Skip to content

chore: empty PR — enumerate which checks are currently red - #3190

Closed
ranxianglei wants to merge 1 commit into
masterfrom
chore/empty-red-check
Closed

ranxianglei wants to merge 1 commit into
masterfrom
chore/empty-red-check

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

空 PR,无任何变更。目的:让全套 CI 在当前 master 上跑一遍,定位当前红的检查项。结果见评论。

@github-actions

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: chore/empty-red-check (752d381)

Option A — Install this exact build (recommended)

npm install -g billion-context@0.0.3660-pr.3190.b13cd11

This exact version is immutable and reproducible. The rolling pr tag tracks the newest test build across all PRs.

Option B — Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr3190.tgz
npm install -g package

Install instructions are refreshed on each push.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 收到。正在拉取 chore/empty-red-check(752d381)在 GitHub Actions 上的全套 check-run 结果,枚举当前红的检查项。稍后回贴结论(按「结构化八问」体例)。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b)

空 PR 没有 diff,所以红的就是当前 master 的真实状态。全套 CI(head 752d381)共 39 项检查:32 绿 / 7 红。7 个红全部是 test 分片腿;所有守卫(version/kernel/pi-subagents/lane/config-surface/docs-sync/docs-coverage/unused-exports/lock-sync/test-import-graph)、static-gates、e2e 各线(codex/dsh/opencode/pi/registry/hermetic)、build-artifact 全绿。

背景:这套 test 矩阵是最近 #3143 刚拆成 os × node × shard[1,2,3] = 12 条腿(每条腿跑一个确定性文件切片,组内 c=4)。实测:shard1 全绿;shard2 三条红(windows-22 那条绿);shard3 四条全红。每条红腿恰好 # fail 1。

逐一定位(每个红 job 只挂一个用例):

腿 挂掉的用例 文件 报错 性质
u22s3 / u24s3 hook e2e: manual bili start … attached not refused (#1322/#1335/#1660) tests/claude-native.test.ts:1502 断言期望 proxy attached at http://127.0.0.1:N,实际 stderr 是 native-attach: skip …(attachPolicy=own-lane) + #1335 attach gate refuses it by default 确定性(仅 linux,windows 上该 e2e 被 skip)
u22s2 / u24s2 zone live: marked upgrade-replacement fails fast… (#3044) tests/port-zone-live.test.ts listen EADDRINUSE: address already in use 127.0.0.1:18787 抖动(固定 zone 端口 18787 抢端口,#394 同类)
w22s3 / w24s3 performSelfRestart: a replacement that survives SIGTERM… (#3044) tests/auto-restart.test.ts 期望日志匹配 still alive after SIGTERM — manual cleanup required (kill N),实际 aborted: replacement did not become ready within 0.8s 抖动(windows 慢机 0.8s 就绪窗口太紧)
w24s2 gap 3: non-stream + preflight 429 after early commit… tests/wire-exit-gap-cells.test.ts:231 headers arrived before summarization finished (1610ms)(期望 true 实得 false) 疑似时序竞态(单次命中)

八问:

  1. 是什么问题:master 上 test 矩阵有 7 条腿红,但根因不是 7 个,而是「1 个确定性断言失败 + 若干 live 测试的时序/端口抖动」这一类混合。
  2. 什么原因导致的:① 唯一确定性的那个——src/launcher.ts:3873-3886 的 claude-native hooks must not attach to a lifecycle-less proxy — attach gate + self-managed fallback #1335 own-lane attach 门:claude hook 用 BILI_CLAUDE_NATIVE_PORT 钉死端口(strictPort),遇到一个无 watchdog/lifecycle-less 的手动 bili start 守护进程时,pickAttachable 不认它、claude-native hooks must not attach to a lifecycle-less proxy — attach gate + self-managed fallback #1335 分支把它当 squatter 直接抛错拒接;而 [Design] 定死端口 lane(claude 48787 / zcode 48789)端口被占时的 0 配置解法 #1660 的 e2e 却断言"用户手动起的守护进程应默认 attach"——两者语义打架。② 其余都是 live/process 类用例在 c=4、12 条腿并行下的时序/端口竞态(autoRestartOnUpdate + occupied target port -> orphan process leak: new instance silently hops to the next port, old instance never exits #3044 自重启家族 + 一个 preflight-429 竞态)。
  3. 影响范围与严重程度:影响的是 CI 门禁本身(红检查把合并挡在 blocked),不影响任何已发布产物的运行行为。其中 claude-native 那条是真实的行为契约冲突(linux 必现、windows skip),需判定"有意收紧策略留了旧测试"还是"attach 路径回归";其余是 CI 稳定性问题(非确定,跨腿随机命中不同用例)。
  4. 怎么修复的:本 issue 只做诊断,未改任何代码。建议方向(待你拍板):claude-native 冲突先定意图——若属有意收紧则更新该 e2e+文档,否则修 attach 路径让 user-zone 手动守护进程可默认 attach;autoRestartOnUpdate + occupied target port -> orphan process leak: new instance silently hops to the next port, old instance never exits #3044 抖动家族按机制收敛(放宽/隔离 ready 超时、live 端口测试改用 listen(0) 动态端口而非写死 18787);gap-3 单独确认是否 flake。
  5. 是否有回归问题:本 PR 无 diff,自身不引入回归;但 master 现态存在 1 处疑似回归(attach 门 vs [Design] 定死端口 lane(claude 48787 / zcode 48789)端口被占时的 0 配置解法 #1660,见第 2 问)。另注:ci已经运行超过20分钟了 检查为什么并作优化 #3143(新增 shard 分片)是很近的调度变更,这些 live 测试的排布刚变过,值得核对是否相关(拆到更多 runner 理论上应降低单 runner 端口争用,未证实为因)。
  6. 是否破坏缓存:不涉及(纯 CI 诊断,不碰压缩/缓存逻辑)。
  7. 是否改了设计方向:不涉及(未改代码);仅指出 ci已经运行超过20分钟了 检查为什么并作优化 #3143 分片是近期设计变更。
  8. 是否引入了新的配置:不涉及。

收工判定:诊断完成——7 个红检查已全部定位到「文件 + 用例 + 报错 + 性质」,并区分出 1 个确定性项与若干抖动项。空 PR 本身没问题、已达成目的(暴露了 master 现状)。下一步等你定:要不要我把 ①claude-native attach 冲突、②#3044 自重启/端口抖动家族、③gap-3 时序竞态 分别立项(各带来源标记),或直接在某个 follow-up PR 里内联修掉?我不会擅自开 issue / 改代码。

ranxianglei added a commit that referenced this pull request Oct 14, 2026
…#3182/#3184/#3188)

#3145 sharded the test matrix and concentrated the port-family live suites
into the same shards; with #2816's intra-shard file concurrency of 4 the
files fight over the default zone base 18787 and the shared ephemeral port
pool. port-zone-live pins 18787 (EADDRINUSE), claude-native/auto-restart
lose freePort() picks to concurrent binders (TOCTOU), and master push CI
skips the test lanes entirely so nobody saw it — every PR since #3145 shows
the same 7 red legs (#3182/#3184/#3188, empty-PR proof in #3190).

- scripts/live-files.mjs: shared LIVE_FILES table + ZONE_SLOT_BAND=32;
  zonePortForSlot(slot) = 18787 + (slot+1)*32 (first band above the default
  base so an env-ignoring spawn lands on 18787 and collides with nobody's
  assertions — it fails loudly in its own suite). Max band 20003 < 32768.
- tests/preload-zone-slot.mjs: CI-only preload; node --test runs each file
  in its own process, so scanning argv for tests/*.test.ts identifies this
  file; sets BILI_ZONE_PORT for LIVE_FILES members. npm test is untouched.
- scripts/run-ci-tests.mjs: import the shared table; add the preload to
  PRELOAD_ARGS (main phase + diagnostic re-runs see the same environment).
- port-zone-live/claude-native: shadow the imported ZONE_PORT_BASE with
  resolveZonePortBase() so every assertion follows the injected band;
  claude-native child envs spread ZONE_ENV.
- claude-native squatter + auto-restart stub tests: bounded retries on the
  infra shapes (stolen freePort pick / stub exit 91 on EADDRINUSE) so a
  shared-runner race is retried, never masked as a product failure.

Local proof (this box runs real legacy daemons on 18787/18831/18834):
master fails zone-live 7/9 in isolation; this branch fails 2/9, both
starvation-shaped under a concurrent foreign full-suite run, and the
9-file port-family set at concurrency 4 shows zero cross-file collisions.
@ranxianglei

Copy link
Copy Markdown
Owner Author

Diagnosis PR — purpose fulfilled. This zero-change PR exposed the 7/12 red test legs (port-family collisions from #3145 sharding + #3055's new #3044 port tests). Fix: #3197 — now fully green 39/39 incl. all 12 test legs.

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