机器：BW1000 gfx936 镜像：harbor.sourcefind.cn:5443/dcu/admin/base/vllm:0.11.0-ubuntu22.04-dtk25.04.2-1226-das1.7-py3.10-20251226 容器启动命令： docker run -it --network=host --ipc=host --name test --shm-size=16G --device=/dev/kfd --device=/dev/mkfd --device=/dev/dri --privileged -v /root:/root -v /opt/hyhal:/opt/hyhal --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined localhost:61001/cdc2/dcu_vllm:0.11.0-ubuntu22.04-dtk25.04.2-1226-das1.7-py3.10-20251226 /bin/bash 服务启动： VLLM_HOST_IP=127.0.0.1 MASTER_ADDR=127.0.0.1 MASTER_PORT=29500 GLOO_SOCKET_IFNAME=bond0 NCCL_SOCKET_IFNAME=bond0 HIP_VISIBLE_DEVICES=0,1 python3 -m vllm.entrypoints.openai.api_server --served-model-name qwen3 --model /root/models/Qwen3-8B --tensor-parallel-size 2 --port 9080 --enable-auto-tool-choice --tool-call-parser hermes --gpu-memory-utilization 0.9 --trust-remote-code 报错信息： WARNING: Logging before InitGoogleLogging() is written to STDERR E0116 17:08:02.158560 26638 ProcessGroupGloo.cpp:143] Gloo connectFullMesh failed with [/home/pytorch/third_party/gloo/gloo/transport/tcp/pair.cc:144] no error ERROR 01-16 17:08:02 [multiproc_executor.py:597] WorkerProc failed to start. ERROR 01-16 17:08:02 [multiproc_executor.py:597] Traceback (most recent call last): ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/multiproc_executor.py", line 571, in worker_main ERROR 01-16 17:08:02 [multiproc_executor.py:597] worker = WorkerProc(*args, **kwargs) ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/executor/multiproc_executor.py", line 430, in __init__ ERROR 01-16 17:08:02 [multiproc_executor.py:597] self.worker.init_device() ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/vllm/worker/worker_base.py", line 299, in init_device ERROR 01-16 17:08:02 [multiproc_executor.py:597] self.worker.init_device() # type: ignore ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/worker/gpu_worker.py", line 169, in init_device ERROR 01-16 17:08:02 [multiproc_executor.py:597] init_worker_distributed_environment(self.vllm_config, self.rank, ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/vllm/v1/worker/gpu_worker.py", line 703, in init_worker_distributed_environment ERROR 01-16 17:08:02 [multiproc_executor.py:597] init_distributed_environment(parallel_config.world_size, rank, ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/vllm/distributed/parallel_state.py", line 1085, in init_distributed_environment ERROR 01-16 17:08:02 [multiproc_executor.py:597] _WORLD = init_world_group(ranks, local_rank, backend) ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/vllm/distributed/parallel_state.py", line 910, in init_world_group ERROR 01-16 17:08:02 [multiproc_executor.py:597] return GroupCoordinator( ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/vllm/distributed/parallel_state.py", line 224, in __init__ ERROR 01-16 17:08:02 [multiproc_executor.py:597] cpu_group = torch.distributed.new_group(ranks, backend="gloo") ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/torch/distributed/c10d_logger.py", line 97, in wrapper ERROR 01-16 17:08:02 [multiproc_executor.py:597] func_return = func(*args, **kwargs) ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/torch/distributed/distributed_c10d.py", line 4565, in new_group ERROR 01-16 17:08:02 [multiproc_executor.py:597] return _new_group_with_tag( ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/torch/distributed/distributed_c10d.py", line 4648, in _new_group_with_tag ERROR 01-16 17:08:02 [multiproc_executor.py:597] pg, pg_store = _new_process_group_helper( ERROR 01-16 17:08:02 [multiproc_executor.py:597] File "/usr/local/lib/python3.10/dist-packages/torch/distributed/distributed_c10d.py", line 1744, in _new_process_group_helper ERROR 01-16 17:08:02 [multiproc_executor.py:597] backend_class = ProcessGroupGloo( ERROR 01-16 17:08:02 [multiproc_executor.py:597] RuntimeError: Gloo connectFullMesh failed with [/home/pytorch/third_party/gloo/gloo/transport/tcp/pair.cc:144] no error