Repository Issue Activity (beta)

#1673 Reservation pods assigned identical gpu-group cause invisible GPU accounting error

#1528 Provide per workloads per fractional GPU metrics service for external metrics collectors (e.g. Prometheus) to use

#1524 Support limiting fractional GPU memory ('Dynamic Fractions')

#1523 Support per shard administrative GPU compute sharing mode

#1522 Support requesting GPU compute sharing mode and policy

#1520 Support GPU sharing by requesting fractional GPU memory

#1521 Support requesting number of fractional GPU devices

#1536 fix(chart): kai-config Helm hook recreates SA on upgrade, causing leader election 401 when scheduler pods are not restarted

#1671 Admission block external Block NVIDIA_VISIBLE_DEVICES

#1543 feat(admission): scope nvidia runtime class injection to fraction pods only

#1664 Scheduler panics (nil pointer) in preempt when a scenario has no tasks to allocate

#1365 feat(binder): make binder plugins configurable

#1410 Improve performance for reclaim/preempt scenarios with many victims

#1348 Elastic sub-grouping per grouping level

#1527 podgrouper - declarative pod grouping for arbitrary CRDs via Karta

#1598 Support gpu NUMA topology aware scheduling

#1555 [Design] add support for Karta in pod grouper

#1544 feat: CDI/NRI-native GPU injection for fraction pods

#1400 Design for limits per resource type

#1353 refactor(scheduler): vectorize resource representation for performance at scale

#1603 Research Numa topology solution

#1189 Automatic subgroup creation per JobSet replica in the JobSet podgrouper

#1350 Scheduler Core Logic for minSubGroup

#1623 e2e suite should refuse to run against non-local clusters by default

#1658 Cross ResourceClaim device topology constraints within a PodGroup