Skip to content

[PR] fix: stale split-brain notice clears itself once the peer stops contesting (#4626) - #4630

Merged
ranxianglei merged 4 commits into
masterfrom
2026-10-25_4626-splitbrain-stale-clear
Oct 31, 2026
Merged

ranxianglei merged 4 commits into
masterfrom
2026-10-25_4626-splitbrain-stale-clear

Conversation

@ranxianglei

@ranxianglei ranxianglei commented Oct 25, 2026 •

Copy link
Copy Markdown
Owner

What

Two defects in one failure domain (#4626, both reported with exact timelines):

  1. Stale notice (original report): the restart-handover split-brain card persisted forever after the peer pid was gone (v0.1.206 repro: notice surfaced 16:57, peer dead by then, still served at 22:44 until a manual ack).
  2. False-positive verdict (follow-up report): the verdict itself fires during a clean rotation — the accused process had already exited 8.75 s / 29.95 s before its own [split-brain] log lines (second alert 54 ms under the freshness threshold). Two processes never coexisted.

Root cause (two layers):

  • Detection: dispose() performs a final marker write on graceful shutdown (so peers can attribute this instance's last commits), leaving a tombstone whose mtime equals the death time. findContestedWriters treats "dead pid + marker fresher than the 30 s grace" as possibly-live — so ANY session served within the 10-min window is flagged as dual-written by a peer that definitely exited through our own code path.
  • Persistence: once computed, the verdict is stored in the storage runtime state and replayed verbatim on every /__bili/status read until a manual ack; nothing ever re-checked current evidence.

Fix (four facets, one coherent loop — "the verdict is a function of current evidence"):

  1. Detection side: dispose() stamps disposedAt on its final marker write. Markers carrying it are attribution-only and never enter the contest scan — a peer that ran our dispose code provably initiated orderly exit, so no pid-visibility doubt applies and no grace is given. Crash/killed peers carry no stamp and keep the original freshness grace (defends platforms where pid probing can't see the peer). Additive schema field; absent = legacy behavior. Log wording corrected too: "another live instance" → "another instance" (remaining flagged cases are not always live).
  2. Read side: the server registers the contest predicate so getStorageRuntimeStatus() self-clears the stored entry at the status read when the accused peer no longer contests — the same never-freeze discipline as the sibling secondProcess warning directly above it.
  3. Fail-closed evidence check (integrated from a concurrent fleet fix, original author preserved): the contest definition is extracted into ONE function (contestedServedAt) shared by detection and clearing so the two can never drift; clearing consults only the accused peer's own marker via peerStillContests(), which is tri-state — true = still contested, false = provably stopped (marker absent/reaped, dead past grace, window lapsed), null = UNKNOWN (FS error / torn file) → the notice is KEPT. A transient read error must never drop a real alert.
  4. Cosmetic (same issue): slice(0, 8) of a session-… id renders literally session- — the ack log line now logs the full id (the [split-brain] error line already does) and the web card shows a 24-char prefix so distinct sessions are tellable apart.

Branch is rebased onto current master (e290b4fda, green CI) as a linear history; an intermediate origin/ci-fixes merge became unnecessary once master itself went green and was dropped. An earlier one-word de-export micro-fix became redundant after master absorbed the proper fix upstream and was dropped.

Behavior change (disclosed)

  • Old (detection): any session served within the last 10 min by a peer exited up to 30 s earlier was reported as dual-written — i.e., every graceful proxy/host rotation with live traffic produced 1–N guaranteed false alarms.
  • New (detection): gracefully-shutdown peers are never flagged; crash/killed peers behave exactly as before (freshness grace retained).
  • Old (persistence): a surfaced entry stayed on the wire until a manual POST /__bili/storage/split-brain/ack, even after the peer process exited and its marker went stale/reaped.
  • New (persistence): it also clears automatically at the next status read once the peer is PROVABLY stopped; if the evidence is unreadable the notice is kept (fail closed). A live peer keeps the notice up. Ack semantics untouched (same-contest suppression / newer-contest re-surface both pinned by tests).
  • Cosmetic: ack log + web card show a usable session-id prefix instead of session-; the detection log no longer claims the peer is "live".

No config surface change. No kernel/ change. No persistence-format change (verdict is in-memory per process; writer-marker files gain one optional additive field).

Tests

tests/served-markers.test.ts:

  • disposed marker (dead pid + fresh mtime + recent servedAt) → not contested (direct);
  • e2e rotation: periodic flush leaves unstamped markers, dispose stamps, successor writer serving the same session raises NO contest;
  • negative: crash-shaped marker (no stamp, dead pid, fresh mtime) → STILL contested (grace preserved);
  • existing tests pass unchanged.

tests/issue-split-brain-surface.test.ts:

  • G3: fresh tombstone keeps the notice; dead pid + stale marker clears at the status read with an info line carrying full ids; stays cleared; an evidence-based clear does NOT poison the ack guard (renewed contest re-surfaces).
  • G3-negative: without a registered check the stored entry replays as-is (legacy); live peer + fresh marker keeps it surfaced.
  • fail-closed (integrated): unreadable dir/file keeps the notice; ENOENT clears; torn JSON keeps.
  • web e2e (updated): the recorded contest carries real marker evidence in the server's writers dir (fixture pid provably-dead 999999999), final step backdates the marker to prove auto-clear end-to-end through HTTP without an ack.

Targeted local runs green: both files above (19/19), npm run typecheck, npm run build. Full suite = remote CI.

Closes #4626

@github-actions

github-actions Bot commented Oct 25, 2026 •

Copy link
Copy Markdown
Contributor

📦 Built Package Artifact

Branch: 2026-10-25_4626-splitbrain-stale-clear (4368daf)

Latest build — install this exact version

npm install -g billion-context@0.0.6237-pr.4630.e8c58b5

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

Recent builds (commit → version)

commit version build
4368daf billion-context@0.0.6237-pr.4630.e8c58b5 build
bb69547 billion-context@0.0.6228-pr.4630.c986e3b build
1ac4a82 billion-context@0.0.5925-pr.4630.7092ecc build
92c688b billion-context@0.0.5924-pr.4630.669c388 build
8b7154d billion-context@0.0.5881-pr.4630.fe87ef9 build
7d77759 billion-context@0.0.5814-pr.4630.d3b7c01 build
b8538ee billion-context@0.0.5662-pr.4630.089750e build
a481b2d billion-context@0.0.5628-pr.4630.86701ab build
c962f32 billion-context@0.0.5627-pr.4630.ca1ea79 build

Download artifact

  1. Download the artifact from the Actions run
  2. Extract the tarball and install:
tar xzf billion-context-pr4630.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

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

查重: 不重复 — open+closed 检索 "split-brain stale notice clears" / "4626" / "split-brain" 均无并行条目;仅相关的已关闭 #3923 是该功能(split-brain 检测本身)的原始 issue,本 PR 修的是它的「判决冻结」缺陷,机制不同。
标签: bug / dsh / scope/storage / p2(存储层可观测性缺陷;复现路径为 dsh 重启交接;有手动 ack 绕行 → p2)。

独立审核完成(未照抄 issue/PR 侧结论;全 diff 5 文件逐段读完 + 机制源码交叉验证 + 本地定向复跑)。

本地验证(PR head c962f32)

  • node scripts/check-unused-exports.mjs:唯一失败行 = INTERNAL_ONLY src/agent/client-build-note.ts: ClientBuildCheck —— 与 master HEAD(146b34f)红风暴的规范错误行逐字相同,且该文件本 PR 未触碰;PR 新增的两个导出(registerSplitBrainEvidenceCheck、server.ts 对 findContestedWriters 的引用)均被跨模块使用。CI 上红的 unused-exports = 继承性失败,非本 PR 引入。
  • 定向测试 tests/issue-split-brain-surface.test.ts + tests/served-markers.test.ts:16/16 pass;npm run typecheck(含 kernel/pi-subagents prebuild):exit 0。

回归清点(②,触路径既有行为逐项)

既有行为 结论
secondProcess never-freeze 自清(runtime-status.ts:282) 保持,新块加在其后,互不干扰
ack 语义(G2:同 contest 抑制 servedAt <= acked、新 contest 再浮现) 保持——证据自清不写 splitBrainAcks,不污染 ack 守卫(G3 Phase C 钉住)
无注册 checker 时的 legacy replay(standalone record/ack 用法) 保持(G3-negative 钉住)
检测路径 checkContested/VERDICT_TTL/onContest 未动
writer-marker 文件格式 / reapStaleWriterMarkers 未动
/__bili/status wire 形状 字段形状不变,仅 splitBrain 值的生命周期变化(已披露);新增一条 info 日志
web 卡片 id 前缀 8→24 字符、ack 日志全 id 纯外观,PR body 已披露;escapeHtml/i18n key 不变

高危模式逐项过:无常量复制(复用同一 findContestedWriters 与既有 CONTEST_WINDOW_MS/PEER_FILE_FRESH_MS)、无阈值漂移、无 fail-fast→吞错、无输出格式意外变化。行为变更旧值→新值在 PR body 披露且与我读码结论一致。

结构化八问

  1. 是什么问题:重启交接后 split-brain 提示卡死——peer 进程早死了,通知却永久挂在 status/web 上,只能手动 ack(报告者机器 5 条/天)。
  2. 什么原因导致的:判决在检测时刻算一次就写进存储运行时状态,之后每次 status 读原样回放;而优雅重启时 dispose() 恰好重写 marker 墓碑,mtime 在检测时刻是新鲜的,于是「pid 活或文件新鲜」的证据永远成立、却无人复检。根因是判决与证据脱钩(存的是快照而非函数),不是检测逻辑错。
  3. 影响范围与严重程度:所有走统一存储 writer-marker 的多实例/重启场景;症状级——提示噪音 + 误导(暗示仍有活跃 peer),数据面无损(last-writer-wins 警告本身仍会在真冲突时按 commit 触发)。p2。
  4. 怎么修复的:把存储判决变成当前证据的函数——server 注册产生判决的那个扫描本身(findContestedWriters over 自己的 writers dir,单一「仍在竞争」定义),status 读取时发现被指控 peer 不再出现在扫描结果里就自清,与上方 secondProcess 同一 never-freeze 纪律;新鲜墓碑刻意保留(dead pid 单独不构成缺席证明)。
  5. 是否有回归问题:无。见上表逐项;ack 守卫不被自清污染(再竞争会重新浮现,测试钉住);pid 复用场景方向保守(alive 短路新鲜度检查 → 通知保留,诚实方向);10 分钟窗口过期即使 pid 活着也清——此时双方 commit 已不再重叠,竞争语义本身消失,PR 已披露。
  6. 是否破坏缓存:不涉及(纯可观测性面;不碰 fold/prefix-cache/重试/游标任何机制)。
  7. 是否改了设计方向:不涉及架构转向;是把 The same sessionId is served by two bili processes simultaneously: interleaved logs from global + pi-native dual instanc | 同一 sessionId 被两个 bili 进程同时服务:全局 + pi-native 双实例日志交织、重复 cold-load(split-brain) #3923 既有「never-freeze」纪律从 secondProcess 推广到 splitBrain,同一模式。
  8. 是否引入了新的配置:不涉及。无 env/config/默认值变化;writer-marker 文件不动;persistence 格式不动(判决本就 per-process 内存态)。

测试有效性(④):G3 双向都有牙(新鲜墓碑必须保留=负向;dead+stale 必须清=正向;且断言 info 日志带全 id);G3-negative 钉住 legacy 与 live-peer 两个不触发面;web e2e 用真实 marker 证据走完整 HTTP 链路证明无 ack 自清(fixture pid 从可能存活的 4242 换成超出 pid_max 的 999_999_999,确定性更好)。这些测试在原 bug 下会红,不是恒真。

CI 分账:除 unused-exports(继承,本地已证)外其余 lane 全绿;test/static-gates/e2e lane 因 gate 层 fail-fast 被 skip——全套件判决待 master 红解除后由远程 CI 给出(#4599 总控下,#4606 修复车在途,等 owner 对 A/B/C 拍板)。分支 base = 当前 master HEAD,不陈旧。

判定:代码可合。 mergeable_state:blocked 纯粹来自 master 红风暴的 required-checks 死锁(等 #4606/owner 决策),本分支无需任何改动。master 转绿后 retrigger CI 即可合并。

@JElfferich

Copy link
Copy Markdown

Thanks for turning this around. I checked billion-context@0.0.5881-pr.4630.fe87ef9 (the artifact has the seam) and the diff: src/server.ts registering findContestedWriters as the evidence check and getStorageRuntimeStatus() clearing at read is exactly the option-1 shape from the report. Reusing the same scan instead of a second "is the peer alive" test is the right call — it keeps one definition of "still contested".

Two things I'd want settled before release:

  1. The clear currently fails open. findContestedWriters() returns [] when readdirSync(dir) throws (EACCES, ENOENT during a state-dir move, transient I/O on a network FS) and swallows per-file read/stat errors. Fed straight into the checker, "scan could not run" is indistinguishable from "peer stopped contesting" — so a transient FS error silently drops a real notice and logs peer … no longer contests it, i.e. the one direction that should stay conservative. Either read the accused peer's marker by path and only clear when that file is absent or provably stale (dead pid + mtime past the 30 s grace), or let the checker return tri-state (contested / not-contested / unknown) and keep the notice on unknown.

  2. The CI precondition is stale. [PR] fix(ci): clear the two master-red gates in one branch — unused-exports (#4602) + the #3890 LIVE_FILES tripwire (#4604) #4606 is closed unmerged and master has moved on (bb7c2497). A rebase onto current master plus a re-run would let us see whether the inherited unused-exports / LIVE_FILES reds still exist, rather than carrying a note from 25 Oct.

Also: I'm on 0.1.208, which doesn't contain the seam yet — is this slated for the next release? Happy to re-run my restart-handover repro (supervisor restart → fresh tombstone → notice persists) against the prerelease or the released build and report back with the log lines.

One nit on tests: a case pinning whichever fail-safe direction you pick (unreadable writers dir) would keep it from regressing later.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

CI auto-fix lane (attempt 1/2): windows-latest 22/3 已修,push 92c688bc2,等 CI 重跑。

查重:不重复——此红不是本 PR diff 引入,是继承自 master(master bb7c249 同一子测试同样失败)。

八问结论

  1. 是什么问题:PR 唯一红 shard(windows 22/3)挂在一个子测试:runLaunch opencode-inline: …(tests/launcher-opencode-inline.test.ts:112)——断言期望 spawn 的 command 等于 BILI_CLIENT_BIN(fake-opencode.cmd),实际是 cmd.exe。
  2. 什么原因导致的:[PR] pi/opencode inline changed to manual opt-in: revert the default flip; default behavior is identical to the previous vers | pi/opencode inline 改为手动开启:收回默认翻转,默认行为与旧版完全一致 #4687(opencode inline 模式)写这个测试时只按 POSIX 假设。Windows 上 planClientSpawn(src/launcher.ts:5401,[bug] Windows: a space in the client or -e path truncates the launch (spawn shell:true, unquoted) — and silently kills compression when only the command is quoted #679/[Bug][Windows launcher] 经 codex.cmd 启动破坏 argv:嵌入引号拆分、%变量% 展开和 & 被解释 #2196 既定行为)必然把 .cmd shim 包成 COMSPEC /d /s /c <line>——CreateProcess 无法直接执行批处理文件。launch 器行为是对的,错的是测试断言。
  3. 影响范围与严重程度:纯 CI 假红,零产品行为影响。其余 shard 全绿;master 上同一子测试也红(master 多 shard 红的来源之一)。
  4. 怎么修复的:断言改平台感知——win32 断言 command 以 cmd.exe 结尾、args 前缀 ["/d","/s","/c"]、包装命令行包含 shim 路径;非 Windows 保持原断言。独立标注 commit。本地验证:定向测试 2/2 过(posix)、typecheck/build 干净;win32 形状对照 planClientSpawn 源码与 CI 日志实际值核对过。
  5. 是否有回归问题:无——只动一个测试文件的一处断言,src 未动;posix 路径断言不变(本地通过)。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

备注:

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

CI 复跑确认:head 1ac4a82ad(含本修复 92c688bc2)全部 55 个 check run 无失败——windows-latest 22/3 转绿,其余 shard 保持绿。CI 侧收工。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

Both points verified independently — you're right on both, and both are now settled on the branch.

1. Fail-open clear — confirmed real, now fixed (commit 1ac4a82ad).
I re-read the seam: findContestedWriters() does return [] when readdirSync(dir) throws, and skips per-file read/parse/stat failures silently — fed into the .some(...) checker, "scan could not run" was indeed indistinguishable from "peer stopped contesting", and the only direction that should stay conservative was the one that failed open.

The fix takes your option (a)+(b) hybrid:

  • New peerStillContests() in src/served-markers.ts: reads only the accused peer's own marker file (<dir>/<peerInstanceId>.json) — no full-dir scan needed for the clear decision. Tri-state:
    • file absent (clean ENOENT) → provably stopped → clears;
    • dir missing/unreadable, any other I/O error, torn/malformed JSON, or name/content disagreement → unknown → notice kept, and no misleading "no longer contests" log line;
    • file readable → judged by the same window/freshness/pid test as detection.
  • The per-marker judgment is extracted into one shared helper (contestedServedAt) used by both the dir scan and the direct-path check, so "still contests" keeps its single definition — detection and clearing can't drift apart again.
  • getStorageRuntimeStatus() now clears only on explicit false; unknown keeps the entry (same never-freeze discipline as secondProcess, just fail-closed).

Behavior-change note vs the PR's previous commit: an FS error at status-read time now keeps the notice instead of clearing it — this aligns the code with what the PR body already claimed ("once the peer provably stops contesting"); vs master there is no change either way since the seam didn't exist yet.

Your test nit is in: new G4 pins all three directions — writers dir removed ⇒ kept + no false-clear log line; dir restored with the accused marker reaped ⇒ cleared; torn marker file ⇒ kept. G3/G3-negative now register through the same function the server uses, so they track the real shape.

2. CI precondition — confirmed, and already mostly done by the fleet. #4606 is closed-unmerged (verified via API), master is at bb7c24972. One correction to the premise: the branch had already picked up current master — the #4599 sentinel's help-merge sweep merged ci-fixes into this branch today (which is also where the LIVE_FILES tripwire and the ClientBuildCheck de-export landed), plus a plugin-fork test fix. So rather than rewriting the branch history (which would have clobbered those commits), I fast-forwarded my fix commit onto the true tip 92c688bc2. New head: 1ac4a82ad.

CI on the new head is fully green: 39 success / 16 skipped / 0 failed — the skips are the env-gated e2e lanes (normal state), and notably all 11 full-suite test shards (ubuntu/windows × node 22/24) actually ran this time after being fail-fast-skipped before. Local targeted runs on the same head: split-brain + served-markers suites 17/17, typecheck and build clean.

Release timing: the fix lands on master when this PR merges; cutting releases is owner-approved, so I can't promise a version number — it rides the next release after merge (you're on 0.1.208; the next cut will carry it). Your repro offer is very welcome: run the supervisor-restart → fresh-tombstone sequence against the prerelease built from 1ac4a82ad (same pr.* artifact pattern you used) or the released build once it ships. The two log lines that matter are the [split-brain] … committed by BOTH error at detection and the new [split-brain] session <full-id> notice cleared — peer <id> (pid N) no longer contests it info line ~30 s after the peer exits — if you see the former without ever seeing the latter, that's a regression worth reporting back.

@ranxianglei
ranxianglei force-pushed the 2026-10-25_4626-splitbrain-stale-clear branch from 1ac4a82 to bb69547 Compare October 26, 2026 09:47
ci-check added 4 commits October 26, 2026 18:19
…testing (#4626)

The surfaced contest was persisted into the storage runtime state at
detection time and replayed verbatim on every /__bili/status read until a
manual ack — but detection fires while the restarting peer's tombstone
marker is still seconds fresh (its graceful-shutdown dispose() rewrites
the marker), so a restart handover left an alarming card behind for hours
after the peer pid was gone.

The stored verdict now re-evaluates against CURRENT evidence at the status
read, with the same never-freeze discipline as the sibling second-process
warning: the server registers the very scan that produced the verdict
(findContestedWriters over its own writers dir), and when that scan no
longer reports the accused peer (pid gone + marker past its freshness
grace, or the contest window lapsed) the entry clears itself with an info
log line. Without a registered check the legacy replay behavior stands
(standalone record/ack use). The fresh-tombstone case keeps persisting:
a dead pid alone is not proof of absence on every platform, so the
notice stays honest while marker freshness says the peer may be alive.
…#4626)

slice(0, 8) of a "session-…" id renders literally "session-" — the ack
log line and the web banner both showed an unusable prefix. Log the full
id (the [split-brain] error line already does) and give the card a 24-char
prefix so distinct sessions are tellable apart.
@ranxianglei
ranxianglei force-pushed the 2026-10-25_4626-splitbrain-stale-clear branch from e1d2619 to 4368daf Compare October 26, 2026 10:20
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

上游冲突已解除:PR 分支已与当前 master 对齐,mergeable: true(blocked 仅为 required checks 在跑,非冲突)。

过程

  1. master 自分支基点前进 14 个提交([PR] fix(#4797): root-cause fix for superseded live-key collision — restamp pruning + semantics-preserving guard translation | fix(#4797): superseded 活键碰撞根治——restamp 剪枝 + 守卫保义翻译 + 一次性标志修复 (kernel 0.0.166) #4807/After the agent restarts, this session can no longer be compressed: the ref suggested by acp_status is flagged as "renum | 代理重启后该会话再也压不了缩:acp_status 建议的 ref 被 compress 判为「已重编号」,两面对 ref 的认知不一致 #4797/Plugin history render drops an entire fold (covered rows suppressed AND no summary anchor) when the fold reaches the transcript tail — ok:true, no error #4732/Two existing red tests on master are masked by CI sharding: #4125's give-up reason is preempted by #4516's dead override | master 上两个既有红测试被 CI 分片藏住:#4125 give-up 理由被 #4516 的 dead 覆写抢跑 + #1365 attach 超时 #4824 等),双方共同触碰的文件只有两个 lockfile——PR 侧 dfadc21b3 曾把记录的 acp-kernel 版本同步到 0.0.165(CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 lock-sync),master 随后升到 0.0.166 并重新生成了同样的行 → 两个 lockfile 各一个 version hunk 冲突。源码零重叠(master 动 stream/preflight/persist/native-intercept,本 PR 动 served-markers/server/runtime-status/web),全部自动合并。
  2. 解法:取 master 侧 0.0.166——与自动合并后已是 0.0.166 的 kernel/package.json 一致,且严格新于 PR 的同步意图。合并提交 e1d26192b,本地验证:typecheck + build 干净、定向套件 tests/issue-split-brain-surface.test.ts + tests/served-markers.test.ts 19/19 过,push。
  3. push 期间出现并发 rebase:另一车道把 4 个 fix commit 直接 rebase 到 master head e290b4fda 上并 force-push(新 tip 4368daf64),丢掉的只是两个已被 master 吸收的 CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 过渡 chore commit(lock-sync 0.0.165、gate 重触发)。已验证 git diff e1d26192b..4368daf64 树内容逐字节一致——我的验证结论对现 tip 同样成立,历史更线性,保留该 tip,不再回推。

八问

  1. 是什么问题:PR 分支与 master 冲突(dirty)无法合并——不是代码冲突,是两个 lockfile 里 acp-kernel 记录版本的行双方都改过。
  2. 什么原因导致的:并行车道的 lockfile 版本漂移——本 PR 在 CI Control Center: Centralized CI Failure Handling & Status Sync (Comment all CI issues here; do not open new issues) | CI 总控: CI 失败集中处理与状态同步(所有 CI 问题在此评论, 勿开新 issue) #4599 哨兵下同步到 0.0.165,master 随后把 kernel 升到 0.0.166 重新生成同一行;这是 lock-sync 门禁下的常态竞争,不是 bug。
  3. 影响范围与严重程度:仅本 PR 的可合并性被卡住;解法本身零产品行为变化(只让记录版本跟上 master 已有的 0.0.166)。
  4. 怎么修复的:merge master 进分支,lockfile 取较新的 master 侧;其余自动合并。并发车道的 rebase 产出同树内容的线性历史成为最终 tip,等价且更干净,予以保留。
  5. 是否有回归问题:无。解法只动了记录版本号行;同树内容上 typecheck/build 干净、split-brain + served-markers 定向 19/19 过。master 带入的变更(stream/preflight/kernel restamp 等)属其自身已审范围。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入了新的配置:不涉及。

CI:新 head 4368daf64 上 check runs 进行中(17 success / 15 skipped / 1 running / 8 queued,暂无失败);全套件判决以远程 CI 为准。

判定:冲突收工。 PR 恢复可合并状态,剩 required checks 跑完即可由 owner 合并。

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

Labels

bug Something isn't working dsh p2 scope/storage

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug][split-brain] Restart-handover notice persists forever after the peer pid is gone (stale storage.splitBrain)

2 participants