From 7f0612c8984c58459394acdc33a03bc07b4431cc Mon Sep 17 00:00:00 2001
From: Yingge He <yinggeh@nvidia.com>
Date: Thu, 17 Oct 2024 11:01:21 -0700
Subject: [PATCH] Fix incorrect initialization of shared_ptr

---
 src/infer_response.cc       | 3 ++-
 src/infer_response.h        | 4 ++--
 src/metric_model_reporter.h | 4 +++-
 3 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/src/infer_response.cc b/src/infer_response.cc
index 092338d97..1a3f85175 100644
--- a/src/infer_response.cc
+++ b/src/infer_response.cc
@@ -88,7 +88,8 @@ InferenceResponse::InferenceResponse(
       response_fn_(response_fn), response_userp_(response_userp),
       response_delegator_(delegator),
 #ifdef TRITON_ENABLE_METRICS
-      responses_sent_(responses_sent), infer_start_ns_(infer_start_ns),
+      responses_sent_(std::move(responses_sent)),
+      infer_start_ns_(infer_start_ns),
 #endif  // TRITON_ENABLE_METRICS
       null_response_(false)
 {
diff --git a/src/infer_response.h b/src/infer_response.h
index 281af4740..8d09285ee 100644
--- a/src/infer_response.h
+++ b/src/infer_response.h
@@ -63,7 +63,7 @@ class InferenceResponseFactory {
         is_cancelled_(false)
 #ifdef TRITON_ENABLE_METRICS
         ,
-        responses_sent_(0)
+        responses_sent_(std::make_shared<std::atomic<uint64_t>>(0))
 #endif  // TRITON_ENABLE_METRICS
 #ifdef TRITON_ENABLE_STATS
         ,
@@ -387,7 +387,7 @@ class InferenceResponse {
 
 #ifdef TRITON_ENABLE_METRICS
   // Total number of responses sent created by its response factory.
-  std::shared_ptr<std::atomic<uint64_t>> responses_sent_;
+  const std::shared_ptr<std::atomic<uint64_t>> responses_sent_;
 
   // The start time of associate request in ns.
   const uint64_t infer_start_ns_;
diff --git a/src/metric_model_reporter.h b/src/metric_model_reporter.h
index 5ab9f0201..faeb5f399 100644
--- a/src/metric_model_reporter.h
+++ b/src/metric_model_reporter.h
@@ -57,11 +57,13 @@ struct MetricReporterConfig {
   bool latency_histograms_enabled_ = true;
   // Create and use Summaries for per-model latency related metrics
   bool latency_summaries_enabled_ = false;
+  // Buckets used for any histogram metrics. Each value represents
+  // a bucket boundary.
+  prometheus::Histogram::BucketBoundaries buckets_ = {100, 500, 2000, 5000};
   // Quantiles used for any summary metrics. Each pair of values represents
   // { quantile, error }. For example, {0.90, 0.01} means to compute the
   // 90th percentile with 1% error on either side, so the approximate 90th
   // percentile value will be between the 89th and 91st percentiles.
-  prometheus::Histogram::BucketBoundaries buckets_ = {10, 100, 500, 1000};
   prometheus::Summary::Quantiles quantiles_ = {
       {0.5, 0.05}, {0.9, 0.01}, {0.95, 0.001}, {0.99, 0.001}, {0.999, 0.001}};