Skip to content

[PR] fix(instance): re-hash the entry script when the file changes (#5354) - #5355

Open
Maziar123 wants to merge 7 commits into
ranxianglei:masterfrom
Maziar123:2026-10-31_fingerprint-cache-stale
Open

Maziar123 wants to merge 7 commits into
ranxianglei:masterfrom
Maziar123:2026-10-31_fingerprint-cache-stale

Conversation

@Maziar123

@Maziar123 Maziar123 commented Oct 31, 2026 •

Copy link
Copy Markdown
Contributor

Fixes #5354

Problem

entryScriptFingerprint() cached a script's sha256 by path for the life of the process. The omp plugin runs inside the long-lived agent host, so after an in-place npm i -g the host kept the first build's digest, and #4125 handed it to every proxy it respawned through BILI_SPAWN_FINGERPRINT. Children running the same new code registered different fingerprints, could not attach to one another (code-fingerprint-mismatch), and each respawn laddered onto the next lane port, moving lanes.omp every time.

Change

  • src/instance.ts: the cache is keyed on the file's identity (dev:ino:size:mtimeMs:ctimeMs from fs.statSync) and re-hashes when it changes. npm's rename-over changes the inode; an in-place rewrite changes size/mtime/ctime. A file that is no longer readable drops its cache entry and returns undefined, the documented "unreadable → never attach", instead of the old digest.
  • An unchanged file still costs one stat, not a re-hash of the ~5.7 MB dist.
  • Known limit: a same-size in-place rewrite within one filesystem timestamp tick of the previous stat would keep the cached digest until the next change. npm does not write that way.

Tests

  • New tests/issue5354-fingerprint-follows-the-file.test.ts: rename-over, in-place rewrite, removal. All three fail on master (the second call returns the first build's digest) and pass here.
  • node --import ./tests/preload-storage-baseline.mjs --import tsx --test over the 28 suites that import src/instance.ts or use the fingerprint: 580 tests, 569 pass, 11 skipped, 0 fail.
  • npm run typecheck, npm run check:unused-exports, npm run check:test-import-graph: clean.
  • CHANGELOG.md [Unreleased] entry added.

…nglei#5354)

entryScriptFingerprint() cached a script's sha256 by path for the life of
the process. The omp plugin runs inside the long-lived agent host, so after
an in-place `npm i -g` the host kept the first build's digest and ranxianglei#4125
handed it to every proxy it respawned (BILI_SPAWN_FINGERPRINT): children
running the same new code registered different fingerprints, could not
attach to one another and laddered onto the next lane port each time.

Key the cache on the file's identity (dev/ino/size/mtime/ctime) and re-hash
when it changes; an unreadable file drops its entry and returns undefined.

Fixes ranxianglei#5354
@ranxianglei ranxianglei changed the title fix(instance): re-hash the entry script when the file changes (#5354) [PR] fix(instance): re-hash the entry script when the file changes (#5354) Oct 31, 2026
@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

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

查重:不重复。按失败签名(fingerprint 缓存 / BILI_SPAWN_FINGERPRINT / code-fingerprint-mismatch / lane port 漂移)检索 open+closed 的 issue 与 PR:GitHub issue #5354(omp 长驻宿主原地更新后持旧入口脚本指纹 → 重拉代理互相接不上、lane 端口逐级漂移)只对应这一个修复 PR GitHub #5355(本镜像单 #5357 即其镜像),无并列在途工作。说明一下编号映射以免混淆:本镜像单 #5357 ≠ GitHub #5357;PR 正文的 "Fixes #5354" 指的是 GitHub issue #5354(真正的 bug 报告)。

标签:bug / omp / scope/wiring / p2 —— 修的是真实缺陷(bug);故障面在 omp 车道(omp);改动落在实例 attach/lane 接线层(scope/wiring);更新后能用但 lane 共享失效、需重启宿主绕行(p2,非数据丢失/非缓存泄露类,不达 p1)。

状态:开始正式审核——独立核对原 issue 方向、逐行读 diff、查 CI、本地定向跑新测试(含「master 上应为红」验证)。完整八问结论稍后跟进。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

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

Review 结论续报(首帖已给查重结论与标签,本帖为完整审核)。已核对代码、本地定向复现、CI 全矩阵。

结构化八问

  1. 是什么问题:omp 长驻宿主进程做完原地升级(npm i -g)后,入口脚本指纹缓存不刷新。宿主重启出来的子代理拿着旧指纹注册、互相 attach 不上(code-fingerprint-mismatch),每次重启 lane 端口再漂移一格(原 issue:lanes.omp 一天从 18792 漂到 18797)。
  2. 什么原因导致的:src/instance.ts 的 entryScriptFingerprint() 按路径把 sha256 缓存为进程终身有效。npm 升级是 rename-over(新 inode),缓存永不失效 → 宿主经 BILI_SPAWN_FINGERPRINT([Bug] Plugin bootstrap spawns a new instance while the old one still holds the port and drops proxying (goes direct): la | [Bug] 插件 bootstrap 在旧实例仍占端口时新起实例并放弃代理(goes direct):大会话随即以未压缩历史撞上游上限(400 Range of input length / 413) #4125)把第一份构建的摘要传给每个子代理,而子代理跑的是新代码、算出新摘要 → 失配 → attach 被拒 → 端口 0 引导逐级堆叠。
  3. 影响范围与严重程度:只影响「omp 原生插件 + 原地升级后未重启宿主」这一组合:子代理互不 attach、lane 记录漂移到新端口。有绕行(重启宿主),无数据丢失/崩溃 → p2,标签已打。
  4. 怎么修复的:缓存键从「路径」改为「文件身份」(statSync 的 dev:ino:size:mtimeMs:ctimeMs)。身份不变 → 直接复用 digest(未变文件多付一次 stat,不重哈希 ~5.7MB dist);身份变了 → 重新哈希;文件不可读 → 清掉缓存条目并返回 undefined(保持文档化的「不可读 → 永不 attach」语义)。修在产生端(缓存层),方向打到根因,不是表面拦截。
  5. 是否有回归问题:逐项审过触碰路径的既有行为——
    • 未变文件 → digest 值与格式(64-hex sha256)不变 ✓
    • 无路径/不可读 → undefined(never attach)保持,且读失败时清陈旧条目,严格更正确 ✓
    • 新进程行为与改前逐字节一致(空缓存本来就现哈希)✓
    • 调用点仅三处:启动注册 / attach 判定 / spawn 交接(src/instance.ts:360、src/launcher.ts:4390、src/launcher.ts:5058),全在启动/attach 面,不在请求热路径;每调用 +1 次 statSync,代价可忽略 ✓
    • 有意行为变更披露(旧→新):文件变化后返回新构建的摘要,而不是进程生命周期的首个摘要——这正是修复目标本身。
    • 边界情形:stat 与读之间的竞态下次调用自愈(身份不匹配触发重哈希);自更新窗口(文件已换、进程未重启)符合该函数「将 spawn 脚本的哈希」契约,且 bili updater 换完即重启进程,实际两侧一致。
    • 已知局限(作者已披露):同一时间戳 tick 内的等大小原地重写会保留旧 digest 直到下次变化;npm 不是这么写的(rename-over 必换 inode),接受。
  6. 是否破坏缓存:不涉及(前缀缓存/重试/游标/去重键均未触碰)。
  7. 是否改了设计方向:没有,instance 模块内部实现细节。
  8. 是否引入新的配置:没有。

测试有效性(质量关)

  • 新测试用真实 fs 操作(write/rename/rm)+ 精确 sha256 断言,覆盖 rename-over(npm 方式)/ 原地重写 / 删除三场景;断言有牙。
  • 红绿双向本地验证:换入 master 的 instance.ts → 3 个用例全挂(第二次调用返回旧摘要);PR head 上 3/3 通过。
  • npm run typecheck 干净;相邻指纹/实例套件(issue3436 / issue3654 / issue4125 / issue4811 / port-zone-live)本地 20 过 0 挂 10 跳过(env 门控)。
  • diff 整洁:3 文件 +65/-5,CHANGELOG 按仓库体例,无锁文件/版本行/无关改动;代码命名、注释、风格均达标,无死代码无新依赖。

CI(head d9383cf)

判定

可合(合并动作留给人)。若希望矩阵全绿,合并前对那个红分片点一次 rerun 即可确认;按上面的 flake 判定直接合也成立。

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE — 单机制修复,方向对、改得干净、测试有牙。入口脚本指纹从「按路径永久缓存」改为「按文件身份(dev:ino:size:mtimeMs:ctimeMs)缓存、变了就重算」,正对 omp 长驻宿主原地更新后一直持第一版构建旧摘要、导致重拉代理互不 attach 并逐级漂移 lane 端口的根因(#5354/#4125)。改动仅 src/instance.ts 一个函数 + 回归测试 + CHANGELOG,不碰 kernel/、不扩配置面。

八问结论:

  1. 是什么问题:omp 插件跑在长驻宿主里,宿主进程缓存了入口脚本指纹且永不刷新;原地 npm i -g 后宿主仍拿首版构建的 sha256,经 BILI_SPAWN_FINGERPRINT 传给每个重拉代理。
  2. 什么原因:entryScriptFingerprint 用「解析后路径」当缓存键、命中即返回旧值,隐含假设「同路径=同内容」;npm 原地更新会让同一路径指向新字节(rename-over 换 inode / 就地改写换 size+mtime+ctime)。根因是缓存键没反映文件真实身份,不是「少一次重算」。
  3. 影响面与严重度:omp 车道(长驻宿主+自动重拉)。能用但 lane 共享失效——更新后每个新构建各占一端口、互相接不上,需重启宿主绕行;非数据丢失/非缓存泄露,p2。
  4. 怎么修的:缓存键由「路径」换成「文件身份串 dev:ino:size:mtimeMs:ctimeMs」,一致才复用旧摘要、不一致重读重算;stat 失败(删/不可读)删条目并返回 undefined,保住 #1225「读不到→绝不 attach」。两处调用点 src/launcher.ts:4390(findLiveAttachableInstance)、:5058(主启动 attach-vs-spawn)均受益。
  5. 回归风险:①未变文件多一次 statSync(不重哈希 ~5.7MB dist),调用点在启动/attach 决策而非每请求热路径,开销可忽略;②删除后返回 undefined 而非失效旧摘要,比原行为更正确;③已知边界(作者已披露):同一 fs 时间戳 tick 内等大小就地改写沿用旧摘要至下次变更,npm 不这么写,可接受。无 fail-fast→吞错、无语义漂移。
  6. 缓存:不涉及——改的是实例 attach 的代码指纹判定,不动压缩缓存/重试/增量游标/去重键。
  7. 设计方向:不变,仍是内容身份门,只是让缓存键如实反映文件身份。
  8. 新配置:无新 env/config/默认值。

测试:tests/issue5354-fingerprint-follows-the-file.test.ts 三 case(rename-over/就地改写/删除)断言精确期望(sha256(buildB)/sha256(newer)/undefined);按老逻辑第二次调用无条件返回首份缓存摘要,三断言在 master 必红——真回归钉。CI 现 failed=0(部分腿排队/重跑中)。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b) 审核完成:GitHub PR #5355(入口脚本指纹改为按文件身份 dev:ino:size:mtimeMs:ctimeMs 缓存、变了就重算)已 APPROVE —— 单机制修复、方向对、回归测试有牙、不碰 kernel/,CI failed=0;完整八问结论见 PR 上的 review。

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE(复审)— 新 head e86185a 相对我上一轮已批的 d9383cf,指纹修复本身逐字节相同(src/instance.ts 与新测试文件的 diff 为空);唯一增量是把当前 master(e209109) 合进来 + CHANGELOG 冲突按「两条都留」解决(#5354 条目仍在)。真实 PR footprint 仍是 3 个文件(CHANGELOG +1 / src/instance.ts 21+5− / 新测试 48),不碰 kernel/、不碰 pi-subagents/、不扩配置面。CI 当前 failed=0(仍在跑)。上一轮结论全部成立,继续批准。

八问(针对本 PR 的实质改动,代码与上轮一致):

  1. 是什么问题:omp 插件跑在长驻宿主里,宿主进程缓存入口脚本指纹且永不刷新;原地 npm i -g 后宿主仍拿首版构建的 sha256,经 BILI_SPAWN_FINGERPRINT 传给每个重拉代理。
  2. 什么原因:entryScriptFingerprint 用「解析后路径」当缓存键、命中即返回旧值,隐含「同路径=同内容」;npm 原地更新让同一路径指向新字节(rename-over 换 inode / 就地改写换 size+mtime+ctime)。根因是缓存键没反映文件真实身份,不是「少一次重算」。
  3. 影响面与严重度:omp 车道(长驻宿主+自动重拉)。能用但 lane 共享失效——更新后每个新构建各占一端口、互相接不上,需重启宿主绕行;非数据丢失/非缓存泄露,p2。
  4. 怎么修的:缓存键由「路径」换成「文件身份串 dev:ino:size:mtimeMs:ctimeMs」,一致才复用旧摘要、不一致重读重算;stat 失败(删/不可读)删条目并返回 undefined,保住 #1225「读不到→绝不 attach」。两处调用点 src/launcher.ts:4390(findLiveAttachableInstance)、:5058(主启动 attach-vs-spawn)均受益。
  5. 回归风险:①未变文件多一次 statSync(不重哈希 ~5.7MB dist),调用点在启动/attach 决策而非每请求热路径,开销可忽略;②删除后返回 undefined 而非失效旧摘要,比原行为更正确;③已知边界(作者已披露):同一 fs 时间戳 tick 内等大小就地改写沿用旧摘要至下次变更,npm 不这么写,可接受。无 fail-fast→吞错、无语义漂移。
  6. 缓存:不涉及——改的是实例 attach 的代码指纹判定,不动压缩缓存/重试/增量游标/去重键。
  7. 设计方向:不变,仍是内容身份门,只是让缓存键如实反映文件身份。
  8. 新配置:无新 env/config/默认值。

测试:tests/issue5354-fingerprint-follows-the-file.test.ts 三 case(rename-over/就地改写/删除)断言精确期望(sha256(buildB)/sha256(newer)/undefined);老逻辑第二次调用无条件返回首份缓存摘要,三断言在 master 必红——真回归钉。本次合并未触碰该测试与修复逻辑。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b) 复审完成(新 head e86185a):作者把当前 master 合进来,指纹修复本身与上轮已批的 d9383cf 逐字节相同、仅 CHANGELOG 冲突按「两条都留」解决;真实 footprint 仍 3 文件、不碰 kernel/,CI failed=0 → GitHub PR #5355 已再次 APPROVE(八问结论见 PR review)。

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE(第三次复审)— 新 head b8a4a2f 又是一次纯 master 同步合并(merge(e86185a, 3a2d051)),指纹修复本身与我首次批准的 d9383cf 逐字节相同(src/instance.ts 与新测试在三个 head 间 diff 全空)。真实 PR footprint 仍是 3 个文件(CHANGELOG +1 = #5354 条目 / src/instance.ts 21+5− / 新测试 48),不碰 kernel/、不碰 pi-subagents/、不扩配置面。CI 当前 failed=0(新一轮仍在排队/跑)。无新的可顺手修的小问题——修复干净、无错别字/陈旧注释/错误谓词/缺失的显然测试。结论与前两轮一致,继续批准。

八问(针对本 PR 实质改动,代码三次未变):

  1. 是什么问题:omp 插件跑在长驻宿主里,宿主进程缓存入口脚本指纹且永不刷新;原地 npm i -g 后宿主仍拿首版构建的 sha256,经 BILI_SPAWN_FINGERPRINT 传给每个重拉代理。
  2. 什么原因:entryScriptFingerprint 用「解析后路径」当缓存键、命中即返回旧值,隐含「同路径=同内容」;npm 原地更新让同一路径指向新字节(rename-over 换 inode / 就地改写换 size+mtime+ctime)。根因是缓存键没反映文件真实身份。
  3. 影响面与严重度:omp 车道(长驻宿主+自动重拉)。能用但 lane 共享失效——更新后每个新构建各占一端口、互相接不上,需重启宿主绕行;非数据丢失/非缓存泄露,p2。
  4. 怎么修的:缓存键由「路径」换成「文件身份串 dev:ino:size:mtimeMs:ctimeMs」,一致才复用旧摘要、不一致重读重算;stat 失败(删/不可读)删条目并返回 undefined,保住 #1225「读不到→绝不 attach」。两处调用点 src/launcher.ts:4390、:5058 均受益。
  5. 回归风险:①未变文件多一次 statSync(不重哈希 ~5.7MB dist),调用点在启动/attach 决策而非每请求热路径,开销可忽略;②删除后返回 undefined 而非失效旧摘要,比原行为更正确;③已知边界(作者已披露):同一 fs 时间戳 tick 内等大小就地改写沿用旧摘要至下次变更,npm 不这么写,可接受。无 fail-fast→吞错、无语义漂移。
  6. 缓存:不涉及——改的是实例 attach 的代码指纹判定,不动压缩缓存/重试/增量游标/去重键。
  7. 设计方向:不变,仍是内容身份门,只是让缓存键如实反映文件身份。
  8. 新配置:无新 env/config/默认值。

测试:tests/issue5354-fingerprint-follows-the-file.test.ts 三 case(rename-over/就地改写/删除)断言精确期望(sha256(buildB)/sha256(newer)/undefined);老逻辑第二次调用无条件返回首份缓存摘要,三断言在 master 必红——真回归钉。三轮合并均未触碰该测试与修复逻辑。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b) 第三次复审完成(新 head b8a4a2f):又是一次纯 master 同步合并,指纹修复与首次批准的 d9383cf 逐字节相同、真实 footprint 仍 3 文件不碰 kernel/,CI failed=0,无可顺手修的小问题 → GitHub PR #5355 已再次 APPROVE(八问见 PR review)。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

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

结论先行:这两个红不是本 PR(#5355)引入的回归,是 Windows / node22 环境性抖动(flake)。本 PR 只改了 src/instance.ts 的 entryScriptFingerprint(修 #5354 指纹缓存失效);两个挂掉的测试既不 import 也不引用这个函数。我已在带本 PR 改动的确切 head c0083f4258、用与 CI 同大版本的本地 Linux node v22 复跑:两个测试全绿(codex-mcp-shell 4/4、seq-interval-lane 1/1)。已触发两个失败分片的重跑(当前 in_progress,summary failed:0)。按 CI 总控纪律,这类 Windows 抖动的根治加固归口 #4599 / ci-fixes,不塞进本 PR 的 diff(scope 干净优先)。

结构化八问

  1. 是什么问题:两个 CI 检查红——tests/seq-interval-lane.test.ts 断言折叠后应有 1 个块却读到 0(one block recorded: [],0!==1);tests/codex-mcp-shell.test.ts 期望入口报告文件内容为 repo-layout\n 却读到空串。两者都只在 windows-latest/node22 挂。
  2. 什么原因导致的:均为 Windows 运行时环境抖动,与本 PR 代码无关——
  3. 影响范围与严重程度:只影响本 PR 的 CI 绿灯判定,不波及任何用户行为或产品功能。属「慢速 Windows runner 下 CI 误红」,非可用性缺陷;严重度低(流程/体验层面)。
  4. 怎么修复的:不动产品代码(本 PR diff 正确且最小)。处置=(a)立即重跑两个失败分片,让 flake 自愈;(b)根治项是把这两处等待从「固定睡眠 / 仅判存在」改成「轮询直到条件达成 + 超时上限」,该项按纪律归口 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/ci-fixes 由哨兵统一收敛,不混进指纹 PR。
  5. 是否有回归问题:无。本 PR 唯一语义变更是 entryScriptFingerprint 的缓存键由「路径」升级为「dev:ino:size:mtime:ctime」身份、身份变化才重新哈希——正是修 Long-lived omp host keeps a stale entry-script fingerprint after an in-place update, so respawned lane proxies cannot attach to each other and the lane port walks #5354 的正确方向,未触碰折叠落盘时序、MCP 入口解析、任何 wire/阈值/默认值;两个红灯用例与它零耦合(不 import instance.ts)。
  6. 是否破坏缓存:不涉及。entryScriptFingerprint 是实例脚本的内容指纹(用于注册附着判定),与 KV/前缀缓存、重试游标无关;本次未改任何缓存策略。
  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。

收工判定:本 PR 代码干净、无需改动;CI 红为环境性 flake,已触发重跑(in_progress)。若重跑仍红,即确认是 Windows 侧确定性抖动,届时按 #4599 走 ci-fixes 加固(seq-interval 的 settle 改为轮询直到块行出现;codex-mcp-shell 的 waitFor 改为要求读到非空再放行),不在本 PR 内改。

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE(第四次复审)— 新 head c0083f4 仍是纯 master 同步合并(merge(b8a4a2f, be611a8)),指纹修复与我首次批准的 d9383cf 逐字节相同(src/instance.ts 与新测试在全部四个 head 间 diff 全空)。真实 PR footprint 仍是 3 个文件(CHANGELOG +1 = #5354 条目 / src/instance.ts 21+5− / 新测试 48),不碰 kernel/、不碰 pi-subagents/、不扩配置面。CI 当前 failed=0。无新的可顺手修的小问题——修复干净。结论与前几轮一致,继续批准。

八问(针对本 PR 实质改动,代码四轮未变):

  1. 是什么问题:omp 插件跑在长驻宿主里,宿主进程缓存入口脚本指纹且永不刷新;原地 npm i -g 后宿主仍拿首版构建的 sha256,经 BILI_SPAWN_FINGERPRINT 传给每个重拉代理。
  2. 什么原因:entryScriptFingerprint 用「解析后路径」当缓存键、命中即返回旧值,隐含「同路径=同内容」;npm 原地更新让同一路径指向新字节(rename-over 换 inode / 就地改写换 size+mtime+ctime)。根因是缓存键没反映文件真实身份。
  3. 影响面与严重度:omp 车道(长驻宿主+自动重拉)。能用但 lane 共享失效——更新后每个新构建各占一端口、互相接不上,需重启宿主绕行;非数据丢失/非缓存泄露,p2。
  4. 怎么修的:缓存键由「路径」换成「文件身份串 dev:ino:size:mtimeMs:ctimeMs」,一致才复用旧摘要、不一致重读重算;stat 失败(删/不可读)删条目并返回 undefined,保住 #1225「读不到→绝不 attach」。两处调用点 src/launcher.ts:4390(findLiveAttachableInstance)、:5058(主启动 attach-vs-spawn)均受益。
  5. 回归风险:①未变文件多一次 statSync(不重哈希 ~5.7MB dist),调用点在启动/attach 决策而非每请求热路径,开销可忽略;②删除后返回 undefined 而非失效旧摘要,比原行为更正确;③已知边界(作者已披露):同一 fs 时间戳 tick 内等大小就地改写沿用旧摘要至下次变更,npm 不这么写,可接受。无 fail-fast→吞错、无语义漂移。
  6. 缓存:不涉及——改的是实例 attach 的代码指纹判定,不动压缩缓存/重试/增量游标/去重键。
  7. 设计方向:不变,仍是内容身份门,只是让缓存键如实反映文件身份。
  8. 新配置:无新 env/config/默认值。

测试:tests/issue5354-fingerprint-follows-the-file.test.ts 三 case(rename-over/就地改写/删除)断言精确期望(sha256(buildB)/sha256(newer)/undefined);老逻辑第二次调用无条件返回首份缓存摘要,三断言在 master 必红——真回归钉。四轮合并均未触碰该测试与修复逻辑。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

(ai: vllm-qwen/qwen3.8-27b) 第四次复审完成(新 head c0083f4):仍是纯 master 同步合并,指纹修复与首次批准的 d9383cf 逐字节相同、真实 footprint 仍 3 文件不碰 kernel/,CI failed=0,无可顺手修的小问题 → GitHub PR #5355 已再次 APPROVE(八问见 PR review)。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) 本轮(第四次)复审完成,结论:APPROVE。

做了什么:独立核对原 issue #5354 方向、逐行读 diff、查 CI 镜像缓存、确认修复在四个 head 间未变。

结果:新 head c0083f4 是纯 master 同步合并,指纹修复与首次批准的 d9383cf 逐字节相同;真实改动仍只有 3 个文件(CHANGELOG +1 / src/instance.ts 21+5− / 新测试 48),不碰 kernel/、不扩配置面。CI failed=0。无可顺手修的小问题。已在 GitHub PR #5355 再次 APPROVE(八问见 PR review)。

一句话:omp 长驻宿主原地更新后入口脚本指纹不再被旧缓存钉死——缓存键从「路径」换成「文件身份串(dev:ino:size:mtimeMs:ctimeMs)」,文件一变就重算哈希;删除/不可读返回 undefined 保住「读不到→绝不 attach」。行为变更已写入 CHANGELOG。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

🔶 需要决策 · (ai: vllm-qwen/qwen3.8-27b)

CI is green — the only remaining blocker is a merge conflict I cannot clear without your call.

What I verified live this run (mirror cache + GitHub API, not memory):

  • All 30 checks pass/skipped (failed: 0) — mergeable_state is now purely blocked by the conflict. The two Windows Node22 shards that were red last run (test windows-latest/node22 shard 1 & 3) now show success after the re-run triggered previously. So there is no red CI left to fix.
  • GitHub reports mergeable_state: dirty — the PR conflicts with master. Master advanced 173 commits past the branch point; the PR's actual change is small (3 files: src/instance.ts, a new test, one CHANGELOG line). The messy merge state comes from repeated "merge origin/master into feature" commits tangled into the head history (criss-cross ancestry), which drags ~70 unrelated files into conflict on a plain merge.
  • Why I'm stuck: this is an external contributor's PR — the head branch 2026-10-31_fingerprint-cache-stale lives on fork Maziar123/billion-context. Clearing the conflict requires updating that head branch, but my credential has no write access to that fork (permissions.push: false, collaborator check → 403). There is no self-service path available to me. Retargeting/replacing the contribution or opening a competing PR is not something I'll do without explicit owner approval (external-contribution policy).

The resolution itself is mechanical and ready (final tree = master + exactly the 3-file change; take master's version for the ~70 phantom files, reconcile only CHANGELOG to keep both entries, keep the auto-merged instance.ts + test). It just needs a write path to the head branch.

Options:

  • A (recommended): Author Maziar123 updates their own fork — in their clone: git fetch <upstream> master && git checkout 2026-10-31_fingerprint-cache-stale && git merge <upstream>/master && git push. CI is already green, so once the conflict clears the PR is mergeable. Preserves their ownership; zero maintainer overhead.
  • B (fastest bot-executable): You grant ranxianglei collaborator (write) access to Maziar123/billion-context. I then push an additive merge of master into their branch (keeps all their commits, no history rewrite), verify the tree equals master + the 3-file change, and confirm the PR goes clean.
  • C: You authorize landing the fix another way (maintainer-side branch + retarget, or cherry-pick the single feature commit d9383cfcbd onto master) — needs explicit sign-off since it changes how the external contribution lands.

Default if undecided: leave the PR as-is — CI is green, only the conflict remains; no destructive action taken. Awaiting your call.


结构化八问(中文):

  1. 是什么问题:PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355(外部贡献者 Maziar123 的 fork)与 master 产生合并冲突,GitHub 判定 merge 被阻塞;CI 已全部通过,唯一剩余阻塞就是这次冲突。
  2. 什么原因导致的:master 在 PR 分支点之后前进了 173 个提交;PR 实际改动只有 3 个文件(src/instance.ts、一个新测试、CHANGELOG 一行),但 head 历史里反复"把 origin/master 合并进特性分支"形成了 criss-cross 纠缠的祖先关系,普通 merge 会拖出约 70 个无关文件进冲突。
  3. 影响范围与严重程度:仅这条 PR 无法合入;不影响任何用户或生产行为(代码未变、CI 绿)。严重程度低——纯流程阻塞。
  4. 怎么修复的:本轮无可自行修复路径——head 分支只在外部 fork 上,我的凭据对该 fork 无写权限(push:false、collaborator 403),无法更新 head 来解冲突;已备好机械化解法(最终树 = master + 恰好 3 文件改动),按车道规则升级为需决策并停手。
  5. 是否有回归问题:不涉及——本轮未改任何代码/测试/配置,只做只读核查与状态确认。
  6. 是否破坏缓存:不涉及——无任何运行时/缓存相关改动。
  7. 是否改了设计方向:不涉及——未动架构/边界/数据流。
  8. 是否引入了新的配置:不涉及——无新 env/config/默认值。

判定:本轮无代码改动;因外部 fork 写权限缺失被硬阻塞,已列 A/B/C 选项请 owner 拍板;默认不动作。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Outcome — CI green; blocked on a merge conflict I can't clear without your call.

Verified live this run (mirror cache + GitHub API, not memory):

  • CI fully green — all 30 checks pass/skipped (failed: 0). The two Windows Node22 shards red last run now show success after the re-run. No red CI remains.
  • Only blocker = merge conflict with master (mergeable_state: dirty). Root cause: master advanced 173 commits past the branch point; the PR's real change is 3 files, but tangled criss-cross ancestry drags ~70 unrelated files into conflict on a plain merge.
  • Hard blocker: the head branch 2026-10-31_fingerprint-cache-stale lives on external fork Maziar123/billion-context; my credential has no write access (permissions.push: false, collaborator check 403). Clearing the conflict requires updating that head branch → no self-service path available to me. Retargeting/replacing the external contribution isn't something I'll do unilaterally.

Actions taken this run:

  • Posted the decision request (options A/B/C + recommendation + default-if-undecided) in the comment above.
  • Applied the needs-decision label via add-label.sh.
  • No code changed, no push performed; git left clean.

The mechanical resolution is ready (final tree = master + exactly the 3-file change) to execute the moment option B (grant ranxianglei write access to the fork) or C is approved. Default if undecided: leave the PR as-is.

(结构化八问与 A/B/C 选项详见上一条评论。)

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Re-checked live this run (post-restart) — no change since the last escalation:

  • CI still green — all 30 checks pass/skipped (failed: 0, pending: 0).
  • Merge still blocked — mergeable_state: dirty; PR head unchanged at c0083f4258.
  • No decision action yet — the author has not updated their fork (option A), and I still have no write access to Maziar123/billion-context (push: false), so option B is not yet executable.

The only remaining blocker is therefore still the merge conflict on the external fork, which I cannot clear without one of the options in my earlier 🔶 需要决策 comment:

  • A (recommended) — author Maziar123 merges master into their branch and pushes;
  • B — you grant ranxianglei write access to the fork → I push the ready additive merge;
  • C — you authorize an alternative landing path.

Default if undecided: leave the PR as-is. No code changed this run, no push performed, git left clean. Awaiting your call — I'll execute immediately once B or C is approved.

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE(第五次复审)— 新 head dc25251 仍是纯 master 同步合并(merge(c0083f4, dc041dd)),指纹修复与我首次批准的 d9383cf 逐字节相同(src/instance.ts 的 entryScriptFingerprint 与新测试在全部五个 head 间未变)。真实 PR footprint(相对并入的 master 父提交 dc041dd)仍是 3 个文件:CHANGELOG +1(#5354 条目)/ src/instance.ts 21+5−(仅指纹函数段)/ 新测试 48,不碰 kernel/、不碰 pi-subagents/、不扩配置面。CI 当前 failed=0。无可顺手修的小问题、无开放尾巴。结论与前几轮一致,继续批准。

说明(透明起见):跨 head 的朴素 diff 会看到 atomicWriteJson 由私有变 export 并加「#5955 wedged-holder ledger」注释——那是本次并入的 master(dc041dd) 自身改动,不是本 PR 的;已用真实 footprint(对 master 父提交)核对,本 PR 对 instance.ts 的贡献只有指纹函数那段。

八问(针对本 PR 实质改动,代码五轮未变):

  1. 是什么问题:omp 插件跑在长驻宿主里,宿主进程缓存入口脚本指纹且永不刷新;原地 npm i -g 后宿主仍拿首版构建的 sha256,经 BILI_SPAWN_FINGERPRINT 传给每个重拉代理。
  2. 什么原因:entryScriptFingerprint 用「解析后路径」当缓存键、命中即返回旧值,隐含「同路径=同内容」;npm 原地更新让同一路径指向新字节(rename-over 换 inode / 就地改写换 size+mtime+ctime)。根因是缓存键没反映文件真实身份。
  3. 影响面与严重度:omp 车道(长驻宿主+自动重拉)。能用但 lane 共享失效——更新后每个新构建各占一端口、互相接不上,需重启宿主绕行;非数据丢失/非缓存泄露,p2。
  4. 怎么修的:缓存键由「路径」换成「文件身份串 dev:ino:size:mtimeMs:ctimeMs」,一致才复用旧摘要、不一致重读重算;stat 失败(删/不可读)删条目并返回 undefined,保住 #1225「读不到→绝不 attach」。两处调用点 src/launcher.ts:4390(findLiveAttachableInstance)、:5058(主启动 attach-vs-spawn)均受益。
  5. 回归风险:①未变文件多一次 statSync(不重哈希 ~5.7MB dist),调用点在启动/attach 决策而非每请求热路径,开销可忽略;②删除后返回 undefined 而非失效旧摘要,比原行为更正确;③已知边界(作者已披露):同一 fs 时间戳 tick 内等大小就地改写沿用旧摘要至下次变更,npm 不这么写,可接受。无 fail-fast→吞错、无语义漂移。
  6. 缓存:不涉及——改的是实例 attach 的代码指纹判定,不动压缩缓存/重试/增量游标/去重键。
  7. 设计方向:不变,仍是内容身份门,只是让缓存键如实反映文件身份。
  8. 新配置:无新 env/config/默认值。

测试:tests/issue5354-fingerprint-follows-the-file.test.ts 三 case(rename-over/就地改写/删除)断言精确期望(sha256(buildB)/sha256(newer)/undefined);老逻辑第二次调用无条件返回首份缓存摘要,三断言在 master 必红——真回归钉。五轮合并均未触碰该测试与修复逻辑。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) 第五轮复审完成,结论:APPROVE(GitHub PR #5355,head dc25251)。

说明:上一轮我处理到一半进程被强杀(卡住 35 分钟),现已重新核完并补齐——不再空转。

做了什么:独立核对原 issue #5354 方向、逐行读 diff、查 CI 镜像缓存、确认修复在五个 head 间未变。

结果:新 head dc25251 是纯 master 同步合并(merge(c0083f4, dc041dd)),指纹修复与首次批准的 d9383cf 逐字节相同;真实 footprint 仍只有 3 个文件(CHANGELOG +1 / src/instance.ts 21+5− 仅指纹函数段 / 新测试 48),不碰 kernel/、不扩配置面。CI failed=0。无可顺手修的小问题、无开放尾巴。已在 GitHub 再次 APPROVE(八问见 PR review)。

一个透明说明:跨 head 的朴素 diff 里能看到 atomicWriteJson 由私有变 export 并加「#5955 wedged-holder ledger」注释——那是本次并入的 master 自身改动,不是本 PR 的;已用真实 footprint(对 master 父提交)核对确认。

一句话:omp 长驻宿主原地更新后入口脚本指纹不再被旧缓存钉死——缓存键从「路径」换成「文件身份串(dev:ino:size:mtimeMs:ctimeMs)」,文件一变就重算哈希;删除/不可读返回 undefined 保住「读不到→绝不 attach」。行为变更已写入 CHANGELOG。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Update — the author has acted (option A), CI is green, and the remaining dirty flag looks stale. Verified live this run (mirror cache + anonymous ls-remote + fetched the head + a real local merge):

  • Author updated their fork: head is now dc252516 (was c0083f4258). Its CHANGELOG carries recent master entries (the #6132 compress-at-fold-point removal, AGENTS.md slimming), i.e. they merged current master in — exactly option A from my earlier 🔶 需要决策 comment.
  • CI fully green on dc252516: all 30 checks pass (failed: 0, pending: 0).
  • The #5354 fix is intact in the new head (entryScriptFingerprint identity logic at src/instance.ts:333–342, plus its regression test).
  • No actual conflict exists: I fetched the latest master (c26c31ace4) and did a real local merge of it into dc252516 → 0 conflicting files (the branch is only ~1 master-commit behind and auto-merges cleanly, including instance.ts). Yet GitHub/mirror still reports mergeable_state: dirty. On a very fast-moving master, GitHub's async mergeability computation lags — so this dirty is most likely a stale read, not a live conflict.

Where that leaves us: I still have no write access to Maziar123/billion-context (push: false), so I can't push anything to force a recompute either way. Two realistic paths:

  • Likely self-clears: as GitHub recomputes mergeability against the settled base, the stale dirty should flip to clean/blocked with no action needed — worth a short wait before treating it as a real blocker.
  • If it persists: option B — you grant ranxianglei write access to the fork, and I push a trivial refresh (merge the one newer master commit) to force GitHub to recompute and clear it. (Or the author does a quick re-push.)

No code changed, no push performed this run. Holding at needs-decision; I'll act immediately if access is granted or if CI actually goes red.


结构化八问(中文·简):

  1. 是什么问题:任务要求解 PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 与 master 的合并冲突;本轮发现作者已按选项 A 更新 fork(head→dc252516),且 GitHub 的 dirty 标记疑似过期。
  2. 什么原因导致的:作者把近期 master 并入分支后推送;master 仍在快速推进,GitHub 异步可合并性计算滞后,故仍显示 dirty(我本地真实合并为 0 冲突)。
  3. 影响范围与严重程度:仅该 PR 合入流程;不影响用户/生产。当前无真实冲突、CI 全绿——实际风险低,主要是状态标记滞后的误报阻塞。
  4. 怎么修复的:非我修复——作者按选项 A 自行解决;我只做只读核验(新 head 含 Long-lived omp host keeps a stale entry-script fingerprint after an in-place update, so respawned lane proxies cannot attach to each other and the lane port walks #5354 修复+测试、CI 绿、本地合并 0 冲突)。
  5. 是否有回归问题:不涉及——未改任何代码/测试/配置。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新的配置:不涉及。

判定:作者已解决冲突、CI 全绿、本地实测无冲突,GitHub 的 dirty 大概率为过期标记;我对外部 fork 无写权限无法主动刷新。建议先短等观察其自愈;若持续则走选项 B(授予 ranxianglei 该 fork 写权限)由我推一次极小更新强制重算。等待 owner 决定。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Closing note for the record (full detail in my previous comment): nothing new to act on — state is stable and holding.

Verified live ~20 min ago (mirror cache + ls-remote + fetched head + real local merge):

  • Head dc252516 (author already merged current master in = option A), #5354 fix + test intact.
  • CI 30/30 green (failed: 0, pending: 0).
  • Real local merge of latest master (c26c31ace4) into dc252516 → 0 conflicts; branch only ~1 master-commit behind. The GitHub dirty flag is a stale async-mergeability read, not a live conflict.
  • I have no write access to Maziar123/billion-context (push: false), so I cannot push a refresh to force a recompute.

What remains / risk: low. The stale dirty should self-clear as GitHub recomputes against the settled base. If it persists, the only unblock is option B (you grant ranxianglei write access to the fork → I push a one-commit refresh) or the author re-pushes. I am not re-running checks in a loop; I'll act immediately only if access is granted or CI actually goes red.


结构化八问(中文·简,本轮无代码改动):

  1. 是什么问题:解 PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 与 master 的合并冲突;实际作者已自行解决,GitHub 的 dirty 为过期标记。
  2. 什么原因导致的:作者已并入近期 master 并推送;master 快速推进致 GitHub 异步可合并性计算滞后。
  3. 影响范围与严重程度:仅该 PR 合入流程;无真实冲突、CI 全绿,风险低(状态误报阻塞)。
  4. 怎么修复的:作者按选项 A 解决;我仅只读核验(head 含修复+测试、CI 绿、本地合并 0 冲突)。
  5. 是否有回归问题:不涉及——未改任何代码/测试/配置。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新的配置:不涉及。

判定:冲突已由作者解决、CI 全绿、本地实测无冲突;我无 fork 写权限,维持 needs-decision 待命,除非授权访问或 CI 变红,否则不再动作。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Live re-check (this run): unchanged — and the stale dirty is NOT self-clearing.

  • Mirror [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355: head dc252516, CI 30/30 green (failed: 0, pending: 0), mergeable_state: dirty.
  • Fork head still dc252516 (author's option-A push; #5354 fix + test intact).
  • My real local merge of latest master into dc252516 = 0 conflicts (branch only ~1 master-commit behind) → no actual conflict; GitHub's dirty is a stuck mergeable-state read.

Revised conclusion (correcting my earlier "likely self-clears"): this dirty has persisted across multiple checks over several hours and has not recomputed on base movement alone. A clean local merge proves there's no real conflict, so the only reliable way to clear a stuck GitHub mergeable read is a head push (even a no-op, or merging the one newer master commit). I cannot do that — I have no write access to Maziar123/billion-context (push: false).

Actionable ask (recommendation now firm):

  • Option B (recommended): grant ranxianglei write access to the fork → I push a one-commit refresh to force GitHub to recompute and clear dirty.
  • Option C (equally fast): the author does a trivial re-push of the branch.
  • Waiting for self-clear is no longer viable.

needs-decision label already applied. No code changed, no push this run. I'll act the moment access is granted (or if CI goes red), and I won't repeat this analysis on every re-dispatch cycle.


结构化八问(中文·简,本轮无代码改动):

  1. 是什么问题:PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 与 master 的合并冲突;实测无真实冲突,GitHub dirty 为卡死的过期状态。
  2. 什么原因导致的:作者已并入近期 master 并推送(head dc252516),本地真实合并 0 冲突;GitHub 可合并性状态未在 base 变动时重算,卡在 dirty。
  3. 影响范围与严重程度:仅该 PR 合入流程(被误报阻塞);不影响用户/生产,CI 全绿。
  4. 怎么修复的:非我修复;作者已解决冲突。我确认需一次 head push 才能清除卡死的 dirty,而我无 fork 写权限。
  5. 是否有回归问题:不涉及——未改任何代码/测试/配置。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新的配置:不涉及。

判定:冲突已解决、CI 全绿、本地无冲突;dirty 为卡死的 GitHub 状态,需一次 head push 清除。推荐选项 B(授予 ranxianglei fork 写权限)由我推一次极小刷新,或作者重新推送。等待 owner 决定。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Re-dispatch — verified live this run, nothing changed; still blocked on the same stuck flag.

  • Mirror [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355: head dc252516, CI 30/30 green (failed: 0, pending: 0), mergeable_state: dirty (unchanged for ~14h).
  • Fork head still dc252516; my real local merge of latest master into it = 0 conflicts → no actual conflict.
  • I still have no write access to Maziar123/billion-context (push: false) — re-verified this run.
  • No new owner/author decision in the thread. The reviewer has already APPROVED the PR (earlier floor), so the only thing between this PR and merge is the stuck GitHub dirty, which clears on a head push.

This keeps re-firing each cycle because the mirror still reports dirty; I can't break that loop without one of: option B (grant ranxianglei write access → I push a one-commit refresh) or the author re-pushes. needs-decision label already applied. No code changed, no push this run.


结构化八问(中文·简,本轮无代码改动):

  1. 是什么问题:PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 被 GitHub 卡死的 dirty 可合并性状态阻塞;实测无真实冲突。
  2. 什么原因导致的:作者已并入近期 master(head dc252516)且本地合并 0 冲突,但 GitHub 的可合并性状态未重算,持续显示 dirty。
  3. 影响范围与严重程度:仅该 PR 合入被误报阻塞;不影响用户/生产,CI 全绿、已获 reviewer APPROVE。
  4. 怎么修复的:非我修复;需一次 head push 清除卡死状态,而我无 fork 写权限,故维持待命。
  5. 是否有回归问题:不涉及——未改任何代码/测试/配置。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新的配置:不涉及。

判定:PR 已 APPROVE、CI 全绿、本地无冲突;唯一阻塞是卡死的 GitHub dirty,需 head push 清除。推荐选项 B(授予 ranxianglei fork 写权限)由我推一次极小刷新,或作者重新推送。未获授权前不再动作。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Reporting after the idle-restart: I am in a stable BLOCKED state and need your decision to proceed. Re-verified live just now (nothing has changed in ~20h):

The blocker (permission): the PR is already APPROVED by the reviewer and CI is fully green; my local merge of latest master into the head = 0 conflicts, so there is no real conflict. The only thing left is the stuck GitHub dirty flag, which clears on a head push — and I cannot push to the external fork because I have no write access. That permission is the hard blocker; I have no self-service path around it (retargeting/replacing the external contribution is not something I'll do unilaterally).

To unblock (one action):

  • Option B (recommended): grant ranxianglei write access to Maziar123/billion-context → I push a one-commit refresh and the flag clears.
  • Option C: the author Maziar123 re-pushes the branch.

needs-decision label already applied. No code changed, no push this run. Until one of the above happens, further re-dispatches will find the same state — I'm holding and will act the instant access is granted or CI goes red.


结构化八问(中文·简,本轮无代码改动):

  1. 是什么问题:PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 被 GitHub 卡死的 dirty 状态阻塞,而我清除它所需的 fork 写权限缺失。
  2. 什么原因导致的:作者已并入 master(head dc252516,本地合并 0 冲突);GitHub 可合并性状态未重算卡在 dirty,清除需 head push,而我对外部 fork 无写权限。
  3. 影响范围与严重程度:仅该 PR 合入被阻塞 + 车道反复空转/重启;不影响用户/生产(CI 全绿、代码已 APPROVE)。
  4. 怎么修复的:非我修复——受阻于权限,无法执行 head push;已按车道规则停在 needs-decision。
  5. 是否有回归问题:不涉及——未改任何代码/测试/配置。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新的配置:不涉及。

判定:代码已批、CI 全绿、无真实冲突;唯一阻塞是卡死的 GitHub dirty + 我无 fork 写权限。需 owner 授予写权限(选项 B)或作者重新推送(选项 C)才能推进。等待授权。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Updated diagnosis (correcting my earlier "stale flag" read): there IS now a real conflict — master moved. Verified live this run:

  • Mirror [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355: head dc252516, CI 30/30 green (failed: 0), mergeable_state: dirty.
  • Master advanced 161 commits since my last check (c26c31ace4 → e1f31ad9d5). Re-testing the merge against the current master now yields 1 conflict: CHANGELOG.md (my earlier "0 conflicts / stale" was measured against the older master tip). So GitHub's dirty is correct, not stale.
  • The conflict is isolated to CHANGELOG.md — src/instance.ts and the new test auto-merge cleanly even 161 commits behind. The fix is mechanical: merge current master into the branch and keep both the #5354 entry and master's entries in CHANGELOG.

Still blocked on the same permission: I have no write access to Maziar123/billion-context (push: false, re-confirmed), so I can't push the resolved merge myself.

To unblock (one action):

  • Option A (natural now that there's a real conflict): author Maziar123 merges current master into their branch and pushes — they'll hit the same single CHANGELOG.md conflict and resolve it by keeping both entries.
  • Option B (bot-executable): grant ranxianglei write access to the fork → I merge e1f31ad9d5, resolve CHANGELOG.md (keep both entries), verify the tree = master + the 3-file change, and push.

needs-decision label already applied. No code changed, no push this run.


结构化八问(中文·简,本轮无代码改动):

  1. 是什么问题:PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 与 master 的真实合并冲突(更正我此前"过期标记"的误判)。
  2. 什么原因导致的:master 自上次检查后又前进 161 个提交(c26c31ace4→e1f31ad9d5),分支落后;对当前 master 合并实测仅 CHANGELOG.md 冲突(instance.ts 与新测试自动合并干净)。
  3. 影响范围与严重程度:仅该 PR 合入被阻塞;不影响用户/生产(CI 全绿、代码已 APPROVE)。严重度低(流程阻塞)。
  4. 怎么修复的:非我修复——修法明确(合并当前 master + CHANGELOG 保留两条),但我无 fork 写权限,无法推送。
  5. 是否有回归问题:不涉及——未改任何代码/测试/配置。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新的配置:不涉及。

判定:更正——现对当前 master 存在真实冲突(仅 CHANGELOG.md);修法明确但受 fork 写权限所阻。推荐选项 A(作者合并 master 并推送)或选项 B(授予 ranxianglei 写权限由我推送)。等待 owner 决定。

@Maziar123

Copy link
Copy Markdown
Contributor Author

Merged current master into the branch and kept both entries in the one CHANGELOG.md conflict.

The diff against master is still the same 3 files: src/instance.ts, tests/issue5354-fingerprint-follows-the-file.test.ts, and one CHANGELOG.md line.

Local check on the new head: tests/issue5354-fingerprint-follows-the-file.test.ts passes 3/3. CI is the full check.

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Verified — your option-A push cleared the blocker. Checked live this run (mirror cache + fork + a real local merge of current master into the head), not from memory:

  • New head 1b8594f913 (was dc252516d0). You merged current master — master tip e7a42babec is now an ancestor of the head, so the branch is fully up to date (merge-base = master tip).
  • Conflict resolved. GitHub mergeable_state flipped dirty → blocked. blocked here means "required CI checks not yet green" — the branch is not behind master — so there is no merge conflict left. That was the single CHANGELOG.md conflict you described, now resolved by keeping both entries (the #5354 entry is present under ### Fixed; all master entries are intact).
  • Footprint unchanged — still exactly 3 files (65+/5−): CHANGELOG.md +1, src/instance.ts 21+/5− (only the entryScriptFingerprint region), tests/issue5354-fingerprint-follows-the-file.test.ts +48.
  • Fix + test byte-identical to the version the reviewer already APPROVED (d9383cfcbd): I diffed the entryScriptFingerprint function and the whole test file against that commit — zero difference. The test's only non-builtin deps are tests/tmp-rm.ts (identical to master) and src/instance.ts (master + only the fingerprint change), so the freshly-merged master introduced no new coupling.

CI on 1b8594f913 is running: 0 failed so far (11 success / 11 skipped / 2 in progress / 6 queued — the test shards + e2e lanes still going). It's the authoritative full-suite gate and will land shortly.

Where that leaves it: the only thing that was between this PR and merge — the conflict — is gone. The PR is ready to merge once CI is green (merging is human-only). The reviewer already APPROVED. If any CI shard goes red I'll be woken up and handle it; given the fix + test are byte-identical to what passed CI 30/30 across the five prior heads and the merge is clean, I expect green.

Housekeeping: cleared the now-stale needs-decision label (the option A/B/C decision point is resolved — you did A).


结构化八问(中文):

  1. 是什么问题:外部贡献者 PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 此前被「与 master 的合并冲突」连续卡住多轮;本轮作者已按选项 A 把当前 master 并入分支并推送,需核实冲突是否真的解决、diff 是否仍干净、CI 状态。
  2. 什么原因导致的:此前 master 大幅前移(161+ 提交)导致分支落后、CHANGELOG.md 冲突;作者本次已合并当前 master 并保留双方条目,冲突根因消除。
  3. 影响范围与严重程度:仅该 PR 的合入流程;不涉及任何用户/生产行为(代码与已批准版本逐字节一致)。此前严重度低(纯流程阻塞),现已解除。
  4. 怎么修复的:非我修复——作者按选项 A 自行合并 master 并解决 CHANGELOG 冲突;我做了只读核验(新 head 含修复+测试、master 为祖先、净 diff 仍 3 文件、修复与测试与已批准版 d9383cfcbd 逐字节一致、CI 在跑 0 failed)。
  5. 是否有回归问题:无。净 diff 只含 entryScriptFingerprint 一处(缓存键由路径改为文件身份串 dev/ino/size/mtime/ctime,变了重算、读不到返回 undefined),与已批准版逐字节相同;未触碰 wire/阈值/默认值/kernel/。
  6. 是否破坏缓存:不涉及(前缀缓存/重试/游标/去重键均未触碰)。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新的配置:不涉及。

判定:冲突已解除(dirty→blocked,master 为祖先、净 diff 仍 3 文件、修复与测试与已批准版逐字节一致);CI 在跑(0 failed)。待 CI 全绿即可由 owner 合并(合并动作留给人)。

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE(第六次复审)— 新 head 1b8594f 仍是纯 master 同步合并(merge(dc25251, e7a42ba)),指纹修复与我首次批准的 d9383cf 逐字节相同(src/instance.ts 的 entryScriptFingerprint 与新测试在全部六个 head 间未变)。真实 PR footprint(相对并入的 master 父提交 e7a42ba)仍是 3 个文件:CHANGELOG +1(#5354 条目)/ src/instance.ts 21+5−(仅指纹函数段)/ 新测试 48,不碰 kernel/、不碰 pi-subagents/、不扩配置面。CI 当前 failed=0。无可顺手修的小问题、无开放尾巴。结论与前几轮一致,继续批准。

说明(透明起见):跨 head 的朴素 diff 会看到 atomicWriteJson 由私有变 export 并加「#5955 wedged-holder ledger」注释——那是本次并入的 master 自身改动,不是本 PR 的;已用真实 footprint(对 master 父提交)核对,本 PR 对 instance.ts 的贡献只有指纹函数那段。

八问(针对本 PR 实质改动,代码六轮未变):

  1. 是什么问题:omp 插件跑在长驻宿主里,宿主进程缓存入口脚本指纹且永不刷新;原地 npm i -g 后宿主仍拿首版构建的 sha256,经 BILI_SPAWN_FINGERPRINT 传给每个重拉代理。
  2. 什么原因:entryScriptFingerprint 用「解析后路径」当缓存键、命中即返回旧值,隐含「同路径=同内容」;npm 原地更新让同一路径指向新字节(rename-over 换 inode / 就地改写换 size+mtime+ctime)。根因是缓存键没反映文件真实身份。
  3. 影响面与严重度:omp 车道(长驻宿主+自动重拉)。能用但 lane 共享失效——更新后每个新构建各占一端口、互相接不上,需重启宿主绕行;非数据丢失/非缓存泄露,p2。
  4. 怎么修的:缓存键由「路径」换成「文件身份串 dev:ino:size:mtimeMs:ctimeMs」,一致才复用旧摘要、不一致重读重算;stat 失败(删/不可读)删条目并返回 undefined,保住 #1225「读不到→绝不 attach」。两处调用点 src/launcher.ts:4390(findLiveAttachableInstance)、:5058(主启动 attach-vs-spawn)均受益。
  5. 回归风险:①未变文件多一次 statSync(不重哈希 ~5.7MB dist),调用点在启动/attach 决策而非每请求热路径,开销可忽略;②删除后返回 undefined 而非失效旧摘要,比原行为更正确;③已知边界(作者已披露):同一 fs 时间戳 tick 内等大小就地改写沿用旧摘要至下次变更,npm 不这么写,可接受。无 fail-fast→吞错、无语义漂移。
  6. 缓存:不涉及——改的是实例 attach 的代码指纹判定,不动压缩缓存/重试/增量游标/去重键。
  7. 设计方向:不变,仍是内容身份门,只是让缓存键如实反映文件身份。
  8. 新配置:无新 env/config/默认值。

测试:tests/issue5354-fingerprint-follows-the-file.test.ts 三 case(rename-over/就地改写/删除)断言精确期望(sha256(buildB)/sha256(newer)/undefined);老逻辑第二次调用无条件返回首份缓存摘要,三断言在 master 必红——真回归钉。六轮合并均未触碰该测试与修复逻辑。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) 第六轮复审完成,结论:APPROVE(GitHub PR #5355,head 1b8594f)。

新 head 是纯 master 同步合并(merge(dc25251, e7a42ba)),指纹修复与首次批准的 d9383cf 逐字节相同;真实 footprint 仍只有 3 个文件(CHANGELOG +1 / src/instance.ts 21+5− 仅指纹函数段 / 新测试 48),不碰 kernel/、不扩配置面。CI failed=0。无可顺手修的小问题、无开放尾巴。已在 GitHub 再次 APPROVE(八问见 PR review)。

透明说明:跨 head 朴素 diff 里的 atomicWriteJson 私有→export(#5955 wedged-holder ledger)是本次并入的 master 自身改动,非本 PR;已用真实 footprint 核对确认。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 🔶 需要决策 · (ai: vllm-qwen/qwen3.8-27b)

Same recurring blocker — re-verified live this run, and I still cannot clear it without your call. (This is the Nth round of the loop from floors 25–63; the author's option-A push on 2026-11-15 cleared it once, but master has moved again.)

Verified now (mirror cache + fork + a real local merge of current master into the head — not from memory):

  • CI fully green on head 1b8594f913: 18 success / 12 skipped, 0 failures (30/30).
  • Master moved e7a42babec → a259e0b8fa over the 2 days since the author's last push. The head is now behind master.
  • Real local merge = 1 conflict, isolated to CHANGELOG.md — HEAD carries the #5354 entry, master added a new pi-nudge entry; resolution is keep both (exactly as before). Every other file auto-merges cleanly; the #5354 fix + test remain byte-identical to the 6×-approved d9383cfcbd.
  • Hard blocker (unchanged): I have no write access to the fork Maziar123/billion-context (push: false, only pull: true) — re-confirmed this run via the API. So I cannot push the resolved merge. The fix is done, approved, and CI-green; only the fork write path is missing.

Options:

  • A: author Maziar123 merges current master into their branch and pushes (they have write access to their own fork) — what they did on 2026-11-15. Low owner effort, preserves the PR, but recurs on the next master move.
  • B (recommended): you grant ranxianglei write access to Maziar123/billion-context → I push the resolved merge (mechanical, ready) and can keep the branch current on every future master move, breaking the loop. Preserves the contributor's authorship (merge commit stays on their branch).
  • C (most decisive): you land the fix directly — e.g. cherry-pick the single approved commit d9383cfcbd onto master and close this PR. Ends the ~7-day churn now; it changes how the external contribution lands, so it needs your explicit sign-off.

Recommendation: B — one-time grant, stops the recurring churn, keeps the contribution intact. C if you'd rather just land it now regardless of the PR.
Default if undecided: leave the PR as-is (no destructive action) — it stays blocked on the fork write path.

No code changed this run, no push performed, git left clean (the local merge test ran in a throwaway worktree, since removed).


结构化八问(中文):

  1. 是什么问题:外部贡献者 PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 与 master 的合并冲突再次出现(master 两天前移),而我无法更新外部 fork 上的 head 分支来解冲突。
  2. 什么原因导致的:master 自作者上次 option-A 推送(floor 64,head 1b8594f913 含 master 至 e7a42babec)后又前移到 a259e0b8fa;分支落后,本地真实合并仅 CHANGELOG.md 一处冲突(HEAD 的 Long-lived omp host keeps a stale entry-script fingerprint after an in-place update, so respawned lane proxies cannot attach to each other and the lane port walks #5354 条目 vs master 新增的 pi-nudge 条目,保留两条即可);我对外部 fork 无写权限(push:false)。
  3. 影响范围与严重程度:仅该 PR 合入被阻塞 + 车道反复空转;不影响用户/生产(CI 全绿、代码已 6 次 APPROVE、与批准版逐字节一致)。严重度低(纯流程阻塞),但已持续约 7 天。
  4. 怎么修复的:非我修复——修法机械且已备好(合并当前 master + CHANGELOG 保留两条),但受 fork 写权限所阻,无法推送;已按车道规则停在 needs-decision(标签已重新打上)。
  5. 是否有回归问题:不涉及——本轮未改任何代码/测试/配置,只做只读核查(本地合并测试后已清理 worktree)。
  6. 是否破坏缓存:不涉及。
  7. 是否改了设计方向:不涉及。
  8. 是否引入新的配置:不涉及。

判定:CI 全绿、冲突仅 CHANGELOG.md 一处(机械可解)、修复与测试与 6 次批准版逐字节一致;唯一阻塞是我对外部 fork 无写权限。需 owner 在 A/B/C 中拍板(推荐 B 授权我推送以终结 churn,或 C 直接落地);默认不动作。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Re-dispatch of the same auto-fix lane. Re-verified live just now (status freshness) — nothing has changed since the decision request above:

  • Fork push: false (option B — owner grant — not yet taken).
  • Fork head still 1b8594f913 (option A — author push — not yet taken).
  • Master still a259e0b8fa; the mirror now freshly recomputes mergeable_state: dirty (the single CHANGELOG.md conflict); CI still green (18 success / 12 skipped / 0 failures).

The blocker is unchanged: the #5354 fix is done, 6×-approved, and CI-green, but I have no write access to the external fork to push the (mechanical, ready) resolved merge. The A/B/C decision request is already posted above and needs-decision is applied — I'm holding for your call and will not re-post the same analysis on every re-dispatch.

Unblocks on B (you grant ranxianglei write access to the fork → I push immediately) or A (author merges current master + pushes). Default: no action, PR stays blocked on the fork write path.

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Status (no change since the 🔶 需要决策 request above): fork Maziar123/billion-context still push: false, head still 1b8594f913, master a259e0b8fa, CI green (18 success / 12 skipped / 0 failures), conflict still the single CHANGELOG.md (keep-both).

Still blocked on the external-fork write path — the #5354 fix is done, 6×-approved, and CI-green, but I have no write access to the fork to push the (mechanical, ready) resolved merge. Holding for your A/B/C call (recommend B: grant ranxianglei fork write access → I push immediately) or the author's push (A).

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Re-dispatch (attempt 1/6). Re-verified live — the only change since my last note is master advanced (a259e0b8fa → 548e396022); everything else is unchanged:

  • Fork Maziar123/billion-context still push: false (owner hasn't granted access — option B not taken).
  • Head still 1b8594f913 (author hasn't pushed — option A not taken).
  • CI still green (30 total / 0 failed); mergeable_state still dirty.
  • Re-ran the real merge test against the new master: the conflict is still the single CHANGELOG.md (keep-both) — all other files auto-merge cleanly.

So the blocker is unchanged: the #5354 fix is done, 6×-approved, and CI-green, but I have no write access to the external fork to push the (mechanical, ready) resolved merge. The 🔶 需要决策 request (A/B/C, recommend B: grant ranxianglei fork write access → I push immediately) is still pending with the owner. Holding.

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Status confirmed (full detail at floor 78) — no change this run: master advanced to 548e396022, but fork Maziar123/billion-context still push: false, head still 1b8594f913, CI green (0 failed), mergeable_state still dirty, and the conflict is still the single CHANGELOG.md (keep-both) per a fresh local merge test.

Still blocked on the external-fork write path — the #5354 fix is done, 6×-approved, and CI-green, but I have no write access to the fork to push the (mechanical, ready) resolved merge. The 🔶 需要决策 request (A/B/C, recommend B: grant ranxianglei fork write access → I push immediately) is pending with the owner. Holding.

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

No change — still holding. Re-verified live this run: fork Maziar123/billion-context still push: false (option B not taken), head still 1b8594f913 (option A not taken), CI green (30/30, 0 failed). Only change: master advanced 548e396022 → a446a3c1c8; a fresh local merge confirms the conflict is still the single CHANGELOG.md (keep-both), src/instance.ts auto-merges cleanly. Still blocked on the external-fork write path; the 🔶 需要决策 request (A/B/C, recommend B: grant ranxianglei fork write access → I push immediately) is pending with the owner.

结构化八问(中文·本轮无代码改动)

  1. 是什么问题:PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 被外部 fork 写权限缺失阻塞(push: false),无法推入已备好的解冲突合并;与上轮相比无变化。
  2. 什么原因导致的:我对 Maziar123/billion-context 无写权限,作者也未自行推送;master 又前进(548e396022→a446a3c1c8),但冲突仍是单条 CHANGELOG.md(两条都留)。
  3. 影响范围与严重程度:仅该 PR 合入流程;不影响用户/生产,CI 全绿、已获 reviewer 多次 APPROVE。严重度低(纯流程阻塞)。
  4. 怎么修复的:非我修复——需 owner 授予 ranxianglei fork 写权限(选项 B)或作者自推(选项 A);我仅做只读核验。
  5. 是否有回归问题:不涉及——本轮未改任何代码/测试/配置。
  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。

判定:阻塞点(外部 fork 写路径)无变化,维持 needs-decision 待命;未获授权前不再动作。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

No change — still holding. Re-verified live: fork Maziar123/billion-context still push: false (option B not taken), head still 1b8594f913 (option A not taken), CI green (30/30, 0 failed), master unchanged (a446a3c1c8). Blocker (external-fork write path) unchanged; the 🔶 需要决策 request (A/B/C, recommend B: grant ranxianglei fork write access → I push immediately) is still pending with the owner.

结构化八问(中文·本轮无任何改动)

  1. 是什么问题:PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 被外部 fork 写权限缺失阻塞(push: false),与上轮完全一致。
  2. 什么原因导致的:我对 fork 无写权限、作者也未自行推送;master 本轮未移动,冲突仍为单条 CHANGELOG.md(两条都留)。
  3. 影响范围与严重程度:仅该 PR 合入流程;不影响用户/生产,CI 全绿、已获 reviewer 多次 APPROVE。严重度低(纯流程阻塞)。
  4. 怎么修复的:非我修复——需 owner 授予 ranxianglei fork 写权限(选项 B)或作者自推(选项 A);我仅做只读核验。
  5. 是否有回归问题:不涉及——本轮未改任何代码/测试/配置。
  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。

判定:阻塞点(外部 fork 写路径)无变化,维持 needs-decision 待命;未获授权前不再动作。

@Maziar123

Copy link
Copy Markdown
Contributor Author

Merged current master into the PR branch and resolved the only conflict in CHANGELOG.md, keeping both entries. The implementation and regression test are unchanged.

Validation: the targeted fingerprint test passed 3/3, and npm run typecheck passed. The branch is now updated; no fork-access grant is needed for this conflict resolution.

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

The author resolved the conflict (option A) — the PR is now mergeable, pending CI green + your merge.

Verified live this run:

What remains: the 2 pending checks to finish, then your merge. No fork-access grant was needed — the author pushed the resolved merge themselves. I'll confirm the 2 pending checks green when they land.

结构化八问(中文)

  1. 是什么问题:PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 此前被「外部 fork 写权限缺失 + CHANGELOG.md 冲突」阻塞;本轮作者已自行解决。
  2. 什么原因导致的:作者把当前 master 合入分支、按「两条都留」解决唯一的 CHANGELOG.md 冲突后推送(选项 A),冲突随之清除。
  3. 影响范围与严重程度:仅该 PR 合入流程;不影响用户/生产。冲突已清、修复完整、CI 基本全绿。
  4. 怎么修复的:非我修复——作者自行合并 master 并解冲突;我只做只读核验(新 head 为规范 merge、master 已是其祖先、Long-lived omp host keeps a stale entry-script fingerprint after an in-place update, so respawned lane proxies cannot attach to each other and the lane port walks #5354 修复与测试完整)。
  5. 是否有回归问题:不涉及——本轮我未改任何代码;修复本身与已批的 d9383cf 逐字节一致。
  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。

判定:冲突已清、PR 现可合(待 2 个无关检查跑绿 + owner 合并);我未改任何代码。待 2 个 pending 检查落定后我会补一句确认全绿。

@Maziar123

Copy link
Copy Markdown
Contributor Author

All required checks are now green on d0434b3f1, including every platform test shard and the static gates. The opt-in end-to-end jobs remain skipped by their existing gates. This is ready for maintainer review and merge.

@ranxianglei ranxianglei left a comment

Copy link
Copy Markdown
Owner

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

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

APPROVE(第七次复审)— 新 head d0434b3 仍是纯 master 同步合并(merge(1b8594f, 04ecbb4)),指纹修复与我首次批准的 d9383cf 逐字节相同(src/instance.ts 的 entryScriptFingerprint 与新测试在全部七个 head 间未变)。真实 PR footprint(相对并入的 master 父提交 04ecbb4)仍是 3 个文件:CHANGELOG +1(#5354 条目)/ src/instance.ts 21+5−(仅指纹函数段)/ 新测试 48,不碰 kernel/、不碰 pi-subagents/、不扩配置面。CI 当前 failed=0(success=18 / skipped=11 / in_progress=1)。无可顺手修的小问题、无开放尾巴。结论与前几轮一致,继续批准。

说明(透明起见):跨 head 的朴素 diff 会看到 atomicWriteJson 由私有变 export 并加「#5955 wedged-holder ledger」注释——那是本次并入的 master 自身改动,不是本 PR 的;已用真实 footprint(对 master 父提交)核对,本 PR 对 instance.ts 的贡献只有指纹函数那段。

八问(针对本 PR 实质改动,代码七轮未变):

  1. 是什么问题:omp 插件跑在长驻宿主里,宿主进程缓存入口脚本指纹且永不刷新;原地 npm i -g 后宿主仍拿首版构建的 sha256,经 BILI_SPAWN_FINGERPRINT 传给每个重拉代理。
  2. 什么原因:entryScriptFingerprint 用「解析后路径」当缓存键、命中即返回旧值,隐含「同路径=同内容」;npm 原地更新让同一路径指向新字节(rename-over 换 inode / 就地改写换 size+mtime+ctime)。根因是缓存键没反映文件真实身份。
  3. 影响面与严重度:omp 车道(长驻宿主+自动重拉)。能用但 lane 共享失效——更新后每个新构建各占一端口、互相接不上,需重启宿主绕行;非数据丢失/非缓存泄露,p2。
  4. 怎么修的:缓存键由「路径」换成「文件身份串 dev:ino:size:mtimeMs:ctimeMs」,一致才复用旧摘要、不一致重读重算;stat 失败(删/不可读)删条目并返回 undefined,保住 #1225「读不到→绝不 attach」。两处调用点 src/launcher.ts:4390(findLiveAttachableInstance)、:5058(主启动 attach-vs-spawn)均受益。
  5. 回归风险:①未变文件多一次 statSync(不重哈希 ~5.7MB dist),调用点在启动/attach 决策而非每请求热路径,开销可忽略;②删除后返回 undefined 而非失效旧摘要,比原行为更正确;③已知边界(作者已披露):同一 fs 时间戳 tick 内等大小就地改写沿用旧摘要至下次变更,npm 不这么写,可接受。无 fail-fast→吞错、无语义漂移。
  6. 缓存:不涉及——改的是实例 attach 的代码指纹判定,不动压缩缓存/重试/增量游标/去重键。
  7. 设计方向:不变,仍是内容身份门,只是让缓存键如实反映文件身份。
  8. 新配置:无新 env/config/默认值。

测试:tests/issue5354-fingerprint-follows-the-file.test.ts 三 case(rename-over/就地改写/删除)断言精确期望(sha256(buildB)/sha256(newer)/undefined);老逻辑第二次调用无条件返回首份缓存摘要,三断言在 master 必红——真回归钉。七轮合并均未触碰该测试与修复逻辑。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) 第七轮复审完成,结论:APPROVE(GitHub PR #5355,head d0434b3)。

新 head 是纯 master 同步合并(merge(1b8594f, 04ecbb4)),指纹修复与首次批准的 d9383cf 逐字节相同;真实 footprint 仍只有 3 个文件(CHANGELOG +1 / src/instance.ts 21+5− 仅指纹函数段 / 新测试 48),不碰 kernel/、不扩配置面。CI failed=0(success=18 / skipped=11 / in_progress=1)。无可顺手修的小问题、无开放尾巴。已在 GitHub 再次 APPROVE(八问见 PR review)。

透明说明:跨 head 朴素 diff 里的 atomicWriteJson 私有→export(#5955 wedged-holder ledger)是本次并入的 master 自身改动,非本 PR;已用真实 footprint 核对确认。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

CI update: 29/30 green — the two slow lanes (mysql-contract, docs-sync) landed green. Only 1 check still running: test (ubuntu-latest, 22, 2). 0 failed. The PR remains mergeable (conflict cleared); I'll confirm full green when the last shard lands.

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b) Thanks @Maziar123. Review is complete — I read the full diff end-to-end and posted APPROVE on d0434b3f1 (review id 5973344796). The fingerprint fix (identity-keyed entryScriptFingerprint cache + the 3-case regression test) is correct, minimal, and unchanged across all seven heads; the true footprint is 3 files, no kernel/, no config-surface change. CI matches your note: the mirror cache shows failed=0 (platform test shards + static gates green, opt-in e2e jobs skipped by their existing gates).

One thing I can't do: per repo rules (AGENTS.md) PR merges are human-only, so I'm not able to merge it myself. The merge decision belongs to the owner (@ranxianglei). Everything on the review side is done and green, so PR #5355 is ready for the owner to merge whenever.

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

CI fully green — the PR is ready to merge. All 30 checks pass (0 failed, 0 pending) on the new head d0434b3f10. The conflict is cleared (the author resolved it via option A — merged current master, kept both CHANGELOG.md entries), the #5354 fix is intact, and CI is green.

Note: the mirror's mergeable_state still reads blocked even with all checks green — that's the mirror's async recompute lagging (the conflict is genuinely cleared; origin/master is an ancestor of the head). Ready for your merge whenever.

结构化八问(中文·本轮无代码改动)

  1. 是什么问题:PR [PR] fix(instance): re-hash the entry script when the file changes (#5354) #5355 此前被「外部 fork 写权限缺失 + CHANGELOG.md 冲突」阻塞;本轮作者已自行解决,CI 已全绿。
  2. 什么原因导致的:作者把当前 master 合入分支、按「两条都留」解唯一的 CHANGELOG.md 冲突后推送(选项 A),冲突清除、CI 在新 head 上跑绿。
  3. 影响范围与严重程度:仅该 PR 合入流程;不影响用户/生产。冲突已清、修复完整、CI 全绿。
  4. 怎么修复的:非我修复——作者自行合并 master 并解冲突;我只做只读核验(新 head 为规范 merge、master 已是其祖先、Long-lived omp host keeps a stale entry-script fingerprint after an in-place update, so respawned lane proxies cannot attach to each other and the lane port walks #5354 修复与测试完整、CI 全绿)。
  5. 是否有回归问题:不涉及——本轮我未改任何代码;修复本身与已批的 d9383cf 逐字节一致。
  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。

判定:冲突已清、CI 全绿、PR 现可合;我未改任何代码。合并动作留给你。

@ranxianglei

Copy link
Copy Markdown
Owner

🤖 Powered by ework · qwen3.8-27b

[bot] 🏷 (ai: vllm-qwen/qwen3.8-27b)

Re-verified live — the conflict is back (master moved), still blocked on fork write access.

  • Master moved 04ecbb42dd → 98fa9575; fork head unchanged (d0434b3f10).
  • Real merge test: the only conflict is CHANGELOG.md again (keep-both); all other files auto-merge cleanly.
  • CI fully green (30/30, 0 failed, 0 pending) on the current head.
  • Fork still push:false (re-verified live) — I cannot push the resolution to the fork.

The decision request (recommend B: grant ranxianglei write access to the fork so I can push the keep-both resolution) is still pending with the owner. Holding — no code changed this run.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

2 participants