From ec015dc7af4e8e0c38504a30c47efc037050e1a9 Mon Sep 17 00:00:00 2001
From: Matthew Kotila <matthew.r.kotila@gmail.com>
Date: Fri, 9 Aug 2024 14:05:16 -0700
Subject: [PATCH] Add functionality to continue benchmarking in Triton C API
 mode if server logging support is disabled

---
 src/client_backend/triton_c_api/triton_loader.cc | 11 +++++++----
 src/client_backend/triton_c_api/triton_loader.h  |  5 +++--
 2 files changed, 10 insertions(+), 6 deletions(-)

diff --git a/src/client_backend/triton_c_api/triton_loader.cc b/src/client_backend/triton_c_api/triton_loader.cc
index 0c561cc2..27e1eda6 100644
--- a/src/client_backend/triton_c_api/triton_loader.cc
+++ b/src/client_backend/triton_c_api/triton_loader.cc
@@ -242,7 +242,8 @@ TritonLoader::StartTriton()
   // Check API version.
   uint32_t api_version_major, api_version_minor;
   REPORT_TRITONSERVER_ERROR(
-      api_version_fn_(&api_version_major, &api_version_minor));
+      api_version_fn_(&api_version_major, &api_version_minor),
+      "unable to get api version");
   if ((TRITONSERVER_API_VERSION_MAJOR != api_version_major) ||
       (TRITONSERVER_API_VERSION_MINOR > api_version_minor)) {
     std::stringstream sstream;
@@ -264,10 +265,10 @@ TritonLoader::StartTriton()
   RETURN_IF_TRITONSERVER_ERROR(
       set_cuda_memory_pool_byte_size_(server_options, 0, 1073741824),
       "setting cuda memory pool byte size failed.");
-  RETURN_IF_TRITONSERVER_ERROR(
+  REPORT_TRITONSERVER_ERROR(
       set_log_verbose_fn_(server_options, verbose_level_),
       "setting verbose logging level");
-  RETURN_IF_TRITONSERVER_ERROR(
+  REPORT_TRITONSERVER_ERROR(
       set_log_info_fn_(server_options, verbose_),
       "setting if log verbose level is true");
   RETURN_IF_TRITONSERVER_ERROR(
@@ -1050,7 +1051,9 @@ TritonLoader::InferResponseCompleteAsync(
     TRITONSERVER_InferenceResponse* response, const uint32_t flags,
     AsyncRequestInfo* async_request_info)
 {
-  REPORT_TRITONSERVER_ERROR(inference_response_error_fn_(response));
+  REPORT_TRITONSERVER_ERROR(
+      inference_response_error_fn_(response),
+      "unable to get inference response error");
 
   if (async_request_info->enable_stats) {
     tc::RequestTimers timer{*async_request_info->timer};
diff --git a/src/client_backend/triton_c_api/triton_loader.h b/src/client_backend/triton_c_api/triton_loader.h
index 993e610f..4ea840f1 100644
--- a/src/client_backend/triton_c_api/triton_loader.h
+++ b/src/client_backend/triton_c_api/triton_loader.h
@@ -68,11 +68,12 @@
     }                                                                       \
   } while (false)
 
-#define REPORT_TRITONSERVER_ERROR(E)                                      \
+#define REPORT_TRITONSERVER_ERROR(E, MSG)                                 \
   do {                                                                    \
     TRITONSERVER_Error* err__ = (E);                                      \
     if (err__ != nullptr) {                                               \
-      std::cout << GetSingleton()->error_message_fn_(err__) << std::endl; \
+      std::cerr << "error: " << (MSG) << ": "                             \
+                << GetSingleton()->error_message_fn_(err__) << std::endl; \
       GetSingleton()->error_delete_fn_(err__);                            \
     }                                                                     \
   } while (false)