fix nccl 报超时的问题，原因是双线程调度导致提前关闭远端worker，warmup后远端worker执行异常，引发rank0超时。

http://hpczentao.sugon.com/bug-view-93029.html

vllm/engine/llm_engine.py

@@ -1328,6 +1328,8 @@ class LLMEngine:
             while True:
                 self.sem_m2s.acquire()
                 if not self.thread_running:
+                    logger.debug("Stopping remote worker execution loop.")
+                    self.model_executor.stop_remote_worker_execution_loop()
                     break
 
                 virtual_engine = 0
@@ -1438,8 +1440,9 @@ class LLMEngine:
             # torch.distributed ops which may otherwise timeout, and unblocks
             # the RPC thread in the workers so that they can process any other
             # queued control plane messages, such as add/remove lora adapters.
-            logger.debug("Stopping remote worker execution loop.")
-            self.model_executor.stop_remote_worker_execution_loop()
+            # logger.debug("Stopping remote worker execution loop.")
+            # self.model_executor.stop_remote_worker_execution_loop()
+            self.finish_thread()
         return ctx.request_outputs
 
     def step(self) -> List[Union[RequestOutput, PoolingRequestOutput]]: