请求进入到waitting队列后，增加request.is_finished()判断

vllm/v1/core/sched/scheduler.py

@@ -339,6 +339,10 @@ class Scheduler(SchedulerInterface):
 
                 request = self.waiting.peek_request()
                 
+                if request.is_finished():
+                    self.waiting.pop_request()
+                    continue
+
                 # KVTransfer: skip request if still waiting for remote kvs.
                 if request.status == RequestStatus.WAITING_FOR_REMOTE_KVS:
                     is_ready = self._update_waiting_for_remote_kv(request)