Skip to content

flaky: plugin-audit zh-CN 摘要与 cloud-connection marketplace seed 在整仓并发下偶发失败(单跑必过) #3785

Description

@os-zhuang

#3723 / PR #3747 期间两轮整仓 npx turbo test 各偶发失败不同的包,单独重跑及干净工作树上均通过。记录下来,免得下一个 agent 重新排查一遍 —— 我为排除自己的改动做了两轮 stash 对照才确认无关。

观测(2026-07-28,worktree 基于 #3747 分支,改动不涉及这两个包)

第一轮(turbo test,99 任务):

FAIL src/audit-writers.test.ts > audit writers — localized activity summaries (framework#3039)
     > localizes verb + object label to the workspace locale (zh-CN)
@objectstack/plugin-audit: Tests 1 failed | 45 passed (46)

第二轮(turbo test --continue,132 任务)—— plugin-audit 这轮通过,换 cloud-connection 挂:

FAIL src/marketplace-install-local-seed-lookup.test.ts > marketplace install — seed lookup resolution
     > writes target record ids (never raw externalId strings) for package objects unknown to the metadata service
FAIL src/marketplace-install-local-state-machine-exempt.test.ts > marketplace install — state_machine initialStates exemption (#3433)
     > lands every mid-lifecycle seed row (no initialStates rejection on the marketplace seam)
@objectstack/cloud-connection: Tests 2 failed | 62 passed (64)

对照:

  • turbo test --filter=@objectstack/plugin-audit --force → 46/46 通过
  • turbo test --filter=@objectstack/cloud-connection --force → 64/64 通过
  • git stash -u 后干净树上两包同跑 → 全过

归类

#3060 / #3662 同一类:仅在整仓满并行负载下出现,单包必绿。两处都是会启动真实 kernel 的重型测试,疑似负载下的时序/资源争用,但未定位根因 —— 本 issue 首先是记录,其次才是修复请求。

失败断言本身值得看一眼的线索:

建议

按前例(#3662 的处理方式)排查:若是超时类,对齐兄弟用例的显式 timeout;若是真竞态,则是产品 bug 不是测试 bug,需要升级处理。

按 AGENTS.md Prime Directive #10 立此 issue;发现于 #3747 的验证过程,与该 PR 改动无关。

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions