Skip to content

ci(nightly): retire the dead macOS Colima job#422

Merged
haksungjang merged 1 commit into
mainfrom
ci/retire-macos-colima-job
Jul 16, 2026
Merged

ci(nightly): retire the dead macOS Colima job#422
haksungjang merged 1 commit into
mainfrom
ci/retire-macos-colima-job

Conversation

@haksungjang

Copy link
Copy Markdown
Member

Why

nightly's "macOS real scan (Colima)" was chartered as best-effort because Colima startup was believed "slow and flaky (infra)". Measurement says otherwise:

  • Last 19 runs: 19 failures, all deterministic, all in 17–40 seconds (nowhere near the 45-min timeout), all at colima start --vm-type=vz — the scan step never executed once.
  • vz backend: the VM exits immediately after boot (Exiting:true, empty error array) — hosted Apple Silicon runners do not provide nested virtualization.
  • qemu backend: verified by a one-off experiment on a throwaway branch (run 29487846488, 2026-07-16) — lima 2.1.3's QEMU driver panics (send on closed channel, qemu_driver.go:412) before the VM starts.

So the job cannot succeed on macos-latest with either backend, and its continue-on-error + exclusion from the failure alarm made it invisible: a permanently red job burning a runner slot nightly while its comment claimed the result was "visible in the step summary" (the summary only ever printed the failure branch).

What

  • Remove the job; leave the evidence and the re-add condition (hosted arm64 nested virtualization, or a lima qemu driver that boots there) as an in-place comment.
  • Drop the now-moot exclusion note on report-failure.
  • macOS real-scan coverage remains a maintainer-run local check on real Apple Silicon hardware, where Colima works (release verification already runs there).

The experiment workflow lives only on a throwaway branch and is deleted with it — nothing lands in main.

The job carried a "slow and flaky (infra)" charter, but the measured
record disagrees: its last 19 runs all failed deterministically in
17-40 seconds at `colima start`, and the scan step never executed. On
the hosted arm64 runner neither backend boots — vz exits immediately
(no nested virtualization on hosted Apple Silicon), and a one-off
experiment (run 29487846488) showed lima 2.1.3's qemu driver panics
("send on closed channel") before the VM starts.

A permanently red best-effort job burns runner time and reads as
"occasionally passing" when it never does; its step summary promised a
scan result that was never produced. Remove it and record the evidence
and the re-add condition (hosted arm64 nested virtualization) in
place. macOS real-scan coverage remains a maintainer-run local check
on real Apple Silicon hardware, where Colima works.
@haksungjang
haksungjang merged commit 551d720 into main Jul 16, 2026
33 checks passed
@haksungjang
haksungjang deleted the ci/retire-macos-colima-job branch July 16, 2026 14:16
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant