Merge branch 'v0.11.0-dev_scheduler' into 'v0.11.0-dev'

修复schedule_default，schedule_split_pd，因kv cache耗尽，服务挂掉的问题

See merge request dcutoolkit/deeplearing/vllm!266

vllm/v1/core/sched/scheduler.py

@@ -344,7 +344,9 @@ class Scheduler(SchedulerInterface):
                     break
 
                 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)
@@ -672,7 +674,9 @@ class Scheduler(SchedulerInterface):
                 break
 
             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)