Checklist

• I searched related issues but found no solution.
• The bug persists in the latest version.
• Issues without environment info and a minimal reproducible demo are hard to resolve and may receive no feedback.
• If this is not a bug report but a general question, please start a discussion at https://github.com/sgl-project/sglang/discussions. Otherwise, it will be closed.
• Please use English. Otherwise, it will be closed.

Describe the bug

Environment Description

I am using SGLang 0.5.9. My machine does not have RDMA devices; everything runs over TCP.

When I enable the L3 KV cache (i.e., enable hicache-storage-backend) and simultaneously run a benchmark program to stress test the service, the memory usage of the sglang::scheduler process continuously increases.

Based on my observations:

• Memory usage increases by approximately 10 GB per hour
• There appears to be no upper limit
• The process memory eventually exceeds 200 GB
• The process is then killed by the system (OOM)

The 200 GB process memory I mentioned refers to the RES value observed for that process in the top command. In addition, after I stop the benchmark program, the memory usage neither continues to increase nor decreases.

I tested both:

• hicache-storage-backend=file
• hicache-storage-backend=mooncake

The issue occurs in both cases, regardless of whether the backend is local disk or Mooncake.

Additionally, I tested all versions from 0.5.5 to 0.5.9, and the problem exists in all of them.

Versions

• SGLang version: 0.5.6～0.5.9
• Mooncake version: 0.3.7～0.3.9

Reproduction

SGLang Launch Command (File Backend)

export SGLANG_HICACHE_FILE_BACKEND_STORAGE_DIR=/root/kvcache

nohup python3 -m sglang.launch_server \
  --model-path Qwen/Qwen3-4B \
  --trust-remote-code \
  --host 0.0.0.0 \
  --port 30000 \
  --mem-fraction-static 0.85 \
  --enable-hierarchical-cache \
  --hicache-size 100 \
  --page-size 128 \
  --hicache-io-backend kernel \
  --hicache-mem-layout page_first \
  --hicache-storage-backend file &

SGLang Launch Command (Mooncake Backend)

export MOONCAKE_TE_META_DATA_SERVER="etcd://etcdn1.mooncake-c1.dns.org:2379;etcd://etcdn2.mooncake-c1.dns.org:2379;etcd://etcdn3.mooncake-c1.dns.org:2379;etcd://etcdn4.mooncake-c1.dns.org:2379;etcd://etcdn5.mooncake-c1.dns.org:2379"
export MOONCAKE_MASTER="etcd://etcdn1.mooncake-c1.dns.org:2379;etcd://etcdn2.mooncake-c1.dns.org:2379;etcd://etcdn3.mooncake-c1.dns.org:2379;etcd://etcdn4.mooncake-c1.dns.org:2379;etcd://etcdn5.mooncake-c1.dns.org:2379"
export MOONCAKE_PROTOCOL="tcp"
export MOONCAKE_DEVICE=""

nohup python3 -m sglang.launch_server \
  --model-path Qwen/Qwen3-4B \
  --trust-remote-code \
  --host 0.0.0.0 \
  --port 30000 \
  --mem-fraction-static 0.85 \
  --enable-hierarchical-cache \
  --hicache-size 100 \
  --page-size 128 \
  --hicache-io-backend kernel \
  --hicache-mem-layout page_first \
  --hicache-storage-backend mooncake \
  --enable-metrics &

Benchmark Command

python3 bench_multiturn.py \
    --host localhost \
    --port 30000 \
    --model-path Qwen/Qwen3-4B \
    --dataset-path /root/sglang-src/data/ShareGPT_V3_unfiltered_cleaned_split.json \
    --disable-random-sample \
    --num-clients 24 \
    --num-rounds 12 \
    --max-parallel 4 \
    --request-rate 8 \
    --request-length 2048 \
    --output-length 1 \
    --ready-queue-policy random \
    --disable-auto-run \
    --enable-round-barrier

Environment

Python: 3.12.3 (main, Nov 6 2025, 13:44:16) [GCC 13.3.0]
CUDA available: True
GPU 0: NVIDIA GeForce RTX 3090
GPU 0 Compute Capability: 8.6
CUDA_HOME: /usr/local/cuda
NVCC: Cuda compilation tools, release 12.9, V12.9.86
CUDA Driver Version: 575.64.03
PyTorch: 2.9.1+cu129
sglang: 0.5.9
sgl_kernel: 0.3.21
flashinfer_python: 0.6.3
flashinfer_cubin: 0.6.3
flashinfer_jit_cache: Module Not Found
triton: 3.5.1
transformers: 4.57.1
torchao: 0.9.0
numpy: 2.3.5
aiohttp: 3.13.3
fastapi: 0.124.2
hf_transfer: 0.1.9
huggingface_hub: 0.36.0
interegular: 0.3.3
modelscope: 1.33.0
orjson: 3.11.5
outlines: 0.1.11
packaging: 25.0
psutil: 7.1.3
pydantic: 2.12.5
python-multipart: 0.0.20
pyzmq: 27.1.0
uvicorn: 0.38.0
uvloop: 0.22.1
vllm: Module Not Found
xgrammar: 0.1.27
openai: 2.6.1
tiktoken: 0.12.0
anthropic: 0.75.0
litellm: Module Not Found
decord2: 2.0.0
NVIDIA Topology:
GPU0 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X 0-31 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

ulimit soft: 1048576