diff --git a/include/dsn/utils/latency_tracer.h b/include/dsn/utils/latency_tracer.h
index 573b5c10bd..f3b68e090d 100644
--- a/include/dsn/utils/latency_tracer.h
+++ b/include/dsn/utils/latency_tracer.h
@@ -69,10 +69,14 @@ DSN_DECLARE_bool(enable_latency_tracer);
 class latency_tracer
 {
 public:
+    //-is_sub:
+    //  if `is_sub`=true means its points will be dumped by parent tracer and won't be dumped
+    //  repeatedly in destructor
+    //-threshold:
     //  threshold < 0: don't dump any trace points
     //  threshold = 0: dump all trace points
     //  threshold > 0: dump the trace point when time_used > threshold
-    latency_tracer(const std::string &name, uint64_t threshold = 0);
+    latency_tracer(const std::string &name, bool is_sub = false, uint64_t threshold = 0);
 
     ~latency_tracer();
 
diff --git a/src/replica/mutation.cpp b/src/replica/mutation.cpp
index 30f84a1ead..f980b9ab8a 100644
--- a/src/replica/mutation.cpp
+++ b/src/replica/mutation.cpp
@@ -59,8 +59,10 @@ mutation::mutation()
     _appro_data_bytes = sizeof(mutation_header);
     _create_ts_ns = dsn_now_ns();
     _tid = ++s_tid;
-    tracer = std::make_shared<dsn::utils::latency_tracer>(
-        fmt::format("{}[{}]", "mutation", _tid), FLAGS_abnormal_write_trace_latency_threshold);
+    tracer =
+        std::make_shared<dsn::utils::latency_tracer>(fmt::format("{}[{}]", "mutation", _tid),
+                                                     false,
+                                                     FLAGS_abnormal_write_trace_latency_threshold);
 }
 
 mutation_ptr mutation::copy_no_reply(const mutation_ptr &old_mu)
diff --git a/src/utils/latency_tracer.cpp b/src/utils/latency_tracer.cpp
index 1896d5cfd9..09d7c19ea4 100644
--- a/src/utils/latency_tracer.cpp
+++ b/src/utils/latency_tracer.cpp
@@ -25,8 +25,8 @@ namespace utils {
 
 DSN_DEFINE_bool("replication", enable_latency_tracer, false, "whether enable the latency tracer");
 
-latency_tracer::latency_tracer(const std::string &name, uint64_t threshold)
-    : _name(name), _threshold(threshold), _is_sub(false), _start_time(dsn_now_ns())
+latency_tracer::latency_tracer(const std::string &name, bool is_sub, uint64_t threshold)
+    : _name(name), _threshold(threshold), _is_sub(is_sub), _start_time(dsn_now_ns())
 {
 }
 
@@ -56,7 +56,7 @@ void latency_tracer::set_sub_tracer(const std::shared_ptr<latency_tracer> &trace
     if (!FLAGS_enable_latency_tracer) {
         return;
     }
-    tracer->_is_sub = true;
+
     _sub_tracer = tracer;
 }
 
diff --git a/src/utils/test/latency_tracer_test.cpp b/src/utils/test/latency_tracer_test.cpp
index 88f15d730d..2a905b84b0 100644
--- a/src/utils/test/latency_tracer_test.cpp
+++ b/src/utils/test/latency_tracer_test.cpp
@@ -41,12 +41,12 @@ class latency_tracer_test : public testing::Test
 
     void init_trace_points()
     {
-        _tracer1 = std::make_shared<latency_tracer>("name1", true);
+        _tracer1 = std::make_shared<latency_tracer>("name1");
         for (int i = 0; i < _tracer1_stage_count; i++) {
             ADD_CUSTOM_POINT(_tracer1, fmt::format("stage{}", i));
         }
 
-        _tracer2 = std::make_shared<latency_tracer>("name2", true);
+        _tracer2 = std::make_shared<latency_tracer>("name2");
 
         for (int i = 0; i < _tracer2_stage_count; i++) {
             ADD_CUSTOM_POINT(_tracer2, fmt::format("stage{}", i));