Correct message when is_generating hits 500 code (#2889)

Author: JaySmithWpg

CHANGELOG.md

@@ -5,6 +5,8 @@
 
 ## Bug Fixes:
 * Fixed bug where setting `default_enabled=False` made it so that the entire queue did not start by [@freddyaboulton](https://github.com/freddyaboulton) in [PR 2876](https://github.com/gradio-app/gradio/pull/2876)  
+* Fixed bug where an error raised after yielding iterative output would not be displayed in the browser by 
+[@JaySmithWpg](https://github.com/JaySmithWpg) in [PR 2889](https://github.com/gradio-app/gradio/pull/2889)  
 
 ## Documentation Changes:
 * Added a Guide on using Google Sheets to create a real-time dashboard with Gradio's `DataFrame` and `LinePlot` component, by [@abidlabs](https://github.com/abidlabs) in [PR 2816](https://github.com/gradio-app/gradio/pull/2816) 

gradio/queue.py

@@ -330,12 +330,17 @@ async def process_events(self, events: List[Event], batch: bool) -> None:
                         return
                     response = await self.call_prediction(awake_events, batch)
                 for event in awake_events:
+                    if response.status != 200:
+                        relevant_response = response
+                    else:
+                        relevant_response = old_response
+
                     await self.send_message(
                         event,
                         {
                             "msg": "process_completed",
-                            "output": old_response.json,
-                            "success": old_response.status == 200,
+                            "output": relevant_response.json,
+                            "success": relevant_response.status == 200,
                         },
                     )
             else:

test/test_queue.py

@@ -223,6 +223,38 @@ async def test_process_event_handles_exception_in_call_prediction_request(
         mock_event.disconnect.assert_called_once()
         assert queue.clean_event.call_count >= 1
 
+    @pytest.mark.asyncio
+    async def test_process_event_handles_exception_in_is_generating_request(
+        self, queue: Queue, mock_event: Event
+    ):
+        # We need to return a good response with is_generating=True first,
+        # setting up the function to expect further iterative responses.
+        # Then we provide a 500 response.
+        side_effects = [
+            MagicMock(has_exception=False, status=200, json=dict(is_generating=True)),
+            MagicMock(has_exception=False, status=500, json=dict(error="Foo")),
+        ]
+        mock_event.disconnect = AsyncMock()
+        queue.gather_event_data = AsyncMock(return_value=True)
+        queue.clean_event = AsyncMock()
+        queue.send_message = AsyncMock(return_value=True)
+        queue.call_prediction = AsyncMock(side_effect=side_effects)
+
+        queue.active_jobs = [[mock_event]]
+        await queue.process_events([mock_event], batch=False)
+        queue.send_message.assert_called_with(
+            mock_event,
+            {
+                "msg": "process_completed",
+                "output": {"error": "Foo"},
+                "success": False,
+            },
+        )
+
+        assert queue.call_prediction.call_count == 2
+        mock_event.disconnect.assert_called_once()
+        assert queue.clean_event.call_count >= 1
+
     @pytest.mark.asyncio
     async def test_process_event_handles_error_sending_process_completed_msg(
         self, queue: Queue, mock_event: Event