Merge pull request #13933 from michal-otmianowski/ignore-expired-s3-cache-entries

Verify if cache entry has expired prior to serving it to client

litellm/caching/s3_cache.py

@@ -13,6 +13,7 @@ import asyncio
 import json
 from functools import partial
 from typing import Optional
+from datetime import datetime, timezone, timedelta
 
 from litellm._logging import print_verbose, verbose_logger
 
@@ -69,11 +70,9 @@ class S3Cache(BaseCache):
 
             if ttl is not None:
                 cache_control = f"immutable, max-age={ttl}, s-maxage={ttl}"
-                import datetime
 
                 # Calculate expiration time
-                expiration_time = datetime.datetime.now() + datetime.timedelta(seconds=ttl)
-
+                expiration_time = datetime.now(timezone.utc) + timedelta(seconds=ttl)
                 # Upload the data to S3 with the calculated expiration time
                 self.s3_client.put_object(
                     Bucket=self.bucket_name,
@@ -126,6 +125,13 @@ class S3Cache(BaseCache):
             )
 
             if cached_response is not None:
+                if "Expires" in cached_response:
+                    expires_time = cached_response['Expires']
+                    current_time = datetime.now(expires_time.tzinfo)
+
+                    if current_time > expires_time:
+                        return None
+
                 # cached_response is in `b{} convert it to ModelResponse
                 cached_response = (
                     cached_response["Body"].read().decode("utf-8")

tests/test_litellm/caching/test_s3_cache.py

@@ -56,7 +56,7 @@ def test_s3_cache_set_cache_with_ttl(mock_s3_dependencies):
     assert "max-age=3600" in call_args[1]["CacheControl"]
 
 
-def test_s3_cache_get_cache(mock_s3_dependencies):
+def test_s3_cache_get_cache_no_expires_info_in_response(mock_s3_dependencies):
     """Test basic get_cache functionality"""
     cache = S3Cache("test-bucket")
 
@@ -75,6 +75,54 @@ def test_s3_cache_get_cache(mock_s3_dependencies):
 
     assert result == {"key": "value", "number": 42}
 
+def test_s3_cache_get_cache_with_expires_valid(mock_s3_dependencies):
+    """Test get_cache when response contains Expires and cache entry is still valid"""
+    cache = S3Cache("test-bucket")
+
+    # Create a future expiration time (1 hour from now)
+    future_time = datetime.datetime.now(datetime.timezone.utc) + datetime.timedelta(hours=1)
+
+    mock_response = {
+        "Body": MagicMock(),
+        "Expires": future_time
+    }
+    mock_response["Body"].read.return_value = b'{"key": "value", "number": 42}'
+    cache.s3_client.get_object.return_value = mock_response
+
+    result = cache.get_cache("test_key")
+
+    cache.s3_client.get_object.assert_called_once_with(
+        Bucket="test-bucket",
+        Key="test_key"
+    )
+
+    # Should return the cached value since it's not expired
+    assert result == {"key": "value", "number": 42}
+
+
+def test_s3_cache_get_cache_with_expires_expired(mock_s3_dependencies):
+    """Test get_cache when response contains Expires and cache entry is no longer valid"""
+    cache = S3Cache("test-bucket")
+
+    # Create a past expiration time (1 hour ago)
+    past_time = datetime.datetime.now(datetime.timezone.utc) - datetime.timedelta(hours=1)
+
+    mock_response = {
+        "Body": MagicMock(),
+        "Expires": past_time
+    }
+    mock_response["Body"].read.return_value = b'{"key": "value", "number": 42}'
+    cache.s3_client.get_object.return_value = mock_response
+
+    result = cache.get_cache("test_key")
+
+    cache.s3_client.get_object.assert_called_once_with(
+        Bucket="test-bucket",
+        Key="test_key"
+    )
+
+    # Should return None since the cache entry is expired
+    assert result is None
 
 def test_s3_cache_get_cache_not_found(mock_s3_dependencies):
     """Test get_cache when key is not found"""
@@ -126,7 +174,6 @@ def test_s3_cache_initialization():
     cache_with_path = S3Cache("test-bucket", s3_path="my/cache/path")
     assert cache_with_path.key_prefix == "my/cache/path/"
 
-
 # ============================================================================
 # ASYNC TESTS
 # ============================================================================