Open
Description
Model Series
Qwen3
What are the models used?
Qwen3-235B-A22B
What is the scenario where the problem happened?
vllm
Is this a known issue?
- I have followed the GitHub README.
- I have checked the Qwen documentation and cannot find an answer there.
- I have checked the documentation of the related framework and cannot find useful information.
- I have searched the issues and there is not a similar one.
Information about environment
vllm/vllm-openai:v0.8.5
Log output
打印日志如下：
INFO 04-30 02:37:31 [init.py:239] Automatically detected platform cuda.
INFO 04-30 02:37:37 [api_server.py:1043] vLLM API server version 0.8.5
INFO 04-30 02:37:37 [api_server.py:1044] args: Namespace(subparser='serve', model_tag='/root/.cache/huggingface/Qwen3-235B-A22B', config='', host='0.0.0.0', port=8000, uvicorn_log_level='info', disable_uvicorn_access_log=False, allow_credentials=False, allowed_origins=[''], allowed_methods=[''], allowed_headers=['*'], api_key=None, lora_modules=None, prompt_adapters=None, chat_template=None, chat_template_content_format='auto', response_role='assistant', ssl_keyfile=None, ssl_certfile=None, ssl_ca_certs=None, enable_ssl_refresh=False, ssl_cert_reqs=0, root_path=None, middleware=[], return_tokens_as_token_ids=False, disable_frontend_multiprocessing=False, enable_request_id_headers=False, enable_auto_tool_choice=False, tool_call_parser=None, tool_parser_plugin='', model='/root/.cache/huggingface/Qwen3-235B-A22B', task='auto', tokenizer=None, hf_config_path=None, skip_tokenizer_init=False, revision=None, code_revision=None, tokenizer_revision=None, tokenizer_mode='auto', trust_remote_code=True, allowed_local_media_path=None, load_format='auto', download_dir=None, model_loader_extra_config={}, use_tqdm_on_load=True, config_format=<ConfigFormat.AUTO: 'auto'>, dtype='auto', max_model_len=None, guided_decoding_backend='auto', reasoning_parser='deepseek_r1', logits_processor_pattern=None, model_impl='auto', distributed_executor_backend=None, pipeline_parallel_size=2, tensor_parallel_size=8, data_parallel_size=1, enable_expert_parallel=True, max_parallel_loading_workers=None, ray_workers_use_nsight=False, disable_custom_all_reduce=False, block_size=None, gpu_memory_utilization=0.9, swap_space=4, kv_cache_dtype='auto', num_gpu_blocks_override=None, enable_prefix_caching=None, prefix_caching_hash_algo='builtin', cpu_offload_gb=0, calculate_kv_scales=False, disable_sliding_window=False, use_v2_block_manager=True, seed=None, max_logprobs=20, disable_log_stats=False, quantization=None, rope_scaling=None, rope_theta=None, hf_token=None, hf_overrides=None, enforce_eager=False, max_seq_len_to_capture=8192, tokenizer_pool_size=0, tokenizer_pool_type='ray', tokenizer_pool_extra_config={}, limit_mm_per_prompt={}, mm_processor_kwargs=None, disable_mm_preprocessor_cache=False, enable_lora=None, enable_lora_bias=False, max_loras=1, max_lora_rank=16, lora_extra_vocab_size=256, lora_dtype='auto', long_lora_scaling_factors=None, max_cpu_loras=None, fully_sharded_loras=False, enable_prompt_adapter=None, max_prompt_adapters=1, max_prompt_adapter_token=0, device='auto', speculative_config=None, ignore_patterns=[], served_model_name=None, qlora_adapter_name_or_path=None, show_hidden_metrics_for_version=None, otlp_traces_endpoint=None, collect_detailed_traces=None, disable_async_output_proc=False, max_num_batched_tokens=None, max_num_seqs=None, max_num_partial_prefills=1, max_long_partial_prefills=1, long_prefill_token_threshold=0, num_lookahead_slots=0, scheduler_delay_factor=0.0, preemption_mode=None, num_scheduler_steps=1, multi_step_stream_outputs=True, scheduling_policy='fcfs', enable_chunked_prefill=None, disable_chunked_mm_input=False, scheduler_cls='vllm.core.scheduler.Scheduler', override_neuron_config=None, override_pooler_config=None, compilation_config=None, kv_transfer_config=None, worker_cls='auto', worker_extension_cls='', generation_config='auto', override_generation_config=None, enable_sleep_mode=False, additional_config=None, enable_reasoning=True, disable_cascade_attn=False, disable_log_requests=False, max_log_len=None, disable_fastapi_docs=False, enable_prompt_tokens_details=False, enable_server_load_tracking=False, dispatch_function=<function ServeSubcommand.cmd at 0x7f99737c7e20>)
INFO 04-30 02:37:45 [config.py:717] This model supports multiple tasks: {'embed', 'reward', 'score', 'generate', 'classify'}. Defaulting to 'generate'.
WARNING 04-30 02:37:45 [arg_utils.py:1658] Pipeline Parallelism without Ray distributed executor is not supported by the V1 Engine. Falling back to V0.
WARNING 04-30 02:37:45 [arg_utils.py:1525] Chunked prefill is enabled by default for models with max_model_len > 32K. Chunked prefill might not work with some features or models. If you encounter any issues, please disable by launching with --enable-chunked-prefill=False.
INFO 04-30 02:37:45 [config.py:1770] Defaulting to use ray for distributed inference
INFO 04-30 02:37:45 [confi
7A50
g.py:2003] Chunked prefill is enabled with max_num_batched_tokens=2048.
INFO 04-30 02:37:45 [llm_engine.py:240] Initializing a V0 LLM engine (v0.8.5) with config: model='/root/.cache/huggingface/Qwen3-235B-A22B', speculative_config=None, tokenizer='/root/.cache/huggingface/Qwen3-235B-A22B', skip_tokenizer_init=False, tokenizer_mode=auto, revision=None, override_neuron_config=None, tokenizer_revision=None, trust_remote_code=True, dtype=torch.bfloat16, max_seq_len=40960, download_dir=None, load_format=LoadFormat.AUTO, tensor_parallel_size=8, pipeline_parallel_size=2, disable_custom_all_reduce=False, quantization=None, enforce_eager=False, kv_cache_dtype=auto, device_config=cuda, decoding_config=DecodingConfig(guided_decoding_backend='auto', reasoning_backend='deepseek_r1'), observability_config=ObservabilityConfig(show_hidden_metrics=False, otlp_traces_endpoint=None, collect_model_forward_time=False, collect_model_execute_time=False), seed=None, served_model_name=/root/.cache/huggingface/Qwen3-235B-A22B, num_scheduler_steps=1, multi_step_stream_outputs=True, enable_prefix_caching=None, chunked_prefill_enabled=True, use_async_output_proc=False, disable_mm_preprocessor_cache=False, mm_processor_kwargs=None, pooler_config=None, compilation_config={"splitting_ops":[],"compile_sizes":[],"cudagraph_capture_sizes":[256,248,240,232,224,216,208,200,192,184,176,168,160,152,144,136,128,120,112,104,96,88,80,72,64,56,48,40,32,24,16,8,4,2,1],"max_capture_size":256}, use_cached_outputs=False,
2025-04-30 02:37:46,000 INFO worker.py:1654 -- Connecting to existing Ray cluster at address: 192.168.40.11:6379...
2025-04-30 02:37:46,012 INFO worker.py:1841 -- Connected to Ray cluster.
INFO 04-30 02:37:48 [ray_utils.py:335] No current placement group found. Creating a new placement group.
INFO 04-30 02:37:48 [ray_distributed_executor.py:176] use_ray_spmd_worker: False
(pid=717) INFO 04-30 02:37:52 [init.py:239] Automatically detected platform cuda.
(pid=289, ip=192.168.40.19) INFO 04-30 02:37:58 [init.py:239] Automatically detected platform cuda. [repeated 8x across cluster] (Ray deduplicates logs by default. Set RAY_DEDUP_LOGS=0 to disable log deduplication, or see https://docs.ray.io/en/master/ray-observability/user-guides/configure-logging.html#log-deduplication for more options.)
INFO 04-30 02:38:01 [ray_distributed_executor.py:352] non_carry_over_env_vars from config: set()
INFO 04-30 02:38:01 [ray_distributed_executor.py:354] Copying the following environment variables to workers: ['LD_LIBRARY_PATH', 'VLLM_USAGE_SOURCE', 'VLLM_WORKER_MULTIPROC_METHOD', 'VLLM_USE_V1']
INFO 04-30 02:38:01 [ray_distributed_executor.py:357] If certain env vars should NOT be copied to workers, add them to /root/.config/vllm/ray_non_carry_over_env_vars.json file
INFO 04-30 02:38:01 [cuda.py:292] Using Flash Attention backend.
(RayWorkerWrapper pid=722) INFO 04-30 02:38:02 [cuda.py:292] Using Flash Attention backend.
(pid=295, ip=192.168.40.19) INFO 04-30 02:37:58 [init.py:239] Automatically detected platform cuda. [repeated 7x across cluster]
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] Error executing method 'init_device'. This might cause deadlock in distributed execution.
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] Traceback (most recent call last):
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/vllm/worker/worker_base.py", line 612, in execute_method
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] return run_method(self, method, args, kwargs)
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/vllm/utils.py", line 2456, in run_method
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] return func(*args, **kwargs)
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/ray/util/tracing/tracing_helper.py", line 463, in _resume_span
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] return method(self, *_args, **_kwargs)
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/vllm/worker/worker_base.py", line 604, in init_device
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] self.worker.init_device() # type: ignore
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/vllm/worker/worker.py", line 186, in init_device
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] init_worker_distributed_environment(self.vllm_config, self.rank,
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/vllm/worker/worker.py", line 525, in init_worker_distributed_environment
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] init_distributed_environment(parallel_config.world_size, rank,
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/vllm/distributed/parallel_state.py", line 909, in init_distributed_environment
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] _WORLD = init_world_group(ranks, local_rank, backend)
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/vllm/distributed/parallel_state.py", line 771, in init_world_group
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] return GroupCoordinator(
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/vllm/distributed/parallel_state.py", line 225, in init
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] cpu_group = torch.distributed.new_group(ranks, backend="gloo")
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/torch/distributed/c10d_logger.py", line 95, in wrapper
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] func_return = func(*args, **kwargs)
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/torch/distributed/distributed_c10d.py", line 4981, in new_group
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] return _new_group_with_tag(
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/torch/distributed/distributed_c10d.py", line 5071, in _new_group_with_tag
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] pg, pg_store = _new_process_group_helper(
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] File "/usr/local/lib/python3.12/dist-packages/torch/distributed/distributed_c10d.py", line 1953, in _new_process_group_helper
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] backend_class = ProcessGroupGloo(
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] ^^^^^^^^^^^^^^^^^
(RayWorkerWrapper pid=721) ERROR 04-30 02:38:09 [worker_base.py:620] RuntimeError: [enforce fail at /pytorch/third_party/gloo/gloo/transport/tcp/device.cc:83] ifa != nullptr. Unable to find address for: enp199s0f0
(RayWorkerWrapper pid=295, ip=192.168.40.19) INFO 04-30 02:38:03 [cuda.py:292] Using Flash Attention backend. [repeated 14x across cluster]Description
镜像为vllm/vllm-openai:v0.8.5
通过examples/online_serving/run_cluster.sh (https://github.com/vllm-project/vllm/blob/main/examples/online_serving/run_cluster.sh) 创建了2个node
然后在主节点运行NCCL_SOCKET_IFNAME=enp199s0f0 GLOO_SOCKET_IFNAME=enp199s0f0 vllm serve /root/.cache/huggingface/Qwen3-235B-A22B --tensor-parallel-size 8 --pipeline-parallel-size 2 --enable-reasoning --reasoning-parser deepseek_r1 --trust-remote-code --host 0.0.0.0 --enable-expert-parallel
执行ifconfig，得到两台机器的网卡都是 enp199s0f0