CTR prediction profiling

demo-serving/op/ctr_prediction_op.cpp

@@ -23,6 +23,9 @@
 #include "predictor/framework/kv_manager.h"
 #include "predictor/framework/memory.h"
 
+// Flag where enable profiling mode
+DECLARE_bool(enable_ctr_profiling);
+
 namespace baidu {
 namespace paddle_serving {
 namespace serving {
@@ -46,6 +49,11 @@ const int CTR_PREDICTION_DENSE_SLOT_ID = 26;
 const int CTR_PREDICTION_DENSE_DIM = 13;
 const int CTR_PREDICTION_EMBEDDING_SIZE = 10;
 
+bthread::Mutex CTRPredictionOp::mutex_;
+int64_t CTRPredictionOp::cube_time_us_ = 0;
+int32_t CTRPredictionOp::cube_req_num_ = 0;
+int32_t CTRPredictionOp::cube_req_key_num_ = 0;
+
 void fill_response_with_message(Response *response,
                                 int err_code,
                                 std::string err_msg) {
@@ -135,7 +143,28 @@ int CTRPredictionOp::inference() {
     return 0;
   } else if (kvinfo->sparse_param_service_type ==
              configure::EngineDesc::REMOTE) {
-    int ret = cube->seek(table_name, keys, &values);
+    struct timeval start;
+    struct timeval end;
+
+    int ret;
+
+    if (FLAGS_enable_ctr_profiling) {
+      gettimeofday(&start, NULL);
+      ret = cube->seek(table_name, keys, &values);
+      gettimeofday(&end, NULL);
+      uint64_t usec =
+          end.tv_sec * 1e6 + end.tv_usec - start.tv_sec * 1e6 - start.tv_usec;
+
+      // Statistics
+      mutex_.lock();
+      cube_time_us_ += usec;
+      ++cube_req_num_;
+      cube_req_key_num_ += keys.size();
+      mutex_.unlock();
+    } else {
+      ret = cube->seek(table_name, keys, &values);
+    }
+
     if (ret != 0) {
       fill_response_with_message(res, -1, "Query cube for embeddings error");
       LOG(ERROR) << "Query cube for embeddings error";

demo-serving/op/ctr_prediction_op.h

@@ -55,6 +55,7 @@ static const char* CTR_PREDICTION_MODEL_NAME = "ctr_prediction";
  * and modifications we made
  *
  */
+
 class CTRPredictionOp
     : public baidu::paddle_serving::predictor::OpWithChannel<
           baidu::paddle_serving::predictor::ctr_prediction::Response> {
@@ -64,6 +65,12 @@ class CTRPredictionOp
   DECLARE_OP(CTRPredictionOp);
 
   int inference();
+
+ private:
+  static bthread::Mutex mutex_;
+  static int64_t cube_time_us_;
+  static int32_t cube_req_num_;
+  static int32_t cube_req_key_num_;
 };
 
 }  // namespace serving

predictor/common/constant.h

@@ -40,8 +40,6 @@ DECLARE_int32(reload_interval_s);
 DECLARE_bool(enable_model_toolkit);
 DECLARE_string(enable_protocol_list);
 DECLARE_bool(enable_cube);
-DECLARE_string(cube_config_path);
-DECLARE_string(cube_config_file);
 
 // STATIC Variables
 extern const char* START_OP_NAME;

predictor/src/pdserving.cpp

@@ -69,6 +69,9 @@ static bvar::PassiveStatus<std::string> s_predictor_revision(
 
 DEFINE_bool(V, false, "print version, bool");
 DEFINE_bool(g, false, "user defined gflag path");
+DEFINE_bool(enable_ctr_profiling,
+            false,
+            "Enable profiling in CTR prediction demo");
 DECLARE_string(flagfile);
 
 namespace bthread {