ci: run perf jobs on bamboo

[jdx]

Summary

• run Tak measurement jobs on the isolated bamboo-perf runner
• identify the Bamboo image/toolchain as a distinct Tak runner class
• leave report and publish jobs on hosted runners

Validation

• actionlint (with bamboo-perf registered as the expected custom label)
• git diff --check
• trusted main baseline queued on the disposable runner

AI-assisted — Tool: Codex; model: unavailable; version: unavailable.

---

Note

Medium Risk
Perf baselines and PR gates now depend on a self-hosted runner and a new Tak series ID, and perf-pr only runs for one author until that guard is removed—wrong runner availability or label config would break CI signal without touching app code.

Overview
Moves Tak instruction-count measurement in perf.yml and perf-pr.yml from GitHub-hosted ubuntu-24.04 to the self-hosted bamboo-perf runner, and registers that label in .github/actionlint.yaml so workflow linting accepts it.

Benchmark metadata now uses TAK_RUNNER: bamboo-v2-ubuntu24.04-x64-30vcpu-24gb-rust1.97.1-tokio1, explicitly starting a new measurement series (comments call out incompatibility with the old hosted-runner baseline). The workflows drop RUSTUP_TOOLCHAIN and the manual rustup toolchain install step, relying on mise-action on the Bamboo image instead.

perf-pr additionally gates measure and report with github.event.pull_request.user.login == 'jdx' so only that author’s PRs run the expensive compare/comment path; report/publish-style jobs that stay on ubuntu-latest are unchanged aside from that condition.

^{Reviewed by Cursor Bugbot for commit cefaafa. Bugbot is set up for automated code reviews on this repo. Configure here.}

Summary by CodeRabbit

• Chores
  • Updated performance testing workflows to use a standardized, pinned execution environment.
  • Improved consistency and comparability across performance test runs.
  • Simplified workflow setup by removing redundant environment configuration and installation steps.
  • Improved the reliability and repeatability of performance measurements across runs.
  • Added dedicated runner configuration for performance testing to support more predictable execution.

[coderabbitai]

📝 Walkthrough

Walkthrough

Both performance workflows now run measurement jobs on bamboo-perf. They select the pinned Bamboo Rust and Tokio environment through TAK_RUNNER. The pull-request workflow no longer installs a benchmark Rust toolchain explicitly. Actionlint now accepts the runner label.

Changes

Performance workflow runner migration

Layer / File(s)	Summary
Configure Bamboo performance runners .github/workflows/perf-pr.yml, .github/workflows/perf.yml, .github/actionlint.yaml	The measurement jobs now use bamboo-perf and the pinned Bamboo TAK_RUNNER identifier. GitHub-hosted runner settings and explicit benchmark toolchain installation were removed. Actionlint now defines the self-hosted runner label.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Possibly related PRs

• jdx/fnox#656: Updates the performance workflow runners and Rust environment.
• jdx/fnox#670: Modifies the performance workflow runner and Rust toolchain configuration.

Poem

A rabbit hops on Bamboo lanes,
Pinned Rust runs benchmark trains.
TAK_RUNNER leads the way,
And workflows start the day.
Green checks greet the hay!

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: running performance CI jobs on the Bamboo runner.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[greptile-apps]

Greptile Summary

The PR moves Tak measurement jobs to the dedicated Bamboo performance runner while leaving reporting and publishing on hosted runners.

• Registers bamboo-perf as an actionlint-approved self-hosted runner label.
• Introduces a Bamboo-specific Tak measurement series identifier.
• Restricts pull-request performance jobs to PRs authored by jdx.
• Relies on the Bamboo image’s installed Rust toolchain rather than installing one in each workflow.

Confidence Score: 5/5

This follow-up appears safe to merge because no eligible blocking failure remains.

No blocking failure remains.

Important Files Changed

Filename	Overview
.github/actionlint.yaml	Registers the new bamboo-perf custom runner label for workflow linting.
.github/workflows/perf-pr.yml	Moves PR measurements to Bamboo, adopts a new Tak runner class, removes explicit Rust installation, and author-gates measurement and reporting.
.github/workflows/perf.yml	Moves main-branch measurements to Bamboo and starts a distinct benchmark series tied to the new image and toolchain.

_{Reviews (5): Last reviewed commit: "ci: restrict bamboo perf prs to jdx" | Re-trigger Greptile}

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.github/workflows/perf-pr.yml:
- Line 40: Create .github/actionlint.yaml and declare bamboo-perf under
self-hosted-runner.labels so actionlint recognizes the custom runner label. This
configuration change covers both references in
.github/workflows/perf-pr.yml#L40-L40 and .github/workflows/perf.yml#L40-L40; no
workflow changes are required.

In @.github/workflows/perf.yml:
- Around line 35-40: Split the workflow so the Bamboo-backed measure job remains
read-only and no longer runs tak push or requires contents: write. Add a
separate publish job on ubuntu-latest that receives the generated Git notes and
required measurement state from measure, then performs the Push measurements/tak
push step with contents: write and its GITHUB_TOKEN available only to that job.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Central YAML (base), Organization UI (inherited)

Review profile: CHILL

Plan: Pro Plus

Run ID: b59720f2-262e-4f21-ad9b-f8a0b838a137

📥 Commits

Reviewing files that changed from the base of the PR and between 13e2a7b and f6a881e.

📒 Files selected for processing (2)

• .github/workflows/perf-pr.yml
• .github/workflows/perf.yml

[coderabbitai]

🔒 Security & Privacy | 🟠 Major | 🏗️ Heavy lift

Keep tak push on a hosted publish job.

Push measurements is still a step inside measure, so this runner change also moves the job-level contents: write permission and the GITHUB_TOKEN used by tak push onto bamboo-perf. That does not meet the PR objective that publish jobs remain on hosted runners. Keep measure read-only on Bamboo. Transfer the generated Git notes and required measurement state to a separate publish job on ubuntu-latest, and grant contents: write only there.

🧰 Tools

🪛 actionlint (1.7.12)

[error] 40-40: label "bamboo-perf" is unknown. available labels are "windows-latest", "windows-latest-8-cores", "windows-2025", "windows-2025-vs2026", "windows-2022", "windows-11-arm", "ubuntu-slim", "ubuntu-latest", "ubuntu-latest-4-cores", "ubuntu-latest-8-cores", "ubuntu-latest-16-cores", "ubuntu-24.04", "ubuntu-24.04-arm", "ubuntu-22.04", "ubuntu-22.04-arm", "macos-latest", "macos-latest-xlarge", "macos-latest-large", "macos-26-intel", "macos-26-xlarge", "macos-26-large", "macos-26", "macos-15-intel", "macos-15-xlarge", "macos-15-large", "macos-15", "macos-14-xlarge", "macos-14-large", "macos-14", "self-hosted", "x64", "arm", "arm64", "linux", "macos", "windows". if it is a custom label for self-hosted runner, set list of labels in actionlint.yaml config file

(runner-label)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In @.github/workflows/perf.yml around lines 35 - 40, Split the workflow so the
Bamboo-backed measure job remains read-only and no longer runs tak push or
requires contents: write. Add a separate publish job on ubuntu-latest that
receives the generated Git notes and required measurement state from measure,
then performs the Push measurements/tak push step with contents: write and its
GITHUB_TOKEN available only to that job.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit cefaafa. Configure here.}

[cursor]

PR perf jobs skip other authors

Medium Severity

The perf-pr workflow still triggers on every pull request, but the new if conditions on measure and report run those jobs only when the PR author is jdx. Other contributors get no instruction-count comparison, sticky comment, or regression gate, despite the workflow’s stated purpose of gating regressions on the PR that caused them.

Additional Locations (1)

• .github/workflows/perf-pr.yml#L144-L145

 

^{Reviewed by Cursor Bugbot for commit cefaafa. Configure here.}

[github-actions]

Instruction counts

benchmark	trend	instructions	Δ	wall (min)	Δ
schema	▁█	7,061,558 → 7,062,951	+0.02%	3.27 → 3.22ms	-1.62%
usage	▁█	10,869,387 → 10,870,612	+0.01%	3.51 → 3.41ms	-2.91%

No instruction-count regression above 1.25%.

Measured on the base but not here — a benchmark that stops running also stops gating: schema on gha-ubuntu-24.04-x64-rust-1.91.1-tokio1, usage on gha-ubuntu-24.04-x64-rust-1.91.1-tokio1

_{Only instruction counts gate. Wall clock is shown for context — on identical hardware it moves 4-20% run to run.}

_{Measured by tak — instruction-counted CLI benchmarks, stored in this repository's git notes.}

_{cefaafa4b147 vs 13e2a7bbc6af · measured on the runner, not pushed to the history.}