From b0305d0a31a4eb4dd1ff6204ef2e1e5f84db07c4 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Mon, 21 Sep 2026 14:48:42 -0700
Subject: [PATCH] test(router): call both mid-stream fallback attempt functions
 directly

The router coverage gate wants every router.py function reached by name from
a router test. The two per-endpoint attempt functions were only reached through
their callers, so each now has a direct test proving the per-request controls
carrier never reaches the provider call and every hop's stream comes back wrapped.
---
 ...st_router_aresponses_streaming_fallback.py | 38 ++++++++++++++++
 tests/test_litellm/test_router.py             | 44 +++++++++++++++++++
 2 files changed, 82 insertions(+)

diff --git a/tests/router_unit_tests/test_router_aresponses_streaming_fallback.py b/tests/router_unit_tests/test_router_aresponses_streaming_fallback.py
index 18b10e9c8e5..5370089eef5 100644
--- a/tests/router_unit_tests/test_router_aresponses_streaming_fallback.py
+++ b/tests/router_unit_tests/test_router_aresponses_streaming_fallback.py
@@ -423,6 +423,44 @@ async def test_aresponses_per_request_fallbacks_survive_into_hop_streams():
     assert collected == [completed_event]
 
 
+@pytest.mark.asyncio
+async def test_aresponses_attempt_strips_the_controls_carrier_and_wraps_every_hop_stream():
+    """Each attempt of the chain, not only the primary's, comes back wrapped for mid-stream
+    failover, and the per-request controls carrier rides into the wrapper's re-entry kwargs
+    without ever reaching the provider call."""
+    from types import MappingProxyType
+
+    from litellm.router_utils.fallback_event_handlers import (
+        MID_STREAM_FALLBACK_CONTROLS_KEY,
+        MidStreamFallbackControls,
+    )
+
+    router = _make_three_tier_router()
+    completed_event = _make_completed_event(1, 1, 2)
+    hop_stream = _scripted_responses_stream([completed_event])
+    seen: dict = {}
+
+    async def fake_original(**kwargs):
+        seen.update(kwargs)
+        return hop_stream
+
+    controls = MidStreamFallbackControls(MappingProxyType({"fallbacks": [{"primary": ["fb1", "fb2"]}]}))
+    stream = await router._ageneric_api_call_with_fallbacks_responses_attempt(
+        model="fb1",
+        original_generic_function=fake_original,
+        stream=True,
+        input="hi",
+        **{MID_STREAM_FALLBACK_CONTROLS_KEY: controls},
+    )
+    collected = [event async for event in stream]
+
+    assert seen["model"] == "openai/fb1-model"
+    assert MID_STREAM_FALLBACK_CONTROLS_KEY not in seen
+    assert "fallbacks" not in seen
+    assert stream is not hop_stream
+    assert collected == [completed_event]
+
+
 @pytest.mark.asyncio
 async def test_aresponses_fallback_on_in_stream_error_event():
     """A retriable in-stream error event (429) must trigger the router's mid-stream
diff --git a/tests/test_litellm/test_router.py b/tests/test_litellm/test_router.py
index 1ff04a1722f..9d2e37b4fb5 100644
--- a/tests/test_litellm/test_router.py
+++ b/tests/test_litellm/test_router.py
@@ -14222,6 +14222,50 @@ async def test_anthropic_messages_hop_stream_failure_reaches_second_fallback_ent
     assert b"overloaded_error" not in body
 
 
+@pytest.mark.asyncio
+async def test_anthropic_messages_attempt_strips_the_controls_carrier_and_wraps_every_hop_stream():
+    """Each attempt of the chain, not only the primary's, comes back wrapped for mid-stream
+    failover, and the per-request controls carrier never reaches the provider call."""
+    from types import MappingProxyType
+
+    from litellm.router_utils.fallback_event_handlers import (
+        MID_STREAM_FALLBACK_CONTROLS_KEY,
+        MidStreamFallbackControls,
+    )
+
+    router = Router(
+        model_list=[
+            {"model_name": "fb1", "litellm_params": {"model": "anthropic/fb1-model", "api_key": "sk-test"}},
+        ],
+        num_retries=0,
+    )
+    hop_stream = _AnthropicMessagesFakeByteStream(
+        [_anthropic_messages_message_start_chunk(), _anthropic_messages_content_chunk("from fb1")]
+    )
+    seen: dict = {}
+
+    async def fake_original(**kwargs):
+        seen.update(kwargs)
+        return hop_stream
+
+    controls = MidStreamFallbackControls(MappingProxyType({"fallbacks": [{"primary": ["fb1", "fb2"]}]}))
+    stream = await router._ageneric_api_call_with_fallbacks_anthropic_messages_attempt(
+        model="fb1",
+        original_generic_function=fake_original,
+        stream=True,
+        messages=[{"role": "user", "content": "hi"}],
+        max_tokens=10,
+        **{MID_STREAM_FALLBACK_CONTROLS_KEY: controls},
+    )
+    body = b"".join([chunk async for chunk in stream])
+
+    assert seen["model"] == "anthropic/fb1-model"
+    assert MID_STREAM_FALLBACK_CONTROLS_KEY not in seen
+    assert "fallbacks" not in seen
+    assert stream is not hop_stream
+    assert b"from fb1" in body
+
+
 @pytest.mark.asyncio
 async def test_anthropic_messages_fallback_triggers_after_lifecycle_only_frame():
     """Regression: Anthropic routinely sends a message_start lifecycle frame