diff --git a/tests/v1/core/test_kv_cache_utils.py b/tests/v1/core/test_kv_cache_utils.py index 2c4dab3f85..08463a2800 100644 --- a/tests/v1/core/test_kv_cache_utils.py +++ b/tests/v1/core/test_kv_cache_utils.py @@ -202,6 +202,18 @@ def test_kv_cache_block(): assert block.block_hash is None +def test_kv_cache_block_uses_slots(): + block = KVCacheBlock(block_id=0) + + # Slots eliminate per-instance __dict__, saving ~264 bytes per block. + # At 100K+ blocks this avoids tens of MB of overhead and GC pressure. + assert not hasattr(block, "__dict__") + + # Verify that slots actually prevent dynamic attribute assignment. + with pytest.raises(AttributeError): + block.unexpected_field = True + + def test_free_kv_cache_block_queue_initialization(): # Test with a single block block = KVCacheBlock(block_id=0) diff --git a/vllm/v1/core/kv_cache_utils.py b/vllm/v1/core/kv_cache_utils.py index cfaa37074d..2ed7ef7e08 100644 --- a/vllm/v1/core/kv_cache_utils.py +++ b/vllm/v1/core/kv_cache_utils.py @@ -106,7 +106,7 @@ def init_none_hash(hash_fn: Callable[[Any], bytes]): NONE_HASH = BlockHash(hash_fn(hash_seed)) -@dataclass +@dataclass(slots=True) class KVCacheBlock: """KV-cache block metadata."""