vllm - 💡(How to fix) Fix [Bug]: Qwen3.5-122B-A10B Engine hangs at Prefill phase under high concurrency (40 reqs) with multi-node PP=2 on dual 4090D nodes [1 participants]

vllm2026-04-16 07:57:02

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

vllm-project/vllm#39985•Fetched 2026-04-17 08:27:57

View on GitHub

Comments

Participants

Timeline

Reactions

Author

ShuZihan

Participants

ShuZihan

Timeline (top)

labeled ×1

Code Example

Collecting environment information...
==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0
Clang version                : Could not collect
CMake version                : Could not collect
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.10.0+cu129
Is debug build               : False
CUDA used to build PyTorch   : 12.9
ROCM used to build PyTorch   : N/A
XPU used to build PyTorch    : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.13 (main, Mar  4 2026, 09:23:07) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-4.18.0-372.32.1.90.po1.x86_64-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 12.9.86
CUDA_MODULE_LOADING set to   :
GPU models and configuration :
GPU 0: NVIDIA GeForce RTX 4090 D
GPU 1: NVIDIA GeForce RTX 4090 D
GPU 2: NVIDIA GeForce RTX 4090 D
GPU 3: NVIDIA GeForce RTX 4090 D
GPU 4: NVIDIA GeForce RTX 4090 D
GPU 5: NVIDIA GeForce RTX 4090 D
GPU 6: NVIDIA GeForce RTX 4090 D
GPU 7: NVIDIA GeForce RTX 4090 D

Nvidia driver version        : 580.105.08
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:                   52 bits physical, 57 bits virtual
Byte Order:                      Little Endian
CPU(s):                          224
On-line CPU(s) list:             0-223
Vendor ID:                       AuthenticAMD
BIOS Vendor ID:                  Advanced Micro Devices, Inc.
Model name:                      AMD EPYC 9654 96-Core Processor
BIOS Model name:                 AMD EPYC 9654 96-Core Processor
CPU family:                      25
Model:                           17
Thread(s) per core:              2
Core(s) per socket:              56
Socket(s):                       2
Stepping:                        1
Frequency boost:                 enabled
CPU max MHz:                     3707.8120
CPU min MHz:                     1500.0000
BogoMIPS:                        4800.50
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 pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 invpcid_single hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local avx512_bf16 clzero irperf xsaveerptr wbnoinvd amd_ppin arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid overflow_recov succor smca fsrm flush_l1d
Virtualization:                  AMD-V
L1d cache:                       3.5 MiB (112 instances)
L1i cache:                       3.5 MiB (112 instances)
L2 cache:                        112 MiB (112 instances)
L3 cache:                        512 MiB (16 instances

---

# ========== Leader Node (Rank 0) ==========
NCCL_IB_DISABLE=1 GLOO_SOCKET_IFNAME=eth0 NCCL_SOCKET_IFNAME=eth0 VLLM_HOST_IP=<LEADER_POD_IP> \
vllm serve /path/to/Qwen3---5-122B-A10B-V1 \
  --port 31111 \
  --tensor-parallel-size 8 \
  --pipeline-parallel-size 2 \
  --trust-remote-code \
  --served-model-name Qwen3.5-122B-A10B \
  --gpu-memory-utilization 0.9 \
  --max-model-len 16384 \
  --enforce-eager \
  --nnodes 2 \
  --node-rank 0 \
  --master-addr <LEADER_POD_IP>

# ========== Worker Node (Rank 1) ==========
NCCL_IB_DISABLE=1 GLOO_SOCKET_IFNAME=eth0 NCCL_SOCKET_IFNAME=eth0 VLLM_HOST_IP=<WORKER_POD_IP> \
vllm serve /path/to/Qwen3---5-122B-A10B-V1 \
  --tensor-parallel-size 8 \
  --pipeline-parallel-size 2 \
  --trust-remote-code \
  --served-model-name Qwen3.5-122B-A10B \
  --gpu-memory-utilization 0.9 \
  --max-model-len 16384 \
  --enforce-eager \
  --nnodes 2 \
  --node-rank 1 \
  --master-addr <LEADER_POD_IP> \
  --headless

RAW_BUFFERClick to expand / collapse

Your current environment

<details> <summary>The output of <code>python collect_env.py</code></summary>

Collecting environment information...
==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0
Clang version                : Could not collect
CMake version                : Could not collect
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.10.0+cu129
Is debug build               : False
CUDA used to build PyTorch   : 12.9
ROCM used to build PyTorch   : N/A
XPU used to build PyTorch    : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.13 (main, Mar  4 2026, 09:23:07) [GCC 11.4.0] (64-bit runtime)
Python platform              : Linux-4.18.0-372.32.1.90.po1.x86_64-x86_64-with-glibc2.35

==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 12.9.86
CUDA_MODULE_LOADING set to   :
GPU models and configuration :
GPU 0: NVIDIA GeForce RTX 4090 D
GPU 1: NVIDIA GeForce RTX 4090 D
GPU 2: NVIDIA GeForce RTX 4090 D
GPU 3: NVIDIA GeForce RTX 4090 D
GPU 4: NVIDIA GeForce RTX 4090 D
GPU 5: NVIDIA GeForce RTX 4090 D
GPU 6: NVIDIA GeForce RTX 4090 D
GPU 7: NVIDIA GeForce RTX 4090 D

Nvidia driver version        : 580.105.08
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:                   52 bits physical, 57 bits virtual
Byte Order:                      Little Endian
CPU(s):                          224
On-line CPU(s) list:             0-223
Vendor ID:                       AuthenticAMD
BIOS Vendor ID:                  Advanced Micro Devices, Inc.
Model name:                      AMD EPYC 9654 96-Core Processor
BIOS Model name:                 AMD EPYC 9654 96-Core Processor
CPU family:                      25
Model:                           17
Thread(s) per core:              2
Core(s) per socket:              56
Socket(s):                       2
Stepping:                        1
Frequency boost:                 enabled
CPU max MHz:                     3707.8120
CPU min MHz:                     1500.0000
BogoMIPS:                        4800.50
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 pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 invpcid_single hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local avx512_bf16 clzero irperf xsaveerptr wbnoinvd amd_ppin arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq la57 rdpid overflow_recov succor smca fsrm flush_l1d
Virtualization:                  AMD-V
L1d cache:                       3.5 MiB (112 instances)
L1i cache:                       3.5 MiB (112 instances)
L2 cache:                        112 MiB (112 instances)
L3 cache:                        512 MiB (16 instances

</details>

🐛 Describe the bug

Description: I am running a multi-node vLLM setup using two RTX 4090D nodes (8 GPUs each, total 16 GPUs) serving the Qwen3.5-122B-A10B model. The cluster is configured with tensor-parallel-size=8 and pipeline-parallel-size=2 over standard Ethernet (NCCL_IB_DISABLE=1).

When I send a single streaming request, the engine processes it perfectly and streams the output as expected. However, when I send 40 concurrent requests, the engine completely hangs/freezes during the Prefill phase. No tokens are generated, and the system deadlocks. During this hang, I observed an abnormally high CPU utilization (hovering around 200%) on the host machine.

Commands used to start the engine:

# ========== Leader Node (Rank 0) ==========
NCCL_IB_DISABLE=1 GLOO_SOCKET_IFNAME=eth0 NCCL_SOCKET_IFNAME=eth0 VLLM_HOST_IP=<LEADER_POD_IP> \
vllm serve /path/to/Qwen3---5-122B-A10B-V1 \
  --port 31111 \
  --tensor-parallel-size 8 \
  --pipeline-parallel-size 2 \
  --trust-remote-code \
  --served-model-name Qwen3.5-122B-A10B \
  --gpu-memory-utilization 0.9 \
  --max-model-len 16384 \
  --enforce-eager \
  --nnodes 2 \
  --node-rank 0 \
  --master-addr <LEADER_POD_IP>

# ========== Worker Node (Rank 1) ==========
NCCL_IB_DISABLE=1 GLOO_SOCKET_IFNAME=eth0 NCCL_SOCKET_IFNAME=eth0 VLLM_HOST_IP=<WORKER_POD_IP> \
vllm serve /path/to/Qwen3---5-122B-A10B-V1 \
  --tensor-parallel-size 8 \
  --pipeline-parallel-size 2 \
  --trust-remote-code \
  --served-model-name Qwen3.5-122B-A10B \
  --gpu-memory-utilization 0.9 \
  --max-model-len 16384 \
  --enforce-eager \
  --nnodes 2 \
  --node-rank 1 \
  --master-addr <LEADER_POD_IP> \
  --headless

Observed Results:

1 curl request: Works fine.
40 concurrent requests: The Python client hangs indefinitely waiting for the first chunk. The vLLM server logs stop updating, and the CPU utilization abnormally spikes and stays at ~200%.

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.

extent analysis

TL;DR

The issue can be resolved by adjusting the configuration to reduce CPU utilization and improve concurrency handling, potentially by optimizing the tensor-parallel-size and pipeline-parallel-size parameters or exploring other concurrency control mechanisms.

Guidance

Review the current configuration of tensor-parallel-size=8 and pipeline-parallel-size=2 to determine if these settings are optimal for the given hardware and workload, considering the high CPU utilization observed.
Consider implementing concurrency control mechanisms to manage the 40 concurrent requests more efficiently, potentially through batching or rate limiting, to prevent overloading the system.
Investigate the --enforce-eager flag's impact on performance under high concurrency, as it might be contributing to the increased CPU utilization by forcing eager execution.
Monitor system resources closely during testing to identify any bottlenecks or areas for optimization, focusing on CPU, memory, and network utilization.

Example

No specific code example is provided due to the complexity of the issue and the need for a more detailed analysis of the system's configuration and performance metrics.

Notes

The provided information suggests a performance and concurrency issue rather than a straightforward bug. The solution may involve a combination of configuration adjustments, code optimizations, and potentially exploring different concurrency models to efficiently handle the workload.

Recommendation

Apply a workaround by adjusting the concurrency settings and monitoring system performance to find an optimal balance between throughput and resource utilization, as upgrading to a fixed version is not explicitly implied in the given context.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#api #prompt issue #agent setup #task chaining #parallel task

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Data

Security

Network

Code

UI/UX

Text

System

Multimedia

Protocol

API

Engineering

vllm - 💡(How to fix) Fix [Bug]: Qwen3.5-122B-A10B Engine hangs at Prefill phase under high concurrency (40 reqs) with multi-node PP=2 on dual 4090D nodes [1 participants]

Recommended Tools

GitHub issue graph ai analysis

Code Example

Your current environment

🐛 Describe the bug

Before submitting a new issue...

extent analysis

TL;DR

Guidance

Example

Notes

Recommendation

Still need to ship something?

TRENDING

vllm - 💡(How to fix) Fix [Bug]: Qwen3.5-122B-A10B Engine hangs at Prefill phase under high concurrency (40 reqs) with multi-node PP=2 on dual 4090D nodes [1 participants]

Recommended Tools

GitHub issue graph ai analysis

Code Example

Your current environment

🐛 Describe the bug

Before submitting a new issue...

extent analysis

TL;DR

Guidance

Example

Notes

Recommendation

Still need to ship something?

RELATED_DISCOVERY

TRENDING