vllm: [Bug]: 0.8.0(V1) RayChannelTimeoutError when inferencing DeepSeekV3 on 16 H20 with large batch size
Your current environment
The output of `python collect_env.py`
Collecting environment information...
PyTorch version: 2.6.0+cu124
Is debug build: False
CUDA used to build PyTorch: 12.4
ROCM used to build PyTorch: N/A
OS: Ubuntu 22.04.4 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.31.6
Libc version: glibc-2.35
Python version: 3.12.9 (main, Mar 17 2025, 21:01:58) [Clang 20.1.0 ] (64-bit runtime)
Python platform: Linux-5.15.0-130-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.4.131
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA H20
GPU 1: NVIDIA H20
GPU 2: NVIDIA H20
GPU 3: NVIDIA H20
GPU 4: NVIDIA H20
GPU 5: NVIDIA H20
GPU 6: NVIDIA H20
GPU 7: NVIDIA H20
Nvidia driver version: 550.127.05
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 52 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 180
On-line CPU(s) list: 0-179
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Platinum 8457C
CPU family: 6
Model: 143
Thread(s) per core: 2
Core(s) per socket: 45
Socket(s): 2
Stepping: 8
BogoMIPS: 5200.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology nonstop_tsc cpuid tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch cpuid_fault invpcid_single ssbd ibrs ibpb stibp ibrs_enhanced fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd arat avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid cldemote movdiri movdir64b fsrm md_clear serialize tsxldtrk arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 4.2 MiB (90 instances)
L1i cache: 2.8 MiB (90 instances)
L2 cache: 180 MiB (90 instances)
L3 cache: 195 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-89
NUMA node1 CPU(s): 90-179
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Unknown: No mitigations
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI SW loop, KVM SW loop
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Mitigation; TSX disabled
Versions of relevant libraries:
[pip3] flashinfer-python==0.2.1.post2+cu124torch2.6
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.4.5.8
[pip3] nvidia-cuda-cupti-cu12==12.4.127
[pip3] nvidia-cuda-nvrtc-cu12==12.4.127
[pip3] nvidia-cuda-runtime-cu12==12.4.127
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cufft-cu12==11.2.1.3
[pip3] nvidia-curand-cu12==10.3.5.147
[pip3] nvidia-cusolver-cu12==11.6.1.9
[pip3] nvidia-cusparse-cu12==12.3.1.170
[pip3] nvidia-cusparselt-cu12==0.6.2
[pip3] nvidia-nccl-cu12==2.21.5
[pip3] nvidia-nvjitlink-cu12==12.4.127
[pip3] nvidia-nvtx-cu12==12.4.127
[pip3] pyzmq==26.3.0
[pip3] torch==2.6.0
[pip3] torchaudio==2.6.0
[pip3] torchvision==0.21.0
[pip3] transformers==4.49.0
[pip3] triton==3.2.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.8.0
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 NIC0 NIC1 NIC2 NIC3 NIC4 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X NV18 NV18 NV18 NV18 NV18 NV18 NV18 SYS PIX NODE SYS SYS 0-89 0 N/A
GPU1 NV18 X NV18 NV18 NV18 NV18 NV18 NV18 SYS PIX NODE SYS SYS 0-89 0 N/A
GPU2 NV18 NV18 X NV18 NV18 NV18 NV18 NV18 SYS NODE PIX SYS SYS 0-89 0 N/A
GPU3 NV18 NV18 NV18 X NV18 NV18 NV18 NV18 SYS NODE PIX SYS SYS 0-89 0 N/A
GPU4 NV18 NV18 NV18 NV18 X NV18 NV18 NV18 SYS SYS SYS PIX NODE 90-179 1 N/A
GPU5 NV18 NV18 NV18 NV18 NV18 X NV18 NV18 SYS SYS SYS PIX NODE 90-179 1 N/A
GPU6 NV18 NV18 NV18 NV18 NV18 NV18 X NV18 SYS SYS SYS NODE PIX 90-179 1 N/A
GPU7 NV18 NV18 NV18 NV18 NV18 NV18 NV18 X SYS SYS SYS NODE PIX 90-179 1 N/A
NIC0 SYS SYS SYS SYS SYS SYS SYS SYS X SYS SYS SYS SYS
NIC1 PIX PIX NODE NODE SYS SYS SYS SYS SYS X NODE SYS SYS
NIC2 NODE NODE PIX PIX SYS SYS SYS SYS SYS NODE X SYS SYS
NIC3 SYS SYS SYS SYS PIX PIX NODE NODE SYS SYS SYS X NODE
NIC4 SYS SYS SYS SYS NODE NODE PIX PIX SYS SYS SYS 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
NVIDIA_VISIBLE_DEVICES=all
NVIDIA_REQUIRE_CUDA=cuda>=12.4 brand=tesla,driver>=470,driver<471 brand=unknown,driver>=470,driver<471 brand=nvidia,driver>=470,driver<471 brand=nvidiartx,driver>=470,driver<471 brand=geforce,driver>=470,driver<471 brand=geforcertx,driver>=470,driver<471 brand=quadro,driver>=470,driver<471 brand=quadrortx,driver>=470,driver<471 brand=titan,driver>=470,driver<471 brand=titanrtx,driver>=470,driver<471 brand=tesla,driver>=525,driver<526 brand=unknown,driver>=525,driver<526 brand=nvidia,driver>=525,driver<526 brand=nvidiartx,driver>=525,driver<526 brand=geforce,driver>=525,driver<526 brand=geforcertx,driver>=525,driver<526 brand=quadro,driver>=525,driver<526 brand=quadrortx,driver>=525,driver<526 brand=titan,driver>=525,driver<526 brand=titanrtx,driver>=525,driver<526 brand=tesla,driver>=535,driver<536 brand=unknown,driver>=535,driver<536 brand=nvidia,driver>=535,driver<536 brand=nvidiartx,driver>=535,driver<536 brand=geforce,driver>=535,driver<536 brand=geforcertx,driver>=535,driver<536 brand=quadro,driver>=535,driver<536 brand=quadrortx,driver>=535,driver<536 brand=titan,driver>=535,driver<536 brand=titanrtx,driver>=535,driver<536
NCCL_VERSION=2.20.5-1
NCCL_SOCKET_IFNAME=eth0
NVIDIA_DRIVER_CAPABILITIES=compute,utility
NCCL_IB_HCA=mlx5
NVIDIA_PRODUCT_NAME=CUDA
VLLM_USAGE_SOURCE=production-docker-image
NCCL_IB_GID_INDEX=3
CUDA_VERSION=12.4.0
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
LD_LIBRARY_PATH=/opt/venv/lib/python3.12/site-packages/cv2/../../lib64:/usr/local/nvidia/lib:/usr/local/nvidia/lib64
NCCL_IB_DISABLE=0
VLLM_HOST_IP=10.99.48.142
NCCL_CUMEM_ENABLE=0
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY
š Describe the bug
Firstly I follow the doc https://docs.vllm.ai/en/latest/serving/distributed_serving.html to setup the distributed environment(2 nodes with 8 GPUs per node), and then run the api_server as below:
python3 -m vllm.entrypoints.openai.api_server --port 18011 --model /models/DeepSeek-V3 --tensor-parallel-size 16 --gpu-memory-utilization 0.92 --dtype auto --served-model-name deepseekv3 --max-num-seqs 50 --max-model-len 16384 --trust-remote-code --disable-log-requests --enable-chunked-prefill --enable-prefix-caching
Then I got the RayChannelTimeoutError in Ray module within the call execute_model to run ray dag.
INFO 03-14 00:00:55 [async_llm.py:169] Added request cmpl-49612d570051487899170dc9fc843162-0.
INFO 03-14 00:00:59 [loggers.py:80] Avg prompt throughput: 102.5 tokens/s, Avg generation throughput: 0.1 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.2%, Prefix cache hit rate: 13.4%
ERROR 03-14 00:01:05 [core.py:337] EngineCore hit an exception: Traceback (most recent call last):
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/ray/dag/compiled_dag_node.py", line 2344, in _execute_until
ERROR 03-14 00:01:05 [core.py:337] result = self._dag_output_fetcher.read(timeout)
ERROR 03-14 00:01:05 [core.py:337] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/ray/experimental/channel/common.py", line 318, in read
ERROR 03-14 00:01:05 [core.py:337] outputs = self._read_list(timeout)
ERROR 03-14 00:01:05 [core.py:337] ^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/ray/experimental/channel/common.py", line 409, in _read_list
ERROR 03-14 00:01:05 [core.py:337] raise e
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/ray/experimental/channel/common.py", line 391, in _read_list
ERROR 03-14 00:01:05 [core.py:337] result = c.read(min(remaining_timeout, iteration_timeout))
ERROR 03-14 00:01:05 [core.py:337] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/ray/experimental/channel/shared_memory_channel.py", line 776, in read
ERROR 03-14 00:01:05 [core.py:337] return self._channel_dict[self._resolve_actor_id()].read(timeout)
ERROR 03-14 00:01:05 [core.py:337] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/ray/experimental/channel/shared_memory_channel.py", line 480, in read
ERROR 03-14 00:01:05 [core.py:337] ret = self._worker.get_objects(
ERROR 03-14 00:01:05 [core.py:337] ^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/ray/_private/worker.py", line 893, in get_objects
ERROR 03-14 00:01:05 [core.py:337] ] = self.core_worker.get_objects(
ERROR 03-14 00:01:05 [core.py:337] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 03-14 00:01:05 [core.py:337] File "python/ray/_raylet.pyx", line 3189, in ray._raylet.CoreWorker.get_objects
ERROR 03-14 00:01:05 [core.py:337] File "python/ray/includes/common.pxi", line 106, in ray._raylet.check_status
ERROR 03-14 00:01:05 [core.py:337] ray.exceptions.RayChannelTimeoutError: System error: Timed out waiting for object available to read. ObjectID: 00d95966d8a9e2f5795e7e010e186d6a031a70380100000002e1f505
ERROR 03-14 00:01:05 [core.py:337]
ERROR 03-14 00:01:05 [core.py:337] The above exception was the direct cause of the following exception:
ERROR 03-14 00:01:05 [core.py:337]
ERROR 03-14 00:01:05 [core.py:337] Traceback (most recent call last):
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/vllm/v1/engine/core.py", line 330, in run_engine_core
ERROR 03-14 00:01:05 [core.py:337] engine_core.run_busy_loop()
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/vllm/v1/engine/core.py", line 364, in run_busy_loop
ERROR 03-14 00:01:05 [core.py:337] outputs = step_fn()
ERROR 03-14 00:01:05 [core.py:337] ^^^^^^^^^
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/vllm/v1/engine/core.py", line 192, in step
ERROR 03-14 00:01:05 [core.py:337] output = self.model_executor.execute_model(scheduler_output)
ERROR 03-14 00:01:05 [core.py:337] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/vllm/v1/executor/ray_distributed_executor.py", line 57, in execute_model
ERROR 03-14 00:01:05 [core.py:337] return refs[0].get()
ERROR 03-14 00:01:05 [core.py:337] ^^^^^^^^^^^^^
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/ray/experimental/compiled_dag_ref.py", line 124, in get
ERROR 03-14 00:01:05 [core.py:337] self._dag._execute_until(
ERROR 03-14 00:01:05 [core.py:337] File "/usr/local/lib/python3.12/dist-packages/ray/dag/compiled_dag_node.py", line 2350, in _execute_until
ERROR 03-14 00:01:05 [core.py:337] raise RayChannelTimeoutError(
ERROR 03-14 00:01:05 [core.py:337] ray.exceptions.RayChannelTimeoutError: System error: If the execution is expected to take a long time, increase RAY_CGRAPH_get_timeout which is currently 10 seconds. Otherwise, this may indicate that the execution is hanging.
ERROR 03-14 00:01:05 [core.py:337]
INFO 03-14 00:01:05 [ray_distributed_executor.py:127] Shutting down Ray distributed executor. If you see error log from logging.cc regarding SIGTERM received, please ignore because this is the expected termination process in Ray.
CRITICAL 03-14 00:01:05 [core_client.py:260] Got fatal signal from worker processes, shutting down. See stack trace above for root cause issue.
2025-03-14 00:01:05,920 INFO compiled_dag_node.py:2109 -- Tearing down compiled DAG
Before submitting a new issueā¦
- Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.
About this issue
- Original URL
- State: open
- Created 2 years ago
- Reactions: 2
- Comments: 36 (12 by maintainers)
After more investigation, the problem seems to be the use of torch.compile with
tensor_model_parallel_all_reduce.My guess is based on the following two observations:
tensor_model_parallel_all_reduce, the hang of DeepSeek-R1 would be fixed.RowParallelLinear, can also fix the bug in our environment. Since all_reduce communication ops are used in RowParallelLinear, https://github.com/vllm-project/vllm/pull/17484 actually removes the torch.compile oftensor_model_parallel_all_reduce.After further debugging, sometimes it hangs at the linear operatorļ¼
gdb stack:
pystack:
sometimes hangs at MemcpyAsync:
sometimes hang at cudaGraphLaunch:
It feels like something is blocking with stream.
Through my somewhat clumsy process of adding logs step by step (I had previously tried nsys and torch_profile, but neither provided the information I needed), I discovered that the timeout occurs when it gets stuck at this bellow location. Setting
VLLM_USE_FLASHINFER_SAMPLER=0allows me to circumvent this issue. Currently, V0 does not use Flashinfer by default, while V1 uses Flashinfer by default.Therefore, I suspect that the issue is caused by the Flashinfer library, and I still need to further investigate and pinpoint the problem.
When running with
--enforce-eager, it runs fine for TP=16. Looks like it may be related to cuda graph. cc: @youkaichaoThey are environment variables. Use export or set them when you start vllm.
@youkaichao Could you help take a look at this issue? Iām ready to provide any additional debugging information needed at any time. Thanks so much.
The deeper underlying cause has not yet been identified and is still being pinpointed. Additionally, this setting will result in a performance decrease of approximately 7%.
I just ran a test, and the combination of V1 + TP=8 + PP=2 does not exhibit this issue.
Yes, the default is now V1. If using V0 with the same parameters, no issues were encountered.
I will test the scenario with V1 and TP=8, PP=2. Based on previous tests, this combination is expected to perform worse than TP=16.
Here is the logs, it essentially occurs at the same location as what @jeffye-dev encountered.:
Yes, this is the average outcome; there are inputs ranging from tens of thousands to just a few hundred.
Hi @longqu , can you increase the
RAY_CGRAPH_submit_timeoutandRAY_CGRAPH_get_timeout? The default timeout right now (10s) may not have the best default value, and we will adjust it in the future.