fix: stamp failed_deployment_id in generic API path for weighted failover

_ageneric_api_call_with_fallbacks_helper (used by /v1/messages,
image_edit, generate_content, and other factory_function endpoints)
did not call _set_failed_deployment_id_on_exception on failures.
This meant the weighted-failover logic in
_maybe_run_weighted_failover always bailed out early (failed_id
was None), causing requests on these paths to skip healthy
same-group siblings and jump straight to cross-group fallbacks.

The fix adds the same two calls (_set_deployment_num_retries and
_set_failed_deployment_id) that _acompletion and _completion
already make in their exception handlers.

litellm/router.py

@@ -4313,6 +4313,7 @@ class Router:
 
         passthrough_on_no_deployment = kwargs.pop("passthrough_on_no_deployment", False)
         function_name = "_ageneric_api_call_with_fallbacks"
+        deployment = None
         try:
             parent_otel_span = _get_parent_otel_span_from_kwargs(kwargs)
             try:
@@ -4392,6 +4393,9 @@ class Router:
             )
             if model is not None:
                 self.fail_calls[model] += 1
+            if deployment is not None:
+                self._set_deployment_num_retries_on_exception(e, deployment)
+                self._set_failed_deployment_id_on_exception(e, deployment)
             raise e
 
     async def _aresponses_with_streaming_fallbacks(

tests/test_litellm/test_router_weighted_failover.py

@@ -769,3 +769,85 @@ async def test_failover_falls_through_to_external_fallback_when_remaining_in_coo
         )
 
     assert response._hidden_params["model_id"] == "fallback"
+
+
+# ---------------------------------------------------------------------------
+# Regression: generic API path (used by /v1/messages, image_edit, etc.)
+# must also stamp failed_deployment_id for weighted failover to work
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_generic_api_path_sets_failed_deployment_id():
+    """_ageneric_api_call_with_fallbacks_helper (used by /v1/messages and
+    other factory_function endpoints) did not call
+    _set_failed_deployment_id_on_exception, so weighted failover always
+    bailed out. Verify the attribute is now stamped."""
+    router = Router(
+        model_list=[
+            {
+                "model_name": "test-model",
+                "litellm_params": {
+                    "model": "gpt-4o",
+                    "api_key": "bad",
+                    "mock_response": Exception("region down"),
+                    "weight": 1,
+                },
+                "model_info": {"id": "dep-A"},
+            },
+        ],
+        routing_strategy="simple-shuffle",
+        num_retries=0,
+        enable_weighted_failover=True,
+    )
+
+    with pytest.raises(Exception) as exc_info:
+        await router._ageneric_api_call_with_fallbacks_helper(
+            model="test-model",
+            original_generic_function=router._acompletion,
+            messages=[{"role": "user", "content": "hi"}],
+        )
+
+    assert getattr(exc_info.value, "failed_deployment_id", None) == "dep-A"
+
+
+@pytest.mark.asyncio
+async def test_weighted_failover_works_on_generic_api_path():
+    """End-to-end: weighted failover via the _ageneric_api_call_with_fallbacks
+    path (used by /v1/messages, image_edit, generate_content, etc.) should
+    re-pick a healthy sibling in the same model group before falling back
+    to a cross-group fallback. Uses anthropic_messages to exercise the
+    exact code path reported in the bug."""
+    router = Router(
+        model_list=[
+            {
+                "model_name": "test-model",
+                "litellm_params": {
+                    "model": "gpt-4o",
+                    "api_key": "bad",
+                    "mock_response": Exception("region-A down"),
+                    "weight": 1,
+                },
+                "model_info": {"id": "A"},
+            },
+            {
+                "model_name": "test-model",
+                "litellm_params": {
+                    "model": "gpt-4o",
+                    "api_key": "good",
+                    "mock_response": "ok from B",
+                    "weight": 0,
+                },
+                "model_info": {"id": "B"},
+            },
+        ],
+        routing_strategy="simple-shuffle",
+        num_retries=0,
+        enable_weighted_failover=True,
+    )
+
+    response = await router.anthropic_messages(
+        model="test-model",
+        messages=[{"role": "user", "content": "hi"}],
+    )
+    assert response._hidden_params["model_id"] == "B"