Fix: _handle_failure method getting called 2 times

litellm/responses/streaming_iterator.py

@@ -170,7 +170,6 @@ class BaseResponsesAPIStreamingIterator:
             return None
         except Exception as e:
             # Ensure failures trigger failure hooks
-            self._handle_failure(e)
             raise
 
     def _handle_logging_completed_response(self):

tests/llm_responses_api_testing/test_base_responses_api_streaming_iterator.py

@@ -246,4 +246,52 @@ class TestBaseResponsesAPIStreamingIterator:
         
         # Test with None chunk
         result = iterator._process_chunk(None)
-        assert result is None
+        assert result is None
+
+    def test_process_chunk_exception_does_not_call_handle_failure(self):
+        """
+        Test that _process_chunk raises exceptions without calling _handle_failure.
+        
+        This ensures _handle_failure is only called once in the outer exception handler
+        (in __next__ or __anext__), preventing duplicate failure logging.
+        
+        Previously, _handle_failure was called both in _process_chunk and in the outer
+        exception handler, causing duplicate logs. This test verifies the fix.
+        """
+        # Mock dependencies
+        mock_response = Mock()
+        mock_response.headers = {}
+        mock_logging_obj = Mock(spec=LiteLLMLoggingObj)
+        mock_logging_obj.model_call_details = {"litellm_params": {}}
+        mock_config = Mock(spec=BaseResponsesAPIConfig)
+        
+        # Set up the mock transform method to raise an exception
+        test_exception = ValueError("Test exception in transform")
+        mock_config.transform_streaming_response.side_effect = test_exception
+        
+        # Create the iterator instance
+        iterator = BaseResponsesAPIStreamingIterator(
+            response=mock_response,
+            model="gpt-4",
+            responses_api_provider_config=mock_config,
+            logging_obj=mock_logging_obj
+        )
+        
+        # Mock _handle_failure to track if it's called
+        with patch.object(iterator, '_handle_failure') as mock_handle_failure:
+            # Prepare valid JSON chunk that will trigger transform_streaming_response
+            test_chunk_data = {
+                "type": "response.output_text.delta",
+                "delta": "Hello"
+            }
+            
+            # _process_chunk should raise the exception without calling _handle_failure
+            with pytest.raises(ValueError) as exc_info:
+                iterator._process_chunk(json.dumps(test_chunk_data))
+            
+            # Verify the exception was raised
+            assert str(exc_info.value) == "Test exception in transform"
+            
+            # Verify _handle_failure was NOT called in _process_chunk
+            # It should only be called by the outer exception handler in __next__/__anext__
+            mock_handle_failure.assert_not_called()