Issue Station
Sign In
Tools hereToolbox

vllm - 💡(How to fix) Fix [Bug]: Async double streaming_update can break the request ID -> previous batch row mapping and crash the engine

Official PRs (…)
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.

Helpful · Quick feedback

Loading…

I found a double-streaming_update bug in the AsyncLLM path that can break the worker's request ID -> previous batch row mapping for a resumed request and then kill the engine.

The reproducer submits one resumable request and then applies two legal streaming_update operations to that same live request while async scheduling and speculative decoding are both active.

Under that combination, the scheduler can continue issuing speculative decode positions for the resumed request, but the worker-side request ID -> previous batch row mapping no longer matches the live request state. The verified AsyncLLM runs then reach a worker update step where the request id is missing from prev_req_id_to_index, and the engine dies with:

  • scheduled_spec_decode_tokens={...: [-1, ...]}
  • KeyError: 'req2-...'
  • EngineDeadError

This is not the same bug as:

  • the async same-request-id reuse underflow bug
  • the resumed prompt-width overflow bug

Error Message

KeyError: 'req2-...'

Root Cause

This is a resumed-request / previous-batch-row state reconstruction bug, not a malformed request bug.

The important distinction is:

  • both streaming_updates are legal
  • the same live request is resumed twice
  • async speculative state persists between those updates
  • the request ID -> previous batch row mapping no longer matches the live request state

So the real problem is that async speculative state reconstruction can continue after double streaming_update even when the request ID -> previous batch row mapping no longer matches the live request state, and that mismatch then kills the engine.

Fix Action

Fix / Workaround

============================== 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): 384 On-line CPU(s) list: 0-383 Vendor ID: AuthenticAMD Model name: AMD EPYC 9654 96-Core Processor CPU family: 25 Model: 17 Thread(s) per core: 2 Core(s) per socket: 96 Socket(s): 2 Stepping: 1 Frequency boost: enabled CPU max MHz: 3707.8120 CPU min MHz: 1500.0000 BogoMIPS: 4792.57 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 amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic 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 perfmon_v2 ibrs ibpb stibp ibrs_enhanced 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 rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi 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: 6 MiB (192 instances) L1i cache: 6 MiB (192 instances) L2 cache: 192 MiB (192 instances) L3 cache: 768 MiB (24 instances) NUMA node(s): 2 NUMA node0 CPU(s): 0-95,192-287 NUMA node1 CPU(s): 96-191,288-383 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 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; Enhanced / Automatic IBRS; IBPB conditional; STIBP always-on; RSB filling; PBRSB-eIBRS Not affected; BHI Not affected Vulnerability Srbds: Not affected Vulnerability Tsx async abort: Not affected

Code Example

Collecting environment information...
==============================
        System Info
==============================
OS                           : Ubuntu 22.04.5 LTS (x86_64)
GCC version                  : (Ubuntu 13.1.0-8ubuntu1~22.04) 13.1.0
Clang version                : 16.0.6 (++20231112100510+7cbf1a259152-1~exp1~20231112100554.106)
CMake version                : version 3.22.1
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.10.0+cu128
Is debug build               : False
CUDA used to build PyTorch   : 12.8
ROCM used to build PyTorch   : N/A
XPU used to build PyTorch    : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.9 | packaged by Anaconda, Inc. | (main, Feb  6 2025, 18:56:27) [GCC 11.2.0] (64-bit runtime)
Python platform              : Linux-6.5.0-35-generic-x86_64-with-glibc2.35
    
==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 12.8.61
CUDA_MODULE_LOADING set to   : 
GPU models and configuration : 
GPU 0: NVIDIA GeForce RTX 4090
GPU 1: NVIDIA GeForce RTX 4090

Nvidia driver version        : 570.86.10
cuDNN version                : Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.3.0
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):                             384
On-line CPU(s) list:                0-383
Vendor ID:                          AuthenticAMD
Model name:                         AMD EPYC 9654 96-Core Processor
CPU family:                         25
Model:                              17
Thread(s) per core:                 2
Core(s) per socket:                 96
Socket(s):                          2
Stepping:                           1
Frequency boost:                    enabled
CPU max MHz:                        3707.8120
CPU min MHz:                        1500.0000
BogoMIPS:                           4792.57
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 amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic 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 perfmon_v2 ibrs ibpb stibp ibrs_enhanced 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 rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi 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:                          6 MiB (192 instances)
L1i cache:                          6 MiB (192 instances)
L2 cache:                           192 MiB (192 instances)
L3 cache:                           768 MiB (24 instances)
NUMA node(s):                       2
NUMA node0 CPU(s):                  0-95,192-287
NUMA node1 CPU(s):                  96-191,288-383
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 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; Enhanced / Automatic IBRS; IBPB conditional; 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==0.6.4
[pip3] numpy==2.0.2
[pip3] nvidia-cublas-cu12==12.8.4.1
[pip3] nvidia-cuda-cupti-cu12==12.8.90
[pip3] nvidia-cuda-nvrtc-cu12==12.8.93
[pip3] nvidia-cuda-runtime-cu12==12.8.90
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cudnn-frontend==1.18.0
[pip3] nvidia-cufft-cu12==11.3.3.83
[pip3] nvidia-cufile-cu12==1.13.1.3
[pip3] nvidia-curand-cu12==10.3.9.90
[pip3] nvidia-cusolver-cu12==11.7.3.90
[pip3] nvidia-cusparse-cu12==12.5.8.93
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-cutlass-dsl==4.4.2
[pip3] nvidia-cutlass-dsl-libs-base==4.4.2
[pip3] nvidia-ml-py==13.590.48
[pip3] nvidia-nccl-cu12==2.27.5
[pip3] nvidia-nvjitlink-cu12==12.8.93
[pip3] nvidia-nvshmem-cu12==3.4.5
[pip3] nvidia-nvtx-cu12==12.8.90
[pip3] optree==0.15.0
[pip3] pyzmq==27.1.0
[pip3] torch==2.10.0+cu128
[pip3] torch_c_dlpack_ext==0.1.5
[pip3] torchaudio==2.10.0+cu128
[pip3] torchvision==0.25.0+cu128
[pip3] transformers==4.56.1
[pip3] triton==3.6.0
[conda] flashinfer-python         0.6.4                    pypi_0    pypi
[conda] numpy                     2.0.2                    pypi_0    pypi
[conda] nvidia-cublas-cu12        12.8.4.1                 pypi_0    pypi
[conda] nvidia-cuda-cupti-cu12    12.8.90                  pypi_0    pypi
[conda] nvidia-cuda-nvrtc-cu12    12.8.93                  pypi_0    pypi
[conda] nvidia-cuda-runtime-cu12  12.8.90                  pypi_0    pypi
[conda] nvidia-cudnn-cu12         9.10.2.21                pypi_0    pypi
[conda] nvidia-cudnn-frontend     1.18.0                   pypi_0    pypi
[conda] nvidia-cufft-cu12         11.3.3.83                pypi_0    pypi
[conda] nvidia-cufile-cu12        1.13.1.3                 pypi_0    pypi
[conda] nvidia-curand-cu12        10.3.9.90                pypi_0    pypi
[conda] nvidia-cusolver-cu12      11.7.3.90                pypi_0    pypi
[conda] nvidia-cusparse-cu12      12.5.8.93                pypi_0    pypi
[conda] nvidia-cusparselt-cu12    0.7.1                    pypi_0    pypi
[conda] nvidia-cutlass-dsl        4.4.2                    pypi_0    pypi
[conda] nvidia-cutlass-dsl-libs-base 4.4.2                    pypi_0    pypi
[conda] nvidia-ml-py              13.590.48                pypi_0    pypi
[conda] nvidia-nccl-cu12          2.27.5                   pypi_0    pypi
[conda] nvidia-nvjitlink-cu12     12.8.93                  pypi_0    pypi
[conda] nvidia-nvshmem-cu12       3.4.5                    pypi_0    pypi
[conda] nvidia-nvtx-cu12          12.8.90                  pypi_0    pypi
[conda] optree                    0.15.0                   pypi_0    pypi
[conda] pyzmq                     27.1.0                   pypi_0    pypi
[conda] torch                     2.10.0+cu128             pypi_0    pypi
[conda] torch-c-dlpack-ext        0.1.5                    pypi_0    pypi
[conda] torchaudio                2.10.0+cu128             pypi_0    pypi
[conda] torchvision               0.25.0+cu128             pypi_0    pypi
[conda] transformers              4.56.1                   pypi_0    pypi
[conda] triton                    3.6.0                    pypi_0    pypi

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.17.1
vLLM Build Flags:
  CUDA Archs: 8.9; ROCm: Disabled; XPU: Disabled
GPU Topology:
        GPU0    GPU1    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NODE    96-191,288-383  1               N/A
GPU1    NODE     X      96-191,288-383  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
==============================
TORCH_CUDA_ARCH_LIST=8.9
CUDA_PATH=/usr/local/cuda
LD_LIBRARY_PATH=/usr/local/cuda/lib64:/home/neil/code/llm/llama.cpp/build-cuda/bin
CUDA_HOME=/usr/local/cuda
CUDA_HOME=/usr/local/cuda
CUDAToolkit_ROOT=/usr/local/cuda
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_neil

---

# vllm/v1/engine/output_processor.py
   def apply_streaming_update(self, update: StreamingUpdate) -> None:
       if self.prompt_token_ids:
           self.prompt_token_ids.extend(update.prompt_token_ids or ())
       else:
           self.prompt_token_ids = update.prompt_token_ids or []
       self.prompt_len = len(self.prompt_token_ids)

---

# vllm/v1/core/sched/async_scheduler.py
   cur_num_spec_tokens = len(spec_decode_tokens.get(req_id, ()))
   request.num_output_placeholders += 1 + cur_num_spec_tokens
   request.spec_token_ids = self._spec_token_placeholders

---

# vllm/v1/worker/gpu_model_runner.py
   if req_state.prev_num_draft_len and self.use_async_scheduling:
       ...
       req_state.output_token_ids.extend([-1] * num_accepted)

---

# vllm/v1/worker/gpu_model_runner.py
   prev_req_id_to_index: dict[str, int] = {}
   for i, req_id in enumerate(self.input_batch.req_ids):
       if i in discard_req_indices_set:
           continue
       prev_req_id_to_index[req_id] = i
   self.input_batch.prev_req_id_to_index = prev_req_id_to_index

---

# vllm/v1/worker/gpu_model_runner.py
   assert self.input_batch.prev_req_id_to_index is not None
   prev_req_index = self.input_batch.prev_req_id_to_index[req_id]
   num_accepted = valid_sampled_token_count[prev_req_index] - 1
   req_state.output_token_ids.extend([-1] * num_accepted)

---

scheduled_spec_decode_tokens={req2-...: [-1, -1, -1, -1, -1, -1, -1, -1], req3-...: [-1, -1, -1, -1, -1, -1, -1, -1]}

---

KeyError: 'req2-...'

---

vllm.v1.engine.exceptions.EngineDeadError: EngineCore encountered an issue.

---

export POC_PY=/path/to/python3
export G7_ASYNC=/path/to/repro_g7_doubleupdate_rowmap_official_asyncllm.py
export VLLM_POC_G7_MODEL=/path/to/Qwen2-0.5B-official

CUDA_VISIBLE_DEVICES=0 "$POC_PY" "$G7_ASYNC" \
  --model "$VLLM_POC_G7_MODEL" \
  --run-name g7_doubleupdate_rowmap_official_asyncllm

---

CUDA_VISIBLE_DEVICES=0 CUDA_LAUNCH_BLOCKING=1 "$POC_PY" "$G7_ASYNC" \
  --model "$VLLM_POC_G7_MODEL" \
  --run-name g7_doubleupdate_rowmap_official_asyncllm_blocking \
  --blocking

---

{
  "returncode": 1,
  "error_tail": "vllm.v1.engine.exceptions.EngineDeadError: EngineCore encountered an issue. See stack trace (above) for the root cause."
}

---

scheduled_spec_decode_tokens={req2-...: [-1, -1, -1, -1, -1, -1, -1, -1], req3-...: [-1, -1, -1, -1, -1, -1, -1, -1]}

---

KeyError: 'req2-...'

---

vllm.v1.engine.exceptions.EngineDeadError: EngineCore encountered an issue.
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 13.1.0-8ubuntu1~22.04) 13.1.0
Clang version                : 16.0.6 (++20231112100510+7cbf1a259152-1~exp1~20231112100554.106)
CMake version                : version 3.22.1
Libc version                 : glibc-2.35

==============================
       PyTorch Info
==============================
PyTorch version              : 2.10.0+cu128
Is debug build               : False
CUDA used to build PyTorch   : 12.8
ROCM used to build PyTorch   : N/A
XPU used to build PyTorch    : N/A

==============================
      Python Environment
==============================
Python version               : 3.12.9 | packaged by Anaconda, Inc. | (main, Feb  6 2025, 18:56:27) [GCC 11.2.0] (64-bit runtime)
Python platform              : Linux-6.5.0-35-generic-x86_64-with-glibc2.35
    
==============================
       CUDA / GPU Info
==============================
Is CUDA available            : True
CUDA runtime version         : 12.8.61
CUDA_MODULE_LOADING set to   : 
GPU models and configuration : 
GPU 0: NVIDIA GeForce RTX 4090
GPU 1: NVIDIA GeForce RTX 4090

Nvidia driver version        : 570.86.10
cuDNN version                : Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.3.0
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):                             384
On-line CPU(s) list:                0-383
Vendor ID:                          AuthenticAMD
Model name:                         AMD EPYC 9654 96-Core Processor
CPU family:                         25
Model:                              17
Thread(s) per core:                 2
Core(s) per socket:                 96
Socket(s):                          2
Stepping:                           1
Frequency boost:                    enabled
CPU max MHz:                        3707.8120
CPU min MHz:                        1500.0000
BogoMIPS:                           4792.57
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 amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic 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 perfmon_v2 ibrs ibpb stibp ibrs_enhanced 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 rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi 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:                          6 MiB (192 instances)
L1i cache:                          6 MiB (192 instances)
L2 cache:                           192 MiB (192 instances)
L3 cache:                           768 MiB (24 instances)
NUMA node(s):                       2
NUMA node0 CPU(s):                  0-95,192-287
NUMA node1 CPU(s):                  96-191,288-383
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 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; Enhanced / Automatic IBRS; IBPB conditional; 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==0.6.4
[pip3] numpy==2.0.2
[pip3] nvidia-cublas-cu12==12.8.4.1
[pip3] nvidia-cuda-cupti-cu12==12.8.90
[pip3] nvidia-cuda-nvrtc-cu12==12.8.93
[pip3] nvidia-cuda-runtime-cu12==12.8.90
[pip3] nvidia-cudnn-cu12==9.10.2.21
[pip3] nvidia-cudnn-frontend==1.18.0
[pip3] nvidia-cufft-cu12==11.3.3.83
[pip3] nvidia-cufile-cu12==1.13.1.3
[pip3] nvidia-curand-cu12==10.3.9.90
[pip3] nvidia-cusolver-cu12==11.7.3.90
[pip3] nvidia-cusparse-cu12==12.5.8.93
[pip3] nvidia-cusparselt-cu12==0.7.1
[pip3] nvidia-cutlass-dsl==4.4.2
[pip3] nvidia-cutlass-dsl-libs-base==4.4.2
[pip3] nvidia-ml-py==13.590.48
[pip3] nvidia-nccl-cu12==2.27.5
[pip3] nvidia-nvjitlink-cu12==12.8.93
[pip3] nvidia-nvshmem-cu12==3.4.5
[pip3] nvidia-nvtx-cu12==12.8.90
[pip3] optree==0.15.0
[pip3] pyzmq==27.1.0
[pip3] torch==2.10.0+cu128
[pip3] torch_c_dlpack_ext==0.1.5
[pip3] torchaudio==2.10.0+cu128
[pip3] torchvision==0.25.0+cu128
[pip3] transformers==4.56.1
[pip3] triton==3.6.0
[conda] flashinfer-python         0.6.4                    pypi_0    pypi
[conda] numpy                     2.0.2                    pypi_0    pypi
[conda] nvidia-cublas-cu12        12.8.4.1                 pypi_0    pypi
[conda] nvidia-cuda-cupti-cu12    12.8.90                  pypi_0    pypi
[conda] nvidia-cuda-nvrtc-cu12    12.8.93                  pypi_0    pypi
[conda] nvidia-cuda-runtime-cu12  12.8.90                  pypi_0    pypi
[conda] nvidia-cudnn-cu12         9.10.2.21                pypi_0    pypi
[conda] nvidia-cudnn-frontend     1.18.0                   pypi_0    pypi
[conda] nvidia-cufft-cu12         11.3.3.83                pypi_0    pypi
[conda] nvidia-cufile-cu12        1.13.1.3                 pypi_0    pypi
[conda] nvidia-curand-cu12        10.3.9.90                pypi_0    pypi
[conda] nvidia-cusolver-cu12      11.7.3.90                pypi_0    pypi
[conda] nvidia-cusparse-cu12      12.5.8.93                pypi_0    pypi
[conda] nvidia-cusparselt-cu12    0.7.1                    pypi_0    pypi
[conda] nvidia-cutlass-dsl        4.4.2                    pypi_0    pypi
[conda] nvidia-cutlass-dsl-libs-base 4.4.2                    pypi_0    pypi
[conda] nvidia-ml-py              13.590.48                pypi_0    pypi
[conda] nvidia-nccl-cu12          2.27.5                   pypi_0    pypi
[conda] nvidia-nvjitlink-cu12     12.8.93                  pypi_0    pypi
[conda] nvidia-nvshmem-cu12       3.4.5                    pypi_0    pypi
[conda] nvidia-nvtx-cu12          12.8.90                  pypi_0    pypi
[conda] optree                    0.15.0                   pypi_0    pypi
[conda] pyzmq                     27.1.0                   pypi_0    pypi
[conda] torch                     2.10.0+cu128             pypi_0    pypi
[conda] torch-c-dlpack-ext        0.1.5                    pypi_0    pypi
[conda] torchaudio                2.10.0+cu128             pypi_0    pypi
[conda] torchvision               0.25.0+cu128             pypi_0    pypi
[conda] transformers              4.56.1                   pypi_0    pypi
[conda] triton                    3.6.0                    pypi_0    pypi

==============================
         vLLM Info
==============================
ROCM Version                 : Could not collect
vLLM Version                 : 0.17.1
vLLM Build Flags:
  CUDA Archs: 8.9; ROCm: Disabled; XPU: Disabled
GPU Topology:
        GPU0    GPU1    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      NODE    96-191,288-383  1               N/A
GPU1    NODE     X      96-191,288-383  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
==============================
TORCH_CUDA_ARCH_LIST=8.9
CUDA_PATH=/usr/local/cuda
LD_LIBRARY_PATH=/usr/local/cuda/lib64:/home/neil/code/llm/llama.cpp/build-cuda/bin
CUDA_HOME=/usr/local/cuda
CUDA_HOME=/usr/local/cuda
CUDAToolkit_ROOT=/usr/local/cuda
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1
TORCHINDUCTOR_CACHE_DIR=/tmp/torchinductor_neil
</details>

🐛 Describe the bug

Describe the bug

Version: vLLM 0.17.1
Model: Qwen/Qwen2-0.5B
Hardware reproduced on: NVIDIA GeForce RTX 4090, single GPU

Summary

I found a double-streaming_update bug in the AsyncLLM path that can break the worker's request ID -> previous batch row mapping for a resumed request and then kill the engine.

The reproducer submits one resumable request and then applies two legal streaming_update operations to that same live request while async scheduling and speculative decoding are both active.

Under that combination, the scheduler can continue issuing speculative decode positions for the resumed request, but the worker-side request ID -> previous batch row mapping no longer matches the live request state. The verified AsyncLLM runs then reach a worker update step where the request id is missing from prev_req_id_to_index, and the engine dies with:

  • scheduled_spec_decode_tokens={...: [-1, ...]}
  • KeyError: 'req2-...'
  • EngineDeadError

This is not the same bug as:

  • the async same-request-id reuse underflow bug
  • the resumed prompt-width overflow bug

Trigger chain

  1. Submit a live mixed batch so async scheduling is active.
  2. Submit one resumable late request.
  3. Send a first legal streaming_update for that same live request.
  4. Let async queued work and speculative state persist.
  5. Send a second legal streaming_update for the same live request.
  6. After the second streaming_update, the live request state no longer matches the worker state carried over from the previous batch for that request.
  7. The scheduler still issues speculative decode positions for that request.
  8. The worker later tries to update cached request state through prev_req_id_to_index, but that request id is no longer present in the request ID -> previous batch row mapping, and the engine dies.

Details

Trigger path in code

  1. On the output-processor side, each legal streaming_update extends the live request prompt state. 
    # vllm/v1/engine/output_processor.py
def apply_streaming_update(self, update: StreamingUpdate) -> None:
    if self.prompt_token_ids:
        self.prompt_token_ids.extend(update.prompt_token_ids or ())
    else:
        self.prompt_token_ids = update.prompt_token_ids or []
    self.prompt_len = len(self.prompt_token_ids)
  2. In async scheduling mode, vLLM adds placeholder credit for future output and installs placeholder speculative token ids on the request object. 
    # vllm/v1/core/sched/async_scheduler.py
cur_num_spec_tokens = len(spec_decode_tokens.get(req_id, ()))
request.num_output_placeholders += 1 + cur_num_spec_tokens
request.spec_token_ids = self._spec_token_placeholders
  3. On the worker side, async speculative carry-over can append -1 placeholders into the per-request cached output-token state. 
    # vllm/v1/worker/gpu_model_runner.py
if req_state.prev_num_draft_len and self.use_async_scheduling:
    ...
    req_state.output_token_ids.extend([-1] * num_accepted)
  4. The worker also constructs prev_req_id_to_index so later async iterations can map req_id -> previous batch row. 
    # vllm/v1/worker/gpu_model_runner.py
prev_req_id_to_index: dict[str, int] = {}
for i, req_id in enumerate(self.input_batch.req_ids):
    if i in discard_req_indices_set:
        continue
    prev_req_id_to_index[req_id] = i
self.input_batch.prev_req_id_to_index = prev_req_id_to_index
  5. During the later worker-state update, the code assumes the request id is still present in that request ID -> previous batch row mapping. 
    # vllm/v1/worker/gpu_model_runner.py
assert self.input_batch.prev_req_id_to_index is not None
prev_req_index = self.input_batch.prev_req_id_to_index[req_id]
num_accepted = valid_sampled_token_count[prev_req_index] - 1
req_state.output_token_ids.extend([-1] * num_accepted)
  6. In this issue, after the second streaming_update, the live request state no longer agrees with the worker state carried over from the previous batch, but speculative decode positions are still scheduled for that request.
  7. In the verified official-model AsyncLLM runs, the last scheduler dump before the sink still shows: 
    scheduled_spec_decode_tokens={req2-...: [-1, -1, -1, -1, -1, -1, -1, -1], req3-...: [-1, -1, -1, -1, -1, -1, -1, -1]}
  8. The worker then raises: 
    KeyError: 'req2-...'
    and AsyncLLM surfaces: 
    vllm.v1.engine.exceptions.EngineDeadError: EngineCore encountered an issue.

Standalone AsyncLLM script breakdown

repro_g7_doubleupdate_rowmap_official_asyncllm.py is the official-model AsyncLLM reproducer.

  • It defaults to VLLM_POC_G7_MODEL when set, otherwise to the official local clone path, and also accepts --model.
  • It enables:
    • async scheduling
    • speculative decoding
    • one resumable late request with two legal streaming_update chunks
  • It writes:
    • repro_config.json
    • request_payloads.json
    • engine_args.json
    • req0_outputs.json
    • req1_outputs.json
    • req2_outputs.json
    • req3_outputs.json
    • abort_result.json
    • task_results.json
    • post_run_summary.json
    • error.txt on failure

AsyncLLM repro

  • Official Qwen2-0.5B model, you can find here
  • Anyone wants to reproduce the bug with my PoC needs to download Qwen2-0.5B first
  • Here is my reproduce script

repro_g7_doubleupdate_rowmap_official_asyncllm.py

export POC_PY=/path/to/python3
export G7_ASYNC=/path/to/repro_g7_doubleupdate_rowmap_official_asyncllm.py
export VLLM_POC_G7_MODEL=/path/to/Qwen2-0.5B-official

CUDA_VISIBLE_DEVICES=0 "$POC_PY" "$G7_ASYNC" \
  --model "$VLLM_POC_G7_MODEL" \
  --run-name g7_doubleupdate_rowmap_official_asyncllm

Optional blocking attribution:

CUDA_VISIBLE_DEVICES=0 CUDA_LAUNCH_BLOCKING=1 "$POC_PY" "$G7_ASYNC" \
  --model "$VLLM_POC_G7_MODEL" \
  --run-name g7_doubleupdate_rowmap_official_asyncllm_blocking \
  --blocking

Reproduce Environment

ItemValue
OSUbuntu 22.04.5 LTS
KernelLinux 6.5.0-35-generic
GPU2 x NVIDIA GeForce RTX 4090
GPU memory24564 MiB each
Driver570.86.10
CUDA runtime12.8 (nvidia-smi)
CUDA toolkit12.8.61 (nvcc)
Python3.12.9
vLLM0.17.1
PyTorch2.10.0+cu128
transformers4.56.1
tokenizers0.22.0
flash_attn2.8.3
triton3.6.0
numpy2.0.2

Observed result

post_run_summary.json from the re-verified official-model AsyncLLM run contains:

{
  "returncode": 1,
  "error_tail": "vllm.v1.engine.exceptions.EngineDeadError: EngineCore encountered an issue. See stack trace (above) for the root cause."
}

The same run also saves the full traceback into error.txt.

On stderr, the same run first dumps scheduler state with:

scheduled_spec_decode_tokens={req2-...: [-1, -1, -1, -1, -1, -1, -1, -1], req3-...: [-1, -1, -1, -1, -1, -1, -1, -1]}

Then the worker fatally trips while looking up the previous batch row for that request:

KeyError: 'req2-...'

At the user-facing AsyncLLM surface, that becomes:

vllm.v1.engine.exceptions.EngineDeadError: EngineCore encountered an issue.

So this is a deterministic reproducer that advances through legal double streaming_update state, keeps speculative placeholder decode alive for the resumed request, and then kills the engine when the worker's request ID -> previous batch row mapping no longer matches that request.

Root cause

This is a resumed-request / previous-batch-row state reconstruction bug, not a malformed request bug.

The important distinction is:

  • both streaming_updates are legal
  • the same live request is resumed twice
  • async speculative state persists between those updates
  • the request ID -> previous batch row mapping no longer matches the live request state

So the real problem is that async speculative state reconstruction can continue after double streaming_update even when the request ID -> previous batch row mapping no longer matches the live request state, and that mismatch then kills the engine.

Attachments

The attachment bundle for this report should contain:

  • repro_g7_doubleupdate_rowmap_official_asyncllm.py

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.

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?
Loading…
#api#GPU compatibility#latency issue#model loading#environment variable

Still need to ship something?

×6

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

Back to top recommendations

TRENDING

vllm - 💡(How to fix) Fix [Bug]: Async double streaming_update can break the request ID -> previous batch row mapping and crash the engine