Skip to content

ci: retry Windows suite on silent test-process exits - #422

Closed
yimi528 wants to merge 1 commit into
QwenAudio:mainfrom
yimi528:codex/ci-windows-subprocess-413
Closed

yimi528 wants to merge 1 commit into
QwenAudio:mainfrom
yimi528:codex/ci-windows-subprocess-413

Conversation

@yimi528

@yimi528 yimi528 commented Sep 12, 2026

Copy link
Copy Markdown
Contributor

Summary

  • isolate the flaky Windows npm test behavior behind a Windows-only single retry
  • keep non-Windows jobs on the existing single-run path
  • emit warning annotations when the first attempt fails or only the retry passes
  • keep the job red when both attempts fail

This addresses the CI mitigation requested in #413. It does not claim to fix the underlying Windows child-process failure; it keeps the failure visible while avoiding unrelated intermittent red jobs.

Verification

  • git diff --check origin/main...HEAD
  • branch contains only .github/workflows/ci.yml changes
  • Windows CI remains the authoritative full validation

Fixes #413

windows-latest intermittently kills an entire test file's child process: the
runner reports `✖ <file> ... 'test failed'` with a non-zero exit, no failing
assertion and no diagnostics. It lands on a different heavyweight example suite
each run (smart-cockpit gateway composition, AI Passport device relay), fails the
same way on main, and never reproduced in ~120 targeted reruns on
windows-latest / Node 24.15.0 (3 shards x [15 smart-cockpit groups + 15 AI
Passport groups + 10 composition runs] plus three full `npm test` runs).

Retry `npm test` once on Windows only and announce the retry as a warning
annotation, so a flaky pass stays visible in the job instead of hiding a real
regression. Two failed attempts still fail the job; other platforms keep the
single-run behaviour.
@x-lixu

x-lixu commented Sep 17, 2026

Copy link
Copy Markdown
Collaborator

感谢针对 Windows 偶发测试进程退出提供缓解方案,也感谢此前详细的复现调查。维护者评估后,暂不采用当前对所有 npm test 失败统一重试的策略:实现尚未区分已知进程退出与真实测试失败,可能掩盖间歇性回归。我们会优先评估后续诊断日志方案,保留故障现场后再决定是否需要有针对性的重试,因此先关闭本 PR。感谢理解和贡献!

@x-lixu x-lixu closed this Sep 17, 2026
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.

[Bug] windows-latest 偶发 "test failed" 且无诊断信息(测试文件子进程静默非零退出)

2 participants