Issues

• [serve][llm] Support data_parallel_size>1 for SGLang P/D disaggregation

  community-backlogllmperformanceservetech-debt
  ray-project/ray#66634 · Filter by author limarkdcunha opened on Oct 1, 2026

• [RLlib] Potential infinite Env Runer sampling loop with non default algorithm parameters

  bugcommunity-backlogrllibstabilitytriage
  ray-project/ray#66622 · Filter by author corentinlger opened on Oct 1, 2026

• [serve] Support rack-aware / domain-aware scheduling

  performanceserve
  Feature

  ray-project/ray#66583 · Filter by author jeffreywang88 opened on Sep 29, 2026
  1

  

• [Core] Late RESOURCE_VIEW messages can reinsert dead nodes into the raylet resource map

  bugcommunity-backlogcoreP0performancestabilitytriage
  ray-project/ray#66565 · Filter by author hhhhjun opened on Sep 29, 2026

• [Core] image_uri / container runtime env workers can't get the token under Kubernetes token auth

  coregood-first-issuekuberay
  ray-project/ray#66561 · Filter by author MortalHappiness opened on Sep 29, 2026
  1

  

• [RFC][Serve][LLM] Pluggable Model-Level Routing for Multi-Model Applications

  community-backlogenhancementllmperformanceservetriageusability
  ray-project/ray#66535 · Filter by author suppagoddo opened on Sep 28, 2026
  4

• [Data] Add ORC scanner and reader for DataSourceV2

  community-backlogdatausability
  ray-project/ray#66526 · Filter by author jiangxt2 opened on Sep 28, 2026

• [Core] Blocking ray.get can stall unrelated ray.get calls through the shared PlasmaClient

  community-backlogcoreperformance
  ray-project/ray#66525 · Filter by author MissiontoMars opened on Sep 28, 2026

• [Serve] Allow app-defined health routes to bypass max_ongoing_requests

  community-backlogservestabilityusability
  ray-project/ray#66509 · Filter by author codechrl opened on Sep 26, 2026
  1

• [Data] Support commit-aware checkpointing for transactional Datasinks (Paimon)

  community-backlogdataquestionstability
  ray-project/ray#66491 · Filter by author yujiabin123 opened on Sep 25, 2026

• [Serve] VLLM_API_KEY / engine api_key argument is not enforced by Ray Serve LLM OpenAI endpoints

  community-backlogllmsecurityserve
  ray-project/ray#66487 · Filter by author dhiedjaid-stack opened on Sep 25, 2026
  4

• [Serve] ray-llm:2.58.0 ships vLLM 0.26.0, which does not support moonshotai/Kimi-K3

  community-backlogllmservestability
  ray-project/ray#66486 · Filter by author dhiedjaid-stack opened on Sep 25, 2026

• [Serve] ray-llm:2.58.0 ships quack-kernels 0.4.1 with vLLM 0.26.0, causing broken LLM warmup on SM90+ (AttributeError: module 'cutlass.cute.core' has no attribute 'ThrMma')

  community-backlogllmservestability
  ray-project/ray#66485 · Filter by author dhiedjaid-stack opened on Sep 25, 2026

• [Data] train_test_split returns wrong split sizes: float truncation in split_proportionately, and stratify can yield an empty test set

  bugcommunity-backlogdatastabilitytriage
  ray-project/ray#66472 · Filter by author lucaluo925 opened on Sep 24, 2026
  1

• [Core] Worker SIGSEGV in PyTraceBack_Here when LocalGC arrives before worker.core_worker is assigned

  bugcommunity-backlogcoreP0stabilitytriage
  ray-project/ray#66454 · Filter by author Xargonus opened on Sep 24, 2026
  3

• [Data] Discuss full schema inference beyond the DSV2 16-file sample

  community-backlogdataquestionusability
  ray-project/ray#66444 · Filter by author jiangxt2 opened on Sep 24, 2026
  1

• [core][sandbox] Hooks that modify the rootfs don't take effect with the EROFS rootfs

  community-backlogcorestability
  ray-project/ray#66408 · Filter by author klueska opened on Sep 23, 2026

• Async actor fiber stack protection is re-anchored after the first Python call on CPython 3.14

  community-backlogcorestability
  ray-project/ray#66399 · Filter by author guoshengCS opened on Sep 23, 2026

• Resource-aware scheduling based on actual node utilization and estimated task demand

  community-backlogcoreenhancementperformanceschedulertriage
  ray-project/ray#66397 · Filter by author jazzii36 opened on Sep 23, 2026

• [Serve] Convert Controller health probes from a pull to a push model

  enhancementperformanceserve
  ray-project/ray#66396 · Filter by author johntaylor-cell opened on Sep 23, 2026
  1

  

• [Serve] Tracing config changes don't take effect on live proxies/replicas (OpenTelemetry set-once provider)

  community-backlogenhancementobservabilityservestabilitytriageusability
  ray-project/ray#66385 · Filter by author suppagoddo opened on Sep 22, 2026

• [Serve] Avoid extra controller RPC on replica startup for tracing config

  bugcommunity-backlogperformanceservetriage
  ray-project/ray#66382 · Filter by author suppagoddo opened on Sep 22, 2026
  2

• [Data] CUDA allocations on DGX Spark (unified memory) are not reflected in average_max_uss_per_task

  bugcommunity-backlogdatadocsgpuobservabilitystabilitytriageusability
  ray-project/ray#66335 · Filter by author xuwenyihust opened on Sep 20, 2026

• [core][gcs] HandleReportAutoscalingState success callback reads a stack local after return (stack-use-after-return)

  community-backlogcorestability
  ray-project/ray#66333 · Filter by author LuciferYang opened on Sep 20, 2026

• [core][gcs] GCS crashes on UpdateWorkerDebuggerPort/UpdateWorkerNumPausedThreads for an unknown worker

  community-backlogcorestability
  ray-project/ray#66331 · Filter by author LuciferYang opened on Sep 20, 2026