[Bug] DeepSeek-R1 NCCL WatchDog Timeout Error

[sitabulaixizawaluduo]

Checklist

• [x] 1. I have searched related issues but cannot get the expected help.
• [x] 2. The bug has not been fixed in the latest version.
• [x] 3. Please note that if the bug-related issue you submitted lacks corresponding environment info and a minimal reproducible demo, it will be challenging for us to reproduce and resolve the issue, reducing the likelihood of receiving feedback.
• [x] 4. If the issue you raised is not a bug but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• [x] 5. Please use English, otherwise it will be closed.

Describe the bug

When I used H800 to deploy DeepSeek-R1, I found a very strange bug. I used 4 machines to deploy 2 instances, each with two nodes. On one of the instances, when I turned on the prefix cache, when the prefix cache reached two tokens, I found that the process would be blocked until watchdog timeout. When I turned off the prefix cache, some specific cases must also trigger this behavior. But for other requests, this problem does not occur. Is there something wrong with the current multi-machine communication?

Reproduction

Server Start with prefix caching node rank 0 python3 -m sglang.launch_server --mem-fraction-static 0.85 --disable-custom-all-reduce --trust-remote-code --chunked-prefill-size -1 --host 0.0.0.0 --port 8000 --model-path /models/DeepSeek-R1 --tensor-parallel-size 16 --log-requests --dist-init-addr maas-deepseek-r1-rank0:29500 --nnodes 2 --node-rank 0 node rank 1 python3 -m sglang.launch_server --mem-fraction-static 0.85 --disable-custom-all-reduce --trust-remote-code --chunked-prefill-size -1 --host 0.0.0.0 --port 8000 --model-path /models/DeepSeek-R1 --tensor-parallel-size 16 --log-requests --dist-init-addr maas-deepseek-r1-rank0:29500 --nnodes 2 --node-rank 1

Server Start without prefix caching node rank 0 python3 -m sglang.launch_server --mem-fraction-static 0.85 --disable-custom-all-reduce --trust-remote-code --chunked-prefill-size -1 --host 0.0.0.0 --port 8000 --model-path /models/DeepSeek-R1 --tensor-parallel-size 16 --log-requests --dist-init-addr maas-deepseek-r1-rank0:29500 --nnodes 2 --node-rank 0 --disable-radix-cache node rank 1 python3 -m sglang.launch_server --mem-fraction-static 0.85 --disable-custom-all-reduce --trust-remote-code --chunked-prefill-size -1 --host 0.0.0.0 --port 8000 --model-path /models/DeepSeek-R1 --tensor-parallel-size 16 --log-requests --dist-init-addr maas-deepseek-r1-rank0:29500 --nnodes 2 --node-rank 1 --disable-radix-cache

When prefix caching is enabled, when I execute first `import requests import json import time from openai import OpenAI

model = "/models/deepseek-r1" client = OpenAI( base_url="url/v1", api_key="NOKEY", )

completion = client.chat.completions.create( model=model, messages = [{"role":"user","content":"理财产品投入500，7日年化收益1.7070%，能赚多少"}], temperature=0.8, max_tokens=10, ) print(completion)Then runimport requests import json import time from openai import OpenAI

model = "/models/deepseek-r1" client = OpenAI( base_url="url/v1", api_key="NOKEY", )

completion = client.completions.create( model=model, #messages = [{"role":"user","content":"世界上最古老的已知宗教之一是佛教，它起源于哪个国家？现在信徒分布在哪里？"}], temperature=0.8, max_tokens=10, ) print(completion)`

It will definitely trigger a blocking situation

Environment

`Python: 3.10.16 (main, Dec 4 2024, 08:53:37) [GCC 9.4.0] CUDA available: True GPU 0,1,2,3,4,5,6,7: NVIDIA H800 GPU 0,1,2,3,4,5,6,7 Compute Capability: 9.0 CUDA_HOME: /usr/local/cuda NVCC: Cuda compilation tools, release 12.4, V12.4.131 CUDA Driver Version: 550.90.07 PyTorch: 2.5.1+cu124 sglang: 0.4.1.post7 flashinfer: 0.1.6+cu124torch2.4 triton: 3.1.0 transformers: 4.48.0 torchao: 0.8.0 numpy: 1.26.4 aiohttp: 3.11.11 fastapi: 0.115.6 hf_transfer: 0.1.9 huggingface_hub: 0.27.1 interegular: 0.3.3 modelscope: 1.22.3 orjson: 3.10.15 packaging: 24.2 psutil: 6.1.1 pydantic: 2.10.5 multipart: 0.0.20 zmq: 26.2.0 uvicorn: 0.34.0 uvloop: 0.21.0 vllm: 0.6.4.post1 openai: 1.59.8 anthropic: 0.43.1 decord: 0.6.0 NVIDIA Topology: GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 NIC6 NIC7 NIC8 CPU Affinity NUMA Affinity GPU NUMA ID GPU0 X NV8 NV8 NV8 NV8 NV8 NV8 NV8 PIX NODE NODE NODE SYS SYS SYS SYS SYS 0-55,112-167 N/A GPU1 NV8 X NV8 NV8 NV8 NV8 NV8 NV8 NODE PIX NODE NODE SYS SYS SYS SYS SYS 0-55,112-167 N/A GPU2 NV8 NV8 X NV8 NV8 NV8 NV8 NV8 NODE NODE PIX NODE SYS SYS SYS SYS SYS 0-55,112-167 N/A GPU3 NV8 NV8 NV8 X NV8 NV8 NV8 NV8 NODE NODE NODE PIX SYS SYS SYS SYS SYS 0-55,112-167 N/A GPU4 NV8 NV8 NV8 NV8 X NV8 NV8 NV8 SYS SYS SYS SYS PIX NODE NODE NODE NODE 56-111,168-223 N/A GPU5 NV8 NV8 NV8 NV8 NV8 X NV8 NV8 SYS SYS SYS SYS NODE PIX NODE NODE NODE 56-111,168-223 N/A GPU6 NV8 NV8 NV8 NV8 NV8 NV8 X NV8 SYS SYS SYS SYS NODE NODE PIX NODE NODE 56-111,168-223 N/A GPU7 NV8 NV8 NV8 NV8 NV8 NV8 NV8 X SYS SYS SYS SYS NODE NODE NODE PIX NODE 56-111,168-223 N/A NIC0 PIX NODE NODE NODE SYS SYS SYS SYS X NODE NODE NODE SYS SYS SYS SYS SYS NIC1 NODE PIX NODE NODE SYS SYS SYS SYS NODE X NODE NODE SYS SYS SYS SYS SYS NIC2 NODE NODE PIX NODE SYS SYS SYS SYS NODE NODE X NODE SYS SYS SYS SYS SYS NIC3 NODE NODE NODE PIX SYS SYS SYS SYS NODE NODE NODE X SYS SYS SYS SYS SYS NIC4 SYS SYS SYS SYS PIX NODE NODE NODE SYS SYS SYS SYS X NODE NODE NODE NODE NIC5 SYS SYS SYS SYS NODE PIX NODE NODE SYS SYS SYS SYS NODE X NODE NODE NODE NIC6 SYS SYS SYS SYS NODE NODE PIX NODE SYS SYS SYS SYS NODE NODE X NODE NODE NIC7 SYS SYS SYS SYS NODE NODE NODE PIX SYS SYS SYS SYS NODE NODE NODE X NODE NIC8 SYS SYS SYS SYS NODE NODE NODE NODE SYS SYS SYS SYS NODE NODE NODE NODE X

Legend:

X = Self SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI) NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU) PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge) PIX = Connection traversing at most a single PCIe bridge NV# = Connection traversing a bonded set of # NVLinks

NIC Legend:

NIC0: mlx5_0 NIC1: mlx5_1 NIC2: mlx5_2 NIC3: mlx5_3 NIC4: mlx5_4 NIC5: mlx5_7 NIC6: mlx5_8 NIC7: mlx5_9 NIC8: mlx5_bond_0

ulimit soft: 2108278520`