Skip to content

perf(controller): measure the presence-aware cap's effect on melo-lap-001 and tune it #421

Description

@kyle-sexton

Summary

v0.1.30 enabled the presence-aware cap (#410) on melo-lap-001 with interactiveMaximumConcurrentWorkers: 8 and interactiveIdleThreshold: 10m. Both values are judgment, and #391 never ran its paused-fleet A/B, so neither the per-job hook-latency effect nor the cap's benefit has been measured. Re-measure after about a week of traffic and tune or remove the cap.

Evidence

Fix

After roughly a week on v0.1.30 (from 2026-09-28):

  1. From jobs.json and the worker logs, measure the queue delay added while interactiveCapActive was true against the perf(controller): fleet job load doubles Claude Code hook latency on dev hosts; v0.1.24 idle controller cost (melo-lap-001) #391 replay estimate.
  2. From the Claude Code OTEL store (C:\ProgramData\local-otel\cc-store), compare hook latency at 9-12 running jobs (idle periods, cap off) against 8 or fewer (cap on), using perf(controller): fleet job load doubles Claude Code hook latency on dev hosts; v0.1.24 idle controller cost (melo-lap-001) #391's method.
  3. Decide: keep 8, move it, or turn the cap off; change hosts/melo-lap-001/ci-runner.yaml in provisioning accordingly.

Verification

A comment on this issue with both measurements and the decision, plus the provisioning PR if the value changes.

Related

🤖 Generated with Claude Code

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