fix(google_genai): propagate user from kwargs to logging obj in agenerate_content

The Google-native /v1beta/models/{model}:generateContent route silently
drops the `user` field from the spend log. The handler in
google_endpoints/endpoints.py reads `user` from the request body via
add_litellm_data_to_request, but setup_generate_content_call passes
kwargs to litellm_logging_obj.update_from_kwargs without including
`user`. As a result, `model_call_details["user"]` is None and the
spend log row shows an empty `user` field, blocking per-end-user
attribution on this route.

Pass `user=kwargs.get("user")` so the logging object reflects what the
client sent. OpenAI-compat /v1/chat/completions has always done this
via litellm.completion calling logging.update_environment_variables
with the resolved user value.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

litellm/google_genai/main.py

@@ -188,6 +188,7 @@ class GenerateContentHelper:
         litellm_logging_obj.update_from_kwargs(
             kwargs=kwargs,
             model=model,
+            user=kwargs.get("user"),
             optional_params=dict(generate_content_config_dict),
             litellm_params={
                 "litellm_call_id": litellm_call_id,

tests/test_litellm/google_genai/test_google_genai_main.py

@@ -43,3 +43,68 @@ async def test_agenerate_content_stream():
         )
         mock_post.assert_called_once()
         mock_post.call_args.kwargs["stream"] == True
+
+
+def test_setup_generate_content_call_propagates_user_to_logging_obj():
+    """
+    Regression test for Google-native /v1beta/.../:generateContent route
+    dropping the `user` field from the spend log.
+
+    `setup_generate_content_call` must pass `user` from kwargs to
+    `litellm_logging_obj.update_from_kwargs` so that `logging_obj.user` and
+    `model_call_details["user"]` reflect what the client sent (via header or
+    body) instead of defaulting to None / "".
+    """
+    from unittest.mock import MagicMock, patch
+
+    from litellm.google_genai.main import GenerateContentHelper
+    from litellm.litellm_core_utils.litellm_logging import Logging
+
+    # Use a real Logging instance (required by Pydantic model validation),
+    # but stub out update_from_kwargs so we can assert what it was called with.
+    real_logging_obj = Logging(
+        model="gemini-2.5-pro",
+        messages=[{"role": "user", "content": "hi"}],
+        stream=False,
+        call_type="agenerate_content",
+        start_time=__import__("datetime").datetime.now(),
+        litellm_call_id="call-id-123",
+        function_id="",
+    )
+    mock_update = MagicMock()
+    real_logging_obj.update_from_kwargs = mock_update  # type: ignore[method-assign]
+
+    # Mock the provider config so we hit the update_from_kwargs path (not the
+    # adapter early-return path). Use spec= on a real subclass so Pydantic's
+    # is_instance_of check passes.
+    from litellm.llms.gemini.google_genai.transformation import (
+        GoogleGenAIConfig,
+    )
+
+    mock_provider_config = MagicMock(spec=GoogleGenAIConfig())
+    mock_provider_config.map_generate_content_optional_params.return_value = {}
+    mock_provider_config.transform_generate_content_request.return_value = {}
+
+    with patch(
+        "litellm.get_llm_provider",
+        return_value=("gemini-2.5-pro", "gemini", None, None),
+    ), patch(
+        "litellm.utils.ProviderConfigManager.get_provider_google_genai_generate_content_config",
+        return_value=mock_provider_config,
+    ):
+        GenerateContentHelper.setup_generate_content_call(
+            model="gemini-2.5-pro",
+            contents=[{"parts": [{"text": "hi"}], "role": "user"}],
+            config={},
+            custom_llm_provider="gemini",
+            litellm_logging_obj=real_logging_obj,
+            litellm_call_id="call-id-123",
+            user="my-end-user-uuid-456",
+            metadata={"tags": ["scan_id=abc"]},
+        )
+
+    mock_update.assert_called_once()
+    call_kwargs = mock_update.call_args.kwargs
+    assert (
+        call_kwargs.get("user") == "my-end-user-uuid-456"
+    ), f"Expected user to be propagated, got: {call_kwargs.get('user')!r}"