Handle mid-stream API errors in llama.cpp conversation (#184056)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

homeassistant/components/llama_cpp/entity.py

@@ -396,15 +396,17 @@ class LlamaCppBaseLLMEntity(Entity):
                     cast(ChatCompletion, result).choices[0].message
                 )
 
-            messages.extend(
-                [
-                    msg
-                    async for content in chat_log.async_add_delta_content_stream(
-                        self.entity_id, async_generator
-                    )
-                    if (msg := convert_message(content))
-                ]
-            )
+            # Streamed responses raise API errors while being consumed
+            with api_error_handler():
+                messages.extend(
+                    [
+                        msg
+                        async for content in chat_log.async_add_delta_content_stream(
+                            self.entity_id, async_generator
+                        )
+                        if (msg := convert_message(content))
+                    ]
+                )
 
             if not chat_log.unresponded_tool_results:
                 break

homeassistant/components/llama_cpp/quality_scale.yaml

@@ -34,9 +34,7 @@ rules:
   unique-config-entry: done
 
   # Silver
-  action-exceptions:
-    status: exempt
-    comment: No service actions are registered by this integration.
+  action-exceptions: done
   config-entry-unloading: done
   docs-configuration-parameters: done
   docs-installation-parameters: done

tests/components/llama_cpp/test_conversation.py

@@ -600,6 +600,73 @@ async def test_streaming_response_redundant_role(
     assert content[1].content == "Hello world"
 
 
+@pytest.mark.parametrize(("config_entry_options"), [{CONF_STREAMING: True}])
+@pytest.mark.parametrize(
+    ("error", "expected_speech"),
+    [
+        pytest.param(
+            openai.APIConnectionError(request=httpx.Request(method="POST", url="test")),
+            "Cannot connect to the server: Connection error.",
+            id="connection_error",
+        ),
+        pytest.param(
+            openai.APITimeoutError(request=httpx.Request(method="POST", url="test")),
+            "Connection timed out: Request timed out.",
+            id="timeout",
+        ),
+        pytest.param(
+            openai.APIError(
+                message="An error occurred during streaming",
+                request=httpx.Request(method="POST", url="test"),
+                body=None,
+            ),
+            "API error: An error occurred during streaming",
+            id="api_error",
+        ),
+    ],
+)
+async def test_streaming_response_error(
+    hass: HomeAssistant,
+    mock_chat_log: MockChatLog,
+    mock_config_entry: MockConfigEntry,
+    error: openai.OpenAIError,
+    expected_speech: str,
+) -> None:
+    """Test an API error raised while consuming the stream."""
+
+    async def mock_stream() -> AsyncGenerator[ChatCompletionChunk]:
+        yield ChatCompletionChunk.model_construct(
+            id="chatcmpl-1234567890ABCDEFGHIJKLMNOPQRS",
+            choices=[
+                ChunkChoice.model_construct(
+                    index=0,
+                    delta=ChoiceDelta(role="assistant", content="Hello"),
+                    finish_reason=None,
+                )
+            ],
+            created=1700000000,
+            model="gpt-3.5-turbo-0613",
+            object="chat.completion.chunk",
+        )
+        raise error
+
+    with patch(
+        "openai.resources.chat.completions.AsyncCompletions.create",
+        new_callable=AsyncMock,
+        return_value=mock_stream(),
+    ):
+        result = await conversation.async_converse(
+            hass,
+            "hello",
+            mock_chat_log.conversation_id,
+            Context(),
+            agent_id="conversation.llama_cpp_conversation",
+        )
+
+    assert result.response.response_type == intent.IntentResponseType.ERROR
+    assert result.response.speech["plain"]["speech"] == expected_speech
+
+
 @pytest.mark.parametrize(
     ("config_entry_options"), [{CONF_LLM_HASS_API: ["non-existing"]}]
 )