Skip to content

Supervisor insight: report context and KV capacity pressure #18

Description

@igorls

Problem

Operators cannot currently see how close the active configuration and traffic are to exhausting context capacity.

Required insight

Add one insight object to the existing /api/insights computation and render that same object in the dashboard. It should combine:

  • the prompt-token distribution in the measured request window;
  • configured max context;
  • KV capacity and current/high-water page utilization;
  • max concurrency and active concurrency;
  • representative request IDs near the observed limit.

The insight should distinguish large individual prompts from concurrency-driven KV pressure and mark any recommendation as inferred.

Acceptance

Fixtures cover a near-max-context request, concurrency-driven high KV utilization, a comfortably sized workload, and missing capacity telemetry. Missing data must render unavailable rather than zero pressure.

Split from #12.

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

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions