From 0022ff8bbcc440965cabfccc002ea71ee6aa9356 Mon Sep 17 00:00:00 2001
From: S0ngRu1 <1922909737@qq.com>
Date: Wed, 13 May 2026 14:19:33 +0800
Subject: [PATCH] fix(vertex_ai): harden streaming error code parsing

---
 .../vertex_and_google_ai_studio_gemini.py     | 12 ++-
 ...test_vertex_and_google_ai_studio_gemini.py | 74 ++++++++++++++++++-
 2 files changed, 79 insertions(+), 7 deletions(-)

diff --git a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
index 8ab31791397..f9899197854 100644
--- a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
+++ b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
@@ -3148,14 +3148,20 @@ class ModelResponseIterator:
         if not isinstance(error_data, dict):
             raise VertexAIError(
                 status_code=500,
-                message=f"VertexAIError: unexpected error format: {error_data}",
+                message=f"Unexpected error format in mid-stream chunk: {error_data}",
             )
-        error_code = int(error_data.get("code", 500))
+        raw_code = error_data.get("code", 500)
+        if raw_code is None:
+            raw_code = 500
+        try:
+            error_code = int(raw_code)
+        except (TypeError, ValueError):
+            error_code = 500
         error_message = error_data.get("message", "Unknown error")
         error_status = error_data.get("status", "UNKNOWN")
         raise VertexAIError(
             status_code=error_code,
-            message=f"VertexAIError: {error_status} - {error_message}",
+            message=f"{error_status} - {error_message}",
         )
 
     def _apply_stream_candidates(
diff --git a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
index c679d08b826..c18e309ae3c 100644
--- a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
+++ b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
@@ -4322,8 +4322,8 @@ def test_chunk_parser_raises_on_429_error_chunk():
         streaming_obj.chunk_parser(error_chunk)
 
     assert exc_info.value.status_code == 429
-    assert "RESOURCE_EXHAUSTED" in str(exc_info.value.message)
-    assert "Resource exhausted" in str(exc_info.value.message)
+    assert "RESOURCE_EXHAUSTED" in exc_info.value.message
+    assert "Resource exhausted" in exc_info.value.message
 
 
 def test_chunk_parser_raises_on_500_error_chunk():
@@ -4356,7 +4356,7 @@ def test_chunk_parser_raises_on_500_error_chunk():
         streaming_obj.chunk_parser(error_chunk)
 
     assert exc_info.value.status_code == 500
-    assert "INTERNAL" in str(exc_info.value.message)
+    assert "INTERNAL" in exc_info.value.message
 
 
 def test_chunk_parser_raises_on_error_chunk_with_minimal_fields():
@@ -4454,7 +4454,7 @@ def test_chunk_parser_raises_on_non_dict_error():
         streaming_obj.chunk_parser(error_chunk)
 
     assert exc_info.value.status_code == 500
-    assert "unexpected error format" in str(exc_info.value.message)
+    assert "Unexpected error format" in exc_info.value.message
 
 
 def test_chunk_parser_raises_on_string_error_code():
@@ -4491,6 +4491,72 @@ def test_chunk_parser_raises_on_string_error_code():
     assert isinstance(exc_info.value.status_code, int)
 
 
+def test_chunk_parser_error_chunk_explicit_null_code_uses_500():
+    """JSON null for code must not call int(None); status defaults to 500."""
+    from unittest.mock import Mock
+
+    from litellm.llms.vertex_ai.common_utils import VertexAIError
+    from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
+        ModelResponseIterator,
+    )
+
+    error_chunk = {
+        "error": {
+            "code": None,
+            "message": "Something went wrong.",
+            "status": "UNKNOWN",
+        }
+    }
+
+    logging_obj = Mock()
+    logging_obj.optional_params = {}
+
+    streaming_obj = ModelResponseIterator(
+        streaming_response=iter([]),
+        sync_stream=True,
+        logging_obj=logging_obj,
+    )
+
+    with pytest.raises(VertexAIError) as exc_info:
+        streaming_obj.chunk_parser(error_chunk)
+
+    assert exc_info.value.status_code == 500
+    assert "Something went wrong" in exc_info.value.message
+
+
+def test_chunk_parser_error_chunk_non_numeric_code_defaults_to_500():
+    """Non-numeric code must not become ValueError -> RuntimeError in __next__."""
+    from unittest.mock import Mock
+
+    from litellm.llms.vertex_ai.common_utils import VertexAIError
+    from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
+        ModelResponseIterator,
+    )
+
+    error_chunk = {
+        "error": {
+            "code": "NOT_A_NUMBER",
+            "message": "Malformed.",
+            "status": "INVALID",
+        }
+    }
+
+    logging_obj = Mock()
+    logging_obj.optional_params = {}
+
+    streaming_obj = ModelResponseIterator(
+        streaming_response=iter([]),
+        sync_stream=True,
+        logging_obj=logging_obj,
+    )
+
+    with pytest.raises(VertexAIError) as exc_info:
+        streaming_obj.chunk_parser(error_chunk)
+
+    assert exc_info.value.status_code == 500
+    assert "Malformed" in exc_info.value.message
+
+
 def test_mid_stream_429_error_raises_during_iteration():
     """
     Simulate a full streaming scenario: normal thinking chunks arrive first,