Repository Issue Activity (beta)

#5365 Release v1.16.0 Roadmap

#4195 Is it possible to block podgroups from transitioning from "Completed" to "Inqueue"?

#5317 feat: add queue-level preemptable switch for preempt action

#5972 :bug: StatefulSet PodGroup Can Be Incorrectly Deleted During Rapid Scale-to-Zero/Scale-Up

#5131 TaskEligibleToPreempt blocks gang job tasks that don't need preemption, causing entire gang to fail

#5958 Should gangreclaim consider resource fragmentation when selecting victims?

#5994 gangreclaim: avoid unnecessary evictions when a subset of a SAFE bundle is sufficient

#5984 gangpreempt: avoid unnecessary evictions when a subset of a SAFE bundle is sufficient

#5998 [Feature] Checkpoint-aware task victim ordering for preemption and reclamation

#5846 HAMi DeviceShare can skip Ascend device enforcement after NodeInfo initializes before DeviceShare

#5133 [Feature] Decoupling HyperNode and PodGroup Controllers: Evolving Toward a Diversified AI Orchestration Ecosystem

#5627 addDRAResource: unbounded loop over a user-controlled DRA device count can hang the scheduler

#5989 [release-1.13] Fix Spark E2E image builds with Docker-backed Minikube

#5988 [release-1.14] Fix Spark E2E image builds with Docker-backed Minikube

#5987 [release-1.15] Fix Spark E2E image builds with Docker-backed Minikube

#5973 Recreating a VCJob with the same name may cause its Pods to never be created

#4837 [release-1.12] Fix scheduler panic when metrics are disabled

#5976 e2e-spark fails on every recent PR: buildx cannot inspect moby/buildkit:buildx-stable-1

#5986 Stale Pod informer events can corrupt JobInfo Pod cache after Pod recreation

#5331 Stale PodGroup may be patched to newly recreated Pods after StatefulSet recreation, causing Pods to remain Pending permanently

#5979 [release-1.15] fix: close int64-overflow bypass in DRA device count accounting

#4796 Cherry-pick PR #4786 to release-1.13 branch

#5911 :bug: The capacity plugin may incorrectly prevent a job from being scheduled

#5160 [Proposal] Per-user fair share scheduling plugin with decayed usage tracking

#5710 allocation-rate shard policy assigns almost no nodes when the optional arguments block is omitted