Merge pull request #185 from guru4elephant/add_timeline

add profiling timeline

core/CMakeLists.txt

@@ -26,5 +26,5 @@ endif()
 if (NOT CLIENT_ONLY)
 add_subdirectory(predictor)
 add_subdirectory(general-server)
-add_subdirectory(util)
 endif()
+add_subdirectory(util)

core/general-client/CMakeLists.txt

 if(CLIENT_ONLY)
 add_subdirectory(pybind11)
 pybind11_add_module(serving_client src/general_model.cpp src/pybind_general_model.cpp)
-target_link_libraries(serving_client PRIVATE -Wl,--whole-archive sdk-cpp pybind python -Wl,--no-whole-archive -lpthread -lcrypto -lm -lrt -lssl -ldl -lz)
+target_link_libraries(serving_client PRIVATE -Wl,--whole-archive utils sdk-cpp pybind python -Wl,--no-whole-archive -lpthread -lcrypto -lm -lrt -lssl -ldl -lz)
 endif()

core/general-client/include/general_model.h

@@ -31,6 +31,9 @@
 using baidu::paddle_serving::sdk_cpp::Predictor;
 using baidu::paddle_serving::sdk_cpp::PredictorApi;
 
+DECLARE_bool(profile_client);
+DECLARE_bool(profile_server);
+
 // given some input data, pack into pb, and send request
 namespace baidu {
 namespace paddle_serving {
@@ -45,6 +48,8 @@ class PredictorClient {
   PredictorClient() {}
   ~PredictorClient() {}
 
+  void init_gflags(std::vector<std::string> argv);
+
   int init(const std::string& client_conf);
 
   void set_predictor_conf(const std::string& conf_path,
@@ -87,6 +92,7 @@ class PredictorClient {
   std::map<std::string, std::string> _fetch_name_to_var_name;
   std::vector<std::vector<int>> _shape;
   std::vector<int> _type;
+  std::vector<int64_t> _last_request_ts;
 };
 
 }  // namespace general_model

core/general-client/src/elog

+make: *** No targets specified and no makefile found.  Stop.

core/general-client/src/general_model.cpp

@@ -17,18 +17,42 @@
 #include "core/sdk-cpp/builtin_format.pb.h"
 #include "core/sdk-cpp/include/common.h"
 #include "core/sdk-cpp/include/predictor_sdk.h"
+#include "core/util/include/timer.h"
 
+DEFINE_bool(profile_client, false, "");
+DEFINE_bool(profile_server, false, "");
+
+using baidu::paddle_serving::Timer;
 using baidu::paddle_serving::predictor::general_model::Request;
 using baidu::paddle_serving::predictor::general_model::Response;
 using baidu::paddle_serving::predictor::general_model::Tensor;
 using baidu::paddle_serving::predictor::general_model::FeedInst;
 using baidu::paddle_serving::predictor::general_model::FetchInst;
 
+std::once_flag gflags_init_flag;
+
 namespace baidu {
 namespace paddle_serving {
 namespace general_model {
 using configure::GeneralModelConfig;
 
+void PredictorClient::init_gflags(std::vector<std::string> argv) {
+  std::call_once(gflags_init_flag, [&]() {
+      FLAGS_logtostderr = true;
+      argv.insert(argv.begin(), "dummy");
+      int argc = argv.size();
+      char **arr = new char *[argv.size()];
+      std::string line;
+      for (size_t i = 0; i < argv.size(); i++) {
+        arr[i] = &argv[i][0];
+        line += argv[i];
+        line += ' ';
+      }
+      google::ParseCommandLineFlags(&argc, &arr, true);
+      VLOG(2) << "Init commandline: " << line;
+    });
+}
+
 int PredictorClient::init(const std::string &conf_file) {
   try {
     GeneralModelConfig model_config;
@@ -117,15 +141,20 @@ std::vector<std::vector<float>> PredictorClient::predict(
     return fetch_result;
   }
 
+  Timer timeline;
+  int64_t preprocess_start = timeline.TimeStampUS();
+
   // we save infer_us at fetch_result[fetch_name.size()]
   fetch_result.resize(fetch_name.size() + 1);
 
   _api.thrd_clear();
   _predictor = _api.fetch_predictor("general_model");
+
   VLOG(2) << "fetch general model predictor done.";
   VLOG(2) << "float feed name size: " << float_feed_name.size();
   VLOG(2) << "int feed name size: " << int_feed_name.size();
   VLOG(2) << "fetch name size: " << fetch_name.size();
+
   Request req;
   for (auto & name : fetch_name) {
     req.add_fetch_var_names(name);
@@ -175,16 +204,28 @@ std::vector<std::vector<float>> PredictorClient::predict(
     vec_idx++;
   }
 
-  VLOG(2) << "feed int feed var done.";
+  int64_t preprocess_end = timeline.TimeStampUS();
 
-  // std::map<std::string, std::vector<float> > result;
+  int64_t client_infer_start = timeline.TimeStampUS();
   Response res;
 
+  int64_t client_infer_end = 0;
+  int64_t postprocess_start = 0;
+  int64_t postprocess_end = 0;
+
+  if (FLAGS_profile_client) {
+    if (FLAGS_profile_server) {
+      req.set_profile_server(true);
+    }
+  }
+
   res.Clear();
   if (_predictor->inference(&req, &res) != 0) {
     LOG(ERROR) << "failed call predictor with req: " << req.ShortDebugString();
     exit(-1);
   } else {
+    client_infer_end = timeline.TimeStampUS();
+    postprocess_start = client_infer_end;
     for (auto &name : fetch_name) {
       int idx = _fetch_name_to_idx[name];
       int len = res.insts(0).tensor_array(idx).data_size();
@@ -196,8 +237,29 @@ std::vector<std::vector<float>> PredictorClient::predict(
             *(const float *)res.insts(0).tensor_array(idx).data(i).c_str();
       }
     }
-    fetch_result[fetch_name.size()].resize(1);
-    fetch_result[fetch_name.size()][0] = res.mean_infer_us();
+    postprocess_end = timeline.TimeStampUS();
+  }
+
+  if (FLAGS_profile_client) {
+    std::ostringstream oss;
+    oss << "PROFILE\t"
+        << "prepro_0:" << preprocess_start << " "
+        << "prepro_1:" << preprocess_end << " "
+        << "client_infer_0:" << client_infer_start << " "
+        << "client_infer_1:" << client_infer_end << " ";
+        
+    if (FLAGS_profile_server) {
+      int op_num = res.profile_time_size() / 2;
+      for (int i = 0; i < op_num; ++i) {
+        oss << "op" << i << "_0:" << res.profile_time(i * 2) << " ";
+        oss << "op" << i << "_1:" << res.profile_time(i * 2 + 1) << " ";
+      }
+    }
+    
+    oss << "postpro_0:" << postprocess_start << " ";
+    oss << "postpro_1:" << postprocess_end;
+
+    fprintf(stderr, "%s\n", oss.str().c_str());
   }
 
   return fetch_result;
@@ -308,10 +370,6 @@ std::vector<std::vector<std::vector<float>>> PredictorClient::batch_predict(
         }
       }
     }
-    //last index for infer time
-    fetch_result_batch[batch_size].resize(1);
-    fetch_result_batch[batch_size][0].resize(1);
-    fetch_result_batch[batch_size][0][0] = res.mean_infer_us();
   }
 
   return fetch_result_batch;

core/general-client/src/pybind_general_model.cpp

@@ -31,6 +31,10 @@ PYBIND11_MODULE(serving_client, m) {
        )pddoc";
   py::class_<PredictorClient>(m, "PredictorClient", py::buffer_protocol())
       .def(py::init())
+      .def("init_gflags",
+           [](PredictorClient &self, std::vector<std::string> argv) {
+             self.init_gflags(argv);
+           })
       .def("init",
            [](PredictorClient &self, const std::string &conf) {
              return self.init(conf);

core/general-server/op/general_infer_helper.h

@@ -14,6 +14,7 @@
 
 #pragma once
 
+#include <string.h>
 #include <vector>
 #ifdef BCLOUD
 #ifdef WITH_GPU
@@ -34,8 +35,8 @@ static const char* GENERAL_MODEL_NAME = "general_model";
 
 struct GeneralBlob {
   std::vector<paddle::PaddleTensor> tensor_vector;
-  double infer_time;
-  std::vector<std::string> fetch_name_vector;
+  int64_t time_stamp[20];
+  int p_size = 0;
 
   void Clear() {
     size_t tensor_count = tensor_vector.size();
@@ -60,6 +61,18 @@ struct GeneralBlob {
   std::string ShortDebugString() const { return "Not implemented!"; }
 };
 
+static void AddBlobInfo(GeneralBlob * blob,
+                        int64_t init_value) {
+  blob->time_stamp[blob->p_size] = init_value;
+  blob->p_size++;
+}
+
+static void CopyBlobInfo(const GeneralBlob * src,
+                         GeneralBlob * tgt) {
+  memcpy(&(tgt->time_stamp[0]), &(src->time_stamp[0]),
+         src->p_size * sizeof(int64_t));
+}
+
 }  // namespace serving
 }  // namespace paddle_serving
 }  // namespace baidu

core/general-server/op/general_infer_op.cpp

@@ -52,17 +52,18 @@ int GeneralInferOp::inference() {
   int batch_size = input_blob->GetBatchSize();
 
   VLOG(2) << "infer batch size: " << batch_size;
-  // infer
   Timer timeline;
-  double infer_time = 0.0;
+  int64_t start = timeline.TimeStampUS();
   timeline.Start();
   if (InferManager::instance().infer(GENERAL_MODEL_NAME, in, out, batch_size)) {
     LOG(ERROR) << "Failed do infer in fluid model: " << GENERAL_MODEL_NAME;
     return -1;
   }
-  timeline.Pause();
-  infer_time = timeline.ElapsedUS();
 
+  int64_t end = timeline.TimeStampUS();
+  CopyBlobInfo(input_blob, output_blob);
+  AddBlobInfo(output_blob, start);
+  AddBlobInfo(output_blob, end);
   return 0;
 }
 DEFINE_OP(GeneralInferOp);

core/general-server/op/general_reader_op.cpp

@@ -20,11 +20,13 @@
 #include "core/general-server/op/general_reader_op.h"
 #include "core/predictor/framework/infer.h"
 #include "core/predictor/framework/memory.h"
+#include "core/util/include/timer.h"
 
 namespace baidu {
 namespace paddle_serving {
 namespace serving {
 
+using baidu::paddle_serving::Timer;
 using baidu::paddle_serving::predictor::MempoolWrapper;
 using baidu::paddle_serving::predictor::general_model::Tensor;
 using baidu::paddle_serving::predictor::general_model::Request;
@@ -85,9 +87,10 @@ int GeneralReaderOp::inference() {
     LOG(ERROR) << "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;
-  // read config
 
   VLOG(2) << "start to call load general model_conf op";
   baidu::paddle_serving::predictor::Resource &resource =
@@ -196,6 +199,12 @@ int GeneralReaderOp::inference() {
     }
   }
 
+  timeline.Pause();
+  int64_t end = timeline.TimeStampUS();
+  res->p_size = 0;
+  AddBlobInfo(res, start);
+  AddBlobInfo(res, end);
+
   VLOG(2) << "read data from client success";
   return 0;
 }

core/general-server/op/general_response_op.cpp

@@ -53,6 +53,11 @@ int GeneralResponseOp::inference() {
 
   const Request *req = dynamic_cast<const Request *>(get_request_message());
 
+  Timer timeline;
+  // double response_time = 0.0;
+  // timeline.Start();
+  int64_t start = timeline.TimeStampUS();
+
   VLOG(2) << "start to call load general model_conf op";
   baidu::paddle_serving::predictor::Resource &resource =
       baidu::paddle_serving::predictor::Resource::instance();
@@ -68,11 +73,8 @@ int GeneralResponseOp::inference() {
         model_config->_fetch_alias_name_to_index[req->fetch_var_names(i)];
   }
 
-  // response inst with only fetch_var_names
   Response *res = mutable_data<Response>();
 
-  // res->set_mean_infer_us(infer_time);
-
   for (int i = 0; i < batch_size; ++i) {
     FetchInst *fetch_inst = res->add_insts();
     for (auto & idx : fetch_index) {
@@ -118,6 +120,18 @@ int GeneralResponseOp::inference() {
     }
     var_idx++;
   }
+
+  if (req->profile_server()) {
+    int64_t end = timeline.TimeStampUS();
+    VLOG(2) << "p size for input blob: " << input_blob->p_size;
+    for (int i = 0; i < input_blob->p_size; ++i) {
+      res->add_profile_time(input_blob->time_stamp[i]);
+    }
+    // TODO(guru4elephant): find more elegant way to do this
+    res->add_profile_time(start);
+    res->add_profile_time(end);
+  }
+
   return 0;
 }
 

core/general-server/op/general_text_reader_op.cpp

@@ -19,11 +19,13 @@
 #include "core/general-server/op/general_text_reader_op.h"
 #include "core/predictor/framework/infer.h"
 #include "core/predictor/framework/memory.h"
+#include "core/util/include/timer.h"
 
 namespace baidu {
 namespace paddle_serving {
 namespace serving {
 
+using baidu::paddle_serving::Timer;
 using baidu::paddle_serving::predictor::MempoolWrapper;
 using baidu::paddle_serving::predictor::general_model::Tensor;
 using baidu::paddle_serving::predictor::general_model::Request;
@@ -54,9 +56,11 @@ int GeneralTextReaderOp::inference() {
     return -1;
   }
 
+  Timer timeline;
+  int64_t start = timeline.TimeStampUS();
+
   int var_num = req->insts(0).tensor_array_size();
   VLOG(2) << "var num: " << var_num;
-  // read config
 
   VLOG(2) << "start to call load general model_conf op";
   baidu::paddle_serving::predictor::Resource &resource =
@@ -157,6 +161,10 @@ int GeneralTextReaderOp::inference() {
     }
   }
 
+  int64_t end = timeline.TimeStampUS();
+  AddBlobInfo(res, start);
+  AddBlobInfo(res, end);
+
   VLOG(2) << "read data from client success";
   return 0;
 }

core/general-server/op/general_text_response_op.cpp

@@ -49,10 +49,11 @@ int GeneralTextResponseOp::inference() {
   int batch_size = input_blob->GetBatchSize();
 
   VLOG(2) << "infer batch size: " << batch_size;
-  // infer
-
   const Request *req = dynamic_cast<const Request *>(get_request_message());
 
+  Timer timeline;
+  int64_t start = timeline.TimeStampUS();
+
   VLOG(2) << "start to call load general model_conf op";
   baidu::paddle_serving::predictor::Resource &resource =
       baidu::paddle_serving::predictor::Resource::instance();
@@ -71,8 +72,6 @@ int GeneralTextResponseOp::inference() {
   // response inst with only fetch_var_names
   Response *res = mutable_data<Response>();
 
-  // res->set_mean_infer_us(infer_time);
-
   for (int i = 0; i < batch_size; ++i) {
     FetchInst *fetch_inst = res->add_insts();
     for (auto & idx : fetch_index) {
@@ -118,6 +117,18 @@ int GeneralTextResponseOp::inference() {
     }
     var_idx++;
   }
+  
+  if (req->profile_server()) {
+    int64_t end = timeline.TimeStampUS();
+    
+    for (int i = 0; i < input_blob->p_size; ++i) {
+      res->add_profile_time(input_blob->time_stamp[i]);
+    }
+    // TODO(guru4elephant): find more elegant way to do this
+    res->add_profile_time(start);
+    res->add_profile_time(end);
+  }
+
   return 0;
 }
 DEFINE_OP(GeneralTextResponseOp);

core/general-server/proto/general_model_service.proto

@@ -38,11 +38,12 @@ message FetchInst {
 message Request {
   repeated FeedInst insts = 1;
   repeated string fetch_var_names = 2;
+  optional bool profile_server = 3 [ default = false ];
 };
 
 message Response {
   repeated FetchInst insts = 1;
-  optional float mean_infer_us = 2;
+  repeated int64 profile_time = 2;
 };
 
 service GeneralModelService {

core/predictor/framework/service.cpp

@@ -147,6 +147,7 @@ int InferService::inference(const google::protobuf::Message* request,
   TRACEPRINTF("finish to thread clear");
 
   if (_enable_map_request_to_workflow) {
+    LOG(INFO) << "enable map request == True";
     std::vector<Workflow*>* workflows = _map_request_to_workflow(request);
     if (!workflows || workflows->size() == 0) {
       LOG(ERROR) << "Failed to map request to workflow";
@@ -169,6 +170,7 @@ int InferService::inference(const google::protobuf::Message* request,
       }
     }
   } else {
+    LOG(INFO) << "enable map request == False";
     TRACEPRINTF("start to execute one workflow");
     size_t fsize = _flows.size();
     for (size_t fi = 0; fi < fsize; ++fi) {
@@ -233,6 +235,7 @@ int InferService::_execute_workflow(Workflow* workflow,
   TRACEPRINTF("finish to copy from");
 
   workflow_time.stop();
+  LOG(INFO) << "workflow total time: " << workflow_time.u_elapsed();
   PredictorMetric::GetInstance()->update_latency_metric(
       WORKFLOW_METRIC_PREFIX + dv->full_name(), workflow_time.u_elapsed());
 

core/sdk-cpp/proto/general_model_service.proto

@@ -38,11 +38,12 @@ message FetchInst {
 message Request {
   repeated FeedInst insts = 1;
   repeated string fetch_var_names = 2;
+  optional bool profile_server = 3 [ default = false ];
 };
 
 message Response {
   repeated FetchInst insts = 1;
-  optional float mean_infer_us = 2;
+  repeated int64 profile_time = 2;
 };
 
 service GeneralModelService {

core/util/CMakeLists.txt

 include(src/CMakeLists.txt)
 add_library(utils ${util_srcs})
+

core/util/include/timer.h

@@ -38,6 +38,7 @@ class Timer {
   double ElapsedMS();
   // return elapsed time in sec
   double ElapsedSec();
+  int64_t TimeStampUS();
 
  private:
   struct timeval _start;

core/util/src/CMakeLists.txt

 FILE(GLOB srcs ${CMAKE_CURRENT_LIST_DIR}/*.cc)
 LIST(APPEND util_srcs ${srcs})
+

core/util/src/timer.cc

@@ -54,6 +54,11 @@ double Timer::ElapsedMS() { return _elapsed / 1000.0; }
 
 double Timer::ElapsedSec() { return _elapsed / 1000000.0; }
 
+int64_t Timer::TimeStampUS() {
+  gettimeofday(&_now, NULL);
+  return _now.tv_usec;
+}
+
 int64_t Timer::Tickus() {
   gettimeofday(&_now, NULL);
   return (_now.tv_sec - _start.tv_sec) * 1000 * 1000L +

python/paddle_serving_client/__init__.py

@@ -17,6 +17,7 @@ from .proto import sdk_configure_pb2 as sdk
 from .proto import general_model_config_pb2 as m_config
 import google.protobuf.text_format
 import time
+import sys
 
 int_type = 0
 float_type = 1
@@ -87,6 +88,9 @@ class Client(object):
         # map feed names to index
         self.client_handle_ = PredictorClient()
         self.client_handle_.init(path)
+        read_env_flags = ["profile_client", "profile_server"]
+        self.client_handle_.init_gflags([sys.argv[0]] +
+                                        ["--tryfromenv=" + ",".join(read_env_flags)])
         self.feed_names_ = [var.alias_name for var in model_conf.feed_var]
         self.fetch_names_ = [var.alias_name for var in model_conf.fetch_var]
         self.feed_shapes_ = [var.shape for var in model_conf.feed_var]
@@ -143,9 +147,6 @@ class Client(object):
         for i, name in enumerate(fetch_names):
             result_map[name] = result[i]
 
-        if profile:
-            result_map["infer_time"] = result[-1][0]
-
         return result_map
 
     def batch_predict(self, feed_batch=[], fetch=[], profile=False):