Your current environment

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

PyTorch version : 2.10.0a0+b558c986e8.nv25.11 Is debug build : False CUDA used to build PyTorch : 13.0 ROCM used to build PyTorch : N/A XPU used to build PyTorch : N/A

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

Python version : 3.12.3 (main, Mar 23 2026, 19:04:32) [GCC 13.3.0] (64-bit runtime) Python platform : Linux-5.10.134-18.0.10.lifsea8.x86_64-x86_64-with-glibc2.39

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

Is CUDA available : True CUDA runtime version : 13.0.88 CUDA_MODULE_LOADING set to : LAZY GPU models and configuration : GPU 0: NVIDIA L20 Nvidia driver version : 550.163.01 cuDNN version : Probably one of the following: /usr/lib/x86_64-linux-gnu/libcudnn.so.9.15.0 /usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.15.0 /usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.15.0 /usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.15.0 /usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.15.0 /usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.15.0 /usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.15.0 /usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.15.0 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): 8 On-line CPU(s) list: 0-7 Vendor ID: GenuineIntel Model name: Intel(R) Xeon(R) Gold 6462C CPU family: 6 Model: 143 Thread(s) per core: 2 Core(s) per socket: 4 Socket(s): 1 Stepping: 8 CPU(s) scaling MHz: 100% CPU max MHz: 3900.0000 CPU min MHz: 800.0000 BogoMIPS: 6600.00 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 aperfmperf tsc_known_freq pni pclmulqdq monitor ssse3 fma cx16 pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch cpuid_fault invpcid_single ibrs_enhanced fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves avx_vnni avx512_bf16 wbnoinvd ida arat hwp hwp_notify hwp_act_window hwp_epp hwp_pkg_req avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm uintr md_clear serialize tsxldtrk amx_bf16 avx512_fp16 amx_tile amx_int8 arch_capabilities Hypervisor vendor: KVM Virtualization type: full L1d cache: 192 KiB (4 instances) L1i cache: 128 KiB (4 instances) L2 cache: 8 MiB (4 instances) L3 cache: 60 MiB (1 instance) NUMA node(s): 1 NUMA node0 CPU(s): 0-7 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: Not affected Vulnerability Spec store bypass: Vulnerable Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS, RSB filling, PBRSB-eIBRS SW sequence Vulnerability Srbds: Not affected Vulnerability Tsx async abort: Not affected

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

[pip3] mypy_extensions==1.1.0 [pip3] numpy==1.26.4 [pip3] nvidia-cudnn-frontend==1.15.0 [pip3] nvidia-dali-cuda130==1.52.0 [pip3] nvidia-ml-py==13.580.82 [pip3] nvidia-modelopt==0.37.0 [pip3] nvidia-nvcomp-cu13==5.0.0.6 [pip3] nvidia-nvimgcodec-cu13==0.6.1.37 [pip3] nvidia-nvjpeg==13.0.1.86 [pip3] nvidia-nvjpeg2k-cu13==0.9.0.43 [pip3] nvidia-nvtiff-cu13==0.6.0.78 [pip3] nvidia-resiliency-ext==0.4.1+cuda13 [pip3] onnx==1.18.0 [pip3] onnx-ir==0.1.12 [pip3] onnxscript==0.5.6 [pip3] optree==0.17.0 [pip3] pytorch-triton==3.5.0+gitde3506d2 [pip3] pyzmq==27.1.0 [pip3] torch==2.10.0a0+b558c986e8.nv25.11 [pip3] torch_tensorrt==2.10.0a0 [pip3] torchao==0.14.0+git [pip3] torchdata==0.11.0 [pip3] torchprofile==0.0.4 [pip3] torchtitan==0.1.0 [pip3] torchvision==0.25.0a0+7a13ad0f [pip3] triton_kernels==1.0.0+nv25.11 [conda] Could not collect

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

ROCM Version : Could not collect vLLM Version : 0.1.dev16489+g9efdddca2.d20260525 (git sha: 9efdddca2, date: 20260525) vLLM Build Flags: CUDA Archs: 7.5 8.0 8.6 9.0 10.0 12.0+PTX; ROCm: Disabled; XPU: Disabled GPU Topology: GPU0 CPU Affinity NUMA Affinity GPU NUMA ID GPU0 X 0-7 0 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

PYTORCH_BUILD_NUMBER=0 PYTORCH_HOME=/opt/pytorch/pytorch CUDA_ARCH_LIST=7.5 8.0 8.6 9.0 10.0 12.0 LD_LIBRARY_PATH=/usr/local/cuda/compat/lib.real:/usr/local/lib/python3.12/dist-packages/torch/lib:/usr/local/lib/python3.12/dist-packages/torch_tensorrt/lib:/usr/local/cuda/compat/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64 TORCH_NCCL_USE_COMM_NONBLOCKING=0 NVIDIA_BUILD_ID=231036167 CUDA_VERSION=13.0.2.006 TORCH_ALLOW_TF32_CUBLAS_OVERRIDE=1 CUBLAS_VERSION=13.1.0.3 TORCH_CUDA_ARCH_LIST=7.5 8.0 8.6 9.0 10.0 12.0+PTX CUBLASMP_VERSION=0.6.0.84 CUDA_MODULE_LOADING=LAZY NVIDIA_REQUIRE_CUDA=cuda>=9.0 PYTORCH_BUILD_VERSION=2.10.0a0+b558c98 NVIDIA_DRIVER_CAPABILITIES=compute,utility,video CUDA_COMPONENT_LIST=cccl crt nvrtc driver-dev culibos-dev cudart cudart-dev nvcc NVIDIA_PYTORCH_VERSION=25.11 PYTORCH_VERSION=2.10.0a0+b558c98 CUDA_DRIVER_VERSION=580.95.05 NVIDIA_PRODUCT_NAME=PyTorch CUDNN_VERSION=9.15.0.58 CUDNN_FRONTEND_VERSION=1.15.0 NVIDIA_REQUIRE_JETPACK_HOST_MOUNTS= TORCHINDUCTOR_LOOP_ORDERING_AFTER_FUSION=0 CUDA_HOME=/usr/local/cuda CUDA_HOME=/usr/local/cuda NVIDIA_VISIBLE_DEVICES=0 NCCL_VERSION=2.28.8 PYTORCH_NVML_BASED_CUDA_CHECK=1 TORCHINDUCTOR_COMPILE_THREADS=1

Your output of `python collect_env.py` here

🐛 Describe the bug

Problem

When a request is aborted after get_num_new_matched_tokens submits a lookup but before update_state_after_alloc is called, _cleanup_request_tracker is the only cleanup path. It only pops the tracker — it never calls cleanup_lookup_result or free_lookup_locks. _pending_lookups retains the request_id and read locks on matched KV chunks are never released until TTL expiry, blocking eviction.

Root cause: update_state_after_alloc correctly calls both on the normal path. _cleanup_request_tracker calls neither.

Trigger condition

Request submitted a lookup (maybe_submit_lookup_request → enters _pending_lookups), then aborted before update_state_after_alloc — client disconnect, request timeout.

Reproduction

from unittest.mock import MagicMock
from vllm.distributed.kv_transfer.kv_connector.v1.lmcache_mp_connector import (
    LMCacheMPConnectorUpstream as LMCacheMPConnector,
    LMCacheMPRequestTracker,
)

connector = object.__new__(LMCacheMPConnector)
connector.vllm_block_size = 16
connector.scheduler_adapter = MagicMock()
connector.scheduler_adapter._pending_lookups = {"req-1"}
connector.scheduler_adapter._finished_lookup_results = {"req-1": 32}

req = MagicMock()
req.request_id, req.all_token_ids, req.block_hashes, req.cache_salt = "req-1", list(range(32)), [], ""
req.kv_transfer_params = None

tracker = LMCacheMPRequestTracker(req)
tracker.num_lmcache_hit_blocks = 2
connector.request_trackers = {"req-1": tracker}

# Simulate the full abort path: request_finished calls both
# _cleanup_request_tracker and end_session
connector.request_finished(req, block_ids=[])

# end_session is called but only removes session-level state,
# it does NOT send FREE_LOOKUP_LOCKS to the server
print("end_session called    :", connector.scheduler_adapter.end_session.called)

print("=" * 60)
# True = fixed, False = bug
print("free_lookup_locks called:", connector.scheduler_adapter.free_lookup_locks.called)

Fix

In _cleanup_request_tracker, when the tracker is still in PREFETCHING state, call cleanup_lookup_result and free_lookup_locks — mirroring what update_state_after_alloc already does on the normal path. The PREFETCHING guard is necessary: update_state_after_alloc already handles cleanup for WAITING_FOR_LOAD and READY states, and calling free_lookup_locks again there would release locks on chunks still being retrieved.

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.