fix(rust): preserve messages provider failures

litellm/rust_bridge/runtime.py

@@ -5,7 +5,7 @@ from dataclasses import dataclass
 from enum import Enum
 from typing import Final, Generic, NoReturn, TypeAlias, TypeVar
 
-from litellm.exceptions import APIError
+from litellm.exceptions import APIError, InternalServerError
 from litellm.rust_bridge.bindings import native_exception_types
 
 NativeT = TypeVar("NativeT")
@@ -142,6 +142,12 @@ def raise_upstream(error: BaseException, context: BridgeErrorContext) -> NoRetur
     message_value: Final = args[1] if len(args) > 1 else str(error)
     status: Final = status_value if isinstance(status_value, int) else 0
     message: Final = message_value if isinstance(message_value, str) else str(message_value)
+    if status == 500:
+        raise InternalServerError(
+            message=f"litellm rust {context.route}: {message}",
+            llm_provider=context.provider,
+            model=context.model,
+        ) from error
     raise APIError(
         status_code=status or 500,
         message=f"litellm rust {context.route}: {message}",

tests/test_litellm/rust_bridge/test_runtime.py

@@ -4,7 +4,7 @@ from types import SimpleNamespace
 
 import pytest
 
-from litellm.exceptions import APIError
+from litellm.exceptions import APIError, InternalServerError
 from litellm.rust_bridge import bindings, runtime
 
 
@@ -64,6 +64,22 @@ def test_invoke_translates_upstream_without_fallback() -> None:
     assert caught.value.status_code == 429
 
 
+def test_invoke_translates_internal_server_error_without_fallback() -> None:
+    def fail() -> object:
+        raise RustUpstreamError(500, "provider unavailable")
+
+    with pytest.raises(InternalServerError, match="provider unavailable") as caught:
+        runtime.invoke(
+            native_call=fail,
+            fallback=lambda: pytest.fail("fallback must not run"),
+            adapt=str,
+            mode=runtime.FallbackMode.PYTHON,
+            context=context(),
+        )
+
+    assert caught.value.status_code == 500
+
+
 @pytest.mark.asyncio
 async def test_ainvoke_handles_native_success() -> None:
     async def native() -> int:

tests/test_litellm_rust/test_integrations.py

@@ -62,14 +62,7 @@ async def test_generic_api_logger_exports_success_over_http(route: Route, provid
     "route",
     (
         OCR_ASYNC,
-        pytest.param(
-            MESSAGES_ROUTE,
-            marks=pytest.mark.xfail(
-                strict=True,
-                raises=pytest.fail.Exception,
-                reason="Native Messages provider failure raises TypeError and falls back to a second Python request",
-            ),
-        ),
+        MESSAGES_ROUTE,
     ),
     ids=route_id,
 )

tests/test_litellm_rust/test_messages_callbacks.py

@@ -1,6 +1,5 @@
 import asyncio
 import copy
-import json
 import threading
 from typing import Final
 
@@ -171,7 +170,6 @@ async def test_messages_logging_drain_waits_for_suspended_callback(messages_serv
 
 
 @pytest.mark.asyncio
-@pytest.mark.xfail(strict=True, reason="accepted native Messages errors are replayed through the Python transport")
 async def test_messages_failure_callbacks_receive_original_provider_error(messages_server: RecordingServer) -> None:
     messages_server.default_response = ResponseSpec(body={"error": {"message": "provider unavailable"}}, status=500)
     messages_server.expected_requests = None