[{"data":1,"prerenderedAt":1793},["ShallowReactive",2],{"\u002F2026-06-16-skill":3,"\u002F2026-06-16-skill-rel":530},{"id":4,"title":5,"body":6,"column":513,"date":514,"description":12,"extension":515,"hero_image":516,"meta":517,"navigation":518,"path":519,"seo":520,"series_id":516,"severity":516,"stem":521,"summary":522,"tags":523,"__hash__":529},"posts\u002F2026-06-16-把高风险发版固化成skill.md","不让 AI 碰发版，不等于靠人记住每一步",{"type":7,"value":8,"toc":507},"minimark",[9,13,18,21,24,46,57,60,64,67,70,209,217,220,227,257,260,267,271,274,277,284,354,365,428,431,446,453,456,473,480,483,486,497,500,503],[10,11,12],"p",{},"发版从来不应该由人工记忆驱动。5 月那一轮 OTA 系列故障（FA-005 到 FA-008）和跨版本步进导致的漏 DLL 事故反复说明了这一点：发版是不可逆的、影响全部用户的、细节繁琐的操作，任何一个环节遗漏或顺序错误就导致客户端卡死、文件不一致、激活页死循环。我现在不是让 AI 执行发版——那是禁地——而是把 5 个已经踏过所有坑的发版流程全部固化成 skill，让执行者用最小脑力成本跑完完整链路，而不是依赖文档翻译和记忆。",[14,15,17],"h2",{"id":16},"触发词不该触发时绝对不触发","触发词：不该触发时绝对不触发",[10,19,20],{},"一个 skill 最危险的时刻是被错误调用。\"现在是什么版本\"听起来像发版，\"改一下版本号\"听起来像发版，但它们完全不是。",[10,22,23],{},"Wakou 发版 skill 的触发词清单（必读触发词段）就是为了卡死这条线：",[25,26,27,35],"ul",{},[28,29,30,31],"li",{},"触发：",[32,33,34],"code",{},"发版 \u002F 发布 \u002F release \u002F 打个新版本 \u002F OTA \u002F OTA 死循环 \u002F 版本 bump \u002F \u002Fwakou-release",[28,36,37,38,41,42,45],{},"不触发：",[32,39,40],{},"改一下 package.json 版本号","（用户只想改文件），",[32,43,44],{},"看看现在是什么版本","（只查询）",[10,47,48,49,52,53,56],{},"反面教材我都经历过。改一下配置文件、改一个环境变量、查一个版本号，因为没有明确的触发词界限，最后莫名其妙走到了发版脚本的某一步。Connector OTA skill 同样列明了不触发场景：",[32,50,51],{},"\"Connector 现在是什么版本\"(只查询)"," ",[32,54,55],{},"\"改 tauri.conf 版本号\"(用户只想改文件)"," — 这些都不触发，即使命令里含了 version、release 这类关键字。",[10,58,59],{},"这条最反直觉但最重要：不该触发时被触发，等于主动送一次误操作机会给自动化流程。",[14,61,63],{"id":62},"步骤不可跳过验证链条硬卡","步骤不可跳过：验证链条硬卡",[10,65,66],{},"一旦 skill 确定要执行，每一步都必须有验证关卡，前一步没通过绝不进下一步。我用 Wakou 全自动发版流程作例子。",[10,68,69],{},"预检（Step 1）是整个流程的闸门。跑这些 Bash 检查：",[71,72,77],"pre",{"className":73,"code":74,"language":75,"meta":76,"style":76},"language-bash shiki shiki-themes github-light github-dark","cd \u002FUsers\u002Fxingye\u002FAi\u002Fnew-openclaw\ngit status --short                                          # 工作区干净\ntest -x \u002Fopt\u002Fhomebrew\u002Fbin\u002Fsshpass                           # ssh 工具\ntest -x \u002Fopt\u002Fhomebrew\u002Fbin\u002Fminisign                          # 签名工具\ntest -x \u003C签名 CLI 路径>                                      # 内部签名工具就位\ntest -f \u003C发版签名私钥路径>                                   # 私钥就位\ntest -f \u003CCDN 上传凭据路径>                                   # CDN 凭据就位\n","bash","",[32,78,79,92,109,124,137,167,187],{"__ignoreMap":76},[80,81,84,88],"span",{"class":82,"line":83},"line",1,[80,85,87],{"class":86},"sj4cs","cd",[80,89,91],{"class":90},"sZZnC"," \u002FUsers\u002Fxingye\u002FAi\u002Fnew-openclaw\n",[80,93,95,99,102,105],{"class":82,"line":94},2,[80,96,98],{"class":97},"sScJk","git",[80,100,101],{"class":90}," status",[80,103,104],{"class":86}," --short",[80,106,108],{"class":107},"sJ8bj","                                          # 工作区干净\n",[80,110,112,115,118,121],{"class":82,"line":111},3,[80,113,114],{"class":86},"test",[80,116,117],{"class":86}," -x",[80,119,120],{"class":90}," \u002Fopt\u002Fhomebrew\u002Fbin\u002Fsshpass",[80,122,123],{"class":107},"                           # ssh 工具\n",[80,125,127,129,131,134],{"class":82,"line":126},4,[80,128,114],{"class":86},[80,130,117],{"class":86},[80,132,133],{"class":90}," \u002Fopt\u002Fhomebrew\u002Fbin\u002Fminisign",[80,135,136],{"class":107},"                          # 签名工具\n",[80,138,140,142,144,148,151,154,157,161,164],{"class":82,"line":139},5,[80,141,114],{"class":86},[80,143,117],{"class":86},[80,145,147],{"class":146},"szBVR"," \u003C",[80,149,150],{"class":90},"签名",[80,152,153],{"class":90}," CLI",[80,155,156],{"class":90}," 路",[80,158,160],{"class":159},"sVt8B","径",[80,162,163],{"class":146},">",[80,165,166],{"class":107},"                                      # 内部签名工具就位\n",[80,168,170,172,175,177,180,182,184],{"class":82,"line":169},6,[80,171,114],{"class":86},[80,173,174],{"class":86}," -f",[80,176,147],{"class":146},[80,178,179],{"class":90},"发版签名私钥路",[80,181,160],{"class":159},[80,183,163],{"class":146},[80,185,186],{"class":107},"                                   # 私钥就位\n",[80,188,190,192,194,196,199,202,204,206],{"class":82,"line":189},7,[80,191,114],{"class":86},[80,193,174],{"class":86},[80,195,147],{"class":146},[80,197,198],{"class":90},"CDN",[80,200,201],{"class":90}," 上传凭据路",[80,203,160],{"class":159},[80,205,163],{"class":146},[80,207,208],{"class":107},"                                   # CDN 凭据就位\n",[10,210,211,212,216],{},"任何一条 fail，流程停止并告诉用户修什么。如果 git 工作区脏，不是自动 stash，而是问用户：工作区有未提交改动，要先 commit 还是 stash？发版会自动 bump 三个版本文件并 commit。这样做的目的很明确——让用户",[213,214,215],"strong",{},"清晰意识到","发版会改动工作区，而不是闭着眼睛走入自动流程。",[10,218,219],{},"收三个密码（Step 2）也有安全约定。不要把密码记到内存或 commit 进文件，只读 shell 的 export。如果用户在过去几分钟内说过这些密码（在 conversation context 里能直接拿到），可以直接传 env 跑；否则必须用户主动导入。",[10,221,222,223,226],{},"跑 release.sh（Step 3）用 ",[32,224,225],{},"run_in_background=true"," 避免 5-15 分钟阻塞主线，把输出 tee 到日志。然后（Step 4）用 Monitor 工具监控 10 个 stage 的进度：",[228,229,230,233,236,239,242,245,248,251,254],"ol",{},[28,231,232],{},"preflight",[28,234,235],{},"detect FROM_VERSION",[28,237,238],{},"tar + scp source to Win build host",[28,240,241],{},"build on Win",[28,243,244],{},"stage + diff-pack OTA full.zip",[28,246,247],{},"minisign full.zip + release-manifest.json",[28,249,250],{},"upload to CDN",[28,252,253],{},"admin: create release + PATCH rollout",[28,255,256],{},"git tag + commit version bump",[10,258,259],{},"每个 stage 完成给用户报一句。出现 ✗ FAILED 立刻 surface 错误。没有\"继续试试能不能救\"，就是暴露失败。",[10,261,262,263,266],{},"这整套链条的目的是",[213,264,265],{},"让人工审核点布满全流程","。不是机器自动发版，而是 AI 陪着用户一步步走完发版，每步都看得清、验得出。",[14,268,270],{"id":269},"故障处置内联把坑写在流程里","故障处置内联：把坑写在流程里",[10,272,273],{},"OTA 系列故障的教训不应该沉在 bug 复盘里。我把它们直接铺进 skill 的执行逻辑。",[10,275,276],{},"Wakou 发版 skill 有整整一章叫「🚨 阻塞性 bug 处置 SOP」。列出了从 G19 到 G31 的 11 个已知 bug：",[10,278,279,280,283],{},"G19 是 OTA modal 反复弹。原因是 state base 残留 stale ",[32,281,282],{},"download.zip.minisig","，当时的修复是：",[71,285,287],{"className":73,"code":286,"language":75,"meta":76,"style":76},"# 预检 sanity#3: data\u002F 不删\nsanity#3 强制阻断 deletes 命中 data\u002F。某次发版失败提示 sanity#3 一定是 build 步骤搞坏了 data\u002F 目录的同步，不要绕过 sanity check 去强发，先 debug 为什么 data\u002F 出现在 deletes 列表。\n",[32,288,289,294],{"__ignoreMap":76},[80,290,291],{"class":82,"line":83},[80,292,293],{"class":107},"# 预检 sanity#3: data\u002F 不删\n",[80,295,296,299,302,305,308,311,314,317,320,323,326,329,332,335,338,341,344,346,349,351],{"class":82,"line":94},[80,297,298],{"class":97},"sanity#3",[80,300,301],{"class":90}," 强制阻断",[80,303,304],{"class":90}," deletes",[80,306,307],{"class":90}," 命中",[80,309,310],{"class":90}," data\u002F。某次发版失败提示",[80,312,313],{"class":90}," sanity#3",[80,315,316],{"class":90}," 一定是",[80,318,319],{"class":90}," build",[80,321,322],{"class":90}," 步骤搞坏了",[80,324,325],{"class":90}," data\u002F",[80,327,328],{"class":90}," 目录的同步，不要绕过",[80,330,331],{"class":90}," sanity",[80,333,334],{"class":90}," check",[80,336,337],{"class":90}," 去强发，先",[80,339,340],{"class":90}," debug",[80,342,343],{"class":90}," 为什么",[80,345,325],{"class":90},[80,347,348],{"class":90}," 出现在",[80,350,304],{"class":90},[80,352,353],{"class":90}," 列表。\n",[10,355,356,357,360,361,364],{},"G25 是跨用户机器后 chat 权限错。根因是 ",[32,358,359],{},"sessions\u002Faudit"," 含绝对路径，guardian sync 扩散。这不是\"修了以后的故事\"，而是",[213,362,363],{},"发版前防御 checklist"," 的一部分：",[71,366,368],{"className":73,"code":367,"language":75,"meta":76,"style":76},"# 发版前防御 checklist (每次发版前在 build host 做完整新用户回归)\n# 3. dist 内 grep C:\\Users\\ 应该 0 命中（防 G25 类绝对路径泄露）\nssh CHANCHING@\u003C构建机> 'powershell -Command \"\n  Get-ChildItem D:\\wakou-build\\dist-X.Y.Z\\WakouPanelPortable -Recurse -File -Include *.json,*.jsonl |\n    Where-Object { $_.Length -lt 200KB } |\n    Select-String C:\\\\Users\\\\Administrator -SimpleMatch -List |\n    Select-Object FullName\n\"'\n",[32,369,370,375,380,402,407,412,417,422],{"__ignoreMap":76},[80,371,372],{"class":82,"line":83},[80,373,374],{"class":107},"# 发版前防御 checklist (每次发版前在 build host 做完整新用户回归)\n",[80,376,377],{"class":82,"line":94},[80,378,379],{"class":107},"# 3. dist 内 grep C:\\Users\\ 应该 0 命中（防 G25 类绝对路径泄露）\n",[80,381,382,385,388,391,394,397,399],{"class":82,"line":111},[80,383,384],{"class":97},"ssh",[80,386,387],{"class":90}," CHANCHING@",[80,389,390],{"class":146},"\u003C",[80,392,393],{"class":90},"构建",[80,395,396],{"class":159},"机",[80,398,163],{"class":146},[80,400,401],{"class":90}," 'powershell -Command \"\n",[80,403,404],{"class":82,"line":126},[80,405,406],{"class":90},"  Get-ChildItem D:\\wakou-build\\dist-X.Y.Z\\WakouPanelPortable -Recurse -File -Include *.json,*.jsonl |\n",[80,408,409],{"class":82,"line":139},[80,410,411],{"class":90},"    Where-Object { $_.Length -lt 200KB } |\n",[80,413,414],{"class":82,"line":169},[80,415,416],{"class":90},"    Select-String C:\\\\Users\\\\Administrator -SimpleMatch -List |\n",[80,418,419],{"class":82,"line":189},[80,420,421],{"class":90},"    Select-Object FullName\n",[80,423,425],{"class":82,"line":424},8,[80,426,427],{"class":90},"\"'\n",[10,429,430],{},"任何一项 fail → 不要 GA，修了再 bump 一个 patch 版本。",[10,432,433,434,437,438,441,442,445],{},"Connector OTA skill 的故障字典（§4）则是一张病症与病因的对照表。\"signature verification failed\" 对应的处置是检查 ",[32,435,436],{},"TAURI_SIGNING_PRIVATE_KEY"," 是不是 .sec 文件",[213,439,440],{},"base64 编码后","的内容。\"OTA 不跨级但用户期望直跳\" 的处置是调整 ",[32,443,444],{},"findNextVersion"," 逻辑或在灰度模式放上一个 manifest。",[10,447,448,449,452],{},"关键在",[213,450,451],{},"处置是写死在 skill 里","，不是留在事后复盘让人去翻。OpenClaw Runtime 发版 skill 的故障字典甚至铺了 10 条常见陷阱：BuildKit context 缓存不清导致改动没编进去、gwbridge IP 池满导致 service 无法起、孤儿 docker-proxy 占着 host port、磁盘满导致 openclaw.json 被写成 0 字节。",[10,454,455],{},"每一条都带着：\"现象是什么、根因是什么、怎么修\"。拿 BuildKit 缓存那条：",[457,458,459],"blockquote",{},[10,460,461,462,465,466,469,470,472],{},"症状：rsync\u002FSFTP 把新源推上 prod repo，源文件 grep 确认是新的，但 docker build 出来的镜像里还是旧 bundle。根因：",[32,463,464],{},"--no-cache-filter"," 不可靠，BuildKit 的 build context 层仍可能命中缓存。修法：clawpanel 源改动后，必须用整体 ",[32,467,468],{},"--no-cache","，不要用 ",[32,471,464],{},"。代价：单次 ~20-30 分钟。但这是唯一能确保改动真编进去的方式。",[10,474,475,476,479],{},"这不是\"高级技巧\"或\"经验之谈\"。这是",[213,477,478],{},"发版的必读须知","，写进 skill 里才能确保每次都不遗漏。",[14,481,482],{"id":482},"为什么这样做有效",[10,484,485],{},"高风险操作的安全性来自流程的确定性，不来自执行者的谨慎。人会忘、会打错、会跳步。但固化的流程不会：",[25,487,488,491,494],{},[28,489,490],{},"触发词明确化杀死了\"我不是故意调用发版\"的那类误操作。",[28,492,493],{},"步骤验证链条把每一步的前置条件和验收标准写成 Bash 检查，没法绕过。",[28,495,496],{},"故障处置内联把一个多月里积累的坑提前封死在流程里，新一轮发版不会重蹈覆辙。",[10,498,499],{},"5 个 skill 现在覆盖了：Wakou panel OTA、Connector 桌面端 OTA、OpenClaw runtime 容器全量发布、Codex 助手静默更新、加上客服远程调试的风险边界。它们没有一个是让 AI 执行发版的——都是让 AI 陪着用户，一步步走完一个铺满护栏的流程。",[10,501,502],{},"这是把\"不要让 AI 碰发版\"这条禁区，变成了\"AI 执行发版的框架，确保每一步都验证、每个坑都已知、每种故障都有救法\"的可行路径。",[504,505,506],"style",{},"html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sJ8bj, html code.shiki .sJ8bj{--shiki-default:#6A737D;--shiki-dark:#6A737D}html pre.shiki code .szBVR, html code.shiki .szBVR{--shiki-default:#D73A49;--shiki-dark:#F97583}html pre.shiki code .sVt8B, html code.shiki .sVt8B{--shiki-default:#24292E;--shiki-dark:#E1E4E8}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"title":76,"searchDepth":94,"depth":94,"links":508},[509,510,511,512],{"id":16,"depth":94,"text":17},{"id":62,"depth":94,"text":63},{"id":269,"depth":94,"text":270},{"id":482,"depth":94,"text":482},"交付与更新","2026-06-16","md",null,{},true,"\u002F2026-06-16-skill",{"title":5,"description":12},"2026-06-16-把高风险发版固化成skill","发版是明确不交给 AI 自主执行的操作。通过触发词精确化、步骤验证链条、故障处置内联，把流程固化到 skill，让高风险操作失败概率趋零。",[524,525,526,527,528],"发版流程","OTA","自动化","Tauri","Docker Swarm","WbjJuf61wSEx4d5Iov2KhPdG1N6hik51RAl8rAGgug8",[531,869,1223],{"id":4,"title":5,"body":532,"column":513,"date":514,"description":12,"extension":515,"hero_image":516,"meta":866,"navigation":518,"path":519,"seo":867,"series_id":516,"severity":516,"stem":521,"summary":522,"tags":868,"__hash__":529},{"type":7,"value":533,"toc":860},[534,536,538,540,542,554,560,562,564,566,568,662,666,668,672,692,694,698,700,702,704,708,758,764,812,814,822,826,828,838,842,844,846,854,856,858],[10,535,12],{},[14,537,17],{"id":16},[10,539,20],{},[10,541,23],{},[25,543,544,548],{},[28,545,30,546],{},[32,547,34],{},[28,549,37,550,41,552,45],{},[32,551,40],{},[32,553,44],{},[10,555,48,556,52,558,56],{},[32,557,51],{},[32,559,55],{},[10,561,59],{},[14,563,63],{"id":62},[10,565,66],{},[10,567,69],{},[71,569,570],{"className":73,"code":74,"language":75,"meta":76,"style":76},[32,571,572,578,588,598,608,628,644],{"__ignoreMap":76},[80,573,574,576],{"class":82,"line":83},[80,575,87],{"class":86},[80,577,91],{"class":90},[80,579,580,582,584,586],{"class":82,"line":94},[80,581,98],{"class":97},[80,583,101],{"class":90},[80,585,104],{"class":86},[80,587,108],{"class":107},[80,589,590,592,594,596],{"class":82,"line":111},[80,591,114],{"class":86},[80,593,117],{"class":86},[80,595,120],{"class":90},[80,597,123],{"class":107},[80,599,600,602,604,606],{"class":82,"line":126},[80,601,114],{"class":86},[80,603,117],{"class":86},[80,605,133],{"class":90},[80,607,136],{"class":107},[80,609,610,612,614,616,618,620,622,624,626],{"class":82,"line":139},[80,611,114],{"class":86},[80,613,117],{"class":86},[80,615,147],{"class":146},[80,617,150],{"class":90},[80,619,153],{"class":90},[80,621,156],{"class":90},[80,623,160],{"class":159},[80,625,163],{"class":146},[80,627,166],{"class":107},[80,629,630,632,634,636,638,640,642],{"class":82,"line":169},[80,631,114],{"class":86},[80,633,174],{"class":86},[80,635,147],{"class":146},[80,637,179],{"class":90},[80,639,160],{"class":159},[80,641,163],{"class":146},[80,643,186],{"class":107},[80,645,646,648,650,652,654,656,658,660],{"class":82,"line":189},[80,647,114],{"class":86},[80,649,174],{"class":86},[80,651,147],{"class":146},[80,653,198],{"class":90},[80,655,201],{"class":90},[80,657,160],{"class":159},[80,659,163],{"class":146},[80,661,208],{"class":107},[10,663,211,664,216],{},[213,665,215],{},[10,667,219],{},[10,669,222,670,226],{},[32,671,225],{},[228,673,674,676,678,680,682,684,686,688,690],{},[28,675,232],{},[28,677,235],{},[28,679,238],{},[28,681,241],{},[28,683,244],{},[28,685,247],{},[28,687,250],{},[28,689,253],{},[28,691,256],{},[10,693,259],{},[10,695,262,696,266],{},[213,697,265],{},[14,699,270],{"id":269},[10,701,273],{},[10,703,276],{},[10,705,279,706,283],{},[32,707,282],{},[71,709,710],{"className":73,"code":286,"language":75,"meta":76,"style":76},[32,711,712,716],{"__ignoreMap":76},[80,713,714],{"class":82,"line":83},[80,715,293],{"class":107},[80,717,718,720,722,724,726,728,730,732,734,736,738,740,742,744,746,748,750,752,754,756],{"class":82,"line":94},[80,719,298],{"class":97},[80,721,301],{"class":90},[80,723,304],{"class":90},[80,725,307],{"class":90},[80,727,310],{"class":90},[80,729,313],{"class":90},[80,731,316],{"class":90},[80,733,319],{"class":90},[80,735,322],{"class":90},[80,737,325],{"class":90},[80,739,328],{"class":90},[80,741,331],{"class":90},[80,743,334],{"class":90},[80,745,337],{"class":90},[80,747,340],{"class":90},[80,749,343],{"class":90},[80,751,325],{"class":90},[80,753,348],{"class":90},[80,755,304],{"class":90},[80,757,353],{"class":90},[10,759,356,760,360,762,364],{},[32,761,359],{},[213,763,363],{},[71,765,766],{"className":73,"code":367,"language":75,"meta":76,"style":76},[32,767,768,772,776,792,796,800,804,808],{"__ignoreMap":76},[80,769,770],{"class":82,"line":83},[80,771,374],{"class":107},[80,773,774],{"class":82,"line":94},[80,775,379],{"class":107},[80,777,778,780,782,784,786,788,790],{"class":82,"line":111},[80,779,384],{"class":97},[80,781,387],{"class":90},[80,783,390],{"class":146},[80,785,393],{"class":90},[80,787,396],{"class":159},[80,789,163],{"class":146},[80,791,401],{"class":90},[80,793,794],{"class":82,"line":126},[80,795,406],{"class":90},[80,797,798],{"class":82,"line":139},[80,799,411],{"class":90},[80,801,802],{"class":82,"line":169},[80,803,416],{"class":90},[80,805,806],{"class":82,"line":189},[80,807,421],{"class":90},[80,809,810],{"class":82,"line":424},[80,811,427],{"class":90},[10,813,430],{},[10,815,433,816,437,818,441,820,445],{},[32,817,436],{},[213,819,440],{},[32,821,444],{},[10,823,448,824,452],{},[213,825,451],{},[10,827,455],{},[457,829,830],{},[10,831,461,832,465,834,469,836,472],{},[32,833,464],{},[32,835,468],{},[32,837,464],{},[10,839,475,840,479],{},[213,841,478],{},[14,843,482],{"id":482},[10,845,485],{},[25,847,848,850,852],{},[28,849,490],{},[28,851,493],{},[28,853,496],{},[10,855,499],{},[10,857,502],{},[504,859,506],{},{"title":76,"searchDepth":94,"depth":94,"links":861},[862,863,864,865],{"id":16,"depth":94,"text":17},{"id":62,"depth":94,"text":63},{"id":269,"depth":94,"text":270},{"id":482,"depth":94,"text":482},{},{"title":5,"description":12},[524,525,526,527,528],{"id":870,"title":871,"body":872,"column":513,"date":1211,"description":876,"extension":515,"hero_image":516,"meta":1212,"navigation":518,"path":1213,"seo":1214,"series_id":516,"severity":516,"stem":1215,"summary":1216,"tags":1217,"__hash__":1222},"posts\u002F2026-06-15-Connector两次发版复盘.md","两天连发两版，第二版是为了补第一版",{"type":7,"value":873,"toc":1201},[874,877,881,884,891,902,905,931,941,946,949,954,961,976,987,992,995,1010,1021,1026,1033,1036,1041,1056,1059,1063,1070,1073,1084,1088,1091,1094,1097,1102,1105,1110,1121,1126,1129,1133,1140,1146,1149,1168,1171,1178,1181,1188,1198],[10,875,876],{},"桌面端连接组件在一周内连发两版。1.0.10 修了 WebSocket 端点未被消费导致的 17 分钟周期断线；1.0.11 处理紧随其后暴露的僵尸令牌问题与发版脚本漏洞。这两次发版既是问题的递进修复，也是工程实践上从被动补救走向主动设计的转折。",[14,878,880],{"id":879},"_101017-分钟的周期","1.0.10：17 分钟的周期",[10,882,883],{},"用户反馈从五月中旬开始稳定出现：Connector 约每 17 分钟自动断线一次，断后立即重连。现象本身明确，但根因指向了一个被忽视的设计缺陷。",[10,885,886,887,890],{},"服务端在五月下旬已切换 WebSocket 入口地址。原来的端点经 CDN 反代，该 CDN 的 idle 超时配置在 4-9 分钟，时间不固定。一旦断开，客户端用相同的令牌重连，再次击中 idle 超时，形成周期。这本该不是问题——服务端在 ",[32,888,889],{},"\u002Fauto-pair"," 响应里已返回新的 WebSocket 地址，客户端只需消费这个字段切换端点即可。",[10,892,893,894,897,898,901],{},"但客户端代码从未读取过这个字段。ws_runner 始终硬拼 base_url 生成连接地址，App.tsx 只提取了 ",[32,895,896],{},"device_token","，对 ",[32,899,900],{},"ws_endpoint"," 视而不见。结果是服务端返了新地址，客户端装作没看见，继续走旧路径。旧路径经 CDN，CDN idle 触发，17 分钟周期形成。",[10,903,904],{},"修复需要四处联动：",[25,906,907,913,922,925],{},[28,908,909,910,912],{},"在 AppState 加一个 ",[32,911,900],{}," 字段，用 RwLock 包装以便运行时读取",[28,914,915,918,919,921],{},[32,916,917],{},"set_device_token"," 命令新增参数，收到服务端返的 ",[32,920,900],{}," 时校验并持久化到 config.json",[28,923,924],{},"ws_runner 启动与每次重连前都从 RwLock 读取最新地址，不再拼装",[28,926,927,928,930],{},"App.tsx 消费 ",[32,929,900],{}," 字段，传给 invoke",[10,932,933,934,937,938,940],{},"兼容性处理也做了：env 变量 ",[32,935,936],{},"CPA_WS_ENDPOINT"," 可强制指定（灰度用），服务端没返 ",[32,939,900],{}," 时退化到旧路径拼装，config.json 校验失败时保留现有值不覆盖。",[942,943,945],"h3",{"id":944},"_1010-发版中的五个坑","1.0.10 发版中的五个坑",[10,947,948],{},"代码改好了，但发版流程成了另一个故事。",[10,950,951],{},[213,952,953],{},"坑 1：假 1.0.10",[10,955,956,957,960],{},"发版脚本在 Windows 构建机上只 scp 了 tauri.conf.json 这一个文件，没有同步整个仓源码。构建机的 git HEAD 还停在 1.0.8 的 commit，Cargo.toml workspace.package.version 还是 ",[32,958,959],{},"0.1.0","。结果是 cargo build 出来的二进制是 1.0.8 代码，但 NSIS 打包时用了 tauri.conf.json 里的\"1.0.10\"字符串当文件名——实际上是 1.0.8 代码套上 1.0.10 的皮。",[10,962,963,964,967,968,971,972,975],{},"应急是用 ",[32,965,966],{},"git ls-files | tar over ssh"," 把真 1.0.10 源码同步过去，清掉 macOS metadata 的 ",[32,969,970],{},"._*"," 文件（tar 直接打了这些，Windows 上 tauri-build 把 ",[32,973,974],{},"capabilities\u002F._default.json"," 当 JSON 解析失败），重新 build。",[10,977,978,979,982,983,986],{},"1.0.11 的发版脚本改成了完整源码同步加 ",[32,980,981],{},"tar --exclude='._*'"," 和 Win 端 ",[32,984,985],{},"Remove-Item '._*'"," 双保险。",[10,988,989],{},[213,990,991],{},"坑 2：ssh exit code 谜团",[10,993,994],{},"phase B 跑完整发版脚本时，step 3（Win build）完成了——NSIS bundle 和 minisign 签名都成功写出来了——但 ssh 仍然返 exit code 1，导致脚本静默退出，step 4-10 全跳过。",[10,996,997,998,1001,1002,1005,1006,1009],{},"当时推测是 ",[32,999,1000],{},"ssh ... | tail -10"," 触发 SIGPIPE，但这个推测被现场查证打脸了：tail 会完整消费 stdin，不会向上游发 SIGPIPE。真根因当时没查清楚，可能是 PowerShell 某个 cmdlet 设了非零 ",[32,1003,1004],{},"$LASTEXITCODE","，也可能是 ssh 在大量 escape sequence 输出时本身返非零，也可能是脚本框架的 ",[32,1007,1008],{},"pipefail"," 组合出了问题。",[10,1011,1012,1013,1016,1017,1020],{},"1.0.11 用了一个兜底方案：脚本监测 build 成功的标志——",[32,1014,1015],{},"Finished 1 bundle at:"," 和 ",[32,1018,1019],{},"Finished 1 updater signature at:"," 双命中才认为成功，忽略 ssh exit code。这个兜底已经把\"ssh exit code 非零\"这条问题路径变成了已知但可控的状态：即使 ssh 返错，只要 marker 出现了，就继续执行后续步骤。1.0.11 发版时 log 记录了这次抗性：「ssh exit=1 非零，但 build marker 命中，视为成功」。",[10,1022,1023],{},[213,1024,1025],{},"坑 3：gitlink 边界",[10,1027,1028,1029,1032],{},"发版脚本 step 10 尝试 ",[32,1030,1031],{},"git add cpa-connector\u002F...\u002Ftauri.conf.json","，但因为 cpa-connector 是 gitlink（160000 mode）不是真 submodule，git 会静默跳过这条跨边界的路径。结果 tauri.conf.json 没被 add 进 root 仓的 commit。",[10,1034,1035],{},"这里需要在 cpa-connector 子仓内完成一次独立 commit（包含 Cargo.toml、Cargo.lock、package.json、tauri.conf.json），然后回到 root 仓 add 整个 gitlink 指针。1.0.11 的发版脚本改成了这个流程。",[10,1037,1038],{},[213,1039,1040],{},"坑 4 和 5：参数漏传与版本号不同步",[10,1042,1043,1044,1047,1048,1051,1052,1055],{},"Task 9 重构了前端的 ",[32,1045,1046],{},"tryAutoPairOnce"," 这个 helper 时，body 从 ",[32,1049,1050],{},"{ hostname, os, connector_version }"," 意外退化成了 ",[32,1053,1054],{},"{}","，导致服务端拿不到客户端版本号。同时 package.json 里的 version 字段还停留在 1.0.0，跟 Cargo.toml 的 1.0.9 完全脱节。",[10,1057,1058],{},"这两个都算是发版前的信息丢失，1.0.10 发版后才被指出。",[942,1060,1062],{"id":1061},"_1010-的验证与回滚","1.0.10 的验证与回滚",[10,1064,1065,1066,1069],{},"1.0.10 在 5 月 27 日上午 10:34 开始滚动更新。测试机验证了 23 分钟无断线（vs 之前每 17 分钟必断），服务端日志开始出现 ",[32,1067,1068],{},"host=\u003C直连 WS 域名>"," 的访问。4 小时窗口内，新 WS 入口命中 992 次，旧入口仍有 209566 次（因为 1.0.9 客户端的 OTA 是自然滚动，不是强推）。",[10,1071,1072],{},"但随后用户报了一个投诉，初看像是 1.0.10 的问题。经过排查发现是 openclaw-runtime 1.0.23 的一个独立 schema 问题，与 1.0.10 无关。这件事触发了一个 P0，但因为根因被快速定位到另一个模块，没有对 1.0.10 造成滚动阻断。",[10,1074,1075,1076,1079,1080,1083],{},"1.0.10 本身的回滚预案是改 ",[32,1077,1078],{},"apps\u002Fapi\u002Fsrc\u002Fconnector\u002Fupdates.ts"," 的 RELEASE_CHAIN 数组，删掉 ",[32,1081,1082],{},"'1.0.10'"," 这一项，再重部署服务端。已升级的客户端不会自动回退（Tauri updater 不支持降级），但新检查的 1.0.9 用户就不会继续升级了。这一设计决策埋下了一个伏笔：一旦版本链里出现过某个版本，再想把它从升级路径上彻底抹掉就很困难，因为已升级的用户成了\"污染源\"，他们可能带着该版本的各种遗留问题继续在线。这正是分阶段发版与步进升级设计要反复考量的权衡点。",[14,1085,1087],{"id":1086},"_1011两个小时内的闭环","1.0.11：两个小时内的闭环",[10,1089,1090],{},"1.0.10 发版后的第四个小时，某个客户端用失效的令牌反复击中 WebSocket 4401 拒绝，服务端 30 分钟内记录了 5638 次 reject，来自 2 个孤儿 device，token 前缀稳定但 DB 里查不到。",[10,1092,1093],{},"这是另一个被延期的问题。客户端收到 4401（token 失效）时没有自愈逻辑，只会 exponential backoff 后用同一个失效 token 再试一次，陷入\"僵尸状态\"——在线但永远连不上 WS。与其说这是 1.0.10 的新问题，不如说是 1.0.10 的发版暴露了原本就存在的设计缺陷。",[10,1095,1096],{},"1.0.11 同步推进了三个方向的修复：",[10,1098,1099],{},[213,1100,1101],{},"服务端黑名单",[10,1103,1104],{},"cpa-api 新增一个 TokenBlacklist 类，维护一个 LRU 缓存。收到 4401（token 失效）后，把这个 token 加进黑名单，5 分钟内的重试直接返 4401，不走 DB 查询。这样做的好处是降低 DB 压力，坏处是增加内存占用，但 5 分钟的 TTL 和 LRU 限容使得这个成本可控。",[10,1106,1107],{},[213,1108,1109],{},"客户端自愈",[10,1111,1112,1113,1116,1117,1120],{},"改造 ws.rs 的错误类型，让 4401 close code 能被单独识别为 ",[32,1114,1115],{},"WsError::TokenInvalid","。ws_runner 收到这个错误时，立即清空 in-memory 的 device_token（Mutex 设为 None），emit 一个 ",[32,1118,1119],{},"connector:\u002F\u002Ftoken-invalid"," 事件，不 sleep 直接进入下一轮 loop。frontend 端 listen 这个事件，触发 auto-pair 重新获取 token。这样一来，一次失效就能在 5-15 秒内自愈，不会陷入无限重试。",[10,1122,1123],{},[213,1124,1125],{},"发版脚本全自动化",[10,1127,1128],{},"1.0.10 里 step 4-10 需要手工兜底。1.0.11 把 step 1（版本 bump）改成了在子仓内完成 commit（idempotent skip 检查保证重跑不会失败），step 3 加了 build marker 检查，step 4 也用同样的 exit code 捕获逻辑。最后 step 1-10 全部自动完成。",[942,1130,1132],{"id":1131},"_1011-的数据","1.0.11 的数据",[10,1134,1135,1136,1139],{},"1.0.11 在 5 月 27 日下午 14:30 发版。23 分钟内，那个主要的 zombie token（前缀记录为 ",[32,1137,1138],{},"ct_UAGorMM","）的命中频率从发版前的 47\u002F分钟 降到 0.17\u002F分钟。对比是精确的：TTL 是 5 分钟，日志里看到的恰好是 5 分钟周期的 4 次 first-hit——14:35、14:40、14:45、14:50，说明黑名单在按秒级别的精度工作。",[10,1141,1142,1145],{},[32,1143,1144],{},"invalid or revoked token"," reject 日志整体从 53\u002F分钟 降到 3\u002F分钟，94% 的噪声被消除。",[10,1147,1148],{},"从发版开始到完成的总耗时是 80 分钟（spec 起草 → 发版完成），远快于 1.0.10 的 12 小时——主要是因为没有了\"假版本\"和手工兜底这两个坑。",[10,1150,1151,1152,1155,1156,1159,1160,1163,1164,1167],{},"客户端配置清理的验证是这样的：启动 1.0.11 时，旧配置目录（",[32,1153,1154],{},"%APPDATA%\u002Fcpa\u002Fcpa-connector\u002F","）存在且有历史 session.json，cleanup 把它 rename 成 ",[32,1157,1158],{},".bak.20260527-141556","；第二次启动时我手动造了一个 fake session.json，它又被 rename 成了不同时间戳的 ",[32,1161,1162],{},".bak","；第三次启动时老目录已经不存在，log 输出 ",[32,1165,1166],{},"DEBUG no legacy config dir to clean up","，幂等。软删除策略保留了原文件，方便后续排查。",[14,1169,1170],{"id":1170},"发版复盘该记录什么",[10,1172,1173,1174,1177],{},"这两次发版的共同教训是：",[213,1175,1176],{},"预期效果与实际效果的差异往往来自非功能层面，发版前遗漏的验证会以人工兜底的成本在发版中浮现","。",[10,1179,1180],{},"1.0.10 的预期是一次提交、一次 build、一个发版脚本的无缝运行。实际遇到了\"代码没同步\"、\"metadata 污染\"、\"git 跨边界\"、\"参数漏传\"这四个层次递进的问题，每一个都需要在发版当时即时判断、即时应急、即时修复。如果这些问题在代码审查或发版前的干运行中被抓到，成本会低一个数量级。",[10,1182,1183,1184,1187],{},"1.0.11 的收益是对 1.0.10 的这些漏洞做了结构化的修复，但更重要的是建立了",[213,1185,1186],{},"验收指标清单","。版本号同步检查、发版脚本的每一步都有 success marker、gitlink 操作必须在子仓内完成、参数完整性的代码审查项。这不是一次性的修补，而是对\"发版这件事\"的流程重塑。",[10,1189,1190,1191,1194,1195,1177],{},"回滚判据也值得明确。1.0.10 的触发条件是\"ws rejected 反升或 connector 启动失败率 > 0.1%\"，对应的回滚操作是修改 RELEASE_CHAIN 截断升级路径。但正如前面提到的，这个方案有一个根本限制：",[213,1192,1193],{},"已升级的用户成了污染源，无法通过服务端操作让他们回退","。这意味着一旦发出去的版本有不可接受的 bug，修复也必须通过新版本修补，不能指望用户自动回到前一个版本。这推导出一个硬性要求：",[213,1196,1197],{},"发版前的验证必须足够彻底，因为回滚的代价极高",[10,1199,1200],{},"1.0.10 和 1.0.11 的连续发版正好说明了这一点。1.0.10 解决了一个明确的功能问题，但在过程中埋了四个流程坑；1.0.11 不光修了功能缺陷（4401 自愈），也补上了流程漏洞（发版脚本全自动）。如果 1.0.10 能在发版前避免那些坑，1.0.11 就没必要冲这么紧。如果不能，1.0.11 这一次的\"加固\"就成了对下一轮发版的保险——每一次发版都可能遗留新的坑，但流程上的防御等级在递增。",{"title":76,"searchDepth":94,"depth":94,"links":1202},[1203,1207,1210],{"id":879,"depth":94,"text":880,"children":1204},[1205,1206],{"id":944,"depth":111,"text":945},{"id":1061,"depth":111,"text":1062},{"id":1086,"depth":94,"text":1087,"children":1208},[1209],{"id":1131,"depth":111,"text":1132},{"id":1170,"depth":94,"text":1170},"2026-06-15",{},"\u002F2026-06-15-connector",{"title":871,"description":876},"2026-06-15-Connector两次发版复盘","两天内连发两版解决 WS 端点变更与僵尸令牌问题，从手动兜底到全自动发版的演进。",[1218,1219,1220,527,1221],"Connector","WebSocket","发版","Rust","z88KhEQF4il5fu5KEpSAcE6LACpC1v_CsLysuzrCk3Q",{"id":1224,"title":1225,"body":1226,"column":513,"date":1781,"description":1782,"extension":515,"hero_image":516,"meta":1783,"navigation":518,"path":1784,"seo":1785,"series_id":516,"severity":516,"stem":1786,"summary":1787,"tags":1788,"__hash__":1792},"posts\u002F2026-05-27-步进升级为什么增量包不能跨版本跳.md","跳过一个撤回的版本，客户就少了个 DLL",{"type":7,"value":1227,"toc":1768},[1228,1235,1238,1245,1253,1256,1263,1266,1269,1275,1278,1284,1289,1293,1296,1302,1306,1309,1315,1322,1325,1329,1336,1339,1342,1345,1348,1352,1355,1361,1369,1374,1385,1390,1401,1405,1408,1412,1423,1427,1435,1439,1450,1456,1463,1466,1469,1472,1504,1507,1510,1516,1522,1525,1528,1582,1585,1692,1699,1713,1716,1727,1734,1759,1765],[10,1229,1230,1231,1234],{},"1.0.7 客户升级到 1.0.9 后启动崩溃，报 ",[32,1232,1233],{},"STATUS_DLL_NOT_FOUND","。根因在于增量包的构造逻辑与发版序列的交互：任何\"中间 GA 版本被撤回且后续版本以它作为 baseline\"的组合，都会让跳过撤回版本的客户拿到不完整的增量包。",[14,1236,1237],{"id":1237},"增量包的设计前提",[10,1239,1240,1241,1244],{},"OTA 更新包是二进制差分产物——",[32,1242,1243],{},"release.sh"," 在构造增量包时，只列入「FROM → TO 两个相邻版本之间 sha256 实际不同的文件」。以 1.0.8 → 1.0.9 的增量为例：",[71,1246,1251],{"className":1247,"code":1249,"language":1250},[1248],"language-text","1.0.8 dist ↔ 1.0.9 dist\n  vcruntime140.dll (sha256: abc...)  vcruntime140.dll (sha256: abc...)\n                                    ↑ 一致，不进差分包\n  msvcp140.dll (sha256: def...)     msvcp140.dll (sha256: def...)\n                                    ↑ 一致，不进差分包\n  launcher.exe (sha256: old)        launcher.exe (sha256: new)\n                                    ↑ 不同，进差分包\n","text",[32,1252,1249],{"__ignoreMap":76},[10,1254,1255],{},"相比全量包（500+ MB），增量包通常只有 100-200 MB。对于国内弱网、公网、海外长途网络来说，包体积直接影响下载时间和失败率。选择差分包的代价是必须确保客户的升级链路完整。",[10,1257,1258,1259,1262],{},"这个设计基于一个隐含的假设：",[213,1260,1261],{},"客户严格按发布顺序逐版升级","（1.0.7 → 1.0.8 → 1.0.9 → ...）。每跳一个版本，就叠加一次「FROM→TO」的相邻增量。所有累积变更都会被依次应用，最终到达完整的目标版本。只要这个假设成立，差分机制就是 sound 的。但一旦中间版本被 recall 且升序逻辑允许跳过，假设就被破坏了。",[14,1264,1265],{"id":1265},"发版序列与撤回的交互",[10,1267,1268],{},"关键的发版序列是这样的：",[71,1270,1273],{"className":1271,"code":1272,"language":1250},[1248],"1.0.5  GA   无 VCRuntime DLL\n1.0.6  recalled\n1.0.7  GA   无 VCRuntime DLL\n1.0.8  GA   ↓ 首次 Copy-VcRuntime，dist 含 9 个 VC++ Runtime DLL\n       后 recalled（因 OTA_TRASH_CORRUPT 现场卡死）\n1.0.9  GA   继承 1.0.8 build，含 VCRuntime DLL\n       baseline = 1.0.8（diff 算相邻两版）\n       minClient = 1.0.8（改为 1.0.7 后触发此 bug）\n",[32,1274,1272],{"__ignoreMap":76},[10,1276,1277],{},"当 1.0.9 的 minClient 被改为 1.0.7 后，升序逻辑会为 1.0.7 客户返回 1.0.9 的 manifest。minClient 是一个关键的版本约束字段——它表示\"升级到当前版本前，客户必须至少是什么版本\"。修改 minClient 的动机通常是想扩大升级适配范围（比如从 1.0.8 放宽到 1.0.7），但这个改动隐含了一个假设：\"当前版本及其所有依赖都能兼容从 minClient 开始的所有中间版本\"。在这个案例中，这个假设被打破了。",[71,1279,1282],{"className":1280,"code":1281,"language":1250},[1248],"G26 升序查询（1.0.7 客户，seek current 之后最小 GA）：\n  候选 1.0.8 → recalled，跳过\n  候选 1.0.9 → GA，minClient=1.0.7≤1.0.7 满足 → 命中，返回 1.0.9 manifest\n",[32,1283,1281],{"__ignoreMap":76},[10,1285,1286,1287,1177],{},"1.0.7 客户 apply 1.0.9 的差分包后，系统启动时 launcher.exe 试图加载 VCRUNTIME140.dll，但客户机的 dist 里没有这个文件——1.0.8 引入的 9 个 DLL 永远没有被应用过。Windows 加载器在 LoadLibrary 阶段直接终结进程，报 ",[32,1288,1233],{},[14,1290,1292],{"id":1291},"为什么-107-看起来能启动升级后就崩溃","为什么 1.0.7 看起来能启动，升级后就崩溃",[10,1294,1295],{},"这里需要澄清一个关键点。1.0.7 的 dist 里确实没有 VCRuntime DLL，但 1.0.7 版本的应用仍然能启动。原因在于 launcher.exe 是用 1.0.7 时代的 toolchain build 的，它依赖的 MSVC CRT 版本恰好在干净 Windows 的 System32 里能找到（或者该客户机预装了对应版本的 Visual C++ Redistributable）。这是一个\"侥幸\"——1.0.7 构建时选用的工具链版本较旧，CRT 的 ABI 足够稳定，系统库能满足。",[10,1297,1298,1299,1301],{},"一旦升级到 1.0.9，launcher.exe 是用更新的 toolchain build 的，它依赖的 CRT 版本更新了，链接的 API 集合也有变化。客户机的 System32 里没有这个新版 CRT。Windows PE 加载器会在 LoadLibrary 阶段就找不到 VCRUNTIME140.dll，进程在主函数之前被系统杀掉（",[32,1300,1233],{},"）。应用的错误处理机制根本没机会启动——没有 try-catch，没有日志，只有系统错误弹窗和退出码。",[14,1303,1305],{"id":1304},"为什么会漏-dll","为什么会漏 DLL",[10,1307,1308],{},"这是差分包设计与版本跳过的碰撞。1.0.9 的差分包不包含 vcruntime140.dll，原因看似无害：",[71,1310,1313],{"className":1311,"code":1312,"language":1250},[1248],"1.0.8 dist                1.0.9 dist\n  vcruntime140.dll          vcruntime140.dll\n  sha256: \u003Chash>            sha256: \u003Chash>（同一个 build 产物，哈希相同）\n  → 两边都有，且相同 → 不进差分包\n",[32,1314,1312],{"__ignoreMap":76},[10,1316,1317,1318,1321],{},"但这个逻辑基于一个前提：",[213,1319,1320],{},"从 1.0.8 升级来","。如果客户是从 1.0.7 升级，那么它的 dist 里根本没有 vcruntime140.dll，差分包也不会帮它补上——因为 1.0.9 diff 算法看的是\"1.0.8 有且 1.0.9 也有\" 的文件。",[10,1323,1324],{},"任何「中间 GA 版本引入新文件 + 该版本被 recall + 后续版本用它作 baseline」的组合，都会造成同样的问题。这不是 DLL 特有的现象，而是增量包机制本身的约束。",[14,1326,1328],{"id":1327},"为什么-108-会引入-dll","为什么 1.0.8 会引入 DLL",[10,1330,1331,1332,1335],{},"顺便说一下 1.0.8 为什么成为\"罪魁祸首\"。1.0.8 是第一个在 build-portable.ps1 中加入 ",[32,1333,1334],{},"Copy-VcRuntime"," 的版本，它主动把整套 MSVC redist DLL（9 个）旁置到 dist，这样便携包就不再依赖系统级安装的 Visual C++ Redistributable。这个改动本身是对的，但由于 1.0.8 随后因 OTA_TRASH_CORRUPT 问题被 recall，它成了\"引入新文件但被撤回\"的典型。",[10,1337,1338],{},"1.0.9 继承了 1.0.8 的 build，自然也含有这 9 个 DLL，但 1.0.9 的差分包相对 1.0.8 计算，所以这些 DLL 不会再被列入。结果就是：1.0.7 客户跳过 1.0.8，直接升到 1.0.9，永远收不到这些 DLL。",[10,1340,1341],{},"同时，还有一个独立的关联问题：修复脚本 Fix-VCRuntime.bat 本身因为编码问题（UTF-8 vs GBK 的冲突）无法正常执行，所以即使用户尝试手动修复，脚本也会报错。这是打包流程和脚本编码的两个独立根因，在同一时间窗口内暴露了出来。",[14,1343,1344],{"id":1344},"修复方案的权衡",[10,1346,1347],{},"面对这个问题，有两条主要路径可选：",[942,1349,1351],{"id":1350},"方案-a全量兜底","方案 A：全量兜底",[10,1353,1354],{},"跨版本升级时直接下发完整 full.zip，而不是差分包。",[10,1356,1357,1360],{},[213,1358,1359],{},"优点","：",[25,1362,1363,1366],{},[28,1364,1365],{},"简单粗暴，一个包包含所有文件，不存在漏 DLL 的可能",[28,1367,1368],{},"不需要调整发版流程",[10,1370,1371,1360],{},[213,1372,1373],{},"缺点",[25,1375,1376,1379,1382],{},[28,1377,1378],{},"包体积大（差分包通常 100-200 MB，全量 500+ MB）",[28,1380,1381],{},"客户网络环境差时升级时间长，失败率高",[28,1383,1384],{},"CDN 带宽成本增加",[10,1386,1387,1360],{},[213,1388,1389],{},"适用场景",[25,1391,1392,1395,1398],{},[28,1393,1394],{},"一次性应急（发 hotfix 回滚）",[28,1396,1397],{},"用户基数小、网络条件好",[28,1399,1400],{},"跨度非常大（比如 1.0.1 → 2.0.0）的少见场景",[942,1402,1404],{"id":1403},"方案-b严格步进","方案 B：严格步进",[10,1406,1407],{},"服务端按发布链下发下一个版本，不允许跨级。即使是 recalled 版本也保留在链上，只是不作为升级终点。",[10,1409,1410,1360],{},[213,1411,1359],{},[25,1413,1414,1417,1420],{},[28,1415,1416],{},"包体积小，增量包通常 100-200 MB",[28,1418,1419],{},"每个客户走相同的升级链路，问题复现容易排查",[28,1421,1422],{},"充分利用增量包设计的初衷",[10,1424,1425,1360],{},[213,1426,1373],{},[25,1428,1429,1432],{},[28,1430,1431],{},"升级链路长（1.0.7 → 1.0.9 时需要先升 1.0.8，即使 1.0.8 被 recalled）",[28,1433,1434],{},"如果某个中间版本有严重问题（比如 OTA 卡死），客户升级时仍会卡在它上面",[10,1436,1437,1360],{},[213,1438,1389],{},[25,1440,1441,1444,1447],{},[28,1442,1443],{},"绝大多数常规发版",[28,1445,1446],{},"网络条件多变的客户群",[28,1448,1449],{},"需要严格控制升级质量的生产环境",[10,1451,1452,1455],{},[213,1453,1454],{},"为什么选择 B","：有几个关键因素。首先，客户群分布广（企业内网、弱网、移动网络混合），300+ MB 的全量包在弱网场景下升级失败率会明显增加，反而增加了支持成本。其次，一旦引入全量兜底，就容易形成\"跨版本时总是下发全量\"的惯性，长期来看放弃了增量包的设计初衷。最后，严格步进虽然升级链路长，但每个版本的 DL + apply 成本都是可预测的，失败时也容易重试。",[10,1457,1458,1459,1462],{},"通过让 recalled 版本保持在链上（只改为 non-GA 状态），升序逻辑会自动跳过它们，下一个 GA 版本会包含完整的累积增量。1.0.9 被 recall 后，G26 升序对 1.0.7 客户会自动推荐 1.0.10；1.0.10 的差分是相对 1.0.7 算的（通过 ",[32,1460,1461],{},"FROM_VERSION=1.0.7"," 显式指定），包含了 1.0.8 引入的 9 个 DLL 以及后续的所有变更。这样既保持了包体积小的优势，又确保了跨越 recalled 版本的客户能拿到完整增量。1.0.10 的差分包最关键的是 Removes 为 0——任何跨版本的客户 apply 都不会留下废文件。",[14,1464,1465],{"id":1465},"为什么撤回版本不能从链上摘掉",[10,1467,1468],{},"这里有一个容易忽视的细节：即使某个版本被 recall，也不能从版本链中物理删除。",[10,1470,1471],{},"如果把 1.0.8 从数据库里删掉，升序逻辑会直接跳过它，导致 1.0.7 客户被推荐到 1.0.9——此时就回到了原来的问题。修复的关键是让升序逻辑仍然\"看到\" 1.0.8 的存在（用于确定差分链路），但把它的 rollout 状态改为 non-GA，使得它永远不会被作为升级终点。",[71,1473,1477],{"className":1474,"code":1475,"language":1476,"meta":76,"style":76},"language-sql shiki shiki-themes github-light github-dark","-- 这样的做法是错的（会复现问题）：\nDELETE FROM releases WHERE version = '1.0.8';\n\n-- 正确的做法：\nUPDATE releases SET rollout_status = 'recalled' WHERE version = '1.0.8';\n","sql",[32,1478,1479,1484,1489,1494,1499],{"__ignoreMap":76},[80,1480,1481],{"class":82,"line":83},[80,1482,1483],{},"-- 这样的做法是错的（会复现问题）：\n",[80,1485,1486],{"class":82,"line":94},[80,1487,1488],{},"DELETE FROM releases WHERE version = '1.0.8';\n",[80,1490,1491],{"class":82,"line":111},[80,1492,1493],{"emptyLinePlaceholder":518},"\n",[80,1495,1496],{"class":82,"line":126},[80,1497,1498],{},"-- 正确的做法：\n",[80,1500,1501],{"class":82,"line":139},[80,1502,1503],{},"UPDATE releases SET rollout_status = 'recalled' WHERE version = '1.0.8';\n",[10,1505,1506],{},"换句话说，版本链的完整性是优于减少数据库记录的。每次新版本发布时，baseline 计算和 diff 生成都要以完整的版本序列为基础。所以即使 1.0.8 和 1.0.9 都被 recall，它们仍然需要在数据库中占据一个位置，作为\"这个版本存在过，但用户不应该升级到它\"的标记。升序逻辑看到 recalled 标记后会直接跳过，继续往后找第一个 GA 版本。",[10,1508,1509],{},"修复后的数据库状态是：",[71,1511,1514],{"className":1512,"code":1513,"language":1250},[1248],"1.0.7   GA         minClient=1.0.5\n1.0.8   recalled\n1.0.9   recalled   ← 这一步至关重要\n1.0.10  GA         minClient=1.0.7\n",[32,1515,1513],{"__ignoreMap":76},[10,1517,1518,1519,1521],{},"现在 1.0.7 客户升序时会依次检查 1.0.8（recalled，跳）→ 1.0.9（recalled，跳）→ 1.0.10（GA，命中）。1.0.10 的差分包是通过 ",[32,1520,1461],{}," 显式指定的，所以它包含 1.0.8 和 1.0.9 引入的所有文件。",[14,1523,1524],{"id":1524},"预防和后续",[10,1526,1527],{},"立即修复是发 1.0.10 并 recall 1.0.9，已在 2026-05-11 19:42 完成。发布 1.0.10 时的关键操作是显式指定 baseline：",[71,1529,1531],{"className":73,"code":1530,"language":75,"meta":76,"style":76},"FROM_VERSION=1.0.7 \\\nMINISIGN_PASSWORD='...' ADMIN_PASS='...' BUILD_HOST_PASSWORD='...' \\\nROLLOUT_STATUS=whitelist \\\n  bash scripts\u002Frelease.sh 1.0.10\n",[32,1532,1533,1547,1563,1571],{"__ignoreMap":76},[80,1534,1535,1538,1541,1544],{"class":82,"line":83},[80,1536,1537],{"class":159},"FROM_VERSION",[80,1539,1540],{"class":146},"=",[80,1542,1543],{"class":90},"1.0.7",[80,1545,1546],{"class":97}," \\\n",[80,1548,1549,1552,1555,1558,1561],{"class":82,"line":94},[80,1550,1551],{"class":159},"MINISIGN_PASSWORD=",[80,1553,1554],{"class":90},"'...'",[80,1556,1557],{"class":90}," ADMIN_PASS='...'",[80,1559,1560],{"class":90}," BUILD_HOST_PASSWORD='...'",[80,1562,1546],{"class":86},[80,1564,1565,1568],{"class":82,"line":111},[80,1566,1567],{"class":159},"ROLLOUT_STATUS=whitelist ",[80,1569,1570],{"class":86},"\\\n",[80,1572,1573,1576,1579],{"class":82,"line":126},[80,1574,1575],{"class":90},"  bash",[80,1577,1578],{"class":90}," scripts\u002Frelease.sh",[80,1580,1581],{"class":86}," 1.0.10\n",[10,1583,1584],{},"这样 diff 生成引擎会计算 1.0.7 ↔ 1.0.10 的完整差异，而不是默认的\"上一个 GA\"（1.0.9）。然后执行 PATCH 和 recall 操作：",[71,1586,1588],{"className":73,"code":1587,"language":75,"meta":76,"style":76},"# 1. PATCH 1.0.10 → ga\ncurl -X PATCH \"$ADMIN\u002Fadmin\u002Freleases\u002F$ID\u002Frollout\" -d '{\"status\":\"ga\"}'\n\n# 2. recall 1.0.9（防止升序仍然推荐给 1.0.7 客户）\nbash scripts\u002Frelease\u002Frecall.sh \u003Crelease-id>\n\n# 3. 清 redis rollout-wrapper cache（立即生效，不等待 cache 过期）\nredis-cli --scan --pattern 'cache:rollout-wrapper:*' | xargs -r redis-cli DEL\n",[32,1589,1590,1595,1627,1631,1636,1654,1658,1663],{"__ignoreMap":76},[80,1591,1592],{"class":82,"line":83},[80,1593,1594],{"class":107},"# 1. PATCH 1.0.10 → ga\n",[80,1596,1597,1600,1603,1606,1609,1612,1615,1618,1621,1624],{"class":82,"line":94},[80,1598,1599],{"class":97},"curl",[80,1601,1602],{"class":86}," -X",[80,1604,1605],{"class":90}," PATCH",[80,1607,1608],{"class":90}," \"",[80,1610,1611],{"class":159},"$ADMIN",[80,1613,1614],{"class":90},"\u002Fadmin\u002Freleases\u002F",[80,1616,1617],{"class":159},"$ID",[80,1619,1620],{"class":90},"\u002Frollout\"",[80,1622,1623],{"class":86}," -d",[80,1625,1626],{"class":90}," '{\"status\":\"ga\"}'\n",[80,1628,1629],{"class":82,"line":111},[80,1630,1493],{"emptyLinePlaceholder":518},[80,1632,1633],{"class":82,"line":126},[80,1634,1635],{"class":107},"# 2. recall 1.0.9（防止升序仍然推荐给 1.0.7 客户）\n",[80,1637,1638,1640,1643,1645,1648,1651],{"class":82,"line":139},[80,1639,75],{"class":97},[80,1641,1642],{"class":90}," scripts\u002Frelease\u002Frecall.sh",[80,1644,147],{"class":146},[80,1646,1647],{"class":90},"release-i",[80,1649,1650],{"class":159},"d",[80,1652,1653],{"class":146},">\n",[80,1655,1656],{"class":82,"line":169},[80,1657,1493],{"emptyLinePlaceholder":518},[80,1659,1660],{"class":82,"line":189},[80,1661,1662],{"class":107},"# 3. 清 redis rollout-wrapper cache（立即生效，不等待 cache 过期）\n",[80,1664,1665,1668,1671,1674,1677,1680,1683,1686,1689],{"class":82,"line":424},[80,1666,1667],{"class":97},"redis-cli",[80,1669,1670],{"class":86}," --scan",[80,1672,1673],{"class":86}," --pattern",[80,1675,1676],{"class":90}," 'cache:rollout-wrapper:*'",[80,1678,1679],{"class":146}," |",[80,1681,1682],{"class":97}," xargs",[80,1684,1685],{"class":86}," -r",[80,1687,1688],{"class":90}," redis-cli",[80,1690,1691],{"class":90}," DEL\n",[10,1693,1694,1695,1698],{},"预防层面，在 ",[32,1696,1697],{},"wakou-release"," skill 中加了三条新检测项（G27\u002FG28\u002FG29）：",[228,1700,1701,1704,1707],{},[28,1702,1703],{},"检查 releases 表里有没有「ga → recalled 但后续版本 baseline 还指着它」的组合",[28,1705,1706],{},"评估是否存在客户群可能跨过 recalled 版本拿增量",[28,1708,1709,1710],{},"如果是，强制新版本 specify ",[32,1711,1712],{},"FROM_VERSION=\u003C最老活跃 GA 版本>",[10,1714,1715],{},"这三条检测在每次发版前自动跑，有问题的组合会在 release.sh 阶段被拦住，不允许走到\"上传 CDN\"这一步。",[10,1717,1718,1719,1722,1723,1726],{},"后续架构层改进（未实施）：admin server 的 G26 解析逻辑可以加「baseline 链路完整性检查」。具体是：当升序逻辑返回 manifest 给客户时，校验 ",[32,1720,1721],{},"client.version → release.baseline_version"," 之间的所有版本是否都是 GA（无 recalled）。如果链路中存在 recalled 版本，admin 直接返回错误码（比如 ",[32,1724,1725],{},"EXTRACT_FROM_OLDER_BASELINE","）给 OTA worker，让客户 fallback 到全量包请求，而不是无声给一个不完整的差分。",[10,1728,1729,1730,1733],{},"这个改进的收益是\"防御性更强\"——即使人肉操作出了问题（比如遗漏了 ",[32,1731,1732],{},"FROM_VERSION="," 指定），系统也能主动检测并避免分发坏包。但代价是需要修改：",[25,1735,1736,1742,1749,1752],{},[28,1737,1738,1741],{},[32,1739,1740],{},"release-manifest.json"," 的 schema（添加链路检查标记）",[28,1743,1744,1745,1748],{},"admin 的 ",[32,1746,1747],{},"update.ts::resolveBonjourCliPath"," 逻辑",[28,1750,1751],{},"OTA worker 的错误码处理（识别新的 fallback 信号）",[28,1753,1754,1755,1758],{},"CDN 需要同时提供 ",[32,1756,1757],{},"full-from-empty.zip","（或允许 diff 降级到 full）",[10,1760,1761,1762,1764],{},"鉴于目前的发版流程和人肉检测已经能防住这个问题，架构层改进被标记为后续单独需求，暂时先依赖流程检查和 ",[32,1763,1732],{}," 人肉兜底。",[504,1766,1767],{},"html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html pre.shiki code .sVt8B, html code.shiki .sVt8B{--shiki-default:#24292E;--shiki-dark:#E1E4E8}html pre.shiki code .szBVR, html code.shiki .szBVR{--shiki-default:#D73A49;--shiki-dark:#F97583}html pre.shiki code .sZZnC, html code.shiki .sZZnC{--shiki-default:#032F62;--shiki-dark:#9ECBFF}html pre.shiki code .sScJk, html code.shiki .sScJk{--shiki-default:#6F42C1;--shiki-dark:#B392F0}html pre.shiki code .sj4cs, html code.shiki .sj4cs{--shiki-default:#005CC5;--shiki-dark:#79B8FF}html pre.shiki code .sJ8bj, html code.shiki .sJ8bj{--shiki-default:#6A737D;--shiki-dark:#6A737D}",{"title":76,"searchDepth":94,"depth":94,"links":1769},[1770,1771,1772,1773,1774,1775,1779,1780],{"id":1237,"depth":94,"text":1237},{"id":1265,"depth":94,"text":1265},{"id":1291,"depth":94,"text":1292},{"id":1304,"depth":94,"text":1305},{"id":1327,"depth":94,"text":1328},{"id":1344,"depth":94,"text":1344,"children":1776},[1777,1778],{"id":1350,"depth":111,"text":1351},{"id":1403,"depth":111,"text":1404},{"id":1465,"depth":94,"text":1465},{"id":1524,"depth":94,"text":1524},"2026-05-27","1.0.7 客户升级到 1.0.9 后启动崩溃，报 STATUS_DLL_NOT_FOUND。根因在于增量包的构造逻辑与发版序列的交互：任何\"中间 GA 版本被撤回且后续版本以它作为 baseline\"的组合，都会让跳过撤回版本的客户拿到不完整的增量包。",{},"\u002F2026-05-27",{"title":1225,"description":1782},"2026-05-27-步进升级为什么增量包不能跨版本跳","增量包是相邻两版的差分，跳过中间版本会漏掉某版本引入的文件，导致客户启动崩溃。",[525,1789,1790,1791,524],"增量更新","版本管理","二进制差分","mO5egKlEmvxpoLQ8R1GYBlnXzKl1bq1NoHYzI2X5h-0",1785406912445]