deploy.yml's concurrency group is keyed on the commit SHA (to dedupe a release push's branch and tag runs), so successive pushes to main never cancel each other. Three merges in five minutes ran three full uncached gates plus both image builds back to back on the self-hosted runner, holding a small CI host at its thermal limit for about 20 minutes.
Ref-keyed cancellation isn't safe as-is: a plain main push could cancel a release run mid-publish, and the surviving run has no tag at HEAD, so the release is never published.
Done when: a superseded non-release push to main stops consuming the runner, and a release push is never cancelled or dropped.
deploy.yml's concurrency group is keyed on the commit SHA (to dedupe a release push's branch and tag runs), so successive pushes to main never cancel each other. Three merges in five minutes ran three full uncached gates plus both image builds back to back on the self-hosted runner, holding a small CI host at its thermal limit for about 20 minutes.
Ref-keyed cancellation isn't safe as-is: a plain main push could cancel a release run mid-publish, and the surviving run has no tag at HEAD, so the release is never published.
Done when: a superseded non-release push to main stops consuming the runner, and a release push is never cancelled or dropped.
Pushed as 617a312e. Design: the concurrency group is deploy- for branch pushes (cancel-in-progress true, so a newer main push cancels the superseded run) and deploy-release- for tag pushes (only another push of the same tag at the same commit can cancel one). A new plan job resolves the v* tag at HEAD; a branch run whose commit carries a v* tag skips gate and all downstream jobs, so the tag run is the only one that gates, builds and publishes the release and no later main push can cancel it. dist-zip now reads the version from plan. Not yet verified live: the runner was still busy with three earlier-workflow runs (c8a46fd9, 64bb1957, 98fb9dfd), so the first run on the new workflow (617a312e) was still queued. Leaving open until a superseded run is seen cancelled.
Pushed as 617a312e. Design: the concurrency group is deploy-<ref> for branch pushes (cancel-in-progress true, so a newer main push cancels the superseded run) and deploy-release-<sha> for tag pushes (only another push of the same tag at the same commit can cancel one). A new plan job resolves the v* tag at HEAD; a branch run whose commit carries a v* tag skips gate and all downstream jobs, so the tag run is the only one that gates, builds and publishes the release and no later main push can cancel it. dist-zip now reads the version from plan. Not yet verified live: the runner was still busy with three earlier-workflow runs (c8a46fd9, 64bb1957, 98fb9dfd), so the first run on the new workflow (617a312e) was still queued. Leaving open until a superseded run is seen cancelled.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
deploy.yml's concurrency group is keyed on the commit SHA (to dedupe a release push's branch and tag runs), so successive pushes to main never cancel each other. Three merges in five minutes ran three full uncached gates plus both image builds back to back on the self-hosted runner, holding a small CI host at its thermal limit for about 20 minutes.
Ref-keyed cancellation isn't safe as-is: a plain main push could cancel a release run mid-publish, and the surviving run has no tag at HEAD, so the release is never published.
Done when: a superseded non-release push to main stops consuming the runner, and a release push is never cancelled or dropped.
Pushed as
617a312e. Design: the concurrency group is deploy- for branch pushes (cancel-in-progress true, so a newer main push cancels the superseded run) and deploy-release- for tag pushes (only another push of the same tag at the same commit can cancel one). A new plan job resolves the v* tag at HEAD; a branch run whose commit carries a v* tag skips gate and all downstream jobs, so the tag run is the only one that gates, builds and publishes the release and no later main push can cancel it. dist-zip now reads the version from plan. Not yet verified live: the runner was still busy with three earlier-workflow runs (c8a46fd9,64bb1957,98fb9dfd), so the first run on the new workflow (617a312e) was still queued. Leaving open until a superseded run is seen cancelled.Verified: run 1129 (
617a312e, queued behind the runner backlog) was cancelled when the next main push, run 1130 (af757762), arrived. Closing.