diff --git a/tests/v1/kv_connector/unit/test_offloading_connector.py b/tests/v1/kv_connector/unit/test_offloading_connector.py index 2a365b4dd7..7cf5272574 100644 --- a/tests/v1/kv_connector/unit/test_offloading_connector.py +++ b/tests/v1/kv_connector/unit/test_offloading_connector.py @@ -383,22 +383,6 @@ def test_cpu_offloading_metrics() -> None: total += sample.value return total - # Stats are drained asynchronously — if the transfer finishes - # after the last engine step for that generate() call, the metrics - # won't appear until a subsequent step. Retry with dummy generates - # to force additional stats drains. - deadline = time.monotonic() + _RESET_CACHE_TIMEOUT - while time.monotonic() < deadline: - store_bytes = _get_counter_value("vllm:kv_offload_store_bytes") - load_bytes = _get_counter_value("vllm:kv_offload_load_bytes") - if store_bytes > 0 and load_bytes > 0: - break - llm.generate( - [TokensPrompt(prompt_token_ids=[0])], - SamplingParams(max_tokens=1), - use_tqdm=False, - ) - # New flat counter metrics store_bytes = _get_counter_value("vllm:kv_offload_store_bytes") assert store_bytes > 0, f"Expected store_bytes > 0, got {store_bytes}" diff --git a/vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py b/vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py index 1c5986d515..197beca9ae 100644 --- a/vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py +++ b/vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py @@ -190,11 +190,6 @@ class OffloadingConnector(KVConnectorBase_V1, SupportsHMA): def get_kv_connector_stats(self) -> KVConnectorStats | None: if self.connector_scheduler is not None: return self.connector_scheduler.get_stats() - - # TODO(orozery): Remove once PR #43877 lands - if self.connector_worker is not None: - return OffloadingConnectorStats() - return None @classmethod