From b706bf16f83a5c3daccb20b6152c304da2f144d1 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Tue, 10 Nov 2026 12:11:51 -0800
Subject: [PATCH] revert: drop the LTS-only sync cache read changes

This reverts dfc92ba4cd, d94be668d2 and c2ba33609a. Async speech now awaits the
bridge, so the proxy reads the cache through the async path, which already honors
s-maxage 0 and reads s-max-age first. The sync read goes back to the line's code,
which matches main
---
 litellm/caching/caching.py         |  4 +--
 tests/unit/caching/test_caching.py | 47 ------------------------------
 tests/unit/test_main.py            |  1 -
 3 files changed, 1 insertion(+), 51 deletions(-)

diff --git a/litellm/caching/caching.py b/litellm/caching/caching.py
index 8044a5d24e7..5f5b8210096 100644
--- a/litellm/caching/caching.py
+++ b/litellm/caching/caching.py
@@ -662,9 +662,7 @@ class Cache:
                     cache_key = self.get_cache_key(**kwargs)
                 if cache_key is not None:
                     cache_control_args: Final[DynamicCacheControl] = kwargs.get("cache", {})
-                    max_age: Final = cache_control_args.get(
-                        "s-max-age", cache_control_args.get("s-maxage", float("inf"))
-                    )
+                    max_age = cache_control_args.get("s-maxage") or cache_control_args.get("s-max-age") or float("inf")
                     cache_lookup_kwargs: Final = self._get_safe_cache_lookup_kwargs(kwargs)
                     if dynamic_cache_object is not None:
                         cached_result = dynamic_cache_object.get_cache(cache_key, **cache_lookup_kwargs)
diff --git a/tests/unit/caching/test_caching.py b/tests/unit/caching/test_caching.py
index b0ec0278034..91083921130 100644
--- a/tests/unit/caching/test_caching.py
+++ b/tests/unit/caching/test_caching.py
@@ -1437,53 +1437,6 @@ async def test_caching_with_cache_controls(monkeypatch: pytest.MonkeyPatch) -> N
     assert async_ttl_first.id == async_ttl_second.id
 
 
-@pytest.mark.parametrize("max_age_key", ["s-maxage", "s-max-age"])
-def test_sync_completion_with_zero_s_maxage_skips_the_cached_response(
-    monkeypatch: pytest.MonkeyPatch, max_age_key: str
-) -> None:
-    monkeypatch.setattr(litellm, "cache", Cache(type=LiteLLMCacheType.LOCAL))
-    messages: Final = [{"role": "user", "content": f"zero s-maxage {uuid.uuid4().hex}"}]
-    first: Final = completion(model="gpt-4o-mini", messages=messages, mock_response="first")
-    zero_max_age: Final = completion(
-        model="gpt-4o-mini", messages=messages, cache={max_age_key: 0}, mock_response="second"
-    )
-    open_max_age: Final = completion(
-        model="gpt-4o-mini", messages=messages, cache={max_age_key: 60}, mock_response="third"
-    )
-
-    assert zero_max_age.id != first.id
-    assert open_max_age.id == zero_max_age.id
-
-
-@pytest.mark.asyncio
-@pytest.mark.parametrize(
-    ("cache_controls", "serves_cached"),
-    [
-        ({"s-maxage": 0, "s-max-age": 60}, True),
-        ({"s-maxage": 60, "s-max-age": 0}, False),
-        ({"s-maxage": None, "s-max-age": 0}, False),
-    ],
-)
-async def test_sync_and_async_completion_follow_s_max_age_when_both_max_age_keys_are_sent(
-    monkeypatch: pytest.MonkeyPatch, cache_controls: dict[str, int | None], serves_cached: bool
-) -> None:
-    monkeypatch.setattr(litellm, "cache", Cache(type=LiteLLMCacheType.LOCAL))
-    sync_messages: Final = [{"role": "user", "content": f"sync both max-age keys {uuid.uuid4().hex}"}]
-    sync_first: Final = completion(model="gpt-4o-mini", messages=sync_messages, mock_response="first")
-    sync_second: Final = completion(
-        model="gpt-4o-mini", messages=sync_messages, cache=cache_controls, mock_response="second"
-    )
-    async_messages: Final = [{"role": "user", "content": f"async both max-age keys {uuid.uuid4().hex}"}]
-    async_first: Final = await acompletion(model="gpt-4o-mini", messages=async_messages, mock_response="first")
-    await asyncio.gather(*_PENDING_CACHE_WRITES)
-    async_second: Final = await acompletion(
-        model="gpt-4o-mini", messages=async_messages, cache=cache_controls, mock_response="second"
-    )
-
-    assert (sync_second.id == sync_first.id) is serves_cached
-    assert (async_second.id == async_first.id) is serves_cached
-
-
 @pytest.mark.asyncio
 async def test_embedding_caching_redis_ttl(monkeypatch: pytest.MonkeyPatch, redis_client_without_network: None) -> None:
     mock_pipeline: Final = AsyncMock()
diff --git a/tests/unit/test_main.py b/tests/unit/test_main.py
index 367a67dcfa1..ac22723de8f 100644
--- a/tests/unit/test_main.py
+++ b/tests/unit/test_main.py
@@ -5013,7 +5013,6 @@ async def test_speech_gemini_tts_behind_the_router_waits_as_long_as_the_same_cha
     [
         ({}, {}, 1),
         ({}, {"cache": {"no-cache": True}}, 2),
-        ({}, {"cache": {"s-maxage": 0}}, 2),
         ({"cache": {"no-store": True}}, {}, 2),
         ({}, {"caching": False}, 2),
         ({}, {"caching": "yes"}, 2),