feat: Treat HTTP request completion as final response

src/c++/perf_analyzer/client_backend/openai/openai_client.cc

@@ -63,6 +63,7 @@ namespace openai {
 void
 ChatCompletionRequest::SendResponse(bool is_final, bool is_null)
 {
+  final_response_sent_ = is_final;
   response_callback_(new ChatCompletionResult(
       http_code_, std::move(response_buffer_), is_final, is_null, request_id_));
 }
@@ -172,7 +173,11 @@ ChatCompletionClient::AsyncInfer(
     request->timer_.CaptureTimestamp(
         triton::client::RequestTimers::Kind::REQUEST_END);
     UpdateInferStat(request->timer_);
-    if (!request->is_stream_) {
+
+    // Send final response on request completion
+    // if it has not already been sent.
+    // (e.g. in the case of seeing [DONE] in streaming case)
+    if (!request->IsFinalResponseSent()) {
       request->SendResponse(true /* is_final */, false /* is_null */);
     }
   };

src/c++/perf_analyzer/client_backend/openai/openai_client.h

@@ -121,12 +121,14 @@ class ChatCompletionRequest : public HttpRequest {
         request_id_(request_id)
   {
   }
+  bool IsFinalResponseSent() { return final_response_sent_; };
   void SendResponse(bool is_final, bool is_null);
   bool is_stream_{false};
   std::function<void(InferResult*)> response_callback_{nullptr};
   // The timers for infer request.
   triton::client::RequestTimers timer_;
   const std::string request_id_;
+  bool final_response_sent_{false};
 };
 
 class ChatCompletionClient : public HttpClient {