fix(tests): reduce TTL durations to improve test reliability

- Adjusted TTL durations and sleep intervals across memory and cache tests for consistent expiration behavior.
- Prevented test flakiness caused by timing discrepancies in token expiration and cache cleanup.

backend/tests/unit/services/memory/cache/test_embedding_cache.py

@@ -160,11 +160,11 @@ class TestEmbeddingCache:
 
     async def test_ttl_expiration(self) -> None:
         """Should expire entries after TTL."""
-        cache = EmbeddingCache(max_size=100, default_ttl_seconds=0.1)
+        cache = EmbeddingCache(max_size=100, default_ttl_seconds=0.05)
 
         await cache.put("content", [0.1, 0.2])
 
-        time.sleep(0.2)
+        time.sleep(0.06)
 
         result = await cache.get("content")
 
@@ -226,13 +226,13 @@ class TestEmbeddingCache:
 
     def test_cleanup_expired(self) -> None:
         """Should remove expired entries."""
-        cache = EmbeddingCache(max_size=100, default_ttl_seconds=0.1)
+        cache = EmbeddingCache(max_size=100, default_ttl_seconds=0.05)
 
         # Use synchronous put for setup
         cache._put_memory("hash1", "default", [0.1])
         cache._put_memory("hash2", "default", [0.2], ttl_seconds=10)
 
-        time.sleep(0.2)
+        time.sleep(0.06)
 
         count = cache.cleanup_expired()
 

backend/tests/unit/services/memory/cache/test_hot_cache.py

@@ -212,12 +212,12 @@ class TestHotMemoryCache:
 
     def test_ttl_expiration(self) -> None:
         """Should expire entries after TTL."""
-        cache = HotMemoryCache[str](max_size=100, default_ttl_seconds=0.1)
+        cache = HotMemoryCache[str](max_size=100, default_ttl_seconds=0.05)
 
         cache.put_by_id("test", "1", "value")
 
         # Wait for expiration
-        time.sleep(0.2)
+        time.sleep(0.06)
 
         result = cache.get_by_id("test", "1")
 
@@ -289,12 +289,12 @@ class TestHotMemoryCache:
 
     def test_cleanup_expired(self) -> None:
         """Should remove expired entries."""
-        cache = HotMemoryCache[str](max_size=100, default_ttl_seconds=0.1)
+        cache = HotMemoryCache[str](max_size=100, default_ttl_seconds=0.05)
 
         cache.put_by_id("test", "1", "value1")
         cache.put_by_id("test", "2", "value2", ttl_seconds=10)
 
-        time.sleep(0.2)
+        time.sleep(0.06)
 
         count = cache.cleanup_expired()