Skip to content

[PR] release: hot-disarm the patch lane (v3 workflow file + temp version revert to 0.1.206) - #4671

Closed
ranxianglei wants to merge 1 commit into
bugfix-v0.1.206from
disarm-lane-v3
Closed

ranxianglei wants to merge 1 commit into
bugfix-v0.1.206from
disarm-lane-v3

Conversation

@ranxianglei

Copy link
Copy Markdown
Owner

Emergency disarm of the armed v2 lane after the #4667 self-fire incident (v0.1.207 was published BY the arming merge #4654 itself).

Single commit 7c378558f on top of the lane head, two parts:

  1. Replaces the lane-side .github/workflows/release-bugfix.yml with the v3 gate-branch file — under v3, any close targeting this lane resolves to a millisecond no-op (only a same-repo ship PR with base=bugfix, head=bugfix-vX.Y.Z can fire).
  2. Temporarily reverts package.json 0.1.207 -> 0.1.206, so that IF the OLD v2 base-side copy still governs this very merge event, the monotonic guard refuses (0.1.206 is not newer than npm latest 0.1.207) and nothing ships.

Expected CI behavior: this merge's own publish run may go RED at the version guard depending on which copy governs the event — that red is the disarm working, exactly as stated in the commit message. Nothing publishes under either path. A follow-up restores the 0.1.207 pre-bump once the epoch-1 transition moves forward (under v3 it is then a silent no-op until a ship PR fires).

Verified observation (non-blocking): the v3 file carried here predates three fix commits that landed on master via #4660 (843a37fae gate-head shapes through version-guard, e9b8c0541 publish/auto-cut shell defects, 85c9a5ee8 P4 lane-scan fork point). All three touch paths that are INERT for this disarm's purpose (no ship-shaped PR can exist yet — the bugfix gate branch does not exist), so merging as-is fully disarms the lane. Syncing the file to master's current copy is a trivial optional follow-up.

Follows #4667 (removes the armed-lane exposure; the remaining epoch-1 steps are owner actions listed in that thread).

…to 0.1.206)

EMERGENCY MIGRATION COMMIT, bypasses no direct... actually via reviewed PR as usual. Context: #4654 (v2 lane-side arming) was merged BEFORE the model moved to v3 gate-branch publishing. Under v2 rules, the NEXT merge into this lane would publish 0.1.207 without the advisory note or gate-branch review. This PR replaces the lane-side workflow with the v3 file (whose resolve path no-ops for lane-targeted closes) AND temporarily reverts package.json to 0.1.206 so that if the OLD v2 copy still governs this very merge (it executes the pre-merge base-side file), the monotonic guard refuses (0.1.206 == npm latest) and nothing ships. EXPECTED: this run goes RED at the publish guard. That red is the disarm working. Follow-up PR restores the 0.1.207 pre-bump (under v3 it is then a silent no-op).
@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

查重: 不适用 —— 本条是系统转发的既有 PR #4671 冲突解决任务,非新条目;未新开任何 issue/PR。
标签: bug + scope/ci + p2(保留原 pr)—— 发版管道事故(#4667 自燃)的紧急修复,属 CI/发版域;有绕行且不影响终端用户面,故 p2。

冲突已解决:PR 分支 disarm-lane-v3 已 rebase 到 lane 最新头(bugfix-v0.1.206 @ 22e09b22b)并 force-push(新 head 34aecbe62)。GitHub 合并状态已从 dirty 转为 blocked(冲突消除,仅剩 CI 待跑)。

八问:

  1. 是什么问题:lane 在 PR 切出后合入了 master([PR] align master with v0.1.207 (shipped from bugfix-v0.1.206) #4659 对齐合并),PR 的基线前进;双方相对共同祖先都重写了 .github/workflows/release-bugfix.yml(祖先侧是 v2 armed 版,PR 侧换成旧版 v3、lane 侧被对齐合并换成新版 v3),PR 变 dirty 无法合并。
  2. 什么原因导致的:纯文本三方冲突——PR 从旧 lane 头([PR] release: arm merge-gated lane publish (lane-side release-bugfix.yml) #4654 arming 提交)切出,之后 lane 经对齐合并吸收了 master 上 [PR] release: land the TRUE v3 gate-branch publish model (v2 arm self-fired at #4654) #4660 的三个 v3 修复;不是设计分歧,无需决策。
  3. 影响范围与严重程度:仅影响这个 disarm PR 本身能否落地;不解决则「临时版本回退」这层安全网进不了 lane,旧 v2 副本的暴露窗口继续敞开。不影响任何终端用户功能。
  4. 怎么修复的:rebase,冲突处取 lane 侧(master 当前,v3 含 [PR] release: land the TRUE v3 gate-branch publish model (v2 arm self-fired at #4654) #4660 三修复)的工作流文件。依据是 OP 自己预先认可的两种结局("merging as-is fully disarms the lane";"syncing the file to master's current copy is a trivial optional follow-up")——取新文件等于顺手完成那个可选跟进;v3 语义下 close-targeting-this-lane 两种文件版本都是 no-op,disarm 目的完整保留。版本回退部分(0.1.207→0.1.206)自动干净应用,"若旧 v2 副本治理本次合并则单调守卫拒发"的逻辑原样保持。
  5. 是否有回归问题:无。逐项核过:① 最终工作流文件与 master/lane 逐字节一致(blob 7dd4f449,含 [PR] release: land the TRUE v3 gate-branch publish model (v2 arm self-fired at #4654) #4660 全部修复),无语义漂移;② package.json 净变化只剩 version 一行,单调守卫安全网不变;③ 其余文件零触碰,无 fail-fast 变吞错、无阈值/时序/输出格式漂移。
  6. 是否破坏缓存:不涉及(纯发版流水线变更,不碰缓存/重试/游标/去重键)。
  7. 是否改了设计方向:没有。两处顺手清理,均在此明示:① 原提交里夹带了一行无关编码差异(package.json description 字段 \u2014→—,JSON 字符串值相同,系编辑工具副产物),已还原为全仓统一的 \u2014 形式,使 PR 净 diff 收敛到 version 一行;② 工作流文件取新版,即执行了 OP 注明的可选跟进。
  8. 是否引入新配置:无。

判定:冲突已解、分支干净(mergeable=True,等 CI 绿灯),可合并。PR: #4671

@ranxianglei ranxianglei added bug Something isn't working p2 scope/ci labels Oct 25, 2026
@ranxianglei

Copy link
Copy Markdown
Owner Author

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

SUPERSEDED — please do not merge this PR. Opened as the hot-disarm vehicle for #4667; shortly after, lane bugfix-v0.1.206 moved to 22e09b22b: master was merged INTO the lane as part of #4659's conflict resolution (4cf659966), which already brought the v3 gate-branch workflow file into the lane — byte-identical to master's (including #4660's three fixes 843a37fae / e9b8c0541 / 85c9a5ee8).

Consequences:

  • The lane is already disarmed: v3's on: filter only accepts base=bugfix/master, and its resolve step no-ops every other close — a lane-targeted merge can no longer fire a publish.
  • Merging THIS PR would now revert package.json 0.1.207 -> 0.1.206 on the lane (its second half), undoing the correct pre-bumped state and conflicting with [PR] align master with v0.1.207 (shipped from bugfix-v0.1.206) #4659's tree — wrong on a frozen lane.

Please close it (close rights stay with you per policy). Full incident state is tracked in #4667.

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

查重: 不重复 —— 本 PR 是 #4667 自触发事故的处置动作(follows #4654 arming),无其他并行的 disarm 条目。相关项:#4659(master→lane 合并,改变了局面)、#4660(v3 finalize)、#4655(另一条同样指向该 lane 的 open PR,advisory force-install,范围不同)。

标签: bug, pr, scope/ci, p1(由 p2 改判 p1:事故域是「lane 上任何合并都可能把版本误发到 npm latest」,属核心能力停摆级风险;待 owner 确认处置后可降级)

核验结论:disarm 目标已由 #4659 达成;本 PR 现与 lane head 冲突,按现状合入会把 lane 的发布工作流文件回退到较旧快照。

核实的时序(均为 2026-10-25,+0800):

时刻 事件
15:43 #4654 arming 合并(lane head = v2 armed 副本)
16:14:50 npm 出现 0.1.207(#4667 自触发;发布 job 先跑全套门禁,故晚于合并约 31 分钟落版)
16:15:35 本 PR commit 7c378558f,切自 #4659 之前的 lane 状态
17:33 #4660(v3 gate-branch 模型 + 修复)落 master
17:58 #4659 把 master 并入 lane → lane head 的 .github/workflows/release-bugfix.yml 与 master 最新副本逐字节相同(md5 一致,含 e9b8c05 的 publish/auto-cut shell 修复)
18:08 本 PR 开出

关键机制:同仓 PR 自身的 close 事件执行的是合并后 base 侧副本(#4654 自触发即实证)。v3 文件的触发面只覆盖 branches [master](pull_request)与 [bugfix, master](pull_request_target 双路径)——指向 bugfix-v0.1.206 的 close 根本不命中任何触发分支,即便命中,resolve 步也会毫秒级 no-op。因此 lane 已经惰性化:今后任何指向该 lane 的合并(包括仍 open 的 #4655)都不会产生发布;关闭本 PR 本身也不会触发发布(base 不匹配任何触发分支)。

本 PR 现状(本地实测):

  • 试合并:恰好一个冲突文件 .github/workflows/release-bugfix.yml(package.json 可干净自动合并)
  • 若冲突解成取本 PR 一侧:lane 工作流文件回退到旧快照,恰好丢失 e9b8c05「publish/auto-cut step shell defects」的 21+/10- 修复——自动发布高危区的回归。(PR 正文说「predates three fix commits」;实测三者中只有 e9b8c05 动过这个文件,另两个分别动 ci.yml 与 lane-leak-guard.yml)
  • Part 2(package.json 0.1.207→0.1.206、单调守卫兜底)对本次合并事件仍有效,但治理副本已是 v3,它不再是承重墙;且会让 lane 停在 0.1.206 而 npm latest 是 0.1.207,留下需后续 follow-up 恢复的临时不一致
  • CI:head 7c378558f 尚无任何 check-run(PR 处于 dirty/冲突态)

结构化八问:

  1. 是什么问题:arming 合并后 lane 变成「任何合并即发布」,需要解除该风险,并确认这个 disarm PR 是否还需要、能否安全合入。
  2. 什么原因导致:v2 arming 副本在 v3 gate-branch 模型生效前进了 lane;disarm PR 从旧 lane 快照切出,随后 [PR] align master with v0.1.207 (shipped from bugfix-v0.1.206) #4659 已把最新 v3 文件带进 lane,使 PR 的 Part 1 冗余且冲突。
  3. 影响范围与严重程度:原暴露面 p1(任何版本都可能被误推到 npm latest,波及全部自更新用户);今日 17:58 起 lane 已惰性化,残余风险只剩「本 PR 若以错误方式解冲突合入,lane 丢失发布路径修复」(中危、可控)。
  4. 怎么修复(两选项,对比表在下):推荐 A——直接以 superseded 关闭;B——我来 rebase 到 lane head、删掉整个工作流文件 hunk、只留两行版本回退作额外保险。
  5. 是否有回归问题:A——无(零变更);B——无(冲突 hunk 删除,版本回退为临时态、由承诺的 follow-up 恢复)。两条路都必须避免「取 disarm 侧文件」的解法,否则 lane 丢 e9b8c05 修复。
  6. 是否破坏缓存:不涉及(纯 CI 工作流 + 版本字段,不触碰压缩/缓存路径)。
  7. 是否改了设计方向:无(v3 gate-branch 模型不变;epoch-1 过渡步骤仍为 Release pipeline: merging any PR into the bugfix lane triggers a release — the assembly pipeline's merge triggered itsel | 发版流水线:合并任何 PR 进 bugfix 车道即发布——装配流水线的合并触发了它自己,v0.1.207 意外上线 #4667 线程所列的 owner 动作)。
  8. 是否引入新配置:无。

选项对比:

选项 操作 结果 代价
A(推荐) 以 superseded 关闭 #4671(disarm 目标已由 #4659 达成) lane 保持最新 v3 文件 + 0.1.207,完全惰性 零变更;承诺的「restore 0.1.207」follow-up 也一并免了
B rebase 至 lane head,删掉整个工作流文件 hunk,仅留 package.json 版本回退两行 给本次合并事件加一道实际冗余的保险 lane 留 0.1.206 临时不一致,需 follow-up 恢复

等你拍板:A(直接关,推荐) 或 B(我 rebase 瘦身为纯版本回退)。合并/关闭权都在你手上,我不自行操作。

@ranxianglei

Copy link
Copy Markdown
Owner Author

🤖 Powered by ework · qwen3.8-27b

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

收到,按 SUPERSEDED 处理,不推进合并(合并权也只在人手)。

核对我留下的分支实际状态,与你的判断一致:

  • rebase 后 PR 净 diff 恰好一行:package.json version 0.1.207 → 0.1.206(disarm 那半因冲突解决取了 lane 侧 v3 文件、与 lane 逐字节相同,已是 no-op)。
  • 所以此时合并确实会把 lane 的版本回写成 0.1.206,撤销正确的 pre-bump 状态——你的描述准确。我上一轮「可合并、等 CI」的判定基于 OP 切出时两半都有效的意图,对齐合并落地后第二半性质变了,该判定作废。

按政策我不执行关闭,PR 保持 open 等你关。已补状态标签 superseded 防误合(当前标签集:bug / p1 / scope/ci / pr / superseded,p1 保留不动)。分支 disarm-lane-v3(head 34aecbe62)原样保留,关闭后可随时查看或删除。事故全貌继续以 #4667 为准。

@ranxianglei
ranxianglei deleted the branch bugfix-v0.1.206 October 25, 2026 12:13
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant