Skip to content

[Bug]: NixlConnector should not skip short do_remote_prefill requests in connector metadata #18591

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
1 task done
juncgu opened this issue May 23, 2025 · 3 comments · May be fixed by #18590
Open
1 task done

[Bug]: NixlConnector should not skip short do_remote_prefill requests in connector metadata #18591

juncgu opened this issue May 23, 2025 · 3 comments · May be fixed by #18590
Labels
bug Something isn't working

Comments

@juncgu
Copy link

juncgu commented May 23, 2025

Your current environment

The output of python collect_env.py
CPU(s):                             224
On-line CPU(s) list:                0-223
Vendor ID:                          AuthenticAMD
Model name:                         AMD EPYC 7B13
CPU family:                         25
Model:                              1
Thread(s) per core:                 2
Core(s) per socket:                 56
Socket(s):                          2
Stepping:                           0
BogoMIPS:                           4899.99
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 tsc_known_freq pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw topoext invpcid_single ssbd ibrs ibpb stibp vmmcall fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 clzero xsaveerptr arat npt nrip_save umip vaes vpclmulqdq rdpid fsrm
Hypervisor vendor:                  KVM
Virtualization type:                full
L1d cache:                          3.5 MiB (112 instances)
L1i cache:                          3.5 MiB (112 instances)
L2 cache:                           56 MiB (112 instances)
L3 cache:                           448 MiB (14 instances)
NUMA node(s):                       2
NUMA node0 CPU(s):                  0-55,112-167
NUMA node1 CPU(s):                  56-111,168-223
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; Retpolines, IBPB conditional, IBRS_FW, STIBP conditional, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds:                Not affected
Vulnerability Tsx async abort:      Not affected

Versions of relevant libraries:
[pip3] numpy==2.2.5
[pip3] pyzmq==26.4.0
[pip3] torch==2.8.0.dev20250430+cpu
[pip3] torch-xla==2.8.0+gita7c3e85
[pip3] torchvision==0.22.0.dev20250430+cpu
[pip3] transformers==4.51.3
[pip3] triton==3.3.0
[conda] numpy                     2.2.5                    pypi_0    pypi
[conda] pyzmq                     26.4.0                   pypi_0    pypi
[conda] torch                     2.8.0.dev20250430+cpu          pypi_0    pypi
[conda] torch-xla                 2.8.0+gita7c3e85          pypi_0    pypi
[conda] torchvision               0.22.0.dev20250430+cpu          pypi_0    pypi
[conda] transformers              4.51.3                   pypi_0    pypi
[conda] triton                    3.3.0                    pypi_0    pypi
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.1.dev6496+g4d66549.d20250511 (git sha: 4d66549, date: 20250511)
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
Could not collect

NCCL_CUMEM_ENABLE=0
PYTORCH_NVML_BASED_CUDA_CHECK=1
TORCHINDUCTOR_COMPILE_THREADS=1

🐛 Describe the bug

pytest -v -x v1/kv_connector/unit/test_nixl_connector.py::test_prompt_less_than_block_size was failed (#18490). #18429 wanted to fix it by skipping adding this request into the connector's metadata.

But the failure reason is the unit test has not been updated to due the changes in NixlConnectorSheduler.
The short (promot < block_size) do_remote_prefill request should be copied into the connector's metadata with an empty local_block_ids, so that NixlConnectorWorker can skip (async) reading remote kv blocks, but still be able to send notificaiton to the prefill worker to release its remote kv blocks.

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.
@juncgu juncgu added the bug Something isn't working label May 23, 2025
@juncgu
Copy link
Author

juncgu commented May 23, 2025

@njhill
Copy link
Member

njhill commented May 25, 2025

Thanks @juncgu, I think this should now be fixed by #18631.

@juncgu
Copy link
Author

juncgu commented May 27, 2025

Thanks @juncgu, I think this should now be fixed by #18631.

Hi @njhill,

IMHO, #18631 is not relevant to the case here (when prompt < block_size and no remote kv blocks need to be fetched / read by decode after subtracting the local prefix cache hit). #18631 tries to handle when decode-side has no enough KV space to allocate the new do_remote_prefill request.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working
Projects
None yet
Development

Successfully merging a pull request may close this issue.

2 participants