From fbb21ae0c95ee73c3eda30fa92b7e37259b2d203 Mon Sep 17 00:00:00 2001
From: Genmin <joey@joeyroth.com>
Date: Thu, 30 Apr 2026 14:47:50 -0700
Subject: [PATCH] fix: tolerate invalid midstream fallback status

---
 litellm/exceptions.py                         | 24 ++++++++--
 .../test_exception_header_preservation.py     | 45 +++++++++++++++++++
 2 files changed, 66 insertions(+), 3 deletions(-)

diff --git a/litellm/exceptions.py b/litellm/exceptions.py
index 8b005291556..c04e23a02d0 100644
--- a/litellm/exceptions.py
+++ b/litellm/exceptions.py
@@ -941,6 +941,24 @@ class BlockedPiiEntityError(Exception):
 
 
 class MidStreamFallbackError(ServiceUnavailableError):  # type: ignore
+    @staticmethod
+    def _get_status_code(original_exception: Optional[Exception]) -> int:
+        if original_exception is None:
+            return 503
+
+        for original_status in (
+            getattr(original_exception, "status_code", None),
+            getattr(getattr(original_exception, "response", None), "status_code", None),
+        ):
+            if original_status is None:
+                continue
+            try:
+                return int(original_status)
+            except (TypeError, ValueError):
+                continue
+
+        return 503
+
     def __init__(
         self,
         message: str,
@@ -954,8 +972,8 @@ class MidStreamFallbackError(ServiceUnavailableError):  # type: ignore
         generated_content: str = "",
         is_pre_first_chunk: bool = False,
     ):
-        original_status = getattr(original_exception, "status_code", None)
-        self.status_code = int(original_status) if original_status is not None else 503
+        original_status = self._get_status_code(original_exception)
+        self.status_code = original_status
         self.message = f"litellm.MidStreamFallbackError: {message}"
         self.model = model
         self.llm_provider = llm_provider
@@ -997,7 +1015,7 @@ class MidStreamFallbackError(ServiceUnavailableError):  # type: ignore
         )
 
         # Restore the propagated status and original response/request objects
-        self.status_code = int(original_status) if original_status is not None else 503
+        self.status_code = original_status
         self.response = _saved_response
         self.request = _saved_request
         self.message = _saved_message
diff --git a/tests/test_litellm/test_exception_header_preservation.py b/tests/test_litellm/test_exception_header_preservation.py
index 6ea478c633b..6311dcac13b 100644
--- a/tests/test_litellm/test_exception_header_preservation.py
+++ b/tests/test_litellm/test_exception_header_preservation.py
@@ -254,6 +254,51 @@ class TestExceptionAttributes:
         assert midstream_fallback.response.status_code == 503
         assert str(midstream_fallback.response.request.url) == "https://openai.com/v1/"
 
+    def test_midstream_fallback_error_handles_non_numeric_status_code(self):
+        """
+        Custom/OpenAI-mapped exceptions can expose non-HTTP status codes like
+        "litellm_error"; MidStreamFallbackError should still be constructible.
+        """
+
+        class NonNumericStatusError(Exception):
+            status_code = "litellm_error"
+
+        midstream_error = MidStreamFallbackError(
+            message="stream broke",
+            model="gpt-4o-mini",
+            llm_provider="openai",
+            original_exception=NonNumericStatusError("rate limited"),
+        )
+
+        assert midstream_error.status_code == 503
+        assert midstream_error.response.status_code == 503
+
+    def test_midstream_fallback_error_uses_response_status_code_fallback(self):
+        """
+        If exception.status_code is malformed but response.status_code is valid,
+        preserve the underlying HTTP status.
+        """
+
+        class NonNumericStatusError(Exception):
+            status_code = "litellm_error"
+
+            def __init__(self, response: httpx.Response):
+                self.response = response
+
+        response = httpx.Response(
+            status_code=429,
+            request=httpx.Request("POST", "https://api.openai.com/v1/chat/completions"),
+        )
+        midstream_error = MidStreamFallbackError(
+            message="stream broke",
+            model="gpt-4o-mini",
+            llm_provider="openai",
+            original_exception=NonNumericStatusError(response=response),
+        )
+
+        assert midstream_error.status_code == 429
+        assert midstream_error.response.status_code == 429
+
 
 class TestProxyHeaderExtraction:
     """Test that proxy correctly extracts headers from exceptions."""