fix(exceptions): keep internal_server_error as the public type of an upstream 500

PR #40243 started carrying the upstream error body on InternalServerError so the Responses response.failed event can report the provider's code and message, and openai's APIError.__init__ took the body's type along with it. The proxy then answered an OpenAI-compatible upstream 500 with type server_error while a 502 and a 503 kept internal_server_error, and the integration contract in test_observed_routing.py went red. Pin the type the way RateLimitError pins throttling_error, keeping the body.

litellm/exceptions.py

@@ -787,6 +787,7 @@ class InternalServerError(openai.InternalServerError):
         super().__init__(
             self.message, response=self.response, body=body
         )  # Call the base class constructor with the parameters it needs
+        self.type = "internal_server_error"
 
     def __str__(self):
         _message = self.message

tests/test_litellm/litellm_core_utils/test_exception_mapping_utils.py

@@ -1438,9 +1438,12 @@ def test_openai_compatible_vendor_400_keeps_body_but_not_headers():
 
 
 @pytest.mark.parametrize(
-    ("status_code", "mapped_class"), [(429, litellm.RateLimitError), (500, litellm.InternalServerError)]
+    ("status_code", "mapped_class", "reported_type"),
+    [(429, litellm.RateLimitError, "throttling_error"), (500, litellm.InternalServerError, "internal_server_error")],
 )
-def test_openai_429_and_500_keep_body(status_code: int, mapped_class: type[openai.APIError]):
+def test_openai_429_and_500_keep_body_but_report_litellm_type(
+    status_code: int, mapped_class: type[openai.APIError], reported_type: str
+):
     with pytest.raises(mapped_class) as exc_info:
         exception_type(
             model="gpt-5.4-mini",
@@ -1458,6 +1461,7 @@ def test_openai_429_and_500_keep_body(status_code: int, mapped_class: type[opena
         "code": str(status_code),
         "message": "upstream cannot complete this response",
     }
+    assert exc_info.value.type == reported_type
 
 
 def test_litellm_proxy_repeated_response_header_keeps_each_value():

tests/test_litellm/proxy/common_utils/test_openai_error_payload.py

@@ -147,6 +147,22 @@ def test_a_status_carried_by_an_exception_drives_the_type_it_reports():
     assert openai_error_type(exc, error_status_code(exc, 400)) == "permission_error"
 
 
+def test_an_upstream_5xx_body_does_not_relabel_the_internal_server_error():
+    """The upstream body rides along on the exception for the Responses ``response.failed``
+    event, but a 500 keeps answering the proxy's own ``internal_server_error`` label."""
+    from litellm.exceptions import InternalServerError
+
+    carried = InternalServerError(
+        message="Controlled provider failure",
+        model="gpt-5.4-mini",
+        llm_provider="openai",
+        body={"message": "Controlled provider failure", "type": "server_error", "code": "500"},
+    )
+
+    assert carried.body == {"message": "Controlled provider failure", "type": "server_error", "code": "500"}
+    assert openai_error_type(carried, error_status_code(carried, 400)) == "internal_server_error"
+
+
 def test_a_stringified_none_type_or_param_is_treated_as_absent():
     from litellm.exceptions import BadRequestError