fix(caching): honor s-maxage 0 on the sync cache read so TTS on this line skips cached audio

The sync Cache.get_cache read s-maxage with `or`, so 0 became infinity and
a request asking for no stale audio got the cached result anyway. The async
read already used nested defaults. On lts/1.108.x the TTS bridge's inner chat
call stays on sync completion, so speech hit this path. Main's open #34403
makes the same change for main's sync callers

litellm/caching/caching.py

@@ -662,7 +662,9 @@ 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 = cache_control_args.get("s-maxage") or cache_control_args.get("s-max-age") or float("inf")
+                    max_age: Final = cache_control_args.get(
+                        "s-maxage", cache_control_args.get("s-max-age", 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)

tests/unit/caching/test_caching.py

@@ -1437,6 +1437,24 @@ 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
 async def test_embedding_caching_redis_ttl(monkeypatch: pytest.MonkeyPatch, redis_client_without_network: None) -> None:
     mock_pipeline: Final = AsyncMock()

tests/unit/test_main.py

@@ -4874,6 +4874,7 @@ 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),