Merge remote-tracking branch 'origin/092-scheduler_service' into v0.9.2-dev

vllm/v1/core/sched/scheduler.py

@@ -343,7 +343,9 @@ 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)