ci: cancel superseded runs and trim the PR wheel matrix by kanavsetia · Pull Request #419 · qBraid/pyqasm

kanavsetia · 2026-08-26T16:29:01Z

Why

pyqasm is the largest single contributor to the org's GitHub Actions bill. In the August billing period it accounted for ~$219 of ~$407 gross Actions spend — and almost all of it is macOS:

$196.29   3,166 min @ $0.062/min   Actions macOS 3-core    (~10x the Linux rate)
$ 22.65   2,265 min @ $0.010/min   Actions Windows

The effect shows up clearly in the org-wide effective rate — August used 36% fewer Actions minutes than July but cost more, because the per-minute rate nearly doubled:

Month	Minutes	Gross	$/min
June	38,530	$247.52	$0.00642
July	55,083	$371.60	$0.00675
Aug (MTD)	34,998	$407.32	$0.01164

macOS minutes went from 658 in July to 3,309 in August.

Two causes:

No concurrency guard. synchronize fires on every push to an open PR, so each push started another full 20-entry wheel matrix while the previous one kept running to completion.
Every PR built the complete matrix — 5 Python versions × Windows / Linux / macOS x86_64 / macOS arm64, i.e. 10 macOS jobs per run — even for a one-line docs change.

What this changes

1. Concurrency guard

concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
  cancel-in-progress: ${{ github.event_name == 'pull_request' }}

Superseded PR runs are cancelled. Pushes to main are never cancelled.

2. Per-event matrix selection

Event	Entries	Composition
`pull_request`	7	Linux × 3.10–3.14, macOS arm64 × 3.11, Windows × 3.11
`push` to main / `workflow_dispatch`	20	unchanged — the full original matrix

macOS jobs per PR run drop from 10 → 1.

Invariants preserved

The full 20-entry matrix is byte-for-byte the same set as the previous inline list (verified by diffing the parsed entries), so main and manual runs behave exactly as before.
macOS arm64 / 3.11 is retained in the PR set on purpose — the code-coverage job consumes the artifact that specific entry uploads (if: matrix.os == 'macos-14' && matrix.python == 311). Dropping it would silently break coverage.
Published wheels are unaffected. PyPI wheels are built by release.yml and pre-release.yml, which each carry their own complete matrix. main.yml is CI validation only.

Tradeoff

PRs no longer validate wheel builds on Windows 3.10/3.12/3.13/3.14, macOS x86_64 at all, or macOS arm64 outside 3.11. A platform-specific build break in those combinations would surface on merge to main rather than on the PR — and still before any release, since release.yml builds the full matrix independently.

If that feels too loose, the natural middle ground is adding macos-14 / 3.14 to the PR set to cover the newest Python on arm64; that's one extra macOS job instead of nine.

CI is the largest single line in the org's GitHub Actions bill. In the August billing period pyqasm accounted for ~$219 of ~$407 gross Actions spend, almost all of it macOS: 3,166 macOS minutes at $0.062/min (about 10x the Linux rate) plus 2,265 Windows minutes at $0.010/min. Two things drove that: 1. No concurrency guard. `synchronize` fires on every push to an open PR, so each push started another full 20-entry wheel matrix while the previous one ran to completion. 2. Every PR built the complete matrix -- 5 Python versions across Windows, Linux, macOS x86_64 and macOS arm64, i.e. 10 macOS jobs per run -- even for a one-line change. This adds a concurrency group that cancels superseded pull_request runs (pushes to main are never cancelled), and selects the matrix per event. Pull requests build 7 entries: Linux on all five supported Pythons, macOS arm64 on 3.11, and Windows on 3.11. Pushes to main and manual workflow_dispatch runs still build all 20. Two invariants are preserved deliberately: - The full 20-entry matrix is unchanged from the previous inline list, so main and manual runs behave exactly as before. - macOS arm64 / 3.11 stays in the PR set because the code-coverage job consumes the artifact that entry uploads. Wheels published to PyPI are built by release.yml and pre-release.yml, which each carry their own complete matrix, so this job is CI validation only and the reduction does not change what ships. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

argus-eye · 2026-08-26T16:29:04Z

Argus review

Auto-review is off for this repo. Tick the box below to run a review on this PR.

Trigger Argus review

Estimated cost

Files changed: 1
Diff lines (±): 156
Historical avg: ~243.6k tokens · ~$0.95 · across last 10 review(s)

Tip: you can also comment @argus-eye review at any time.

coderabbitai · 2026-08-26T16:29:12Z

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 97a19a6b-efad-48c5-86d3-a5c02b755e34

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

🔍 Trigger review

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

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

github-actions · 2026-08-26T16:43:57Z

👋 Hey there! It looks like the changelog might need an update.

Please take a moment to edit the CHANGELOG.md with:

A brief, one-to-two sentence summary of your changes.
A link back to this PR for reference.
(Optional) A small working example if you've added new features.

codecov-commenter · 2026-08-26T16:47:53Z

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

TheGupta2012 · 2026-08-27T06:35:49Z

+            {"os":"ubuntu-latest","python":312,"platform_id":"manylinux_x86_64","manylinux_image":"manylinux_2_28"},
+            {"os":"ubuntu-latest","python":313,"platform_id":"manylinux_x86_64","manylinux_image":"manylinux_2_28"},
+            {"os":"ubuntu-latest","python":314,"platform_id":"manylinux_x86_64","manylinux_image":"manylinux_2_28"},
+            {"os":"macos-14","python":311,"platform_id":"macosx_arm64"},


We should at least keep the macos-15-intel image in the tests. Currently, x86 arch is not being tested for MacOS.

TheGupta2012 · 2026-08-27T06:38:15Z

Thanks for catching this @kanavsetia . However, embedding the choice matrix in a string might not be the best path from a maintenance standpoint. Can you figure out if there's a better approach to cleanly switch the python and os matrix based on whether we are running in a PR or from main?

kanavsetia requested a review from TheGupta2012 as a code owner August 26, 2026 16:29

TheGupta2012 reviewed Aug 27, 2026

View reviewed changes

Navigation Menu

Uh oh!

ci: cancel superseded runs and trim the PR wheel matrix - #419

ci: cancel superseded runs and trim the PR wheel matrix#419
kanavsetia wants to merge 1 commit into
mainfrom
ci/trim-wheel-matrix-on-prs

kanavsetia commented Aug 26, 2026

Uh oh!

argus-eye Bot commented Aug 26, 2026

Uh oh!

coderabbitai Bot commented Aug 26, 2026

Review skipped

Uh oh!

github-actions Bot commented Aug 26, 2026

Uh oh!

codecov-commenter commented Aug 26, 2026

Uh oh!

TheGupta2012 Aug 27, 2026

Uh oh!

TheGupta2012 commented Aug 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

kanavsetia commented Aug 26, 2026

Why

What this changes

Invariants preserved

Tradeoff

Uh oh!

argus-eye Bot commented Aug 26, 2026

Argus review

Uh oh!

coderabbitai Bot commented Aug 26, 2026

Review skipped

Uh oh!

github-actions Bot commented Aug 26, 2026

Uh oh!

codecov-commenter commented Aug 26, 2026

Codecov Report

Uh oh!

TheGupta2012 Aug 27, 2026

Choose a reason for hiding this comment

Uh oh!

TheGupta2012 commented Aug 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants