From 53fb0464dd1166cdf93916e292c50c8fa059ef2f Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Tue, 10 Nov 2026 12:11:33 -0800
Subject: [PATCH] fix(logging): port the rest of #43667's internal-call gate

An internal sub-call under an async parent no longer schedules sync success callbacks
of its own, and a cache hit inside an internal sub-call fires no callbacks, since the
parent call that folds it in logs it. This matches main
---
 litellm/caching/caching_handler.py         |  5 ++
 litellm/utils.py                           | 17 +++----
 tests/unit/caching/test_caching_handler.py | 55 +++++++++++++++++++++-
 3 files changed, 68 insertions(+), 9 deletions(-)

diff --git a/litellm/caching/caching_handler.py b/litellm/caching/caching_handler.py
index 09765b1c346..063faa70f2f 100644
--- a/litellm/caching/caching_handler.py
+++ b/litellm/caching/caching_handler.py
@@ -822,7 +822,12 @@ class LLMCachingHandler:
             start_time (datetime): The start time of the operation.
             end_time (datetime): The end time of the operation.
             cache_hit (bool): Whether it was a cache hit.
+
+        An internal sub-call's hit fires nothing, since the parent call that folds it in logs it.
         """
+        if is_internal_call.get():
+            return
+
         from litellm.litellm_core_utils.logging_worker import GLOBAL_LOGGING_WORKER
 
         GLOBAL_LOGGING_WORKER.ensure_initialized_and_enqueue(
diff --git a/litellm/utils.py b/litellm/utils.py
index c2acb89b6e6..ea8726b7f69 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -1375,15 +1375,16 @@ def _dispatch_success_logging(
     is_completion_with_fallbacks: bool,
     is_litellm_internal_call: bool,
 ) -> None:
-    if not is_litellm_internal_call:
-        _schedule_async_success_logging(
-            logging_obj=logging_obj,
-            result=result,
-            start_time=start_time,
-            end_time=end_time,
-            is_completion_with_fallbacks=is_completion_with_fallbacks,
-        )
+    if is_litellm_internal_call:
+        return
 
+    _schedule_async_success_logging(
+        logging_obj=logging_obj,
+        result=result,
+        start_time=start_time,
+        end_time=end_time,
+        is_completion_with_fallbacks=is_completion_with_fallbacks,
+    )
     logging_obj.handle_sync_success_callbacks_for_async_calls(
         result=result,
         start_time=start_time,
diff --git a/tests/unit/caching/test_caching_handler.py b/tests/unit/caching/test_caching_handler.py
index 94b257ff5af..e12c8800dc9 100644
--- a/tests/unit/caching/test_caching_handler.py
+++ b/tests/unit/caching/test_caching_handler.py
@@ -54,7 +54,8 @@ import json
 import httpx
 import respx
 from fastapi.testclient import TestClient
-from litellm._internal_context import current_service_target, in_post_response_phase
+from contextlib import nullcontext
+from litellm._internal_context import current_service_target, in_post_response_phase, internal_sub_call
 from litellm.caching.caching_handler import _PENDING_CACHE_WRITES
 from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler
 
@@ -3088,3 +3089,55 @@ async def test_async_get_cache_answers_400_for_a_single_object_embedding_input()
             call_type=CallTypes.aembedding.value,
             kwargs={"model": model, "custom_llm_provider": "gemini", "input": clip_block, "caching": True},
         )
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("internal", [False, True])
+async def test_a_cache_hit_inside_an_internal_sub_call_fires_no_callbacks_of_its_own(
+    monkeypatch: pytest.MonkeyPatch, internal: bool
+):
+    cache: Final = Cache(type="local")
+    monkeypatch.setattr(litellm, "cache", cache)
+    kwargs: Final = {"model": "gpt-5.4", "messages": [{"role": "user", "content": "hello"}], "caching": True}
+    cache.add_cache(
+        result=litellm.ModelResponse(
+            model="gpt-5.4",
+            choices=[litellm.Choices(message=litellm.Message(role="assistant", content="hi"))],
+            usage=litellm.Usage(prompt_tokens=1000, completion_tokens=500, total_tokens=1500),
+        ),
+        **kwargs,
+    )
+    async_logging: Final = _build_logging_obj(CallTypes.acompletion.value, stream=False)
+    async_logging.async_success_handler = AsyncMock()
+    async_logging.handle_sync_success_callbacks_for_async_calls = MagicMock()
+    sync_logging: Final = _build_logging_obj(CallTypes.completion.value, stream=False)
+    sync_logging.handle_sync_cache_hit_success_callbacks = MagicMock()
+
+    with internal_sub_call() if internal else nullcontext():
+        async_hit: Final = await LLMCachingHandler(
+            original_function=litellm.acompletion, request_kwargs=kwargs, start_time=_FIXED_START
+        ).async_get_cache(
+            model="gpt-5.4",
+            original_function=litellm.acompletion,
+            logging_obj=async_logging,
+            start_time=_FIXED_START,
+            call_type=CallTypes.acompletion.value,
+            kwargs=kwargs,
+            args=(),
+        )
+        sync_hit: Final = LLMCachingHandler(
+            original_function=completion, request_kwargs=kwargs, start_time=_FIXED_START
+        ).sync_get_cache(
+            model="gpt-5.4",
+            original_function=completion,
+            logging_obj=sync_logging,
+            start_time=_FIXED_START,
+            call_type=CallTypes.completion.value,
+            kwargs=kwargs,
+        )
+
+    assert async_hit is not None and async_hit.cached_result is not None
+    assert sync_hit.cached_result is not None
+    assert async_logging.async_success_handler.called is not internal
+    assert async_logging.handle_sync_success_callbacks_for_async_calls.called is not internal
+    assert sync_logging.handle_sync_cache_hit_success_callbacks.called is not internal