Skip to content

[copilot-opt] Investigate 75-94 minute runtimes in CGO, CWI, and License Compliance Check workflows #53450

Description

@github-actions

Problem

Several validation/CI-style workflows (CGO, CWI, License Compliance Check) are the longest-running jobs in the observed window, taking 73–94 minutes to reach completion. Because these are gating checks that feed into PR review and merge decisions, their long tail directly delays feedback to contributors (including Copilot agents iterating on PRs) and lengthens the overall PR cycle time.

Evidence

  • Analysis window: 2026-08-17 16:01 UTC to 2026-08-17 17:49 UTC (50 most recent workflow-run session records)
  • Sessions analyzed: 50
  • Key metrics (duration = updated_at - created_at):
    • CGO: 5627s (~94 min), 5484s (~91 min), 4496s (~75 min), 4365s (~73 min) across 4 runs — all still in_progress or null conclusion at capture time, suggesting these are long-lived/stuck rather than genuinely computing for 90 minutes
    • CWI: 5367s (~89 min), 5366s (~89 min), 4743s (~79 min), 4588s (~76 min) across 4 runs, same pattern
    • License Compliance Check: 5187s (~86 min), completed with success
    • By contrast, comparable validation workflows in the same window (Ponytail Reviewer, Matt Pocock Skills Reviewer, Impeccable Skills Reviewer, PR Data Prefetch) all completed in 15–36 seconds
    • This ~150x latency gap between fast reviewer workflows and CGO/CWI/License Compliance Check strongly suggests these jobs are polling, waiting on an external gate, or running an unbounded/serial step rather than doing 90 minutes of genuine compute

Proposed Change

Instrument CGO, CWI, and License Compliance Check with step-level timing (or enable ACTIONS_RUNNER_DEBUG/step summaries) to identify which specific step accounts for the multi-minute-to-hour duration. If the delay is caused by a serial dependency-install, full-repo scan, or an unbounded wait/poll loop, parallelize the step, cache dependencies, or add a timeout with a fast-fail path so PR authors get feedback in line with the other reviewer workflows (under 1–2 minutes).

Expected Impact

  • Cuts PR feedback latency for these 3 gating checks from ~75–94 minutes toward a target of a few minutes, matching sibling reviewer workflows
  • Reduces the runway during which Copilot agents wait idle on gating results before iterating further on a PR
  • Frees Actions runner minutes that would otherwise be consumed by long-idle/polling jobs

Notes

  • Distinct root cause category: tool/orchestration payload-latency optimization
  • Data quality caveat: exact per-step timing was not available from the session-list summary (only overall run start/update timestamps); the recommendation is based on aggregate duration outliers relative to comparable workflows, and confirming the specific slow step requires the step-level instrumentation proposed above.

Generated by ⚡ Copilot Opt · auto · 45.1 AIC · ⌖ 3.41 AIC · ⊞ 8.2K · ◷

Activity

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

Metadata

Metadata

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions