[{"data":1,"prerenderedAt":1669},["ShallowReactive",2],{"\u002F2026-06-15-connector":3,"\u002F2026-06-15-connector-rel":372},{"id":4,"title":5,"body":6,"column":355,"date":356,"description":12,"extension":357,"hero_image":358,"meta":359,"navigation":360,"path":361,"seo":362,"series_id":358,"severity":358,"stem":363,"summary":364,"tags":365,"__hash__":371},"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":8,"toc":342},"minimark",[9,13,18,21,29,40,43,71,81,86,89,95,102,117,128,133,136,151,162,167,174,177,182,197,200,204,211,214,225,229,232,235,238,243,246,251,262,267,270,274,281,287,290,309,312,319,322,329,339],[10,11,12],"p",{},"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。",[14,15,17],"h2",{"id":16},"_101017-分钟的周期","1.0.10：17 分钟的周期",[10,19,20],{},"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。",[10,22,23,24,28],{},"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 ",[25,26,27],"code",{},"\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。",[10,30,31,32,35,36,39],{},"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 ",[25,33,34],{},"device_token","，对 ",[25,37,38],{},"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。",[10,41,42],{},"修复需要四处联动：",[44,45,46,53,62,65],"ul",{},[47,48,49,50,52],"li",{},"在 AppState 加一个 ",[25,51,38],{}," 字段，用 RwLock 包装以便运行时读取",[47,54,55,58,59,61],{},[25,56,57],{},"set_device_token"," 命令新增参数，收到服务端返的 ",[25,60,38],{}," 时校验并持久化到 config.json",[47,63,64],{},"ws_runner 启动与每次重连前都从 RwLock 读取最新地址，不再拼装",[47,66,67,68,70],{},"App.tsx 消费 ",[25,69,38],{}," 字段，传给 invoke",[10,72,73,74,77,78,80],{},"兼容性处理也做了：env 变量 ",[25,75,76],{},"CPA_WS_ENDPOINT"," 可强制指定（灰度用），服务端没返 ",[25,79,38],{}," 时退化到旧路径拼装，config.json 校验失败时保留现有值不覆盖。",[82,83,85],"h3",{"id":84},"_1010-发版中的五个坑","1.0.10 发版中的五个坑",[10,87,88],{},"代码改好了，但发版流程成了另一个故事。",[10,90,91],{},[92,93,94],"strong",{},"坑 1：假 1.0.10",[10,96,97,98,101],{},"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 ",[25,99,100],{},"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。",[10,103,104,105,108,109,112,113,116],{},"应急是用 ",[25,106,107],{},"git ls-files | tar over ssh"," 把真 1.0.10 源码同步过去，清掉 macOS metadata 的 ",[25,110,111],{},"._*"," 文件（tar 直接打了这些，Windows 上 tauri-build 把 ",[25,114,115],{},"capabilities\u002F._default.json"," 当 JSON 解析失败），重新 build。",[10,118,119,120,123,124,127],{},"1.0.11 的发版脚本改成了完整源码同步加 ",[25,121,122],{},"tar --exclude='._*'"," 和 Win 端 ",[25,125,126],{},"Remove-Item '._*'"," 双保险。",[10,129,130],{},[92,131,132],{},"坑 2：ssh exit code 谜团",[10,134,135],{},"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。",[10,137,138,139,142,143,146,147,150],{},"当时推测是 ",[25,140,141],{},"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 ",[25,144,145],{},"$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 ",[25,148,149],{},"pipefail"," 组合出了问题。",[10,152,153,154,157,158,161],{},"1.0.11 用了一个兜底方案：脚本监测 build 成功的标志——",[25,155,156],{},"Finished 1 bundle at:"," 和 ",[25,159,160],{},"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」。",[10,163,164],{},[92,165,166],{},"坑 3：gitlink 边界",[10,168,169,170,173],{},"发版脚本 step 10 尝试 ",[25,171,172],{},"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。",[10,175,176],{},"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。",[10,178,179],{},[92,180,181],{},"坑 4 和 5：参数漏传与版本号不同步",[10,183,184,185,188,189,192,193,196],{},"Task 9 重构了前端的 ",[25,186,187],{},"tryAutoPairOnce"," 这个 helper 时，body 从 ",[25,190,191],{},"{ hostname, os, connector_version }"," 意外退化成了 ",[25,194,195],{},"{}","，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。",[10,198,199],{},"这两个都算是发版前的信息丢失，1.0.10 发版后才被指出。",[82,201,203],{"id":202},"_1010-的验证与回滚","1.0.10 的验证与回滚",[10,205,206,207,210],{},"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 ",[25,208,209],{},"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）。",[10,212,213],{},"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。",[10,215,216,217,220,221,224],{},"1.0.10 本身的回滚预案是改 ",[25,218,219],{},"apps\u002Fapi\u002Fsrc\u002Fconnector\u002Fupdates.ts"," 的 RELEASE_CHAIN 数组，删掉 ",[25,222,223],{},"'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。",[14,226,228],{"id":227},"_1011两个小时内的闭环","1.0.11：两个小时内的闭环",[10,230,231],{},"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。",[10,233,234],{},"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。",[10,236,237],{},"1.0.11 同步推进了三个方向的修复：",[10,239,240],{},[92,241,242],{},"服务端黑名单",[10,244,245],{},"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。",[10,247,248],{},[92,249,250],{},"客户端自愈",[10,252,253,254,257,258,261],{},"改造 ws.rs 的错误类型，让 4401 close code 能被单独识别为 ",[25,255,256],{},"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 ",[25,259,260],{},"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。",[10,263,264],{},[92,265,266],{},"发版脚本全自动化",[10,268,269],{},"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。",[82,271,273],{"id":272},"_1011-的数据","1.0.11 的数据",[10,275,276,277,280],{},"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 ",[25,278,279],{},"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。",[10,282,283,286],{},[25,284,285],{},"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。",[10,288,289],{},"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。",[10,291,292,293,296,297,300,301,304,305,308],{},"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（",[25,294,295],{},"%APPDATA%\u002Fcpa\u002Fcpa-connector\u002F","）存在且有历史 session.json，cleanup 把它 rename 成 ",[25,298,299],{},".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 ",[25,302,303],{},".bak","；第三次启动时老目录已经不存在，log 输出 ",[25,306,307],{},"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。",[14,310,311],{"id":311},"发版复盘该记录什么",[10,313,314,315,318],{},"这两次发版的共同教训是：",[92,316,317],{},"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","。",[10,320,321],{},"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。",[10,323,324,325,328],{},"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",[92,326,327],{},"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。",[10,330,331,332,335,336,318],{},"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：",[92,333,334],{},"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：",[92,337,338],{},"发版前的验证必须足够彻底，因为回滚的代价极高",[10,340,341],{},"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":343,"searchDepth":344,"depth":344,"links":345},"",2,[346,351,354],{"id":16,"depth":344,"text":17,"children":347},[348,350],{"id":84,"depth":349,"text":85},3,{"id":202,"depth":349,"text":203},{"id":227,"depth":344,"text":228,"children":352},[353],{"id":272,"depth":349,"text":273},{"id":311,"depth":344,"text":311},"交付与更新","2026-06-15","md",null,{},true,"\u002F2026-06-15-connector",{"title":5,"description":12},"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。",[366,367,368,369,370],"Connector","WebSocket","发版","Tauri","Rust","z88KhEQF4il5fu5KEpSAcE6LACpC1v_CsLysuzrCk3Q",[373,885,1099],{"id":374,"title":375,"body":376,"column":355,"date":873,"description":380,"extension":357,"hero_image":358,"meta":874,"navigation":360,"path":875,"seo":876,"series_id":358,"severity":358,"stem":877,"summary":878,"tags":879,"__hash__":884},"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":377,"toc":867},[378,381,385,388,391,410,421,424,428,431,434,570,577,580,587,617,620,627,631,634,637,644,714,725,788,791,806,813,816,833,840,843,846,857,860,863],[10,379,380],{},"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。",[14,382,384],{"id":383},"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",[10,386,387],{},"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。",[10,389,390],{},"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：",[44,392,393,399],{},[47,394,395,396],{},"触发：",[25,397,398],{},"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",[47,400,401,402,405,406,409],{},"不触发：",[25,403,404],{},"改一下 package.json 版本号","（用户只想改文件），",[25,407,408],{},"看看现在是什么版本","（只查询）",[10,411,412,413,416,417,420],{},"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：",[25,414,415],{},"\"Connector 现在是什么版本\"(只查询)"," ",[25,418,419],{},"\"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。",[10,422,423],{},"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。",[14,425,427],{"id":426},"步骤不可跳过验证链条硬卡","步骤不可跳过：验证链条硬卡",[10,429,430],{},"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。",[10,432,433],{},"预检（Step 1）是整个流程的闸门。跑这些 Bash 检查：",[435,436,440],"pre",{"className":437,"code":438,"language":439,"meta":343,"style":343},"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",[25,441,442,455,471,485,498,528,548],{"__ignoreMap":343},[443,444,447,451],"span",{"class":445,"line":446},"line",1,[443,448,450],{"class":449},"sj4cs","cd",[443,452,454],{"class":453},"sZZnC"," \u002FUsers\u002Fxingye\u002FAi\u002Fnew-openclaw\n",[443,456,457,461,464,467],{"class":445,"line":344},[443,458,460],{"class":459},"sScJk","git",[443,462,463],{"class":453}," status",[443,465,466],{"class":449}," --short",[443,468,470],{"class":469},"sJ8bj","                                          # 工作区干净\n",[443,472,473,476,479,482],{"class":445,"line":349},[443,474,475],{"class":449},"test",[443,477,478],{"class":449}," -x",[443,480,481],{"class":453}," \u002Fopt\u002Fhomebrew\u002Fbin\u002Fsshpass",[443,483,484],{"class":469},"                           # ssh 工具\n",[443,486,488,490,492,495],{"class":445,"line":487},4,[443,489,475],{"class":449},[443,491,478],{"class":449},[443,493,494],{"class":453}," \u002Fopt\u002Fhomebrew\u002Fbin\u002Fminisign",[443,496,497],{"class":469},"                          # 签名工具\n",[443,499,501,503,505,509,512,515,518,522,525],{"class":445,"line":500},5,[443,502,475],{"class":449},[443,504,478],{"class":449},[443,506,508],{"class":507},"szBVR"," \u003C",[443,510,511],{"class":453},"签名",[443,513,514],{"class":453}," CLI",[443,516,517],{"class":453}," 路",[443,519,521],{"class":520},"sVt8B","径",[443,523,524],{"class":507},">",[443,526,527],{"class":469},"                                      # 内部签名工具就位\n",[443,529,531,533,536,538,541,543,545],{"class":445,"line":530},6,[443,532,475],{"class":449},[443,534,535],{"class":449}," -f",[443,537,508],{"class":507},[443,539,540],{"class":453},"发版签名私钥路",[443,542,521],{"class":520},[443,544,524],{"class":507},[443,546,547],{"class":469},"                                   # 私钥就位\n",[443,549,551,553,555,557,560,563,565,567],{"class":445,"line":550},7,[443,552,475],{"class":449},[443,554,535],{"class":449},[443,556,508],{"class":507},[443,558,559],{"class":453},"CDN",[443,561,562],{"class":453}," 上传凭据路",[443,564,521],{"class":520},[443,566,524],{"class":507},[443,568,569],{"class":469},"                                   # CDN 凭据就位\n",[10,571,572,573,576],{},"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",[92,574,575],{},"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。",[10,578,579],{},"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。",[10,581,582,583,586],{},"跑 release.sh（Step 3）用 ",[25,584,585],{},"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：",[588,589,590,593,596,599,602,605,608,611,614],"ol",{},[47,591,592],{},"preflight",[47,594,595],{},"detect FROM_VERSION",[47,597,598],{},"tar + scp source to Win build host",[47,600,601],{},"build on Win",[47,603,604],{},"stage + diff-pack OTA full.zip",[47,606,607],{},"minisign full.zip + release-manifest.json",[47,609,610],{},"upload to CDN",[47,612,613],{},"admin: create release + PATCH rollout",[47,615,616],{},"git tag + commit version bump",[10,618,619],{},"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。",[10,621,622,623,626],{},"这整套链条的目的是",[92,624,625],{},"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。",[14,628,630],{"id":629},"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",[10,632,633],{},"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。",[10,635,636],{},"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：",[10,638,639,640,643],{},"G19 是 OTA modal 反复弹。原因是 state base 残留 stale ",[25,641,642],{},"download.zip.minisig","，当时的修复是：",[435,645,647],{"className":437,"code":646,"language":439,"meta":343,"style":343},"# 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",[25,648,649,654],{"__ignoreMap":343},[443,650,651],{"class":445,"line":446},[443,652,653],{"class":469},"# 预检 sanity#3: data\u002F 不删\n",[443,655,656,659,662,665,668,671,674,677,680,683,686,689,692,695,698,701,704,706,709,711],{"class":445,"line":344},[443,657,658],{"class":459},"sanity#3",[443,660,661],{"class":453}," 强制阻断",[443,663,664],{"class":453}," deletes",[443,666,667],{"class":453}," 命中",[443,669,670],{"class":453}," data\u002F。某次发版失败提示",[443,672,673],{"class":453}," sanity#3",[443,675,676],{"class":453}," 一定是",[443,678,679],{"class":453}," build",[443,681,682],{"class":453}," 步骤搞坏了",[443,684,685],{"class":453}," data\u002F",[443,687,688],{"class":453}," 目录的同步，不要绕过",[443,690,691],{"class":453}," sanity",[443,693,694],{"class":453}," check",[443,696,697],{"class":453}," 去强发，先",[443,699,700],{"class":453}," debug",[443,702,703],{"class":453}," 为什么",[443,705,685],{"class":453},[443,707,708],{"class":453}," 出现在",[443,710,664],{"class":453},[443,712,713],{"class":453}," 列表。\n",[10,715,716,717,720,721,724],{},"G25 是跨用户机器后 chat 权限错。根因是 ",[25,718,719],{},"sessions\u002Faudit"," 含绝对路径，guardian sync 扩散。这不是\"修了以后的故事\"，而是",[92,722,723],{},"发版前防御 checklist"," 的一部分：",[435,726,728],{"className":437,"code":727,"language":439,"meta":343,"style":343},"# 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",[25,729,730,735,740,762,767,772,777,782],{"__ignoreMap":343},[443,731,732],{"class":445,"line":446},[443,733,734],{"class":469},"# 发版前防御 checklist (每次发版前在 build host 做完整新用户回归)\n",[443,736,737],{"class":445,"line":344},[443,738,739],{"class":469},"# 3. dist 内 grep C:\\Users\\ 应该 0 命中（防 G25 类绝对路径泄露）\n",[443,741,742,745,748,751,754,757,759],{"class":445,"line":349},[443,743,744],{"class":459},"ssh",[443,746,747],{"class":453}," CHANCHING@",[443,749,750],{"class":507},"\u003C",[443,752,753],{"class":453},"构建",[443,755,756],{"class":520},"机",[443,758,524],{"class":507},[443,760,761],{"class":453}," 'powershell -Command \"\n",[443,763,764],{"class":445,"line":487},[443,765,766],{"class":453},"  Get-ChildItem D:\\wakou-build\\dist-X.Y.Z\\WakouPanelPortable -Recurse -File -Include *.json,*.jsonl |\n",[443,768,769],{"class":445,"line":500},[443,770,771],{"class":453},"    Where-Object { $_.Length -lt 200KB } |\n",[443,773,774],{"class":445,"line":530},[443,775,776],{"class":453},"    Select-String C:\\\\Users\\\\Administrator -SimpleMatch -List |\n",[443,778,779],{"class":445,"line":550},[443,780,781],{"class":453},"    Select-Object FullName\n",[443,783,785],{"class":445,"line":784},8,[443,786,787],{"class":453},"\"'\n",[10,789,790],{},"任何一项 fail → 不要 GA，修了再 bump 一个 patch 版本。",[10,792,793,794,797,798,801,802,805],{},"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 ",[25,795,796],{},"TAURI_SIGNING_PRIVATE_KEY"," 是不是 .sec 文件",[92,799,800],{},"base64 编码后","的内容。\"OTA 不跨级但用户期望直跳\" 的处置是调整 ",[25,803,804],{},"findNextVersion"," 逻辑或在灰度模式放上一个 manifest。",[10,807,808,809,812],{},"关键在",[92,810,811],{},"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。",[10,814,815],{},"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：",[817,818,819],"blockquote",{},[10,820,821,822,825,826,829,830,832],{},"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：",[25,823,824],{},"--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 ",[25,827,828],{},"--no-cache","，不要用 ",[25,831,824],{},"。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。",[10,834,835,836,839],{},"这不是\"高级技巧\"或\"经验之谈\"。这是",[92,837,838],{},"发版的必读须知","，写进 skill 里才能确保每次都不遗漏。",[14,841,842],{"id":842},"为什么这样做有效",[10,844,845],{},"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：",[44,847,848,851,854],{},[47,849,850],{},"触发词明确化杀死了\"我不是故意调用发版\"的那类误操作。",[47,852,853],{},"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。",[47,855,856],{},"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。",[10,858,859],{},"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。",[10,861,862],{},"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。",[864,865,866],"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":343,"searchDepth":344,"depth":344,"links":868},[869,870,871,872],{"id":383,"depth":344,"text":384},{"id":426,"depth":344,"text":427},{"id":629,"depth":344,"text":630},{"id":842,"depth":344,"text":842},"2026-06-16",{},"\u002F2026-06-16-skill",{"title":375,"description":380},"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。",[880,881,882,369,883],"发版流程","OTA","自动化","Docker Swarm","WbjJuf61wSEx4d5Iov2KhPdG1N6hik51RAl8rAGgug8",{"id":4,"title":5,"body":886,"column":355,"date":356,"description":12,"extension":357,"hero_image":358,"meta":1096,"navigation":360,"path":361,"seo":1097,"series_id":358,"severity":358,"stem":363,"summary":364,"tags":1098,"__hash__":371},{"type":7,"value":887,"toc":1086},[888,890,892,894,898,904,906,924,930,932,934,938,942,950,956,960,962,970,976,980,984,986,990,998,1000,1002,1006,1008,1014,1016,1018,1020,1022,1026,1028,1032,1038,1042,1044,1046,1050,1054,1056,1066,1068,1072,1074,1078,1084],[10,889,12],{},[14,891,17],{"id":16},[10,893,20],{},[10,895,23,896,28],{},[25,897,27],{},[10,899,31,900,35,902,39],{},[25,901,34],{},[25,903,38],{},[10,905,42],{},[44,907,908,912,918,920],{},[47,909,49,910,52],{},[25,911,38],{},[47,913,914,58,916,61],{},[25,915,57],{},[25,917,38],{},[47,919,64],{},[47,921,67,922,70],{},[25,923,38],{},[10,925,73,926,77,928,80],{},[25,927,76],{},[25,929,38],{},[82,931,85],{"id":84},[10,933,88],{},[10,935,936],{},[92,937,94],{},[10,939,97,940,101],{},[25,941,100],{},[10,943,104,944,108,946,112,948,116],{},[25,945,107],{},[25,947,111],{},[25,949,115],{},[10,951,119,952,123,954,127],{},[25,953,122],{},[25,955,126],{},[10,957,958],{},[92,959,132],{},[10,961,135],{},[10,963,138,964,142,966,146,968,150],{},[25,965,141],{},[25,967,145],{},[25,969,149],{},[10,971,153,972,157,974,161],{},[25,973,156],{},[25,975,160],{},[10,977,978],{},[92,979,166],{},[10,981,169,982,173],{},[25,983,172],{},[10,985,176],{},[10,987,988],{},[92,989,181],{},[10,991,184,992,188,994,192,996,196],{},[25,993,187],{},[25,995,191],{},[25,997,195],{},[10,999,199],{},[82,1001,203],{"id":202},[10,1003,206,1004,210],{},[25,1005,209],{},[10,1007,213],{},[10,1009,216,1010,220,1012,224],{},[25,1011,219],{},[25,1013,223],{},[14,1015,228],{"id":227},[10,1017,231],{},[10,1019,234],{},[10,1021,237],{},[10,1023,1024],{},[92,1025,242],{},[10,1027,245],{},[10,1029,1030],{},[92,1031,250],{},[10,1033,253,1034,257,1036,261],{},[25,1035,256],{},[25,1037,260],{},[10,1039,1040],{},[92,1041,266],{},[10,1043,269],{},[82,1045,273],{"id":272},[10,1047,276,1048,280],{},[25,1049,279],{},[10,1051,1052,286],{},[25,1053,285],{},[10,1055,289],{},[10,1057,292,1058,296,1060,300,1062,304,1064,308],{},[25,1059,295],{},[25,1061,299],{},[25,1063,303],{},[25,1065,307],{},[14,1067,311],{"id":311},[10,1069,314,1070,318],{},[92,1071,317],{},[10,1073,321],{},[10,1075,324,1076,328],{},[92,1077,327],{},[10,1079,331,1080,335,1082,318],{},[92,1081,334],{},[92,1083,338],{},[10,1085,341],{},{"title":343,"searchDepth":344,"depth":344,"links":1087},[1088,1092,1095],{"id":16,"depth":344,"text":17,"children":1089},[1090,1091],{"id":84,"depth":349,"text":85},{"id":202,"depth":349,"text":203},{"id":227,"depth":344,"text":228,"children":1093},[1094],{"id":272,"depth":349,"text":273},{"id":311,"depth":344,"text":311},{},{"title":5,"description":12},[366,367,368,369,370],{"id":1100,"title":1101,"body":1102,"column":355,"date":1657,"description":1658,"extension":357,"hero_image":358,"meta":1659,"navigation":360,"path":1660,"seo":1661,"series_id":358,"severity":358,"stem":1662,"summary":1663,"tags":1664,"__hash__":1668},"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":1103,"toc":1644},[1104,1111,1114,1121,1129,1132,1139,1142,1145,1151,1154,1160,1165,1169,1172,1178,1182,1185,1191,1198,1201,1205,1212,1215,1218,1221,1224,1228,1231,1237,1245,1250,1261,1266,1277,1281,1284,1288,1299,1303,1311,1315,1326,1332,1339,1342,1345,1348,1380,1383,1386,1392,1398,1401,1404,1458,1461,1568,1575,1589,1592,1603,1610,1635,1641],[10,1105,1106,1107,1110],{},"1.0.7 客户升级到 1.0.9 后启动崩溃，报 ",[25,1108,1109],{},"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。",[14,1112,1113],{"id":1113},"增量包的设计前提",[10,1115,1116,1117,1120],{},"OTA 更新包是二进制差分产物——",[25,1118,1119],{},"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：",[435,1122,1127],{"className":1123,"code":1125,"language":1126},[1124],"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",[25,1128,1125],{"__ignoreMap":343},[10,1130,1131],{},"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。",[10,1133,1134,1135,1138],{},"这个设计基于一个隐含的假设：",[92,1136,1137],{},"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。",[14,1140,1141],{"id":1141},"发版序列与撤回的交互",[10,1143,1144],{},"关键的发版序列是这样的：",[435,1146,1149],{"className":1147,"code":1148,"language":1126},[1124],"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",[25,1150,1148],{"__ignoreMap":343},[10,1152,1153],{},"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。",[435,1155,1158],{"className":1156,"code":1157,"language":1126},[1124],"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",[25,1159,1157],{"__ignoreMap":343},[10,1161,1162,1163,318],{},"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 ",[25,1164,1109],{},[14,1166,1168],{"id":1167},"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",[10,1170,1171],{},"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。",[10,1173,1174,1175,1177],{},"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（",[25,1176,1109],{},"）。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。",[14,1179,1181],{"id":1180},"为什么会漏-dll","为什么会漏 DLL",[10,1183,1184],{},"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：",[435,1186,1189],{"className":1187,"code":1188,"language":1126},[1124],"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",[25,1190,1188],{"__ignoreMap":343},[10,1192,1193,1194,1197],{},"但这个逻辑基于一个前提：",[92,1195,1196],{},"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。",[10,1199,1200],{},"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。",[14,1202,1204],{"id":1203},"为什么-108-会引入-dll","为什么 1.0.8 会引入 DLL",[10,1206,1207,1208,1211],{},"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 ",[25,1209,1210],{},"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。",[10,1213,1214],{},"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。",[10,1216,1217],{},"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。",[14,1219,1220],{"id":1220},"修复方案的权衡",[10,1222,1223],{},"面对这个问题，有两条主要路径可选：",[82,1225,1227],{"id":1226},"方案-a全量兜底","方案 A：全量兜底",[10,1229,1230],{},"跨版本升级时直接下发完整 full.zip，而不是差分包。",[10,1232,1233,1236],{},[92,1234,1235],{},"优点","：",[44,1238,1239,1242],{},[47,1240,1241],{},"简单粗暴，一个包包含所有文件，不存在漏 DLL 的可能",[47,1243,1244],{},"不需要调整发版流程",[10,1246,1247,1236],{},[92,1248,1249],{},"缺点",[44,1251,1252,1255,1258],{},[47,1253,1254],{},"包体积大（差分包通常 100-200 MB，全量 500+ MB）",[47,1256,1257],{},"客户网络环境差时升级时间长，失败率高",[47,1259,1260],{},"CDN 带宽成本增加",[10,1262,1263,1236],{},[92,1264,1265],{},"适用场景",[44,1267,1268,1271,1274],{},[47,1269,1270],{},"一次性应急（发 hotfix 回滚）",[47,1272,1273],{},"用户基数小、网络条件好",[47,1275,1276],{},"跨度非常大（比如 1.0.1 → 2.0.0）的少见场景",[82,1278,1280],{"id":1279},"方案-b严格步进","方案 B：严格步进",[10,1282,1283],{},"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。",[10,1285,1286,1236],{},[92,1287,1235],{},[44,1289,1290,1293,1296],{},[47,1291,1292],{},"包体积小，增量包通常 100-200 MB",[47,1294,1295],{},"每个客户走相同的升级链路，问题复现容易排查",[47,1297,1298],{},"充分利用增量包设计的初衷",[10,1300,1301,1236],{},[92,1302,1249],{},[44,1304,1305,1308],{},[47,1306,1307],{},"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）",[47,1309,1310],{},"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",[10,1312,1313,1236],{},[92,1314,1265],{},[44,1316,1317,1320,1323],{},[47,1318,1319],{},"绝大多数常规发版",[47,1321,1322],{},"网络条件多变的客户群",[47,1324,1325],{},"需要严格控制升级质量的生产环境",[10,1327,1328,1331],{},[92,1329,1330],{},"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。",[10,1333,1334,1335,1338],{},"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 ",[25,1336,1337],{},"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。",[14,1340,1341],{"id":1341},"为什么撤回版本不能从链上摘掉",[10,1343,1344],{},"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。",[10,1346,1347],{},"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。",[435,1349,1353],{"className":1350,"code":1351,"language":1352,"meta":343,"style":343},"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",[25,1354,1355,1360,1365,1370,1375],{"__ignoreMap":343},[443,1356,1357],{"class":445,"line":446},[443,1358,1359],{},"-- 这样的做法是错的（会复现问题）：\n",[443,1361,1362],{"class":445,"line":344},[443,1363,1364],{},"DELETE FROM releases WHERE version = '1.0.8';\n",[443,1366,1367],{"class":445,"line":349},[443,1368,1369],{"emptyLinePlaceholder":360},"\n",[443,1371,1372],{"class":445,"line":487},[443,1373,1374],{},"-- 正确的做法：\n",[443,1376,1377],{"class":445,"line":500},[443,1378,1379],{},"UPDATE releases SET rollout_status = 'recalled' WHERE version = '1.0.8';\n",[10,1381,1382],{},"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。",[10,1384,1385],{},"修复后的数据库状态是：",[435,1387,1390],{"className":1388,"code":1389,"language":1126},[1124],"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",[25,1391,1389],{"__ignoreMap":343},[10,1393,1394,1395,1397],{},"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 ",[25,1396,1337],{}," 显式指定的，所以它包含 1.0.8 和 1.0.9 引入的所有文件。",[14,1399,1400],{"id":1400},"预防和后续",[10,1402,1403],{},"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：",[435,1405,1407],{"className":437,"code":1406,"language":439,"meta":343,"style":343},"FROM_VERSION=1.0.7 \\\nMINISIGN_PASSWORD='...' ADMIN_PASS='...' BUILD_HOST_PASSWORD='...' \\\nROLLOUT_STATUS=whitelist \\\n  bash scripts\u002Frelease.sh 1.0.10\n",[25,1408,1409,1423,1439,1447],{"__ignoreMap":343},[443,1410,1411,1414,1417,1420],{"class":445,"line":446},[443,1412,1413],{"class":520},"FROM_VERSION",[443,1415,1416],{"class":507},"=",[443,1418,1419],{"class":453},"1.0.7",[443,1421,1422],{"class":459}," \\\n",[443,1424,1425,1428,1431,1434,1437],{"class":445,"line":344},[443,1426,1427],{"class":520},"MINISIGN_PASSWORD=",[443,1429,1430],{"class":453},"'...'",[443,1432,1433],{"class":453}," ADMIN_PASS='...'",[443,1435,1436],{"class":453}," BUILD_HOST_PASSWORD='...'",[443,1438,1422],{"class":449},[443,1440,1441,1444],{"class":445,"line":349},[443,1442,1443],{"class":520},"ROLLOUT_STATUS=whitelist ",[443,1445,1446],{"class":449},"\\\n",[443,1448,1449,1452,1455],{"class":445,"line":487},[443,1450,1451],{"class":453},"  bash",[443,1453,1454],{"class":453}," scripts\u002Frelease.sh",[443,1456,1457],{"class":449}," 1.0.10\n",[10,1459,1460],{},"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：",[435,1462,1464],{"className":437,"code":1463,"language":439,"meta":343,"style":343},"# 1. PATCH 1.0.10 → ga\ncurl -X PATCH \"$ADMIN\u002Fadmin\u002Freleases\u002F$ID\u002Frollout\" -d '{\"status\":\"ga\"}'\n\n# 2. recall 1.0.9（防止升序仍然推荐给 1.0.7 客户）\nbash scripts\u002Frelease\u002Frecall.sh \u003Crelease-id>\n\n# 3. 清 redis rollout-wrapper cache（立即生效，不等待 cache 过期）\nredis-cli --scan --pattern 'cache:rollout-wrapper:*' | xargs -r redis-cli DEL\n",[25,1465,1466,1471,1503,1507,1512,1530,1534,1539],{"__ignoreMap":343},[443,1467,1468],{"class":445,"line":446},[443,1469,1470],{"class":469},"# 1. PATCH 1.0.10 → ga\n",[443,1472,1473,1476,1479,1482,1485,1488,1491,1494,1497,1500],{"class":445,"line":344},[443,1474,1475],{"class":459},"curl",[443,1477,1478],{"class":449}," -X",[443,1480,1481],{"class":453}," PATCH",[443,1483,1484],{"class":453}," \"",[443,1486,1487],{"class":520},"$ADMIN",[443,1489,1490],{"class":453},"\u002Fadmin\u002Freleases\u002F",[443,1492,1493],{"class":520},"$ID",[443,1495,1496],{"class":453},"\u002Frollout\"",[443,1498,1499],{"class":449}," -d",[443,1501,1502],{"class":453}," '{\"status\":\"ga\"}'\n",[443,1504,1505],{"class":445,"line":349},[443,1506,1369],{"emptyLinePlaceholder":360},[443,1508,1509],{"class":445,"line":487},[443,1510,1511],{"class":469},"# 2. recall 1.0.9（防止升序仍然推荐给 1.0.7 客户）\n",[443,1513,1514,1516,1519,1521,1524,1527],{"class":445,"line":500},[443,1515,439],{"class":459},[443,1517,1518],{"class":453}," scripts\u002Frelease\u002Frecall.sh",[443,1520,508],{"class":507},[443,1522,1523],{"class":453},"release-i",[443,1525,1526],{"class":520},"d",[443,1528,1529],{"class":507},">\n",[443,1531,1532],{"class":445,"line":530},[443,1533,1369],{"emptyLinePlaceholder":360},[443,1535,1536],{"class":445,"line":550},[443,1537,1538],{"class":469},"# 3. 清 redis rollout-wrapper cache（立即生效，不等待 cache 过期）\n",[443,1540,1541,1544,1547,1550,1553,1556,1559,1562,1565],{"class":445,"line":784},[443,1542,1543],{"class":459},"redis-cli",[443,1545,1546],{"class":449}," --scan",[443,1548,1549],{"class":449}," --pattern",[443,1551,1552],{"class":453}," 'cache:rollout-wrapper:*'",[443,1554,1555],{"class":507}," |",[443,1557,1558],{"class":459}," xargs",[443,1560,1561],{"class":449}," -r",[443,1563,1564],{"class":453}," redis-cli",[443,1566,1567],{"class":453}," DEL\n",[10,1569,1570,1571,1574],{},"预防层面，在 ",[25,1572,1573],{},"wakou-release"," skill 中加了三条新检测项（G27\u002FG28\u002FG29）：",[588,1576,1577,1580,1583],{},[47,1578,1579],{},"检查 releases 表里有没有「ga → recalled 但后续版本 baseline 还指着它」的组合",[47,1581,1582],{},"评估是否存在客户群可能跨过 recalled 版本拿增量",[47,1584,1585,1586],{},"如果是，强制新版本 specify ",[25,1587,1588],{},"FROM_VERSION=\u003C最老活跃 GA 版本>",[10,1590,1591],{},"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。",[10,1593,1594,1595,1598,1599,1602],{},"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 ",[25,1596,1597],{},"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 ",[25,1600,1601],{},"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。",[10,1604,1605,1606,1609],{},"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 ",[25,1607,1608],{},"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：",[44,1611,1612,1618,1625,1628],{},[47,1613,1614,1617],{},[25,1615,1616],{},"release-manifest.json"," 的 schema（添加链路检查标记）",[47,1619,1620,1621,1624],{},"admin 的 ",[25,1622,1623],{},"update.ts::resolveBonjourCliPath"," 逻辑",[47,1626,1627],{},"OTA worker 的错误码处理（识别新的 fallback 信号）",[47,1629,1630,1631,1634],{},"CDN 需要同时提供 ",[25,1632,1633],{},"full-from-empty.zip","（或允许 diff 降级到 full）",[10,1636,1637,1638,1640],{},"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 ",[25,1639,1608],{}," 人肉兜底。",[864,1642,1643],{},"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":343,"searchDepth":344,"depth":344,"links":1645},[1646,1647,1648,1649,1650,1651,1655,1656],{"id":1113,"depth":344,"text":1113},{"id":1141,"depth":344,"text":1141},{"id":1167,"depth":344,"text":1168},{"id":1180,"depth":344,"text":1181},{"id":1203,"depth":344,"text":1204},{"id":1220,"depth":344,"text":1220,"children":1652},[1653,1654],{"id":1226,"depth":349,"text":1227},{"id":1279,"depth":349,"text":1280},{"id":1341,"depth":344,"text":1341},{"id":1400,"depth":344,"text":1400},"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":1101,"description":1658},"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。",[881,1665,1666,1667,880],"增量更新","版本管理","二进制差分","mO5egKlEmvxpoLQ8R1GYBlnXzKl1bq1NoHYzI2X5h-0",1785406912446]