You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Scheduled full test failed: 0e475ceb | 定时全量测试失败: 0e475ceb #7108
Root cause identified for the windows-22-5 / windows-24-5 reds: tests/server-process-listeners.test.ts (the #7026 graceful-shutdown / signal-handler test) — file-level exitCode 1, reproduces in isolation, full retry also fails. CI's own diagnostic: "deterministic break, not a flake".
[PR] fix: release process handlers when an embedded server closes #7026 (894835ac94) merged to master today at 01:46 UTC. Its test's 4th case ("graceful shutdown keeps signal handlers until the process exits") sends real SIGINT then SIGTERM to the process and stubs process.exit; on Windows these signals are emulated, so the file exits non-zero deterministically.
Confirmed NOT PR-specific: the identical failure signature appears in unrelated PR runs #44245921438, #44246105190, #44249183921 (all windows-latest, 22, 5).
ci-fixes has no fix for it yet (and is already an ancestor of the affected PR heads, so merging it is a no-op).
Suggested fix direction: make the #7026 test Windows-compatible (guard the real-signal cases to non-Windows, or assert the Windows-emulated behavior); if the shutdown code genuinely drops handlers on Windows, fix that instead. Belongs on master/ci-fixes, not individual PR branches.
Activity
[-]定时全量测试失败: 0e475ceb[/-][+]Scheduled full test failed: 0e475ceb | 定时全量测试失败: 0e475ceb[/+]ranxianglei commentedon Nov 18, 2026
(ai: vllm-qwen/qwen3.8-27b)
Root cause identified for the windows-22-5 / windows-24-5 reds:
tests/server-process-listeners.test.ts(the #7026 graceful-shutdown / signal-handler test) — file-levelexitCode 1, reproduces in isolation, full retry also fails. CI's own diagnostic: "deterministic break, not a flake".894835ac94) merged to master today at 01:46 UTC. Its test's 4th case ("graceful shutdown keeps signal handlers until the process exits") sends realSIGINTthenSIGTERMto the process and stubsprocess.exit; on Windows these signals are emulated, so the file exits non-zero deterministically.windows-latest, 22, 5).ci-fixeshas no fix for it yet (and is already an ancestor of the affected PR heads, so merging it is a no-op).Suggested fix direction: make the #7026 test Windows-compatible (guard the real-signal cases to non-Windows, or assert the Windows-emulated behavior); if the shutdown code genuinely drops handlers on Windows, fix that instead. Belongs on master/ci-fixes, not individual PR branches.
ranxianglei commentedon Nov 18, 2026
(已收)本单已归入根因聚合 issue(token 见聚合单 body),机器人已停止在本单上的工作,修复走聚合 issue 统一出 PR;本单保留不关,修复验证后统一关闭。(聚合单:[Root Cause] c-i-scheduled-full-test-red (2 items),GitHub #7157)
github-actions commentedon Nov 18, 2026
定时全量已在 a446a3c 恢复绿: https://github.com/ranxianglei/billion-context/actions/runs/44379703370
fix(tests): bound-retry the ranxianglei#5631 post-restart cold-load r…