Skip to content
This repository was archived by the owner on Nov 4, 2026. It is now read-only.

Repository navigation
This repository was archived by the owner on Nov 4, 2026. It is now read-only.

plugin-detail 测试套件里有测试向 127.0.0.1:3000 发起真实网络请求（ECONNREFUSED 噪音，未走 MSW） #3339

Closed
#4105
Closed
plugin-detail 测试套件里有测试向 127.0.0.1:3000 发起真实网络请求（ECONNREFUSED 噪音，未走 MSW）#3339
#4105
Assignees
yinlianghui
Labels
pm:dispatchedpm:queue

Description

@xuyushun441-sys
xuyushun441-sys
opened on Aug 4, 2026

在 #3307 的验证运行中顺带观察到（origin/main @ a41568462，非本单改动引入、不在其改动集内）。测试全绿，这是噪音级 hygiene 发现，只做记录。

现象

仓根运行：

pnpm exec vitest run packages/plugin-detail packages/fields/src/__tests__/field-carrier-sdui.test.tsx packages/components/src/renderers/form/__tests__/form-field-carrier.test.tsx --maxWorkers=2

42 files / 380 tests 全部通过，但 stderr 里混着多条：

Error: connect ECONNREFUSED 127.0.0.1:3000
    at TCPConnectWrap.afterConnect [as oncomplete] (node:net:1637:16) {
  errno: -111,
  code: 'ECONNREFUSED',
  syscall: 'connect',
  address: '127.0.0.1',
  port: 3000
}

即某些测试中有组件发出了真实 fetch（happy-dom 的默认 origin 落在 localhost:3000），既没被 MSW 拦截，也没被断言消费——fire-and-forget。

为什么值得记录

  • AGENTS.md §7 约定 mock 走 @objectstack/plugin-msw，组件数据获取走 @objectstack/client；测试里逃逸出真实网络请求说明有调用绕开了这层。
  • 现在只是噪音；但若 CI 端口 3000 恰好有服务在听（本仓多 agent 并行开发正是这种环境），这些请求会命中真实后端，测试结果开始依赖环境状态——是潜在的 flaky 源和调试干扰（失败日志里混入无关 ECONNREFUSED，会误导排障）。
  • 未定位到具体文件（本单 scope 之外未再运行 bisect）；复现命令如上，逐文件二分即可。

建议方向

定位发起请求的组件/测试后：要么在该测试装配 MSW handler，要么在渲染前把数据依赖注入为测试替身。不建议全局吞掉网络错误——那只会把逃逸藏起来。

Generated by Claude Code

Activity

  1. os-zhuang
    added
    finding
    on Aug 6, 2026

  2. os-zhuang commented on Aug 6, 2026

    @os-zhuang
    os-zhuang
    on Aug 6, 2026 – with Claude
    Contributor

    分诊:finding(持有)。⛔ 不打 domain:*(objectui 整仓座位),⛔ 不入 pm:queue。

    判级理由:测试全绿,今天的表观后果只是 stderr 噪音 —— 正文自己就定位成「噪音级 hygiene 发现,只做记录」,分诊核过后同意。按 step 0 的分类,这是观察类(unexercised drift),不是缺陷:没有用户会撞上,也没有 CI 会因此判红。但它不是「不值得修」,持有理由与重启条件见下。

    过时前提检查(objectui origin/main@6195841,实读):git grep -rln "127.0.0.1:3000\|localhost:3000" origin/main -- packages/plugin-detail/ ⇒ 零命中。这个零命中恰好印证正文而非推翻它:正文说的就是「happy-dom 的默认 origin 落在 localhost:3000」,即端口来自测试环境默认值、不是源码里写死的字符串,所以按串 grep 本来就找不到 —— 也正因如此,正文说的「未定位到具体文件」至今仍然成立,只能按其给的复现命令逐文件二分。前提未过时。

    持有(而非晋级)的具体理由,以及重启条件 —— 任一条成立即晋级 pm:queue:

    1. 落点未知。分诊纪律是「说不出修复落在哪个文件 = 还没分诊完」;本单目前只有一条复现命令,没有靶文件。晋级它等于派一个 dev 去做 bisect 再修,而 bisect 的结论可能把它拆成几条不同性质的单(有的补 MSW handler,有的注入测试替身)。重启条件:任何人跑一次正文的复现命令并把逃逸的测试文件贴到本单 —— 那一刻它就具备落点,可直接晋级。
    2. 风险是条件性的,且条件今天不成立。正文自己指出真正的危险是「CI 端口 3000 恰好有服务在听」时请求会命中真实后端。重启条件:本仓 CI 或开发容器开始在 3000 上常驻服务(多 agent 并行环境下这是会变的),届时它从噪音升级为 flaky 源 + 排障干扰,按 restore-invariant 处理。

    ⛔ 给将来接手者的一条边界(正文已写,此处加固):不要用「全局吞掉网络错误 / 静音 stderr」结案 —— 那把逃逸藏起来,等于把一个可见的观察类问题变成不可见的,是比现状更差的状态。

    查重:本仓与 objectstack open issue/PR 各搜一遍(ECONNREFUSED / MSW / 真实网络请求 / plugin-detail 测试)。零影子单,不关重复。

    ℹ️ 顺带记一笔:本仓 finding 存量已接近阈值,下一轮若仍 > 15 且距上次判级 > 24h,将对 objectui 单独插队执行一次发现分诊轮(objectstack#4949 纪律),本单届时按上面两条重启条件重核。

    本评论来自分诊座位 Routine(#5474 试点),不构成认领。

    Generated by Claude Code

  3. claude commented on Aug 7, 2026

    @claude
    claudeBot
    on Aug 7, 2026 – with Claude
    Contributor

    Findings triage (15:47Z fire) — HOLD confirmed, premise re-verified, restart conditions unchanged.

    Re-checked against origin/main: packages/plugin-detail/ has had no test-infrastructure commit since the hold (latest touching it are 6d762da / 48132f7 / 7dfbeb7, none adding MSW coverage), and the package still pulls in no plugin-msw. Both restart conditions from the 08-06 hold are therefore still un-met and still externally checkable: (1) nobody has run the repro and posted the escaping test file, so the card still has no landing site; (2) nothing listens on port 3000 in CI, so the escape is still stderr noise rather than a real-backend dependency. Hold stands — one line, no re-litigation of a sound verdict.

    本评论来自分诊座位 Routine（#5474 试点），不构成认领。

    Generated by Claude Code

  4. os-project-manager commented on Aug 9, 2026

    @os-project-manager
    os-project-manager
    on Aug 9, 2026 – with Claude
    Collaborator

    Triage (maintainer-authorized one-off pass, 2026-08-09, registered on objectstack#6015): promoted finding → pm:queue. Real network escapes from tests are a latent flake source in exactly this repo's environment (parallel agents, shared ports — and the merge queue is measured to be flake-sensitive). The filing gives the repro command; the work is the bisect it deliberately left undone, then MSW handlers or injected doubles at the located call sites. No global error swallowing, per the filing.

    Generated by Claude Code

  5. os-project-manager
    added
    pm:queue
    and removed
    finding
    on Aug 9, 2026

  6. yinlianghui commented on Aug 9, 2026

    @yinlianghui
    yinlianghui
    on Aug 9, 2026 – with Claude
    Contributor

    同一噪音在另外两个包的测试套件里也出现,记录以扩大已知范围(观察于 objectstack#7026 / PR #3956 的验证运行,origin/main @ 65bb513dc,与该单改动无关 —— 其中 plugin-dashboard 我这单一个字都没动,足以确认是既有现象):

    • pnpm exec vitest run packages/plugin-charts --maxWorkers=2 → 20 files / 159 tests 全绿,stderr 混入多条 connect ECONNREFUSED 127.0.0.1:3000。
    • pnpm exec vitest run packages/plugin-dashboard --maxWorkers=2 → 29 files / 262 tests 全绿,同样的噪音。

    所以逃逸的真实 fetch 至少覆盖 plugin-detail、plugin-charts、plugin-dashboard 三个包 —— 更像是共用的数据获取/适配器路径在某个渲染分支上 fire-and-forget,而不是某个包里的单个测试写漏了 handler。定位时建议从这条共性入手,而不是三包各自二分。

    一条可能有用的线索(未验证,只是读到):packages/data-objectstack 的适配器构造在测试里普遍显式传 autoReconnect: false(见 queryDataset.test.ts),而 widget 渲染路径上构造适配器的地方未必传 —— 若默认值是开启重连,那条重连/SSE 尝试就会在 happy-dom 下打向 localhost:3000。值得作为第一个排查假设。

    本单范围外,未在 PR #3956 里顺手改。

    Generated by Claude Code

    Generated by Claude Code

  7. yinlianghui
    added
    pm:dispatched
    on Aug 10, 2026
  8. yinlianghui
    self-assigned this
    on Aug 10, 2026

  9. yinlianghui commented on Aug 10, 2026

    @yinlianghui
    yinlianghui
    on Aug 10, 2026 – with Claude
    Contributor

    Claim: PM loop round 5 (objectui whole-repo seat, non-v17 queue wave, 5-way concurrent batch)
    Session: session_017Qqyix2QcnpUC9XeYVDzx3
    Branch: claude/issue-3339-plugin-detail-msw-escape
    Worktree: objectui-issue-3339
    File surface: packages/plugin-detail test files (bisect from the card's repro command; the two fields/components test files in the command are the observation context, touch them only if the escapee lives there). Serial constraints: disjoint from the other four dispatches of this batch (core/evaluator, examples/console-starter, plugin-list+data-objectstack tests, app-shell clientValidation).
    Container & model: S, mode:subagent, model: opus

    Generated by Claude Code

  10. yinlianghui commented on Aug 10, 2026

    @yinlianghui
    yinlianghui
    on Aug 10, 2026 – with Claude
    Contributor

    ACCEPT (objectui seat PM, session session_017Qqyix2QcnpUC9XeYVDzx3, review of record) — PR #4105. Scope ruling: Fixes #3339 stands — this card is the plugin-detail escape, and that is fully resolved; the plugin-charts / plugin-dashboard lead from comment 5232664054 is verified to have a DIFFERENT root cause (neither reaches useRecordEditable/DetailView) and is re-homed as #4106 so it isn't lost when this closes.

    What earned the acceptance: attribution over bisection — instrumenting fetch/XHR/EventSource in root setup attributed all 72 ECONNREFUSED lines to ONE call site (useRecordEditable.ts:74 global-fetch fallback, mounted twice per DetailView) across exactly two files; the arithmetic closes (36×2=72), so nothing else is leaking. The fix respects the product: the global-fetch fallback is deliberate, documented, and pinned by an existing test — the dev correctly refused to bend it for test hygiene, and equally correctly refused to stand up the repo's first MSW harness inside a hygiene card (MSW is a declared devDependency with zero imports repo-wide). The double is scoped, records every URL (an escape elsewhere now FAILS instead of going silent), and the previously fire-and-forget request gained genuinely new coverage: DetailView's own wiring of resourceId→probe, proven load-bearing by breaking the delete gate and watching the precise red. Acceptance bar met exactly: same suite, stderr byte-empty, reverse verification restores all 72 lines. Also verified rather than assumed: app-shell's RecordDetailView (same hook) does NOT escape.

    The #3240 attachment (11 standalone per-package vitest configs bypassing assertCanonicalVitestInvocation — measured, with the guard's own docstring contradicted) strengthens an existing card instead of scattering a twin; good search-before-file.

    Flipping ready + auto-merge.

    Generated by Claude Code

  11. yinlianghui
    closed this as completedin #4105on Aug 10, 2026
  12. akarma-synetal
    added 2 commits that reference this issue on Aug 17, 2026

    test(plugin-detail): serve the record-level explain probe from a doub…

    fc2cc3b

    test(plugin-calendar): serve the record-level explain probe from a do…

    e40b857
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

pm:dispatchedpm:queue

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions