WG Workload Aware Scheduling Meeting for 2026-07-30

The Working Group outlined a strategy to standardize topology-aware scheduling via Node Feature Discovery (NFD) and defined the integration boundaries between Kubernetes Core Scheduler and Q, aiming to migrate scheduling logic to Core while Q focuses on capacity management.

Key Takeaways:
• NFD proposes decoupling Nvidia TopoGraph into a generic network topology updater and extending Node Feature Groups to support nested structures for complex cluster topologies 5:34.
• Community feedback highlights the need for generic hardware support beyond Nvidia and suggests starting with standard labels for tree-based structures before implementing complex DAG representations 11:55.
• A major strategic discussion focused on migrating scheduling logic from Q to Core Scheduler, aiming to reduce Q's internal complexity by leveraging the Core Scheduler library for topology and workload management 43:20.
• Q maintainers agreed to eventually retire overlapping features, transitioning to a model where controllers integrate with Core Scheduler Workload APIs rather than managing internal Q workloads 49:40.

The community is moving toward a unified scheduling ecosystem where Core Scheduler handles topology and workload constraints, while Q focuses on capacity management and multi-tenancy.

Sources:

• 5:34 Proposal to integrate TopoGraph into NFD as a generic network topology updater.
• 11:55 Discussion on standardizing labels and supporting generic hardware vendors.
• 43:20 Strategic alignment on migrating scheduling features from Q to Core Scheduler.
• [49:40](https://www.youtube.com/watch?v=TayNLUBx76w&t=