Skip to content

Optimization for heavily CPU-bound workloads leading to severe RAM over-provisioning #576

Description

@taltsafrirpx

When running heavily CPU-bound workloads where memory requests are minimal compared to CPU requests, Karpenter provisions standard instance types (e.g., GCP c4a) that pack requested CPU near 100% while leaving allocated memory utilization/requests under 20%.

Consolidation does not replace these nodes because no smaller standard instance fits the CPU requirements, and Karpenter does not prioritize high-CPU/compute-optimized instance families when calculating optimal node replacement.

  • CPU Requests: ~99% (e.g., 71306m / 72000m)
  • Memory Requests: ~15% - 22%
  • Consolidation state: Idle / No action taken

Improve consolidation heuristics or instance-selection logic to prioritize high-CPU / low-memory instance families when memory request percentage across pods is disproportionately low compared to CPU requests.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions