Build and run agents you can see, understand and trust.
Python agentchatbotlarge-language-modelsllmllm-agentmcpmulti-agentmulti-modalreact-agentrealtime-agentvoice-agentvoice-assistant
fix(model): propagate Responses API stream errors (#3086)
Weng Qiangd283c92fc28f
diff --git a/src/agentscope/model/_openai_response/_model.py b/src/agentscope/model/_openai_response/_model.pyindex 1e8059529..e283e5ba2 100644--- a/src/agentscope/model/_openai_response/_model.py+++ b/src/agentscope/model/_openai_response/_model.py@@ -4,6 +4,7 @@ from collections import OrderedDictfrom datetime import datetimefrom typing import Literal, Any, AsyncGenerator, List, TYPE_CHECKING, Type+import openaifrom pydantic import BaseModel, Fieldfrom ..._utils._common import _generate_id@@ -153,8 +154,6 @@ class OpenAIResponseModel(ChatModelBase):self.formatter = formatter or OpenAIResponseFormatter()self.client_kwargs = client_kwargs or {}- import openai-self.client: openai.AsyncClient = openai.AsyncClient(api_key=self.credential.api_key.get_secret_value(),organization=self.credential.organization,@@ -164,8 +163,6 @@ class OpenAIResponseModel(ChatModelBase):@classmethoddef _get_retryable_exceptions(cls) -> tuple[Type[Exception], ...]:- import openai-return (openai.APIConnectionError,openai.APITimeoutError,@@ -177,8 +174,6 @@ class OpenAIResponseModel(ChatModelBase):def _get_structured_output_fallback_exceptions(cls,) -> tuple[Type[Exception], ...]:- import openai-return (openai.BadRequestError,)async def _call_api(@@ -296,6 +291,19 @@ class OpenAIResponseModel(ChatModelBase):async with response as stream:async for event in stream:event_type = event.type+ # The SDK doesn't raise on these events, so surface them here+ if event_type in ("response.failed", "error"):+ error = (+ event.response.error+ if event_type == "response.failed"+ else event+ )+ raise openai.APIError(+ message=error.message if error else "Response failed",+ request=stream.response.request,+ body=error.model_dump() if error else None,+ )+delta_res = ChatResponse(content=[],is_last=False,diff --git a/tests/model_openai_response_test.py b/tests/model_openai_response_test.pyindex fb36adedd..a1a79226c 100644--- a/tests/model_openai_response_test.py+++ b/tests/model_openai_response_test.py@@ -10,6 +10,7 @@ import unittestfrom unittest import IsolatedAsyncioTestCasefrom unittest.mock import AsyncMock, MagicMock+from openai import APIErrorfrom pydantic import BaseModelfrom utils import AnyString@@ -143,6 +144,7 @@ class _MockAsyncEventStream:self._events = eventsself._index = 0self.exited = False+ self.response = MagicMock()async def __aenter__(self) -> "_MockAsyncEventStream":return self@@ -526,6 +528,39 @@ class TestOpenAIResponseStream(IsolatedAsyncioTestCase):self.mock_client = MagicMock()self.model.client = self.mock_client+ async def test_stream_errors_propagate(self) -> None:+ """Failed stream events raise APIError instead of a partial reply."""+ body = {"code": "server_error", "message": "Generation failed."}+ error = _make_event("error", message="Generation failed.")+ error.model_dump.return_value = body+ for failure in (+ error,+ _make_event("response.failed", response=MagicMock(error=error)),+ ):+ with self.subTest(event_type=failure.type):+ self.mock_client.responses.create = AsyncMock(+ return_value=_MockAsyncEventStream(+ [+ _make_event(+ "response.output_text.delta",+ delta="Partial",+ ),+ failure,+ ],+ ),+ )+ gen = await self.model([])+ await anext(gen)+ with self.assertRaises(APIError) as ctx:+ await anext(gen)+ self.assertDictEqual(+ {+ "message": ctx.exception.message,+ "body": ctx.exception.body,+ },+ {"message": "Generation failed.", "body": body},+ )+async def test_stream_text(self) -> None:"""Stream text yields deltas then final with full content."""completed_resp = MagicMock()