diff --git a/vllm_ascend/distributed/kv_transfer/kv_p2p/mooncake_connector.py b/vllm_ascend/distributed/kv_transfer/kv_p2p/mooncake_connector.py index ee07e5f7..a2323adf 100644 --- a/vllm_ascend/distributed/kv_transfer/kv_p2p/mooncake_connector.py +++ b/vllm_ascend/distributed/kv_transfer/kv_p2p/mooncake_connector.py @@ -449,6 +449,8 @@ class KVCacheRecvingThread(threading.Thread): f"{remote_request_id}: {e}", exc_info=True) finally: + self._send_done_signal_to_free_remote_port(remote_request_id, remote_host, + remote_port_send_num) if all_task_done: self.task_tracker.update_done_task_count(request_id) if request_id in self.proc_not_transfer_request: @@ -460,8 +462,6 @@ class KVCacheRecvingThread(threading.Thread): self._send_done_recv_signal(remote_request_id, remote_host, remote_handshake_port, remote_port_send_num) - self._send_done_signal_to_free_remote_port(remote_request_id, remote_host, - remote_port_send_num) def _send_done_signal_to_free_remote_port( self, request_id: str, remote_host: str,