vllm: [Bug]: Disk I/O Error when using tools due to shared outlines cache database
Your current environment
Collecting environment information...
PyTorch version: 2.1.2+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A
OS: Ubuntu 22.04.3 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.29.0
Libc version: glibc-2.35
Python version: 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.1.79-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA A10G
GPU 1: NVIDIA A10G
GPU 2: NVIDIA A10G
GPU 3: NVIDIA A10G
Nvidia driver version: 535.161.07
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
CPU:
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 7R32
CPU family: 23
Model: 49
Thread(s) per core: 2
Core(s) per socket: 48
Socket(s): 1
Stepping: 0
BogoMIPS: 5599.99
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 sse4_1 sse4_2 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 rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 clzero xsaveerptr rdpru wbnoinvd arat npt nrip_save rdpid
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 1.5 MiB (48 instances)
L1i cache: 1.5 MiB (48 instances)
L2 cache: 24 MiB (48 instances)
L3 cache: 192 MiB (12 instances)
NUMA node(s): 1
NUMA node0 CPU(s): 0-95
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 Retbleed: Mitigation; untrained return thunk; SMT enabled with STIBP protection
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, STIBP always-on, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] torch==2.1.2
[pip3] triton==2.1.0
[conda] Could not collectROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.4.0.post1
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0 GPU1 GPU2 GPU3 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X PHB PHB PHB 0-95 0 N/A
GPU1 PHB X PHB PHB 0-95 0 N/A
GPU2 PHB PHB X PHB 0-95 0 N/A
GPU3 PHB PHB PHB X 0-95 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
š Describe the bug
This was discovered when using the Llama 3 Instruct model with the workaround in #4180 on multiple nodes using a shared filed system for the cache directory.
Requests to the vLLM with tool_calls using the OpenAI compatible tool call API utilize the outlines library. When the filesystem is shared between nodes, e.g.: using AWS Elastic File System, the outlines library opens a SQLite database in the shared cache dir.
This causes I/O errors on at least one node due to each node conflicting on writing to the same SQLite database.
To mitigate this, vLLM could utilize a cache dir in /tmp for Outlines, configured on startup of vLLM:
https://github.com/outlines-dev/outlines/blob/main/outlines/caching.py#L14-L29
Workaround: set the env var OUTLINES_CACHE_DIR to a value such as /tmp/.outlines.
About this issue
- Original URL
- State: closed
- Created 3 years ago
- Reactions: 10
- Comments: 25 (6 by maintainers)
For those running into issues working with Slurm, I found that setting the environment variable OUTLINES_CACHE_DIR to the job directory fixed the problem. Iām not using guided decoding though so not sure if this would affect that.
I installed and tested again; however, I did want to mention that weāre using Lustre for the file system, rather than NFS. Itās possible that others using Slurm are as well.
partially fixed by https://github.com/vllm-project/vllm/pull/7831
closing and wait for outlines to come up with the final solution.
I still encounter this error with vllm 0.6.3.post1 on slurm multi-node env. setting
--guided-decoding-backend=lm-format-enforcerworks for me.@testzer0 I used the environment variable, but Iām not sure whether you can do the same for the Ray processes. Maybe you can try to set the environment variable by using the code line
os.environ["OUTLINES_CACHE_DIR"]={something}in the function youāre using for spawning processes. Still, Iām not sure that this works fine because I donāt know exactly when this variable is used or called.@lapp0 I have been doing some more testing, and now Iām getting a āRead-only file systemā error.
Iām thinking this may have to do with the way that Singularity/Apptainer is mounting directories and is treating its file system. I think that if I build the container with a sandbox/writeable folder for the cache, it could fix it. Itās probably going to be tomorrow before I am able to do that and test, but Iāll report back and let you know.