From 967c4f336743c6e641e599774e08f4e5ab05621e Mon Sep 17 00:00:00 2001
From: yucheng <yucheng@berri.ai>
Date: Wed, 28 Oct 2026 11:10:58 -0700
Subject: [PATCH] test(guardrails): pin the deferred-log flag through
 rate-limit fallback and an A2A step that allows

---
 .../observability/test_guardrail_pipeline_applied_header.py   | 1 +
 tests/unit/proxy/test_common_request_processing.py            | 4 ++++
 2 files changed, 5 insertions(+)

diff --git a/tests/integration/observability/test_guardrail_pipeline_applied_header.py b/tests/integration/observability/test_guardrail_pipeline_applied_header.py
index 6e408d24e9a..55f6f0764c3 100644
--- a/tests/integration/observability/test_guardrail_pipeline_applied_header.py
+++ b/tests/integration/observability/test_guardrail_pipeline_applied_header.py
@@ -756,6 +756,7 @@ def agent(rig: _Rig) -> Iterator[str]:
         pytest.param(
             _RESPONSE_MOVES_PAST_FIRST, [_RESPONSE_BLOCKER, _RESPONSE_SECOND], id="first-step-flags-and-moves-on"
         ),
+        pytest.param(_RESPONSE_STOPS_AFTER_FIRST, [_RESPONSE_FIRST], id="first-step-allows"),
     ],
 )
 def test_an_agent_message_names_every_response_step_that_ran_in_the_spend_log_and_the_trace(
diff --git a/tests/unit/proxy/test_common_request_processing.py b/tests/unit/proxy/test_common_request_processing.py
index 7a2fe623032..7d091327613 100644
--- a/tests/unit/proxy/test_common_request_processing.py
+++ b/tests/unit/proxy/test_common_request_processing.py
@@ -7542,10 +7542,12 @@ class TestPreCallWithFallbacksOnLocalRateLimit:
         processor = ProxyBaseLLMRequestProcessing(data={"model": primary_model})
 
         call_count = 0
+        deferred_per_call = []
 
         async def mock_pre_call_logic(**kwargs):
             nonlocal call_count
             call_count += 1
+            deferred_per_call.append(kwargs["defer_async_logging"])
             model_in_data = processor.data.get("model")
             if model_in_data == primary_model:
                 raise ProxyRateLimitError(
@@ -7578,10 +7580,12 @@ class TestPreCallWithFallbacksOnLocalRateLimit:
                 model=primary_model,
                 route_type="acompletion",
                 llm_router=mock_router,
+                defer_async_logging=True,
             )
 
         assert processor.data["model"] == fallback_model
         assert call_count == 2
+        assert deferred_per_call == [True, True]
 
     @pytest.mark.asyncio
     async def test_raises_when_no_fallbacks_configured(self):