_service import WebService
+elif sys.argv[1] == 'cpu':
+    from paddle_serving_server.web_service import WebService
 import time
 import re
 import base64
@@ -64,8 +67,12 @@ class OCRService(WebService):
 
 ocr_service = OCRService(name="ocr")
 ocr_service.load_model_config("ocr_det_model")
-ocr_service.set_gpus("0")
-ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+if sys.argv[1] == 'gpu':
+    ocr_service.set_gpus("0")
+    ocr_service.prepare_server(
+        workdir="workdir", port=9292, device="gpu", gpuid=0)
+elif sys.argv[1] == 'cpu':
+    ocr_service.prepare_server(workdir="workdir", port=9292)
 ocr_service.init_det()
 ocr_service.run_debugger_service()
 ocr_service.run_web_service()

python/examples/ocr/det_web_server.py

@@ -21,7 +21,10 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import Sequential, ResizeByFactor
 from paddle_serving_app.reader import Div, Normalize, Transpose
 from paddle_serving_app.reader import DBPostProcess, FilterBoxes
-from paddle_serving_server_gpu.web_service import WebService
+if sys.argv[1] == 'gpu':
+    from paddle_serving_server_gpu.web_service import WebService
+elif sys.argv[1] == 'cpu':
+    from paddle_serving_server.web_service import WebService
 import time
 import re
 import base64
@@ -65,8 +68,12 @@ class OCRService(WebService):
 
 ocr_service = OCRService(name="ocr")
 ocr_service.load_model_config("ocr_det_model")
-ocr_service.set_gpus("0")
-ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+if sys.argv[1] == 'gpu':
+    ocr_service.set_gpus("0")
+    ocr_service.prepare_server(
+        workdir="workdir", port=9292, device="gpu", gpuid=0)
+elif sys.argv[1] == 'cpu':
+    ocr_service.prepare_server(workdir="workdir", port=9292, device="cpu")
 ocr_service.init_det()
 ocr_service.run_rpc_service()
 ocr_service.run_web_service()

python/examples/ocr/ocr_debugger_server.py

@@ -22,7 +22,10 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import Sequential, URL2Image, ResizeByFactor
 from paddle_serving_app.reader import Div, Normalize, Transpose
 from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
-from paddle_serving_server_gpu.web_service import WebService
+if sys.argv[1] == 'gpu':
+    from paddle_serving_server_gpu.web_service import WebService
+elif sys.argv[1] == 'cpu':
+    from paddle_serving_server.web_service import WebService
 from paddle_serving_app.local_predict import Debugger
 import time
 import re
@@ -37,8 +40,12 @@ class OCRService(WebService):
                 (2, 0, 1))
         ])
         self.det_client = Debugger()
+        if sys.argv[1] == 'gpu':
             self.det_client.load_model_config(
                 det_model_config, gpu=True, profile=False)
+        elif sys.argv[1] == 'cpu':
+            self.det_client.load_model_config(
+                det_model_config, gpu=False, profile=False)
         self.ocr_reader = OCRReader()
 
     def preprocess(self, feed=[], fetch=[]):
@@ -99,5 +106,8 @@ ocr_service = OCRService(name="ocr")
 ocr_service.load_model_config("ocr_rec_model")
 ocr_service.prepare_server(workdir="workdir", port=9292)
 ocr_service.init_det_debugger(det_model_config="ocr_det_model")
-ocr_service.run_debugger_service(gpu=True)
+if sys.argv[1] == 'gpu':
+    ocr_service.run_debugger_service(gpu=True)
+elif sys.argv[1] == 'cpu':
+    ocr_service.run_debugger_service()
 ocr_service.run_web_service()

python/examples/ocr/ocr_web_server.py

@@ -22,7 +22,10 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import Sequential, URL2Image, ResizeByFactor
 from paddle_serving_app.reader import Div, Normalize, Transpose
 from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
-from paddle_serving_server_gpu.web_service import WebService
+if sys.argv[1] == 'gpu':
+    from paddle_serving_server_gpu.web_service import WebService
+elif sys.argv[1] == 'cpu':
+    from paddle_serving_server.web_service import WebService
 import time
 import re
 import base64
@@ -90,8 +93,12 @@ class OCRService(WebService):
 
 ocr_service = OCRService(name="ocr")
 ocr_service.load_model_config("ocr_rec_model")
-ocr_service.set_gpus("0")
-ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+if sys.argv[1] == 'gpu':
+    ocr_service.set_gpus("0")
+    ocr_service.prepare_server(
+        workdir="workdir", port=9292, device="gpu", gpuid=0)
+elif sys.argv[1] == 'cpu':
+    ocr_service.prepare_server(workdir="workdir", port=9292)
 ocr_service.init_det_client(
     det_port=9293,
     det_client_config="ocr_det_client/serving_client_conf.prototxt")

python/examples/ocr/rec_web_server.py

@@ -22,7 +22,10 @@ from paddle_serving_client import Client
 from paddle_serving_app.reader import Sequential, URL2Image, ResizeByFactor
 from paddle_serving_app.reader import Div, Normalize, Transpose
 from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
-from paddle_serving_server_gpu.web_service import WebService
+if sys.argv[1] == 'gpu':
+    from paddle_serving_server_gpu.web_service import WebService
+elif sys.argv[1] == 'cpu':
+    from paddle_serving_server.web_service import WebService
 import time
 import re
 import base64
@@ -64,8 +67,12 @@ class OCRService(WebService):
 
 ocr_service = OCRService(name="ocr")
 ocr_service.load_model_config("ocr_rec_model")
-ocr_service.set_gpus("0")
 ocr_service.init_rec()
-ocr_service.prepare_server(workdir="workdir", port=9292, device="gpu", gpuid=0)
+if sys.argv[1] == 'gpu':
+    ocr_service.set_gpus("0")
+    ocr_service.prepare_server(
+        workdir="workdir", port=9292, device="gpu", gpuid=0)
+elif sys.argv[1] == 'cpu':
+    ocr_service.prepare_server(workdir="workdir", port=9292, device="cpu")
 ocr_service.run_rpc_service()
 ocr_service.run_web_service()

python/examples/pipeline/imdb_model_ensemble/README_CN.md

+# IMDB model ensemble 样例
+
+## 获取模型
+```
+sh get_data.sh
+```
+
+## 启动服务
+
+```
+python -m paddle_serving_server_gpu.serve --model imdb_cnn_model --port 9292 &> cnn.log &
+python -m paddle_serving_server_gpu.serve --model imdb_bow_model --port 9393 &> bow.log &
+python test_pipeline_server.py &>pipeline.log &
+```
+
+## 启动客户端
+```
+python test_pipeline_client.py
+```
+
+## HTTP 测试
+```
+curl -X POST -k http://localhost:9999/prediction -d '{"key": ["words"], "value": ["i am very sad | 0"]}'
+```

python/examples/pipeline/imdb_model_ensemble/config.yml

-port: 18080
-worker_num: 1
+rpc_port: 18085
+worker_num: 4
 build_dag_each_worker: false
+http_port: 9999
 dag:
-    is_thread_op: true
+    is_thread_op: false
     client_type: brpc
     retry: 1
     use_profile: false
+    tracer:
+        interval_s: 10

python/examples/pipeline/imdb_model_ensemble/test_pipeline_client.py

@@ -11,7 +11,7 @@
 # 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.
-from paddle_serving_client.pipeline import PipelineClient
+from paddle_serving_server.pipeline import PipelineClient
 import numpy as np
 
 client = PipelineClient()
@@ -20,12 +20,16 @@ client.connect(['127.0.0.1:18080'])
 words = 'i am very sad | 0'
 
 futures = []
-for i in range(100):
+for i in range(4):
     futures.append(
         client.predict(
-            feed_dict={"words": words}, fetch=["prediction"], asyn=True))
+            feed_dict={"words": words},
+            fetch=["prediction"],
+            asyn=True,
+            profile=False))
 
 for f in futures:
     res = f.result()
     if res["ecode"] != 0:
         print("predict failed: {}".format(res))
+    print(res)

python/examples/pipeline/imdb_model_ensemble/test_pipeline_server.py

@@ -12,18 +12,21 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 # pylint: disable=doc-string-missing
-
 from paddle_serving_server.pipeline import Op, RequestOp, ResponseOp
 from paddle_serving_server.pipeline import PipelineServer
 from paddle_serving_server.pipeline.proto import pipeline_service_pb2
 from paddle_serving_server.pipeline.channel import ChannelDataEcode
 import numpy as np
-import logging
 from paddle_serving_app.reader import IMDBDataset
-
-logging.basicConfig(level=logging.DEBUG)
+import logging
 
 _LOGGER = logging.getLogger()
+user_handler = logging.StreamHandler()
+user_handler.setLevel(logging.INFO)
+user_handler.setFormatter(
+    logging.Formatter(
+        "%(levelname)s %(asctime)s [%(filename)s:%(lineno)d] %(message)s"))
+_LOGGER.addHandler(user_handler)
 
 
 class ImdbRequestOp(RequestOp):
@@ -76,7 +79,9 @@ bow_op = Op(name="bow",
             client_config="imdb_bow_client_conf/serving_client_conf.prototxt",
             concurrency=1,
             timeout=-1,
-            retry=1)
+            retry=1,
+            batch_size=3,
+            auto_batching_timeout=1000)
 cnn_op = Op(name="cnn",
             input_ops=[read_op],
             server_endpoints=["127.0.0.1:9292"],
@@ -84,13 +89,17 @@ cnn_op = Op(name="cnn",
             client_config="imdb_cnn_client_conf/serving_client_conf.prototxt",
             concurrency=1,
             timeout=-1,
-            retry=1)
+            retry=1,
+            batch_size=1,
+            auto_batching_timeout=None)
 combine_op = CombineOp(
     name="combine",
     input_ops=[bow_op, cnn_op],
-    concurrency=5,
+    concurrency=1,
     timeout=-1,
-    retry=1)
+    retry=1,
+    batch_size=2,
+    auto_batching_timeout=None)
 
 # fetch output of bow_op
 # response_op = ImdbResponseOp(input_ops=[bow_op])

python/examples/pipeline/ocr/README.md

+# OCR Pipeline WebService 
+
+(English|[简体中文](./README_CN.md))
+
+This document will take OCR as an example to show how to use Pipeline WebService to start multi-model tandem services.
+
+## Get Model
+```
+python -m paddle_serving_app.package --get_model ocr_rec
+tar -xzvf ocr_rec.tar.gz
+python -m paddle_serving_app.package --get_model ocr_det
+tar -xzvf ocr_det.tar.gz
+```
+
+## Get Dataset (Optional)
+```
+wget --no-check-certificate https://paddle-serving.bj.bcebos.com/ocr/test_imgs.tar
+tar xf test_imgs.tar
+```
+
+## Start Service
+```
+python web_service.py &>log.txt &
+```
+
+## Test
+```
+python pipeline_http_client.py
+```
+
+
+
+<!--
+## More (PipelineServing)
+
+You can choose one of the following versions to start Service.
+
+### Remote Service Version
+```
+python -m paddle_serving_server_gpu.serve --model ocr_det_model --port 12000 --gpu_id 0 &> det.log &
+python -m paddle_serving_server_gpu.serve --model ocr_rec_model --port 12001 --gpu_id 0 &> rec.log &
+python remote_service_pipeline_server.py &>pipeline.log &
+```
+
+### Local Service Version
+```
+python local_service_pipeline_server.py &>pipeline.log &
+```
+
+### Hybrid Service Version
+```
+python -m paddle_serving_server_gpu.serve --model ocr_rec_model --port 12001 --gpu_id 0 &> rec.log &
+python hybrid_service_pipeline_server.py &>pipeline.log &
+```
+
+## Client Prediction
+
+### RPC
+```
+python pipeline_rpc_client.py
+```
+
+### HTTP
+```
+python pipeline_http_client.py
+```
+-->

python/examples/pipeline/ocr/README_CN.md

+# OCR Pipeline WebService
+
+([English](./README.md)|简体中文)
+
+本文档将以 OCR 为例，介绍如何使用 Pipeline WebService 启动多模型串联的服务。
+
+## 获取模型
+```
+python -m paddle_serving_app.package --get_model ocr_rec
+tar -xzvf ocr_rec.tar.gz
+python -m paddle_serving_app.package --get_model ocr_det
+tar -xzvf ocr_det.tar.gz
+```
+
+## 获取数据集（可选）
+```
+wget --no-check-certificate https://paddle-serving.bj.bcebos.com/ocr/test_imgs.tar
+tar xf test_imgs.tar
+```
+
+## 启动 WebService
+```
+python web_service.py &>log.txt &
+```
+
+## 测试
+```
+python pipeline_http_client.py
+```
+
+<!--
+## 其他 (PipelineServing)
+
+你可以选择下面任意一种版本启动服务。
+
+### 远程服务版本
+```
+python -m paddle_serving_server.serve --model ocr_det_model --port 12000 --gpu_id 0 &> det.log &
+python -m paddle_serving_server.serve --model ocr_rec_model --port 12001 --gpu_id 0 &> rec.log &
+python remote_service_pipeline_server.py &>pipeline.log &
+```
+
+### 本地服务版本
+```
+python local_service_pipeline_server.py &>pipeline.log &
+```
+
+### 混合服务版本
+```
+python -m paddle_serving_server_gpu.serve --model ocr_rec_model --port 12001 --gpu_id 0 &> rec.log &
+python hybrid_service_pipeline_server.py &>pipeline.log &
+```
+
+## 启动客户端
+
+### RPC
+
+```
+python pipeline_rpc_client.py
+```
+
+### HTTP
+
+```
+python pipeline_http_client.py
+```
+-->

python/examples/pipeline/ocr/config.yml

+rpc_port: 18080
+worker_num: 4
+build_dag_each_worker: false
+http_port: 9999
+dag:
+    is_thread_op: false
+    client_type: brpc
+    retry: 1
+    use_profile: false
+op:
+    det:
+        concurrency: 2
+        local_service_conf:
+            model_config: ocr_det_model
+            devices: "0"
+    rec:
+        concurrency: 1
+        timeout: -1
+        retry: 1
+        local_service_conf:
+            model_config: ocr_rec_model
+            devices: "0"

python/examples/pipeline/ocr/hybrid_service_pipeline_server.py

+# 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
+from paddle_serving_server_gpu.pipeline import Op, RequestOp, ResponseOp
+from paddle_serving_server_gpu.pipeline import PipelineServer
+from paddle_serving_server_gpu.pipeline.proto import pipeline_service_pb2
+from paddle_serving_server_gpu.pipeline.channel import ChannelDataEcode
+from paddle_serving_server_gpu.pipeline import LocalRpcServiceHandler
+import numpy as np
+import cv2
+import time
+import base64
+import json
+from paddle_serving_app.reader import OCRReader
+from paddle_serving_app.reader import Sequential, ResizeByFactor
+from paddle_serving_app.reader import Div, Normalize, Transpose
+from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
+import time
+import re
+import base64
+import logging
+
+_LOGGER = logging.getLogger()
+
+
+class DetOp(Op):
+    def init_op(self):
+        self.det_preprocess = Sequential([
+            ResizeByFactor(32, 960), Div(255),
+            Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), Transpose(
+                (2, 0, 1))
+        ])
+        self.filter_func = FilterBoxes(10, 10)
+        self.post_func = DBPostProcess({
+            "thresh": 0.3,
+            "box_thresh": 0.5,
+            "max_candidates": 1000,
+            "unclip_ratio": 1.5,
+            "min_size": 3
+        })
+
+    def preprocess(self, input_dicts):
+        (_, input_dict), = input_dicts.items()
+        data = base64.b64decode(input_dict["image"].encode('utf8'))
+        data = np.fromstring(data, np.uint8)
+        # Note: class variables(self.var) can only be used in process op mode
+        self.im = cv2.imdecode(data, cv2.IMREAD_COLOR)
+        self.ori_h, self.ori_w, _ = self.im.shape
+        det_img = self.det_preprocess(self.im)
+        _, self.new_h, self.new_w = det_img.shape
+        return {"image": det_img}
+
+    def postprocess(self, input_dicts, fetch_dict):
+        det_out = fetch_dict["concat_1.tmp_0"]
+        ratio_list = [
+            float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
+        ]
+        dt_boxes_list = self.post_func(det_out, [ratio_list])
+        dt_boxes = self.filter_func(dt_boxes_list[0], [self.ori_h, self.ori_w])
+        out_dict = {"dt_boxes": dt_boxes, "image": self.im}
+        return out_dict
+
+
+class RecOp(Op):
+    def init_op(self):
+        self.ocr_reader = OCRReader()
+        self.get_rotate_crop_image = GetRotateCropImage()
+        self.sorted_boxes = SortedBoxes()
+
+    def preprocess(self, input_dicts):
+        (_, input_dict), = input_dicts.items()
+        im = input_dict["image"]
+        dt_boxes = input_dict["dt_boxes"]
+        dt_boxes = self.sorted_boxes(dt_boxes)
+        feed_list = []
+        img_list = []
+        max_wh_ratio = 0
+        for i, dtbox in enumerate(dt_boxes):
+            boximg = self.get_rotate_crop_image(im, dt_boxes[i])
+            img_list.append(boximg)
+            h, w = boximg.shape[0:2]
+            wh_ratio = w * 1.0 / h
+            max_wh_ratio = max(max_wh_ratio, wh_ratio)
+        for img in img_list:
+            norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
+            feed = {"image": norm_img}
+            feed_list.append(feed)
+        return feed_list
+
+    def postprocess(self, input_dicts, fetch_dict):
+        rec_res = self.ocr_reader.postprocess(fetch_dict, with_score=True)
+        res_lst = []
+        for res in rec_res:
+            res_lst.append(res[0])
+        res = {"res": str(res_lst)}
+        return res
+
+
+read_op = RequestOp()
+det_op = DetOp(
+    name="det",
+    input_ops=[read_op],
+    local_rpc_service_handler=LocalRpcServiceHandler(
+        model_config="ocr_det_model",
+        workdir="det_workdir",  # defalut: "workdir"
+        thread_num=2,  # defalut: 2
+        devices="0",  # gpu0. defalut: "" (cpu)
+        mem_optim=True,  # defalut: True
+        ir_optim=False,  # defalut: False
+        available_port_generator=None),  # defalut: None
+    concurrency=1)
+rec_op = RecOp(
+    name="rec",
+    input_ops=[det_op],
+    server_endpoints=["127.0.0.1:12001"],
+    fetch_list=["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"],
+    client_config="ocr_rec_client/serving_client_conf.prototxt",
+    concurrency=1)
+response_op = ResponseOp(input_ops=[rec_op])
+
+server = PipelineServer("ocr")
+server.set_response_op(response_op)
+server.prepare_server('config.yml')
+server.run_server()

python/examples/pipeline/ocr/local_service_pipeline_server.py

+# 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
+from paddle_serving_server_gpu.pipeline import Op, RequestOp, ResponseOp
+from paddle_serving_server_gpu.pipeline import PipelineServer
+from paddle_serving_server_gpu.pipeline.proto import pipeline_service_pb2
+from paddle_serving_server_gpu.pipeline.channel import ChannelDataEcode
+from paddle_serving_server_gpu.pipeline import LocalRpcServiceHandler
+import numpy as np
+import cv2
+import time
+import base64
+import json
+from paddle_serving_app.reader import OCRReader
+from paddle_serving_app.reader import Sequential, ResizeByFactor
+from paddle_serving_app.reader import Div, Normalize, Transpose
+from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
+import time
+import re
+import base64
+import logging
+
+_LOGGER = logging.getLogger()
+
+
+class DetOp(Op):
+    def init_op(self):
+        self.det_preprocess = Sequential([
+            ResizeByFactor(32, 960), Div(255),
+            Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), Transpose(
+                (2, 0, 1))
+        ])
+        self.filter_func = FilterBoxes(10, 10)
+        self.post_func = DBPostProcess({
+            "thresh": 0.3,
+            "box_thresh": 0.5,
+            "max_candidates": 1000,
+            "unclip_ratio": 1.5,
+            "min_size": 3
+        })
+
+    def preprocess(self, input_dicts):
+        (_, input_dict), = input_dicts.items()
+        data = base64.b64decode(input_dict["image"].encode('utf8'))
+        data = np.fromstring(data, np.uint8)
+        # Note: class variables(self.var) can only be used in process op mode
+        self.im = cv2.imdecode(data, cv2.IMREAD_COLOR)
+        self.ori_h, self.ori_w, _ = self.im.shape
+        det_img = self.det_preprocess(self.im)
+        _, self.new_h, self.new_w = det_img.shape
+        return {"image": det_img}
+
+    def postprocess(self, input_dicts, fetch_dict):
+        det_out = fetch_dict["concat_1.tmp_0"]
+        ratio_list = [
+            float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
+        ]
+        dt_boxes_list = self.post_func(det_out, [ratio_list])
+        dt_boxes = self.filter_func(dt_boxes_list[0], [self.ori_h, self.ori_w])
+        out_dict = {"dt_boxes": dt_boxes, "image": self.im}
+        return out_dict
+
+
+class RecOp(Op):
+    def init_op(self):
+        self.ocr_reader = OCRReader()
+        self.get_rotate_crop_image = GetRotateCropImage()
+        self.sorted_boxes = SortedBoxes()
+
+    def preprocess(self, input_dicts):
+        (_, input_dict), = input_dicts.items()
+        im = input_dict["image"]
+        dt_boxes = input_dict["dt_boxes"]
+        dt_boxes = self.sorted_boxes(dt_boxes)
+        feed_list = []
+        img_list = []
+        max_wh_ratio = 0
+        for i, dtbox in enumerate(dt_boxes):
+            boximg = self.get_rotate_crop_image(im, dt_boxes[i])
+            img_list.append(boximg)
+            h, w = boximg.shape[0:2]
+            wh_ratio = w * 1.0 / h
+            max_wh_ratio = max(max_wh_ratio, wh_ratio)
+        for img in img_list:
+            norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
+            feed = {"image": norm_img}
+            feed_list.append(feed)
+        return feed_list
+
+    def postprocess(self, input_dicts, fetch_dict):
+        rec_res = self.ocr_reader.postprocess(fetch_dict, with_score=True)
+        res_lst = []
+        for res in rec_res:
+            res_lst.append(res[0])
+        res = {"res": str(res_lst)}
+        return res
+
+
+read_op = RequestOp()
+det_op = DetOp(
+    name="det",
+    input_ops=[read_op],
+    local_rpc_service_handler=LocalRpcServiceHandler(
+        model_config="ocr_det_model",
+        workdir="det_workdir",  # defalut: "workdir"
+        thread_num=2,  # defalut: 2
+        devices="0",  # gpu0. defalut: "" (cpu)
+        mem_optim=True,  # defalut: True
+        ir_optim=False,  # defalut: False
+        available_port_generator=None),  # defalut: None
+    concurrency=1)
+rec_op = RecOp(
+    name="rec",
+    input_ops=[det_op],
+    local_rpc_service_handler=LocalRpcServiceHandler(
+        model_config="ocr_rec_model"),
+    concurrency=1)
+response_op = ResponseOp(input_ops=[rec_op])
+
+server = PipelineServer("ocr")
+server.set_response_op(response_op)
+server.prepare_server('config.yml')
+server.run_server()

python/examples/pipeline/ocr/pipeline_http_client.py

+# 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.
+from paddle_serving_server_gpu.pipeline import PipelineClient
+import numpy as np
+import requests
+import json
+import cv2
+import base64
+import os
+
+
+def cv2_to_base64(image):
+    return base64.b64encode(image).decode('utf8')
+
+
+url = "http://127.0.0.1:9999/ocr/prediction"
+test_img_dir = "imgs/"
+for img_file in os.listdir(test_img_dir):
+    with open(os.path.join(test_img_dir, img_file), 'rb') as file:
+        image_data1 = file.read()
+    image = cv2_to_base64(image_data1)
+
+for i in range(4):
+    data = {"key": ["image"], "value": [image]}
+    r = requests.post(url=url, data=json.dumps(data))
+    print(r.json())

python/examples/pipeline/ocr/pipeline_rpc_client.py

+# 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.
+from paddle_serving_server_gpu.pipeline import PipelineClient
+import numpy as np
+import requests
+import json
+import cv2
+import base64
+import os
+
+client = PipelineClient()
+client.connect(['127.0.0.1:18080'])
+
+
+def cv2_to_base64(image):
+    return base64.b64encode(image).decode('utf8')
+
+
+test_img_dir = "imgs/"
+for img_file in os.listdir(test_img_dir):
+    with open(os.path.join(test_img_dir, img_file), 'rb') as file:
+        image_data = file.read()
+    image = cv2_to_base64(image_data)
+
+for i in range(4):
+    ret = client.predict(feed_dict={"image": image}, fetch=["res"])
+    print(ret)

python/examples/pipeline/ocr/remote_service_pipeline_server.py

+# 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
+from paddle_serving_server_gpu.pipeline import Op, RequestOp, ResponseOp
+from paddle_serving_server_gpu.pipeline import PipelineServer
+from paddle_serving_server_gpu.pipeline.proto import pipeline_service_pb2
+from paddle_serving_server_gpu.pipeline.channel import ChannelDataEcode
+import numpy as np
+import cv2
+import time
+import base64
+import json
+from paddle_serving_app.reader import OCRReader
+from paddle_serving_app.reader import Sequential, ResizeByFactor
+from paddle_serving_app.reader import Div, Normalize, Transpose
+from paddle_serving_app.reader import DBPostProcess, FilterBoxes, GetRotateCropImage, SortedBoxes
+import time
+import re
+import base64
+import logging
+
+_LOGGER = logging.getLogger()
+
+
+class DetOp(Op):
+    def init_op(self):
+        self.det_preprocess = Sequential([
+            ResizeByFactor(32, 960), Div(255),
+            Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]), Transpose(
+                (2, 0, 1))
+        ])
+        self.filter_func = FilterBoxes(10, 10)
+        self.post_func = DBPostProcess({
+            "thresh": 0.3,
+            "box_thresh": 0.5,
+            "max_candidates": 1000,
+            "unclip_ratio": 1.5,
+            "min_size": 3
+        })
+
+    def preprocess(self, input_dicts):
+        (_, input_dict), = input_dicts.items()
+        data = base64.b64decode(input_dict["image"].encode('utf8'))
+        data = np.fromstring(data, np.uint8)
+        # Note: class variables(self.var) can only be used in process op mode
+        self.im = cv2.imdecode(data, cv2.IMREAD_COLOR)
+        self.ori_h, self.ori_w, _ = self.im.shape
+        det_img = self.det_preprocess(self.im)
+        _, self.new_h, self.new_w = det_img.shape
+        return {"image": det_img}
+
+    def postprocess(self, input_dicts, fetch_dict):
+        det_out = fetch_dict["concat_1.tmp_0"]
+        ratio_list = [
+            float(self.new_h) / self.ori_h, float(self.new_w) / self.ori_w
+        ]
+        dt_boxes_list = self.post_func(det_out, [ratio_list])
+        dt_boxes = self.filter_func(dt_boxes_list[0], [self.ori_h, self.ori_w])
+        out_dict = {"dt_boxes": dt_boxes, "image": self.im}
+        return out_dict
+
+
+class RecOp(Op):
+    def init_op(self):
+        self.ocr_reader = OCRReader()
+        self.get_rotate_crop_image = GetRotateCropImage()
+        self.sorted_boxes = SortedBoxes()
+
+    def preprocess(self, input_dicts):
+        (_, input_dict), = input_dicts.items()
+        im = input_dict["image"]
+        dt_boxes = input_dict["dt_boxes"]
+        dt_boxes = self.sorted_boxes(dt_boxes)
+        feed_list = []
+        img_list = []
+        max_wh_ratio = 0
+        for i, dtbox in enumerate(dt_boxes):
+            boximg = self.get_rotate_crop_image(im, dt_boxes[i])
+            img_list.append(boximg)
+            h, w = boximg.shape[0:2]
+            wh_ratio = w * 1.0 / h
+            max_wh_ratio = max(max_wh_ratio, wh_ratio)
+        for img in img_list:
+            norm_img = self.ocr_reader.resize_norm_img(img, max_wh_ratio)
+            feed = {"image": norm_img}
+            feed_list.append(feed)
+        return feed_list
+
+    def postprocess(self, input_dicts, fetch_dict):
+        rec_res = self.ocr_reader.postprocess(fetch_dict, with_score=True)
+        res_lst = []
+        for res in rec_res:
+            res_lst.append(res[0])
+        res = {"res": str(res_lst)}
+        return res
+
+
+read_op = RequestOp()
+det_op = DetOp(
+    name="det",
+    input_ops=[read_op],
+    server_endpoints=["127.0.0.1:12000"],
+    fetch_list=["concat_1.tmp_0"],
+    client_config="ocr_det_client/serving_client_conf.prototxt",
+    concurrency=1)
+rec_op = RecOp(
+    name="rec",
+    input_ops=[det_op],
+    server_endpoints=["127.0.0.1:12001"],
+    fetch_list=["ctc_greedy_decoder_0.tmp_0", "softmax_0.tmp_0"],
+    client_config="ocr_rec_client/serving_client_conf.prototxt",
+    concurrency=1)
+response_op = ResponseOp(input_ops=[rec_op])
+
+server = PipelineServer("ocr")
+server.set_response_op(response_op)
+server.prepare_server('config.yml')
+server.run_server()

python/examples/pipeline/simple_web_service/README.md

+# Simple Pipeline WebService
+
+This document will takes UCI service as an example to introduce how to use Pipeline WebService.
+
+## Get model
+```
+sh get_data.sh
+```
+
+## Start server
+
+```
+python web_service.py &>log.txt &
+```
+
+## Http test
+```
+curl -X POST -k http://localhost:18080/uci/prediction -d '{"key": ["x"], "value": ["0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332"]}'
+```

python/examples/pipeline/simple_web_service/README_CN.md

+# Simple Pipeline WebService
+
+这里以 Uci 服务为例来介绍 Pipeline WebService 的使用。
+
+## 获取模型
+```
+sh get_data.sh
+```
+
+## 启动服务
+
+```
+python web_service.py &>log.txt &
+```
+
+## 测试
+```
+curl -X POST -k http://localhost:18080/uci/prediction -d '{"key": ["x"], "value": ["0.0137, -0.1136, 0.2553, -0.0692, 0.0582, -0.0727, -0.1583, -0.0584, 0.6283, 0.4919, 0.1856, 0.0795, -0.0332"]}'
+```

python/examples/pipeline/simple_web_service/config.yml

+worker_num: 4
+http_port: 18080
+dag:
+    is_thread_op: false
+op:
+    uci:
+        local_service_conf:
+            model_config: uci_housing_model
+            devices: "" # "0,1"

python/examples/pipeline/simple_web_service/get_data.sh

+wget --no-check-certificate https://paddle-serving.bj.bcebos.com/uci_housing.tar.gz
+tar -xzf uci_housing.tar.gz