Skip to content

Add regional affinity to multi-cluster function autoscaling #1068

Description

@nv-lacm

Description

Improve multi-cluster scale-up placement for all NVCF workloads. Prefer an eligible cluster in the same region as current demand. When that region is unavailable, select the nearest eligible region using a documented proximity policy.

Customer placement constraints and existing health, capacity, quota, and instance-type checks remain authoritative. Dynamo Planner continues to manage pods within an NVCF instance; NVCF owns instance creation and cross-cluster placement.

Definition of Done

  • Scale-up can target a specific eligible cluster.
  • A same-region cluster is preferred when available.
  • Cross-region fallback uses a deterministic, documented proximity policy.
  • Explicit customer placement constraints override the default preference.
  • Missing region metadata and unavailable capacity have documented fallback behavior.
  • Single-cluster behavior is unchanged.
  • Logs and metrics identify the selected cluster, policy, and fallback reason.
  • End-to-end tests cover same-region selection, nearest-region fallback, customer overrides, missing metadata, and no eligible cluster.
  • Documentation explains the NVCF and Dynamo Planner responsibility boundary.

Resources (optional)


By submitting this issue, you acknowledge that you are an assigned member of the NVCF development team and agree to follow our code of conduct and our contributing guidelines.

Metadata

Metadata

Assignees

Labels

roadmapRanked roadmap item

Type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions