All issues

Search results

MLA models on vLLM 0.24: KVCacheManager post-init cannot reach worker IPC socket (Errno 2), alloc_page retry storm

Status: Open.

#471 In ovg-project/kvcached;

· rishabhsinha17 opened on Aug 27, 2026

Editable install does not install kvcached_autopatch.pth; KVCACHED_AUTOPATCH silently no-ops

Status: Open.

#470 In ovg-project/kvcached;

· rishabhsinha17 opened on Aug 27, 2026

Design: safe cross-process admission for concurrent physical KV growth

Status: Open.

#467 In ovg-project/kvcached;

· shipiyouniao opened on Aug 27, 2026

Physical KV unmap is not ordered after in-flight vLLM async batches

Status: Open.

#463 In ovg-project/kvcached;

· shipiyouniao opened on Aug 22, 2026

SleepManager has no per-model_name lock, so concurrent sleep/wakeup calls each fire the full upstream call sequence

Status: Open.

#462 In ovg-project/kvcached;

· AmirF194 opened on Aug 21, 2026

PD Disaggregation Support Tracking Issue

Status: Open.

#442 In ovg-project/kvcached;

· RixinLiu opened on Aug 11, 2026

SIGABRT in FTensor::map() — mem_unmap fails with "invalid argument" under repeated large alloc/free churn

Status: Open.

#437 In ovg-project/kvcached;

· jiarong0907 opened on Aug 8, 2026

kvcached support matrix

help wantedExtra attention is neededExtra attention is needed

Status: Open.

#425 In ovg-project/kvcached;

· RixinLiu opened on Aug 5, 2026

Current Roadmap

Status: Open.

#419 In ovg-project/kvcached;

· RixinLiu opened on Jul 30, 2026

Need support for Gemma model which has cross-layer kv sharing

Status: Open.

#417 In ovg-project/kvcached;

· RixinLiu opened on Jul 30, 2026

Partial KV tensor mapping failures can leave inconsistent VMM state

Status: Open.

#412 In ovg-project/kvcached;

· shipiyouniao opened on Jul 27, 2026

SleepManager.wakeup_model() reports failure for a caller whose own wake call actually succeeded, when two wakeups for the same model race

Status: Open.

#408 In ovg-project/kvcached;

· AmirF194 opened on Jul 23, 2026