fix(google_genai): make SDK streaming cost-logging import proxy lazily and tolerantly

A plain pip install litellm has no proxy extras (fastapi, backoff), but
litellm/google_genai/streaming_iterator.py imported litellm.proxy at module scope
and again at end of stream, so google_genai streaming crashed on a SDK-only install
with ImportError instead of degrading cost logging. Defer the proxy imports to
_handle_async_streaming_logging and swallow ImportError there.

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/google_genai/streaming_iterator.py

@@ -2,10 +2,8 @@ import asyncio
 from datetime import datetime
 from typing import TYPE_CHECKING, Any, Final
 
+from litellm._logging import verbose_logger
 from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
-from litellm.proxy.pass_through_endpoints.success_handler import (
-    PassThroughEndpointLogging,
-)
 from litellm.types.passthrough_endpoints.pass_through_endpoints import EndpointType
 
 if TYPE_CHECKING:
@@ -15,8 +13,6 @@ if TYPE_CHECKING:
 else:
     BaseGoogleGenAIGenerateContentConfig = Any
 
-GLOBAL_PASS_THROUGH_SUCCESS_HANDLER_OBJ: Final = PassThroughEndpointLogging()
-
 
 def _encode_google_genai_sse_event(event_lines: list[str]) -> bytes:
     return ("\n".join(event_lines) + "\n\n").encode("utf-8")
@@ -78,15 +74,25 @@ class BaseGoogleGenAIGenerateContentStreamingIterator:
         self,
     ):
         """Handle the logging after all chunks have been collected."""
-        from litellm.proxy.pass_through_endpoints.streaming_handler import (
-            PassThroughStreamingHandler,
-        )
+        try:
+            from litellm.proxy.pass_through_endpoints.streaming_handler import (
+                PassThroughStreamingHandler,
+            )
+            from litellm.proxy.pass_through_endpoints.success_handler import (
+                PassThroughEndpointLogging,
+            )
+        except ImportError as e:
+            verbose_logger.debug(
+                "Skipping google_genai streaming cost logging; proxy extras not installed: %s",
+                e,
+            )
+            return
 
         end_time: Final = datetime.now()
         asyncio.create_task(
             PassThroughStreamingHandler._route_streaming_logging_to_handler(
                 litellm_logging_obj=self.litellm_logging_obj,
-                passthrough_success_handler_obj=GLOBAL_PASS_THROUGH_SUCCESS_HANDLER_OBJ,
+                passthrough_success_handler_obj=PassThroughEndpointLogging(),
                 url_route="/v1/generateContent",
                 request_body=self.request_body or {},
                 endpoint_type=EndpointType.VERTEX_AI,
@@ -190,4 +196,4 @@ class AsyncGoogleGenAIGenerateContentStreamingIterator(BaseGoogleGenAIGenerateCo
             return chunk
         except StopAsyncIteration:
             await self._handle_async_streaming_logging()
-            raise StopAsyncIteration
+            raise StopAsyncIteration

tests/test_litellm/google_genai/test_google_genai_streaming_iterator.py

@@ -1,5 +1,6 @@
 import json
-from unittest.mock import MagicMock
+import sys
+from unittest.mock import MagicMock, patch
 
 import pytest
 
@@ -126,3 +127,51 @@ async def test_async_streaming_iterator_forwards_sse_comment_events():
 
     chunk = await iterator.__anext__()
     assert chunk == b": keepalive\n\n"
+
+
+def test_streaming_iterator_import_does_not_pull_in_proxy():
+    """SDK streaming must not import litellm.proxy at module scope.
+
+    A plain `pip install litellm` has no proxy extras (fastapi, backoff), so a
+    module-level proxy import here crashes google_genai streaming (issue #36043).
+    """
+    import importlib
+
+    import litellm.google_genai.streaming_iterator as mod
+
+    src = importlib.util.find_spec(mod.__name__).origin
+    with open(src) as f:
+        top_level = "".join(
+            line for line in f if not line.startswith((" ", "\t"))
+        )
+    assert "litellm.proxy" not in top_level
+
+
+@pytest.mark.asyncio
+async def test_async_stream_completes_when_proxy_logging_unavailable():
+    """End-of-stream cost logging must degrade, not kill the stream (issue #36043)."""
+    mock_response = MagicMock()
+
+    async def _aiter_lines():
+        yield 'data: {"text":"hi"}'
+        yield ""
+
+    mock_response.aiter_lines = _aiter_lines
+
+    iterator = AsyncGoogleGenAIGenerateContentStreamingIterator(
+        response=mock_response,
+        model="gemini-test",
+        logging_obj=MagicMock(spec=LiteLLMLoggingObj),
+        generate_content_provider_config=MagicMock(),
+        litellm_metadata={},
+        custom_llm_provider="gemini",
+    )
+
+    blocked = {
+        "litellm.proxy.pass_through_endpoints.streaming_handler": None,
+        "litellm.proxy.pass_through_endpoints.success_handler": None,
+    }
+    with patch.dict(sys.modules, blocked):
+        chunks = [chunk async for chunk in iterator]
+
+    assert chunks == [b'data: {"text":"hi"}\n\n']