[Bug]: crash on startup with TP>1 on H100 nodes: AttributeError in ShmRingBuffer.shared_memory

[sanderland]

Your current environment

The output of python collect_env.py

==============================
System Info

OS : Ubuntu 22.04.5 LTS (x86_64)
GCC version : (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version : Could not collect
CMake version : version 3.22.1
Libc version : glibc-2.35

==============================
PyTorch Info

PyTorch version : 2.7.1+cu126
Is debug build : False
CUDA used to build PyTorch : 12.6
ROCM used to build PyTorch : N/A

==============================
Python Environment

Python version : 3.11.13 (main, Aug 28 2025, 17:07:15) [Clang 20.1.4 ] (64-bit runtime)
Python platform : Linux-6.8.0-1033-aws-x86_64-with-glibc2.35

==============================
CUDA / GPU Info

Is CUDA available : True
CUDA runtime version : 12.8.93
CUDA_MODULE_LOADING set to : LAZY
GPU models and configuration :
GPU 0: NVIDIA H100 80GB HBM3
GPU 1: NVIDIA H100 80GB HBM3
GPU 2: NVIDIA H100 80GB HBM3
GPU 3: NVIDIA H100 80GB HBM3
GPU 4: NVIDIA H100 80GB HBM3
GPU 5: NVIDIA H100 80GB HBM3
GPU 6: NVIDIA H100 80GB HBM3
GPU 7: NVIDIA H100 80GB HBM3

Nvidia driver version : 570.172.08
cuDNN version : Could not collect
HIP runtime version : N/A
MIOpen runtime version : N/A
Is XNNPACK available : True

==============================
CPU Info

Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 48 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 96
On-line CPU(s) list: 0-95
Vendor ID: AuthenticAMD
Model name: AMD EPYC 7R13 Processor
CPU family: 25
Model: 1
Thread(s) per core: 1
Core(s) per socket: 48
Socket(s): 2
Stepping: 1
BogoMIPS: 5300.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch topoext perfctr_core ssbd ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 invpcid rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 clzero xsaveerptr rdpru wbnoinvd arat npt nrip_save vaes vpclmulqdq rdpid
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 3 MiB (96 instances)
L1i cache: 3 MiB (96 instances)
L2 cache: 48 MiB (96 instances)
L3 cache: 384 MiB (12 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-47
NUMA node1 CPU(s): 48-95
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: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Mitigation; Safe RET
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines; IBPB conditional; IBRS_FW; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected

==============================
Versions of relevant libraries

[pip3] flashinfer-python==0.3.1
[pip3] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.6.4.1
[pip3] nvidia-cuda-cupti-cu12==12.6.80
[pip3] nvidia-cuda-nvrtc-cu12==12.6.77
[pip3] nvidia-cuda-runtime-cu12==12.6.77
[pip3] nvidia-cudnn-cu12==9.5.1.17
[pip3] nvidia-cudnn-frontend==1.14.1
[pip3] nvidia-cufft-cu12==11.3.0.4
[pip3] nvidia-cufile-cu12==1.11.1.6
[pip3] nvidia-curand-cu12==10.3.7.77
[pip3] nvidia-cusolver-cu12==11.7.1.2
[pip3] nvidia-cusparse-cu12==12.5.4.2
[pip3] nvidia-cusparselt-cu12==0.6.3
[pip3] nvidia-ml-py==12.575.51
[pip3] nvidia-nccl-cu12==2.26.2
[pip3] nvidia-nvjitlink-cu12==12.6.85
[pip3] nvidia-nvtx-cu12==12.6.77
[pip3] pynvml==12.0.0
[pip3] pyzmq==27.0.2
[pip3] torch==2.7.1
[pip3] torchaudio==2.7.1
[pip3] torchvision==0.22.1
[pip3] transformers==4.56.1
[pip3] triton==3.3.1
[conda] Could not collect

==============================
vLLM Info

ROCM Version : Could not collect
vLLM Version : 0.10.1.1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled
GPU Topology:
�[4mGPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 CPU Affinity NUMA Affinity GPU NUMA ID�[0m
GPU0 X NV18 NV18 NV18 NV18 NV18 NV18 NV18 0-15 0 N/A
GPU1 NV18 X NV18 NV18 NV18 NV18 NV18 NV18 0-15 0 N/A
GPU2 NV18 NV18 X NV18 NV18 NV18 NV18 NV18 0-15 0 N/A
GPU3 NV18 NV18 NV18 X NV18 NV18 NV18 NV18 0-15 0 N/A
GPU4 NV18 NV18 NV18 NV18 X NV18 NV18 NV18 48-63 1 N/A
GPU5 NV18 NV18 NV18 NV18 NV18 X NV18 NV18 48-63 1 N/A
GPU6 NV18 NV18 NV18 NV18 NV18 NV18 X NV18 48-63 1 N/A
GPU7 NV18 NV18 NV18 NV18 NV18 NV18 NV18 X 48-63 1 N/A

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

==============================
Environment Variables

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=/usr/local/cuda-12.8/lib:/usr/local/cuda-12.8/lib64:/usr/local/cuda-12.8:/usr/local/cuda-12.8/targets/x86_64-linux/lib/:/usr/local/cuda-12.8/extras/CUPTI/lib64:/opt/amazon/efa/lib:/opt/amazon/openmpi/lib:/opt/amazon/ofi-nccl/lib/x86_64-linux-gnu:/usr/local/lib:/usr/lib:/usr/local/cuda-12.8/lib:/usr/local/cuda-12.8/lib64:/usr/local/cuda-12.8:/usr/local/cuda-12.8/targets/x86_64-linux/lib/:/usr/local/cuda-12.8/extras/CUPTI/lib64:/opt/amazon/efa/lib:/opt/amazon/openmpi/lib:/opt/amazon/ofi-nccl/lib/x86_64-linux-gnu:/usr/local/lib:/usr/lib:/usr/local/cuda-12.8/lib:/usr/local/cuda-12.8/lib64:/usr/local/cuda-12.8:/usr/local/cuda-12.8/targets/x86_64-linux/lib/:/usr/local/cuda-12.8/extras/CUPTI/lib64:/opt/amazon/efa/lib:/opt/amazon/openmpi/lib:/opt/amazon/ofi-nccl/lib/x86_64-linux-gnu:/usr/local/lib:/usr/lib:/usr/local/cuda-12.8/lib:/usr/local/cuda-12.8/lib64:/usr/local/cuda-12.8:/usr/local/cuda-12.8/targets/x86_64-linux/lib/:/usr/local/cuda-12.8/extras/CUPTI/lib64:/opt/amazon/efa/lib:/opt/amazon/openmpi/lib:/opt/amazon/ofi-nccl/lib/x86_64-linux-gnu:/usr/local/lib:/usr/lib
NCCL_CUMEM_ENABLE=0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
CUDA_MODULE_LOADING=LAZY

🐛 Describe the bug

When running vLLM with tensor-parallel size > 1 on nodes with multiple H100s, some models reliably crash during engine startup with:

AttributeError: 'ShmRingBuffer' object has no attribute 'shared_memory'

This appears to stem from the multiprocessing message-queue broadcaster using POSIX shared memory. Logs show reader workers failing to attach to the shared memory segment created by the writer; later the reader crashes while accessing self.shared_memory.

Affected models

• Consistent repro: zai-org/GLM-4.5-Air (BF16) with --tensor-parallel-size 8
• Also seen previously: Qwen/QwQ-32B with --tensor-parallel-size 4

Minimal repro (GLM-4.5-Air BF16)

• Slurm command (debug partition, 8 GPUs):

srun --partition=debug --nodes=1 --ntasks=1 --gres=gpu:8 --cpus-per-task=16 --mem=128G --time=40 \
  bash -lc 'cd /fsx/sander/project; \
    PORT=$(shuf -i 20000-29999 -n1); \
    timeout 1200 uv run vllm serve zai-org/GLM-4.5-Air \
      --port $PORT \
      --tensor-parallel-size 8 \
      --max-model-len 65536 \
      --gpu-memory-utilization 0.9 \
      --reasoning-parser glm45 \
      --tool-call-parser glm45 \
      --enable-auto-tool-choice \
      --max-num-batched-tokens 32768'

Result: crash during engine initialization with SHM error.

  File ".../vllm/distributed/device_communicators/shm_broadcast.py", line 507, in dequeue
    with self.acquire_read(timeout, cancel) as buf:
  File ".../contextlib.py", line 137, in __enter__
    return next(self.gen)
  File ".../vllm/distributed/device_communicators/shm_broadcast.py", line 438, in acquire_read
    with self.buffer.get_metadata(self.current_idx) as metadata_buffer:
  File ".../contextlib.py", line 137, in __enter__
    return next(self.gen)
  File ".../vllm/distributed/device_communicators/shm_broadcast.py", line 192, in get_metadata
    with memoryview(self.shared_memory.buf[start:end]) as buf:
AttributeError: 'ShmRingBuffer' object has no attribute 'shared_memory'

Accompanying resource tracker messages indicating missing SHM segments:

resource_tracker: '/psm_06b11115': [Errno 2] No such file or directory
resource_tracker: '/psm_96d4380e': [Errno 2] No such file or directory
resource_tracker: '/psm_c8893ede': [Errno 2] No such file or directory
...

Suspected cause

Race or env-specific failure in creating/propagating the POSIX SHM name across TP worker processes on H100 nodes. The code suppresses FileNotFoundError when deserializing a buffer in a different-node case, leading to a partially initialized ShmRingBuffer instance without shared_memory, which later triggers the AttributeError.

Questions

• Any recommended workarounds?
• Could ShmRingBuffer fail fast with a helpful error message instead?

Before submitting a new issue...

• 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.

[njhill]

Thanks @sanderland... would you be interested in opening a PR to address the fail-fast thing?

[jakeyap]

+1, encountered same issue with Qwen/Qwen3-235B-A22B-Instruct-2507
Not consistently reproducible, some servers show this error message, others do not. Nominally all the servers have the same environment

[sanderland]

Thanks @sanderland... would you be interested in opening a PR to address the fail-fast thing?

I would not be confident doing this without understanding the cause and any workarounds.