Clean up error recording

instrumentation/opentelemetry-instrumentation-genai-langchain/tests/requirements.oldest.txt

@@ -20,6 +20,12 @@
 # pyproject floors by UV_RESOLUTION=lowest-direct on the oldest tox factor, so they are NOT pinned
 # here to avoid drift between the declared bound and the tested version.
 
+# Exception: this change depends on the fully-qualified error.type behavior added in
+# opentelemetry-util-genai 1.1b0, which is not on PyPI yet. Install it from the workspace so the
+# oldest env exercises the required behavior. Remove this once 1.1b0 is released and bump the
+# pyproject floor to >= 1.1b0.
+-e ./util/opentelemetry-util-genai
+
 langchain-openai==0.2.0
 langchain-aws==0.2.2
 langchain-google-genai==2.0.0

instrumentation/opentelemetry-instrumentation-genai-langchain/tests/test_llm_call.py

@@ -329,7 +329,9 @@ def assert_openai_completion_attributes_with_error(
     assert span is not None
     assert span.name == "chat gpt-3.5-turbo"
     attributes = span.attributes
-    assert attributes[error_attributes.ERROR_TYPE] == "AuthenticationError"
+    assert (
+        attributes[error_attributes.ERROR_TYPE] == "openai.AuthenticationError"
+    )
     assert attributes[gen_ai_attributes.GEN_AI_OPERATION_NAME] == "chat"
     assert (
         attributes[gen_ai_attributes.GEN_AI_REQUEST_MODEL] == "gpt-3.5-turbo"
@@ -490,7 +492,9 @@ def assert_duration_metric_when_error(metric, parent_span):
 
 def assert_duration_metric_attributes_when_error(attributes, parent_span):
     assert len(attributes) == 4
-    assert attributes[error_attributes.ERROR_TYPE] == "AuthenticationError"
+    assert (
+        attributes[error_attributes.ERROR_TYPE] == "openai.AuthenticationError"
+    )
     assert attributes.get(gen_ai_attributes.GEN_AI_PROVIDER_NAME) == "openai"
     assert (
         attributes.get(gen_ai_attributes.GEN_AI_OPERATION_NAME)

instrumentation/opentelemetry-instrumentation-genai-openai/.changelog/282.fixed

@@ -0,0 +1 @@
+Responses that fail now record the actual error type (`error` event code or the response's error code) instead of a generic `RuntimeError`, for both streaming and non-streaming calls. Responses that finish as incomplete (`max_output_tokens`/`content_filter`) are recorded as a finish reason rather than an error.

instrumentation/opentelemetry-instrumentation-genai-openai/src/opentelemetry/instrumentation/genai/openai/patch.py

@@ -17,9 +17,6 @@
     EmbeddingInvocation,
     InferenceInvocation,
 )
-from opentelemetry.util.genai.types import (
-    Error,
-)
 
 from .chat_wrappers import AsyncChatStreamWrapper, ChatStreamWrapper
 from .utils import (
@@ -89,7 +86,7 @@ def traced_method(wrapped, instance, args, kwargs):
             chat_invocation.stop()
             return result
         except Exception as error:
-            chat_invocation.fail(Error(type=type(error), message=str(error)))
+            chat_invocation.fail(error)
             raise
 
     return traced_method
@@ -125,7 +122,7 @@ async def traced_method(wrapped, instance, args, kwargs):
             return result
 
         except Exception as error:
-            chat_invocation.fail(Error(type=type(error), message=str(error)))
+            chat_invocation.fail(error)
             raise
 
     return traced_method
@@ -140,7 +137,7 @@ def traced_method(wrapped, instance, args, kwargs):
         try:
             result = wrapped(*args, **kwargs)
         except Exception as error:
-            invocation.fail(Error(type=type(error), message=str(error)))
+            invocation.fail(error)
             raise
 
         _safe_set_embeddings_response_properties(invocation, result)
@@ -159,7 +156,7 @@ async def traced_method(wrapped, instance, args, kwargs):
         try:
             result = await wrapped(*args, **kwargs)
         except Exception as error:
-            invocation.fail(Error(type=type(error), message=str(error)))
+            invocation.fail(error)
             raise
 
         _safe_set_embeddings_response_properties(invocation, result)

instrumentation/opentelemetry-instrumentation-genai-openai/src/opentelemetry/instrumentation/genai/openai/patch_responses.py

@@ -11,6 +11,7 @@
     apply_request_attributes,
     extract_params,
     get_inference_creation_kwargs,
+    get_response_error,
     set_invocation_response_attributes,
 )
 from .response_wrappers import (
@@ -115,12 +116,15 @@ def traced_method(
                     capture_content,
                 )
 
+            response = cast("ResponseResult", parsed_result)
             set_invocation_response_attributes(
-                invocation,
-                cast("ResponseResult", parsed_result),
-                capture_content,
+                invocation, response, capture_content
             )
-            invocation.stop()
+            error = get_response_error(response)
+            if error is not None:
+                invocation.fail(error)
+            else:
+                invocation.stop()
             return result
         except Exception as error:
             invocation.fail(error)
@@ -194,12 +198,15 @@ async def traced_method(
                     capture_content,
                 )
 
+            response = cast("ResponseResult", parsed_result)
             set_invocation_response_attributes(
-                invocation,
-                cast("ResponseResult", parsed_result),
-                capture_content,
+                invocation, response, capture_content
             )
-            invocation.stop()
+            error = get_response_error(response)
+            if error is not None:
+                invocation.fail(error)
+            else:
+                invocation.stop()
             return result
         except Exception as error:
             invocation.fail(error)

instrumentation/opentelemetry-instrumentation-genai-openai/src/opentelemetry/instrumentation/genai/openai/response_extractors.py

@@ -25,6 +25,7 @@
     from openai.types.responses.response_usage import ResponseUsage
 
     from opentelemetry.util.genai.types import (
+        Error,
         InputMessage,
         OutputMessage,
         Text,
@@ -57,6 +58,7 @@
 
 try:
     from opentelemetry.util.genai.types import (
+        Error,
         InputMessage,
         OutputMessage,
         Reasoning,
@@ -66,6 +68,7 @@
         ToolCallRequest as ToolCall,
     )
 except ImportError:
+    Error = None
     InputMessage = None
     OutputMessage = None
     Reasoning = None
@@ -365,6 +368,21 @@ def extract_finish_reasons(response: "Response | None") -> list[str]:
     return list(dict.fromkeys(finish_reasons))
 
 
+def get_response_error(response: "Response | None") -> "Error | None":
+    """Return an ``Error`` when the response failed, else ``None``.
+
+    A failed response carries a ``ResponseError`` (``code`` + ``message``).
+    Incomplete responses (``incomplete_details``) are *not* errors — they
+    surface as a finish reason instead.
+    """
+    if Response is None or Error is None or not isinstance(response, Response):
+        return None
+    error = response.error
+    if error is None or not getattr(error, "code", None):
+        return None
+    return Error(type=error.code, message=getattr(error, "message", None))
+
+
 def get_inference_creation_kwargs(
     params: ResponseRequestParams,
     client_instance: object,

instrumentation/opentelemetry-instrumentation-genai-openai/src/opentelemetry/instrumentation/genai/openai/response_wrappers.py

@@ -17,9 +17,11 @@
 
 try:
     from opentelemetry.instrumentation.genai.openai.response_extractors import (  # pylint: disable=no-name-in-module
+        get_response_error,
         set_invocation_response_attributes,
     )
 except ImportError:
+    get_response_error = None
     set_invocation_response_attributes = None
 
 if TYPE_CHECKING:
@@ -140,10 +142,10 @@ def _stop(
         self._self_invocation.stop()
         self._self_response_telemetry_finalized = True
 
-    def _fail(self, message: str, error_type: type[BaseException]) -> None:
+    def _fail(self, error: Error) -> None:
         if self._self_response_telemetry_finalized:
             return
-        self._self_invocation.fail(Error(message=message, type=error_type))
+        self._self_invocation.fail(error)
         self._self_response_telemetry_finalized = True
 
     def _process_chunk(
@@ -155,7 +157,10 @@ def _on_stream_end(self) -> None:
         self._stop(None)
 
     def _on_stream_error(self, error: BaseException) -> None:
-        self._fail(str(error), type(error))
+        if self._self_response_telemetry_finalized:
+            return
+        self._self_invocation.fail(error)
+        self._self_response_telemetry_finalized = True
 
     def get_final_response(self) -> "ParsedResponse[TextFormatT]":
         self.until_done()
@@ -188,23 +193,26 @@ def process_event(self, event: "ResponseStreamEvent[TextFormatT]") -> None:
             if model:
                 self._self_invocation.request_model = model
 
-        if event_type == "response.completed":
+        if event_type in {"response.completed", "response.incomplete"}:
             self._stop(response)
             return
 
-        if event_type in {"response.failed", "response.incomplete"}:
+        if event_type == "response.failed":
             _set_response_attributes(
                 self._self_invocation,
                 response,
                 self._self_capture_content,
             )
-            self._fail(event_type, RuntimeError)
+            error = (
+                get_response_error(response) if get_response_error else None
+            )
+            self._fail(error or Error(type=event_type, message=None))
             return
 
-        if event_type == "response.error":
-            error_type = getattr(event, "code", None) or "response.error"
-            message = getattr(event, "message", None) or error_type
-            self._fail(message, RuntimeError)
+        if event_type == "error":
+            error_type = getattr(event, "code", None) or "error"
+            message = getattr(event, "message", None)
+            self._fail(Error(type=error_type, message=message))
 
 
 class ResponseStreamWrapper(

instrumentation/opentelemetry-instrumentation-genai-openai/tests/requirements.oldest.txt

@@ -21,6 +21,12 @@
 # OpenTelemetry SDK and test utilities come transitively from opentelemetry-test-util-genai, which
 # every oldest env installs. Pin here only test-only deps that nothing else already provides.
 
+# Exception: this change depends on the fully-qualified error.type behavior added in
+# opentelemetry-util-genai 1.1b0, which is not on PyPI yet. Install it from the workspace so the
+# oldest env exercises the required behavior. Remove this once 1.1b0 is released and bump the
+# pyproject floor to >= 1.1b0.
+-e ./util/opentelemetry-util-genai
+
 numpy==2.3.1 ; python_version >= "3.11"
 numpy==2.1.0 ; python_version == "3.10"
 pydantic==2.12.5

instrumentation/opentelemetry-instrumentation-genai-openai/tests/test_async_chat_completions.py

@@ -161,7 +161,8 @@ async def test_async_chat_completion_bad_endpoint(
     )
     assert 4242 == spans[0].attributes[ServerAttributes.SERVER_PORT]
     assert (
-        "APIConnectionError" == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
+        "openai.APIConnectionError"
+        == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
     )
 
 
@@ -184,7 +185,10 @@ async def test_async_chat_completion_404(
     assert_all_attributes(
         spans[0], llm_model_value, latest_experimental_enabled
     )
-    assert "NotFoundError" == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
+    assert (
+        "openai.NotFoundError"
+        == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
+    )
 
 
 @pytest.mark.asyncio()

instrumentation/opentelemetry-instrumentation-genai-openai/tests/test_async_embeddings.py

@@ -160,7 +160,10 @@ async def test_async_embeddings_error_handling(
         latest_experimental_enabled,
         operation_name="embeddings",
     )
-    assert "NotFoundError" == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
+    assert (
+        "openai.NotFoundError"
+        == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
+    )
 
 
 @pytest.mark.asyncio

instrumentation/opentelemetry-instrumentation-genai-openai/tests/test_async_responses.py

@@ -357,7 +357,10 @@ async def test_async_responses_create_connection_error(
     )
     assert span.attributes[ServerAttributes.SERVER_ADDRESS] == "localhost"
     assert span.attributes[ServerAttributes.SERVER_PORT] == 4242
-    assert span.attributes[ErrorAttributes.ERROR_TYPE] == "APIConnectionError"
+    assert (
+        span.attributes[ErrorAttributes.ERROR_TYPE]
+        == "openai.APIConnectionError"
+    )
 
 
 @pytest.mark.asyncio()
@@ -381,7 +384,7 @@ async def test_async_responses_create_api_error(
     )
     assert (
         span.attributes[ErrorAttributes.ERROR_TYPE]
-        == type(exc_info.value).__name__
+        == f"openai.{type(exc_info.value).__name__}"
     )
 
 
@@ -437,7 +440,10 @@ async def test_async_responses_stream_connection_error(
     assert (
         span.attributes[GenAIAttributes.GEN_AI_REQUEST_MODEL] == DEFAULT_MODEL
     )
-    assert span.attributes[ErrorAttributes.ERROR_TYPE] == "APIConnectionError"
+    assert (
+        span.attributes[ErrorAttributes.ERROR_TYPE]
+        == "openai.APIConnectionError"
+    )
 
 
 @pytest.mark.asyncio()
@@ -655,7 +661,10 @@ async def test_async_responses_create_streaming_connection_error(
     assert (
         span.attributes[GenAIAttributes.GEN_AI_REQUEST_MODEL] == DEFAULT_MODEL
     )
-    assert span.attributes[ErrorAttributes.ERROR_TYPE] == "APIConnectionError"
+    assert (
+        span.attributes[ErrorAttributes.ERROR_TYPE]
+        == "openai.APIConnectionError"
+    )
 
 
 @pytest.mark.asyncio()

instrumentation/opentelemetry-instrumentation-genai-openai/tests/test_async_structured_outputs.py

@@ -187,4 +187,7 @@ async def test_async_structured_output_404(
     assert_all_attributes(
         spans[0], llm_model_value, latest_experimental_enabled
     )
-    assert "NotFoundError" == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
+    assert (
+        "openai.NotFoundError"
+        == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
+    )

instrumentation/opentelemetry-instrumentation-genai-openai/tests/test_chat_completions.py

@@ -210,7 +210,8 @@ def test_chat_completion_bad_endpoint(
     )
     assert 4242 == spans[0].attributes[ServerAttributes.SERVER_PORT]
     assert (
-        "APIConnectionError" == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
+        "openai.APIConnectionError"
+        == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
     )
 
     metrics = metric_reader.get_metrics_data().resource_metrics
@@ -231,7 +232,7 @@ def test_chat_completion_bad_endpoint(
         duration_metric.data.data_points[0].attributes[
             ErrorAttributes.ERROR_TYPE
         ]
-        == "APIConnectionError"
+        == "openai.APIConnectionError"
     )
 
 
@@ -253,7 +254,10 @@ def test_chat_completion_404(
     assert_all_attributes(
         spans[0], llm_model_value, latest_experimental_enabled
     )
-    assert "NotFoundError" == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
+    assert (
+        "openai.NotFoundError"
+        == spans[0].attributes[ErrorAttributes.ERROR_TYPE]
+    )
 
     metrics = metric_reader.get_metrics_data().resource_metrics
     assert len(metrics) == 1
@@ -273,7 +277,7 @@ def test_chat_completion_404(
         duration_metric.data.data_points[0].attributes[
             ErrorAttributes.ERROR_TYPE
         ]
-        == "NotFoundError"
+        == "openai.NotFoundError"
     )