diff --git a/.gitea/workflows/project-ci.yml b/.gitea/workflows/project-ci.yml index e19b10e7e..2de097ee1 100644 --- a/.gitea/workflows/project-ci.yml +++ b/.gitea/workflows/project-ci.yml @@ -67,7 +67,7 @@ jobs: echo "SPACETIME_SCHEMA_BASE_REF=${base_ref}" >> "${GITHUB_ENV}" - name: Install npm dependencies - run: npm ci + run: bash scripts/ci-npm-ci-with-retry.sh - name: Run repository lint gates run: npm run lint @@ -101,10 +101,10 @@ jobs: run: GENARRATIVE_GITEA_CI_CHECK_RUNTIME=1 bash scripts/check-gitea-ci-job-image.sh - name: Install npm dependencies - run: npm ci + run: bash scripts/ci-npm-ci-with-retry.sh - name: Install AI game creator dependencies - run: npm ci --prefix apps/ai-game-creator-shell + run: bash scripts/ci-npm-ci-with-retry.sh --prefix apps/ai-game-creator-shell - name: Run frontend and script tests run: npm run test @@ -159,7 +159,7 @@ jobs: echo "SPACETIME_SCHEMA_BASE_REF=${base_ref}" >> "${GITHUB_ENV}" - name: Install npm dependencies - run: npm ci + run: bash scripts/ci-npm-ci-with-retry.sh - name: Prepare server-rs Rust dependencies shell: bash @@ -204,10 +204,10 @@ jobs: run: GENARRATIVE_GITEA_CI_CHECK_RUNTIME=1 bash scripts/check-gitea-ci-job-image.sh - name: Install npm dependencies - run: npm ci + run: bash scripts/ci-npm-ci-with-retry.sh - name: Install AI game creator dependencies - run: npm ci --prefix apps/ai-game-creator-shell + run: bash scripts/ci-npm-ci-with-retry.sh --prefix apps/ai-game-creator-shell - name: Prepare native Rust dependencies shell: bash diff --git a/deploy/container/README.md b/deploy/container/README.md index 10ae56971..523f3d4f4 100644 --- a/deploy/container/README.md +++ b/deploy/container/README.md @@ -67,13 +67,13 @@ Linux Docker Engine 若要从宿主机 CLI 连到容器内服务,直接用 `ht ```bash bash scripts/gitea-ci-job-image.sh build bash scripts/gitea-ci-job-image.sh verify -bash scripts/gitea-ci-job-image.sh export /仓库外受控路径/genarrative-gitea-project-ci-20260723.1.tar.zst +bash scripts/gitea-ci-job-image.sh export /仓库外受控路径/genarrative-gitea-project-ci-20260807.1.tar.zst bash scripts/gitea-ci-job-image.sh load-runner ``` -默认构建 tag 为 `genarrative/gitea-project-ci:20260723.1`。脚本通过 NUL 分隔白名单 tar 流只发送 Dockerfile、checkout 脚本、npm manifests/lock 和 Cargo manifests/lock;当前构建 context 约 `1.638 MB`,不会把业务源码、素材或本地私密文件发送给 Docker daemon。镜像除固定工具链外,还按当前根 `package-lock.json`、`server-rs/Cargo.lock` 和桌面壳 `Cargo.lock` 预热 npm / Cargo 下载缓存;两个 `cargo fetch --locked` 最多执行 5 次整命令级有界重试,再以断网 `cargo fetch --locked` 验证缓存闭合,不包含 `node_modules` 或 Cargo `target`。`build` 完成后会自动运行环境校验,`load-runner` 还会比对宿主和 runner 内层的完整 Image ID,并在内层执行 bwrap 与 Chrome headless canary。当前验证镜像约 `1.788 GB`,完整 Image ID 为 `sha256:c04b114b1f145072c9df7842c4c974e1bb2eaaf391d95d84c9212a460546b7d5`。执行这些命令不要求必须使用 root,但执行账号必须有权访问宿主 Docker API 并管理 runner 容器;没有该权限时交给 runner 运维人员执行。 +默认构建 tag 为 `genarrative/gitea-project-ci:20260807.1`。脚本通过 NUL 分隔白名单 tar 流只发送 Dockerfile、checkout 脚本、根与 AI 游戏创作壳的 npm manifests/lock,以及 server-rs、桌面壳和 AI 游戏创作壳的 Cargo manifests/lock;当前构建 context 约 `2.13 MB`,不会把业务源码、素材或本地私密文件发送给 Docker daemon。镜像除固定工具链外,还按上述五份 lock 预热 npm / Cargo 下载缓存;三个 `cargo fetch --locked` 最多执行 5 次整命令级有界重试,再分别以断网 `cargo fetch --locked` 验证缓存闭合,不包含 `node_modules` 或 Cargo `target`。`build` 完成后会自动运行环境校验,`load-runner` 还会比对宿主和 runner 内层的完整 Image ID,并在内层执行 bwrap 与 Chrome headless canary。当前验证镜像约 `1.85 GB`,完整 Image ID 为 `sha256:8b4b30f5a096522942947927b06cf47bdb1a1016dde9a3573f9780e79d8e40cf`。执行这些命令不要求必须使用 root,但执行账号必须有权访问宿主 Docker API 并管理 runner 容器;没有该权限时交给 runner 运维人员执行。 -runner 配置保留原 `ubuntu-latest` 映射,另外增加 `genarrative-ci:docker://sha256:c04b114b1f145072c9df7842c4c974e1bb2eaaf391d95d84c9212a460546b7d5`。内层 Docker 数据必须持久化,`force_pull` 保持 `false`;该精确 Image ID 在内层不存在时 job 应直接失败,不回退到浮动 tag 或现场拉取。四个 job 使用镜像内 `genarrative-gitea-checkout` 直接从当前 Gitea 拉取事件 commit,带 5 次有界重试,不再运行时下载 GitHub checkout action;随后以 `GENARRATIVE_GITEA_CI_CHECK_RUNTIME=1` 执行 `scripts/check-gitea-ci-job-image.sh`,同时校验工具链、缓存锁命中状态、bwrap 和 Chrome headless。它们仍各自运行干净的 `npm ci` 以校验当前 lockfile 并隔离 PR 依赖,但使用镜像内 npm cache 和 `prefer-offline`;锁文件新增依赖时允许经受控网络补齐,本阶段不启用共享 Actions cache。 +runner 配置保留原 `ubuntu-latest` 映射,另外增加 `genarrative-ci:docker://sha256:8b4b30f5a096522942947927b06cf47bdb1a1016dde9a3573f9780e79d8e40cf`。内层 Docker 数据必须持久化,`force_pull` 保持 `false`;该精确 Image ID 在内层不存在时 job 应直接失败,不回退到浮动 tag 或现场拉取。四个 job 使用镜像内 `genarrative-gitea-checkout` 直接从当前 Gitea 拉取事件 commit,带 5 次有界重试,不再运行时下载 GitHub checkout action;随后以 `GENARRATIVE_GITEA_CI_CHECK_RUNTIME=1` 执行 `scripts/check-gitea-ci-job-image.sh`,同时校验工具链、五份缓存锁命中状态、bwrap 和 Chrome headless。锁不匹配时校验会输出 `partial` 和醒目的 Actions warning,提示在可信分支落地后刷新镜像。各 job 仍运行干净的 `npm ci` 以校验当前 lockfile 并隔离 PR 依赖,但统一通过 `scripts/ci-npm-ci-with-retry.sh` 最多执行 3 次整命令级有界重试,并使用镜像内 npm cache 和 `prefer-offline`;锁文件新增依赖时允许经受控网络补齐,本阶段不启用共享 Actions cache。 更新顺序固定为: diff --git a/deploy/container/gitea-ci-job.Dockerfile b/deploy/container/gitea-ci-job.Dockerfile index fdb050e08..0f7e3fdc5 100644 --- a/deploy/container/gitea-ci-job.Dockerfile +++ b/deploy/container/gitea-ci-job.Dockerfile @@ -15,6 +15,7 @@ ENV CARGO_HTTP_MULTIPLEXING=false \ COPY server-rs /tmp/genarrative-cargo-cache/server-rs COPY apps/desktop-shell/src-tauri /tmp/genarrative-cargo-cache/desktop-shell +COPY apps/ai-game-creator-shell/src-tauri /tmp/genarrative-cargo-cache/apps/ai-game-creator-shell/src-tauri RUN find /tmp/genarrative-cargo-cache -name Cargo.toml -exec dirname {} \; \ | while IFS= read -r crate_dir; do \ @@ -39,12 +40,16 @@ RUN find /tmp/genarrative-cargo-cache -name Cargo.toml -exec dirname {} \; \ } \ && cargo_fetch_with_retry /tmp/genarrative-cargo-cache/server-rs/Cargo.toml \ && cargo_fetch_with_retry /tmp/genarrative-cargo-cache/desktop-shell/Cargo.toml \ + && cargo_fetch_with_retry /tmp/genarrative-cargo-cache/apps/ai-game-creator-shell/src-tauri/Cargo.toml \ && CARGO_NET_OFFLINE=true cargo fetch --locked \ --target x86_64-unknown-linux-gnu \ --manifest-path /tmp/genarrative-cargo-cache/server-rs/Cargo.toml \ && CARGO_NET_OFFLINE=true cargo fetch --locked \ --target x86_64-unknown-linux-gnu \ --manifest-path /tmp/genarrative-cargo-cache/desktop-shell/Cargo.toml \ + && CARGO_NET_OFFLINE=true cargo fetch --locked \ + --target x86_64-unknown-linux-gnu \ + --manifest-path /tmp/genarrative-cargo-cache/apps/ai-game-creator-shell/src-tauri/Cargo.toml \ && rm -rf /tmp/genarrative-cargo-cache FROM ${RUNNER_IMAGE} @@ -123,21 +128,31 @@ COPY --from=rust-dependency-cache /usr/local/cargo /usr/local/cargo COPY --from=rust-dependency-cache /usr/local/rustup /usr/local/rustup ARG NPM_LOCK_SHA256 +ARG AGC_NPM_LOCK_SHA256 ARG SERVER_RUST_LOCK_SHA256 ARG DESKTOP_RUST_LOCK_SHA256 +ARG AGC_RUST_LOCK_SHA256 COPY package.json package-lock.json /usr/local/share/genarrative-ci/npm/ +COPY apps/ai-game-creator-shell/package.json apps/ai-game-creator-shell/package-lock.json /usr/local/share/genarrative-ci/agc-npm/ COPY server-rs/Cargo.lock /usr/local/share/genarrative-ci/locks/server-rs.Cargo.lock COPY apps/desktop-shell/src-tauri/Cargo.lock /usr/local/share/genarrative-ci/locks/desktop-shell.Cargo.lock +COPY apps/ai-game-creator-shell/src-tauri/Cargo.lock /usr/local/share/genarrative-ci/locks/ai-game-creator-shell.Cargo.lock COPY deploy/container/gitea-ci-checkout.sh /usr/local/bin/genarrative-gitea-checkout RUN test -n "${NPM_LOCK_SHA256}" \ + && test -n "${AGC_NPM_LOCK_SHA256}" \ && test -n "${SERVER_RUST_LOCK_SHA256}" \ && test -n "${DESKTOP_RUST_LOCK_SHA256}" \ + && test -n "${AGC_RUST_LOCK_SHA256}" \ && printf '%s %s\n' \ "${NPM_LOCK_SHA256}" \ /usr/local/share/genarrative-ci/npm/package-lock.json \ | sha256sum --check --strict \ + && printf '%s %s\n' \ + "${AGC_NPM_LOCK_SHA256}" \ + /usr/local/share/genarrative-ci/agc-npm/package-lock.json \ + | sha256sum --check --strict \ && printf '%s %s\n' \ "${SERVER_RUST_LOCK_SHA256}" \ /usr/local/share/genarrative-ci/locks/server-rs.Cargo.lock \ @@ -146,6 +161,10 @@ RUN test -n "${NPM_LOCK_SHA256}" \ "${DESKTOP_RUST_LOCK_SHA256}" \ /usr/local/share/genarrative-ci/locks/desktop-shell.Cargo.lock \ | sha256sum --check --strict \ + && printf '%s %s\n' \ + "${AGC_RUST_LOCK_SHA256}" \ + /usr/local/share/genarrative-ci/locks/ai-game-creator-shell.Cargo.lock \ + | sha256sum --check --strict \ && chmod 0755 /usr/local/bin/genarrative-gitea-checkout \ && npm ci \ --ignore-scripts \ @@ -154,6 +173,13 @@ RUN test -n "${NPM_LOCK_SHA256}" \ --prefer-offline \ --prefix /usr/local/share/genarrative-ci/npm \ && rm -rf /usr/local/share/genarrative-ci/npm/node_modules \ + && npm ci \ + --ignore-scripts \ + --no-audit \ + --no-fund \ + --prefer-offline \ + --prefix /usr/local/share/genarrative-ci/agc-npm \ + && rm -rf /usr/local/share/genarrative-ci/agc-npm/node_modules \ && npm cache verify RUN install -m 0755 /usr/local/cargo/bin/rustup /usr/local/bin/rustup \ @@ -162,6 +188,8 @@ RUN install -m 0755 /usr/local/cargo/bin/rustup /usr/local/bin/rustup \ done ENV CARGO_HOME=/usr/local/cargo \ + GENARRATIVE_GITEA_CI_AGC_NPM_LOCK_SHA256=${AGC_NPM_LOCK_SHA256} \ + GENARRATIVE_GITEA_CI_AGC_RUST_LOCK_SHA256=${AGC_RUST_LOCK_SHA256} \ GENARRATIVE_GITEA_CI_DESKTOP_RUST_LOCK_SHA256=${DESKTOP_RUST_LOCK_SHA256} \ GENARRATIVE_GITEA_CI_NPM_LOCK_SHA256=${NPM_LOCK_SHA256} \ GENARRATIVE_GITEA_CI_SERVER_RUST_LOCK_SHA256=${SERVER_RUST_LOCK_SHA256} \ @@ -178,6 +206,7 @@ ENV CARGO_HOME=/usr/local/cargo \ ARG IMAGE_REVISION=uncommitted LABEL org.opencontainers.image.vendor="GenarrativeAI" \ + org.opencontainers.image.version="2026.08.07.1" \ org.opencontainers.image.source="https://git.genarrative.world/GenarrativeAI/Genarrative" \ org.opencontainers.image.revision="${IMAGE_REVISION}" \ org.opencontainers.image.base.name="docker.gitea.com/runner-images:ubuntu-latest@sha256:58ea92624c7c09582e05594d95488331045053d3a3f34cf09649f2a32313a614" \ diff --git a/deploy/container/gitea-ci-job.Dockerfile.dockerignore b/deploy/container/gitea-ci-job.Dockerfile.dockerignore index 7a5cb6b15..cb299b2bf 100644 --- a/deploy/container/gitea-ci-job.Dockerfile.dockerignore +++ b/deploy/container/gitea-ci-job.Dockerfile.dockerignore @@ -12,6 +12,12 @@ !server-rs/crates/*/ !server-rs/crates/*/Cargo.toml !apps/ +!apps/ai-game-creator-shell/ +!apps/ai-game-creator-shell/package.json +!apps/ai-game-creator-shell/package-lock.json +!apps/ai-game-creator-shell/src-tauri/ +!apps/ai-game-creator-shell/src-tauri/Cargo.toml +!apps/ai-game-creator-shell/src-tauri/Cargo.lock !apps/desktop-shell/ !apps/desktop-shell/src-tauri/ !apps/desktop-shell/src-tauri/Cargo.toml diff --git a/docs/project-memory/shared-memory/pitfalls.md b/docs/project-memory/shared-memory/pitfalls.md index 75a7464b7..df65b85f0 100644 --- a/docs/project-memory/shared-memory/pitfalls.md +++ b/docs/project-memory/shared-memory/pitfalls.md @@ -4008,11 +4008,18 @@ - 现象:`Repository checks`、`Frontend tests`、`Backend tests` 和 `Native shell tests` 都从全新 job 容器开始,apt、setup-node、rustup 和原生系统库在不同 job 里重复安装;后端与原生壳的安装时间可达数分钟,并把软件源和代理瞬时失败放大为四份。 - 原因:Gitea Actions job 彼此隔离,上一个 job 在容器内安装的包不会自动进入下一个 job;把同一套不随 PR 变化的工具链写在 workflow step 中,必然每次重做。 -- 处理:用 `deploy/container/gitea-ci-job.Dockerfile` 预装 Node 22、Rust 1.96、`rustfmt`、Chrome、`bwrap`、`rg`、`ffmpeg`、`clang/lld` 和 Tauri / 后端系统依赖,并按锁预热根 npm、server-rs 与桌面壳 Cargo 下载缓存。四个 job 统一 `runs-on: genarrative-ci`,先用镜像内脚本直接从 Gitea checkout,再以 runtime 模式运行 `scripts/check-gitea-ci-job-image.sh`,同时检查缓存锁、工具链、完整 bwrap 与 Chrome headless。`RUSTUP_AUTO_INSTALL=0`;`rust-toolchain.toml` 变更时先重建镜像,不把下载 fallback 放回 job。 +- 处理:用 `deploy/container/gitea-ci-job.Dockerfile` 预装 Node 22、Rust 1.96、`rustfmt`、Chrome、`bwrap`、`rg`、`ffmpeg`、`clang/lld` 和 Tauri / 后端系统依赖,并按锁预热根与 AI 游戏创作壳 npm、server-rs、桌面壳与 AI 游戏创作壳 Cargo 下载缓存。四个 job 统一 `runs-on: genarrative-ci`,先用镜像内脚本直接从 Gitea checkout,再以 runtime 模式运行 `scripts/check-gitea-ci-job-image.sh`,同时检查五份缓存锁、工具链、完整 bwrap 与 Chrome headless。`RUSTUP_AUTO_INSTALL=0`;`rust-toolchain.toml` 变更时先重建镜像,不把下载 fallback 放回 job。 - 依赖边界:每个 job 仍必须各自执行 `npm ci`,让当前 lockfile 和 PR 依赖在干净环境中验证;区别是命中镜像 cache 时只做本地解包,锁新增依赖时才走受控网络。不要把 `node_modules` 或 Cargo `target` 烘进镜像,也不要向不受信任 PR 挂载跨 job 可写 cache。 -- 锁漂移边界:runtime 校验输出 `server_rust_cache_lock=partial` 说明镜像内 Cargo lock 与当前 checkout 不同,不代表新增 crate 已经缓存。必须在新镜像中以 `--network none` 对当前 lock 执行真实 `cargo fetch/build --offline`;`cargo metadata --no-deps` 不会证明依赖 archive 可用,不能作为替代。 +- 锁漂移边界:runtime 校验输出任一 `*_cache_lock=partial` 说明镜像内 lock 与当前 checkout 不同,不代表新增依赖已经缓存;必须同时输出 Actions warning,提示可信分支落地后刷新镜像。必须在新镜像中对 server-rs、桌面壳和 AI 游戏创作壳当前 lock 执行真实 `cargo fetch --locked --offline`;`cargo metadata --no-deps` 不会证明依赖 archive 可用,不能作为替代。 - 构建网络边界:`CARGO_NET_RETRY` 只覆盖部分 crate 下载,registry `config.json` / index TLS 握手仍可能直接终止整次 fetch。Dockerfile 对每个 `cargo fetch --locked` 再做最多 5 次整命令级有界重试,最终仍执行断网 fetch,不能降低为无锁重试或省略离线闭合验证。 -- 验证:workflow 不再出现 GitHub checkout action、apt、setup-node 或 rustup 安装 step;镜像在 `--network none` 下能按当前 npm / Cargo lock 完成依赖准备,四个 job 的环境校验、干净 `npm ci` 和原有测试门禁仍全部执行。 +- 验证:workflow 不再出现 GitHub checkout action、apt、setup-node 或 rustup 安装 step;镜像能按五份当前 lock 完成缓存闭合,四个 job 的环境校验、经 3 次整命令级有界重试保护的干净 `npm ci` 和原有测试门禁仍全部执行。 + +## Gitea Actions HTTPS CONNECT 隧道必须双向收束 socket(2026-08-07) + +- 现象:CI 的 `npm ci` 高频出现 `ECONNRESET / network aborted`,Cargo 则出现 crates.io TLS EOF、连接超时或下载失败;同一出口 gateway 容器看似健康,却累计自动重启数百次,日志反复出现 `Socket.ondata -> Writable.write -> write EPIPE -> Unhandled 'error' event`。 +- 原因:HTTPS CONNECT 建立后使用 `upstreamSocket.pipe(clientSocket)` 与反向 pipe,但只监听 upstream `error`;客户端在 DNS 等待、下载或 job 清理期间关闭连接时,pipe 继续向已断开的 client socket 写入,未处理的 EPIPE 会让 Node 进程退出。`unless-stopped` 自动拉起和浅层 healthcheck 会掩盖崩溃,所有并发 npm / Cargo 隧道同时被 reset。 +- 处理:CONNECT 一开始就为 client socket 注册 `error / close`,解析完成后为 upstream socket注册同样的双向销毁处理;DNS 返回、写 200 和开始 pipe 前都检查 client 是否已销毁。任一端 error、close 或 timeout 都幂等 destroy 两端,不把普通客户端 reset 写成错误日志。不要用进程级 `uncaughtException` 吞掉问题,也不要只增加 npm/Cargo 重试掩盖 gateway 崩溃。 +- 验证:在独立 canary 和正式 gateway 上分别并发制造至少 500 次“CONNECT 后立即断开”,随后确认容器仍运行、restart count 不增加、日志无 EPIPE;再通过同一 proxy 对 npm registry 与 crates index 建立完整 TLS 隧道。切换前仍须确认 Gitea 无活跃 run 且 Runner 内层无 job 容器。 ## Gitea CI 预构建镜像不能只靠 tag 判断内容 diff --git a/docs/【开发运维】本地开发验证与生产运维-2026-05-15.md b/docs/【开发运维】本地开发验证与生产运维-2026-05-15.md index 8e1c58a19..10e9b5a79 100644 --- a/docs/【开发运维】本地开发验证与生产运维-2026-05-15.md +++ b/docs/【开发运维】本地开发验证与生产运维-2026-05-15.md @@ -243,20 +243,20 @@ PR checkout 必须保留完整 Git 历史,并把 PR base SHA 传给 `SPACETIME 当前 `genarrative-station` 使用 Gitea `1.26.4` 和基于 Gitea Runner `2.0.0-dind-rootless` 的固定 digest 修补镜像。Runner 2.0.0 会先把 `systempaths=unconfined` 解析为空 `MaskedPaths` / `ReadonlyPaths`,再被 `mergo.WithOverride` 当成 empty value 丢失;站点修补只在 merge 后保留这两个显式空 slice,不改其它 runner 行为。真实 job inspect 必须看到 `MaskedPaths=[]`、`ReadonlyPaths=[]`、`SecurityOpt=[seccomp=unconfined]`、`Privileged=false`、无 CapAdd 且 `Binds=[]`。外层 runner 以 `rootless` 用户运行,`privileged=false`、不增加 `CAP_SYS_ADMIN`,只映射 `/dev/net/tun`,内部 Docker 只监听私有 Unix socket;runner 配置保持 `docker_host: "-"`、`valid_volumes: []`、`bind_workdir: false` 和 `force_pull: false`,防止内部 Docker socket 或宿主 bind mount 进入 job。job 只连接 `gitea-actions` internal network:`genarrative-station` 由只转发 `/git` 到 Gitea 的内部 gateway 解析,公网依赖只经拒绝私网、保留地址和 metadata 的 80/443 egress proxy;绕过 proxy 的公网和 Postgres/Redis 数据网都必须不可达。完整 bwrap canary 需要 rootless DinD 外层的 rootlesskit AppArmor/userns 边界,以及内层 job 的 namespace/proc 挂载支持;相关 `seccomp/systempaths` 放宽只允许存在于这个无宿主 socket 的 rootless DinD 内层,禁止复制回控制宿主 rootful Docker 的 runner。 -CI job 镜像由 `deploy/container/gitea-ci-job.Dockerfile` 定义:Ubuntu job base 固定为 `sha256:58ea92624c7c09582e05594d95488331045053d3a3f34cf09649f2a32313a614`,Rust stage 固定为 `sha256:19817ead3289c8c631c73df281e18b59b172f6a31f4f563290f69cddd06c30e9`,Node `22.23.1` 发行包执行 SHA-256 校验,Google Linux 主签名指纹固定,Chrome 固定为 `150.0.7871.181-1`。构建脚本以 NUL 分隔白名单 tar 流只发送 Dockerfile、checkout 脚本和 npm / Cargo manifests/lock;当前 context 约 `1.638 MB`。镜像按根 npm 锁、server-rs 锁和桌面壳锁预热下载缓存,不包含 `node_modules` 或 Cargo `target`;两个 `cargo fetch --locked` 在 Cargo 自身重试之外再执行最多 5 次整命令级有界重试,处理 registry index 握手失败,最终仍以断网 `cargo fetch --locked` 关闭验证。当前验证镜像约 `1.788 GB`,默认 tag 为 `genarrative/gitea-project-ci:20260723.1`,完整 Image ID 为 `sha256:c04b114b1f145072c9df7842c4c974e1bb2eaaf391d95d84c9212a460546b7d5`;runner 标签保留 `ubuntu-latest`,并将 `genarrative-ci` 映射到 `docker://sha256:c04b114b1f145072c9df7842c4c974e1bb2eaaf391d95d84c9212a460546b7d5`。内层 Docker 数据必须持久化;`force_pull: false` 表示只使用这个已装载的精确内容,Image ID 缺失时 job 必须失败关闭,不得回退浮动 tag 或临时连 registry。 +CI job 镜像由 `deploy/container/gitea-ci-job.Dockerfile` 定义:Ubuntu job base 固定为 `sha256:58ea92624c7c09582e05594d95488331045053d3a3f34cf09649f2a32313a614`,Rust stage 固定为 `sha256:19817ead3289c8c631c73df281e18b59b172f6a31f4f563290f69cddd06c30e9`,Node `22.23.1` 发行包执行 SHA-256 校验,Google Linux 主签名指纹固定,Chrome 固定为 `150.0.7871.181-1`。构建脚本以 NUL 分隔白名单 tar 流只发送 Dockerfile、checkout 脚本,以及根、AI 游戏创作壳、server-rs 与桌面壳所需的 npm / Cargo manifests/lock;当前 context 约 `2.13 MB`。镜像按根 npm 锁、AI 游戏创作壳 npm 锁、server-rs 锁、桌面壳锁和 AI 游戏创作壳 Cargo 锁预热下载缓存,不包含 `node_modules` 或 Cargo `target`;三个 `cargo fetch --locked` 在 Cargo 自身重试之外再执行最多 5 次整命令级有界重试,处理 registry index 握手失败,最终仍分别以断网 `cargo fetch --locked` 关闭验证。当前验证镜像约 `1.85 GB`,默认 tag 为 `genarrative/gitea-project-ci:20260807.1`,完整 Image ID 为 `sha256:8b4b30f5a096522942947927b06cf47bdb1a1016dde9a3573f9780e79d8e40cf`;runner 标签保留 `ubuntu-latest`,并将 `genarrative-ci` 映射到 `docker://sha256:8b4b30f5a096522942947927b06cf47bdb1a1016dde9a3573f9780e79d8e40cf`。内层 Docker 数据必须持久化;`force_pull: false` 表示只使用这个已装载的精确内容,Image ID 缺失时 job 必须失败关闭,不得回退浮动 tag 或临时连 registry。 镜像更新命令: ```bash bash scripts/gitea-ci-job-image.sh build bash scripts/gitea-ci-job-image.sh verify -bash scripts/gitea-ci-job-image.sh export /仓库外受控路径/genarrative-gitea-project-ci-20260723.1.tar.zst +bash scripts/gitea-ci-job-image.sh export /仓库外受控路径/genarrative-gitea-project-ci-20260807.1.tar.zst bash scripts/gitea-ci-job-image.sh load-runner ``` 执行账号只要有权访问宿主 Docker API 并管理 runner 容器即可,不强制使用 root;无该权限时由 runner 运维人员执行。更新顺序必须是 `build/verify -> export 仓库外镜像归档与 SHA-256 sidecar -> load-runner -> 确认无活跃 job -> 备份当前 config -> 增加或替换 label -> docker restart --timeout 660 gitea-runner`。`--timeout 660` 只是停止宽限,不是 drain API;rootless DinD supervisor 可能同时停止内层 dockerd,因此重启前必须确认 Gitea 没有 `in_progress` run 且内层 `docker ps` 为空。config 和镜像归档只保存到仓库外受控位置,不在文档、仓库或日志中记录注册信息。重启后先重跑真实 PR 的四个 job,复核隔离边界并确认全部通过,再清理旧镜像。回滚时先把 workflow 的 `runs-on` 改回 `ubuntu-latest`,再恢复 config 备份并重启 runner。 -四个 job 先运行镜像内 `genarrative-gitea-checkout`,再以 `GENARRATIVE_GITEA_CI_CHECK_RUNTIME=1` 执行 `scripts/check-gitea-ci-job-image.sh`,校验 Node 主版本、仓库 Rust toolchain、受信任 PATH、缓存锁命中状态、原生命令、pkg-config 依赖、完整 bwrap sandbox 和 Chrome headless。`RUSTUP_AUTO_INSTALL=0`,因此仓库 `rust-toolchain.toml` 变更必须先更新镜像,不能让 job 现场下载。每个 job 仍独立运行 `npm ci`,以当前 lockfile 为准验证 PR 依赖;`NPM_CONFIG_PREFER_OFFLINE=true` 且网络重试为 10 次,命中镜像 cache 时只做干净解包,lock 变化时允许补齐差量。不在镜像内烘入 `node_modules`,也不挂载跨 PR 可写缓存。任何 job 的 sandbox canary 失败都必须停止,不允许跳过。Cargo 通过受控 proxy 下载 lock 差量时继续关闭 HTTP multiplexing 并设置 `CARGO_NET_RETRY=10`。 +四个 job 先运行镜像内 `genarrative-gitea-checkout`,再以 `GENARRATIVE_GITEA_CI_CHECK_RUNTIME=1` 执行 `scripts/check-gitea-ci-job-image.sh`,校验 Node 主版本、仓库 Rust toolchain、受信任 PATH、五份缓存锁命中状态、原生命令、pkg-config 依赖、完整 bwrap sandbox 和 Chrome headless。运行时发现锁不匹配时必须输出对应 `*_cache_lock=partial` 和 Actions warning,提示可信分支落地后刷新镜像,不能把陈旧缓存误报为闭合。`RUSTUP_AUTO_INSTALL=0`,因此仓库 `rust-toolchain.toml` 变更必须先更新镜像,不能让 job 现场下载。每个 job 仍独立运行 `npm ci`,以当前 lockfile 为准验证 PR 依赖;统一通过 `scripts/ci-npm-ci-with-retry.sh` 做最多 3 次整命令级有界重试,同时保留 `NPM_CONFIG_PREFER_OFFLINE=true` 和 npm 自身 10 次 fetch retry。命中镜像 cache 时只做干净解包,lock 变化时允许补齐差量。不在镜像内烘入 `node_modules`,也不挂载跨 PR 可写缓存。任何 job 的 sandbox canary 失败都必须停止,不允许跳过。Cargo 通过受控 proxy 下载 lock 差量时继续关闭 HTTP multiplexing,并设置 `CARGO_NET_RETRY=10`。 站点 stack 仍由宿主受控目录管理,`.env`、runner 注册文件和数据库凭据不进入仓库。Compose 必须在 helper/container 内把该目录挂到与宿主相同的绝对路径再执行;挂载到不同路径会让相对 bind source 被 Docker daemon 解析到错误的宿主目录并启动空数据。升级或 runner 迁移前先停止 Gitea 写入,并把 Gitea 冷快照、数据库导出、compose/env 与 runner config/.runner 保存到仓库外受控备份位置。备份文件、绝对宿主配置和注册 token 不得提交 Git,也不在共享文档中记录具体路径或注册内容。 diff --git a/scripts/check-gitea-ci-job-image.sh b/scripts/check-gitea-ci-job-image.sh index 88753b32d..aa409dae3 100644 --- a/scripts/check-gitea-ci-job-image.sh +++ b/scripts/check-gitea-ci-job-image.sh @@ -36,6 +36,7 @@ verify_cache_lock() { fi if [[ "${GENARRATIVE_GITEA_CI_CHECK_RUNTIME:-0}" == '1' ]]; then printf '%s_cache_lock=partial\n' "${cache_name}" + printf '::warning title=CI dependency cache is partial::%s lock differs from the prebuilt image; refresh the trusted CI image after this lock change lands.\n' "${cache_name}" return fi echo "${cache_name} cache lock does not match the verification checkout." >&2 @@ -43,22 +44,34 @@ verify_cache_lock() { } npm_lock_path="${repo_root}/package-lock.json" +agc_npm_lock_path="${repo_root}/apps/ai-game-creator-shell/package-lock.json" server_rust_lock_path="${repo_root}/server-rs/Cargo.lock" desktop_rust_lock_path="${repo_root}/apps/desktop-shell/src-tauri/Cargo.lock" +agc_rust_lock_path="${repo_root}/apps/ai-game-creator-shell/src-tauri/Cargo.lock" if [[ ! -f "${npm_lock_path}" ]]; then npm_lock_path='/usr/local/share/genarrative-ci/npm/package-lock.json' fi +if [[ ! -f "${agc_npm_lock_path}" ]]; then + agc_npm_lock_path='/usr/local/share/genarrative-ci/agc-npm/package-lock.json' +fi if [[ ! -f "${server_rust_lock_path}" ]]; then server_rust_lock_path='/usr/local/share/genarrative-ci/locks/server-rs.Cargo.lock' fi if [[ ! -f "${desktop_rust_lock_path}" ]]; then desktop_rust_lock_path='/usr/local/share/genarrative-ci/locks/desktop-shell.Cargo.lock' fi +if [[ ! -f "${agc_rust_lock_path}" ]]; then + agc_rust_lock_path='/usr/local/share/genarrative-ci/locks/ai-game-creator-shell.Cargo.lock' +fi verify_cache_lock \ npm \ "${GENARRATIVE_GITEA_CI_NPM_LOCK_SHA256:-}" \ "${npm_lock_path}" +verify_cache_lock \ + agc_npm \ + "${GENARRATIVE_GITEA_CI_AGC_NPM_LOCK_SHA256:-}" \ + "${agc_npm_lock_path}" verify_cache_lock \ server_rust \ "${GENARRATIVE_GITEA_CI_SERVER_RUST_LOCK_SHA256:-}" \ @@ -67,6 +80,10 @@ verify_cache_lock \ desktop_rust \ "${GENARRATIVE_GITEA_CI_DESKTOP_RUST_LOCK_SHA256:-}" \ "${desktop_rust_lock_path}" +verify_cache_lock \ + agc_rust \ + "${GENARRATIVE_GITEA_CI_AGC_RUST_LOCK_SHA256:-}" \ + "${agc_rust_lock_path}" for command_name in \ bwrap \ diff --git a/scripts/ci-npm-ci-with-retry.sh b/scripts/ci-npm-ci-with-retry.sh new file mode 100755 index 000000000..5611d50cf --- /dev/null +++ b/scripts/ci-npm-ci-with-retry.sh @@ -0,0 +1,26 @@ +#!/usr/bin/env bash + +set -euo pipefail + +max_attempts="${GENARRATIVE_CI_NPM_CI_ATTEMPTS:-3}" +base_delay_seconds="${GENARRATIVE_CI_NPM_CI_RETRY_DELAY_SECONDS:-5}" + +if [[ ! "${max_attempts}" =~ ^[1-9][0-9]*$ ]]; then + echo 'GENARRATIVE_CI_NPM_CI_ATTEMPTS must be a positive integer.' >&2 + exit 2 +fi +if [[ ! "${base_delay_seconds}" =~ ^[0-9]+$ ]]; then + echo 'GENARRATIVE_CI_NPM_CI_RETRY_DELAY_SECONDS must be a non-negative integer.' >&2 + exit 2 +fi + +for attempt in $(seq 1 "${max_attempts}"); do + if npm ci "$@"; then + exit 0 + fi + if [[ "${attempt}" -eq "${max_attempts}" ]]; then + echo "npm ci failed after ${max_attempts} attempts." >&2 + exit 1 + fi + sleep "$((attempt * base_delay_seconds))" +done diff --git a/scripts/gitea-ci-job-image.sh b/scripts/gitea-ci-job-image.sh index bda4584e8..8554601af 100644 --- a/scripts/gitea-ci-job-image.sh +++ b/scripts/gitea-ci-job-image.sh @@ -4,7 +4,7 @@ set -euo pipefail repo_root="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd -P)" dockerfile_context_path="deploy/container/gitea-ci-job.Dockerfile" -image_tag="${GENARRATIVE_GITEA_CI_IMAGE_TAG:-genarrative/gitea-project-ci:20260723.1}" +image_tag="${GENARRATIVE_GITEA_CI_IMAGE_TAG:-genarrative/gitea-project-ci:20260807.1}" runner_container="${GENARRATIVE_GITEA_RUNNER_CONTAINER:-gitea-runner}" write_build_context_file_list() { @@ -14,6 +14,10 @@ write_build_context_file_list() { deploy/container/gitea-ci-checkout.sh \ package.json \ package-lock.json \ + apps/ai-game-creator-shell/package.json \ + apps/ai-game-creator-shell/package-lock.json \ + apps/ai-game-creator-shell/src-tauri/Cargo.toml \ + apps/ai-game-creator-shell/src-tauri/Cargo.lock \ server-rs/Cargo.toml \ server-rs/Cargo.lock \ apps/desktop-shell/src-tauri/Cargo.toml \ @@ -61,6 +65,10 @@ case "${command_name}" in deploy/container/gitea-ci-checkout.sh \ package.json \ package-lock.json \ + apps/ai-game-creator-shell/package.json \ + apps/ai-game-creator-shell/package-lock.json \ + apps/ai-game-creator-shell/src-tauri/Cargo.toml \ + apps/ai-game-creator-shell/src-tauri/Cargo.lock \ server-rs/Cargo.toml \ server-rs/Cargo.lock \ apps/desktop-shell/src-tauri/Cargo.toml \ @@ -74,10 +82,14 @@ case "${command_name}" in )" npm_lock_sha256="$(sha256sum "${repo_root}/package-lock.json")" npm_lock_sha256="${npm_lock_sha256%% *}" + agc_npm_lock_sha256="$(sha256sum "${repo_root}/apps/ai-game-creator-shell/package-lock.json")" + agc_npm_lock_sha256="${agc_npm_lock_sha256%% *}" server_rust_lock_sha256="$(sha256sum "${repo_root}/server-rs/Cargo.lock")" server_rust_lock_sha256="${server_rust_lock_sha256%% *}" desktop_rust_lock_sha256="$(sha256sum "${repo_root}/apps/desktop-shell/src-tauri/Cargo.lock")" desktop_rust_lock_sha256="${desktop_rust_lock_sha256%% *}" + agc_rust_lock_sha256="$(sha256sum "${repo_root}/apps/ai-game-creator-shell/src-tauri/Cargo.lock")" + agc_rust_lock_sha256="${agc_rust_lock_sha256%% *}" ( cd "${repo_root}" write_build_context_file_list \ @@ -86,8 +98,10 @@ case "${command_name}" in --pull=false \ --build-arg "IMAGE_REVISION=${image_revision}" \ --build-arg "NPM_LOCK_SHA256=${npm_lock_sha256}" \ + --build-arg "AGC_NPM_LOCK_SHA256=${agc_npm_lock_sha256}" \ --build-arg "SERVER_RUST_LOCK_SHA256=${server_rust_lock_sha256}" \ --build-arg "DESKTOP_RUST_LOCK_SHA256=${desktop_rust_lock_sha256}" \ + --build-arg "AGC_RUST_LOCK_SHA256=${agc_rust_lock_sha256}" \ --file "${dockerfile_context_path}" \ --tag "${image_tag}" \ - diff --git a/scripts/project-ci-workflow.test.ts b/scripts/project-ci-workflow.test.ts index cd0e94ee9..f393d30f7 100644 --- a/scripts/project-ci-workflow.test.ts +++ b/scripts/project-ci-workflow.test.ts @@ -7,6 +7,61 @@ const workflow = readFileSync( resolve(process.cwd(), '.gitea/workflows/project-ci.yml'), 'utf8', ); +const imageBuildScript = readFileSync( + resolve(process.cwd(), 'scripts/gitea-ci-job-image.sh'), + 'utf8', +); +const imageCheckScript = readFileSync( + resolve(process.cwd(), 'scripts/check-gitea-ci-job-image.sh'), + 'utf8', +); +const npmCiRetryScript = readFileSync( + resolve(process.cwd(), 'scripts/ci-npm-ci-with-retry.sh'), + 'utf8', +); +const imageDockerfile = readFileSync( + resolve(process.cwd(), 'deploy/container/gitea-ci-job.Dockerfile'), + 'utf8', +); +const imageDockerignore = readFileSync( + resolve( + process.cwd(), + 'deploy/container/gitea-ci-job.Dockerfile.dockerignore', + ), + 'utf8', +); + +const jobNames = [ + 'repository-checks', + 'frontend-tests', + 'backend-tests', + 'native-shell-tests', +] as const; + +function jobSection(jobName: (typeof jobNames)[number]) { + const jobStart = workflow.indexOf(` ${jobName}:`); + expect(jobStart).toBeGreaterThanOrEqual(0); + + const nextJobOffset = workflow + .slice(jobStart + 1) + .search(/^ [a-z][a-z0-9-]+:$/m); + return workflow.slice( + jobStart, + nextJobOffset < 0 ? undefined : jobStart + 1 + nextJobOffset, + ); +} + +function stepSection(jobName: (typeof jobNames)[number], stepName: string) { + const job = jobSection(jobName); + const stepStart = job.indexOf(` - name: ${stepName}`); + expect(stepStart).toBeGreaterThanOrEqual(0); + + const nextStepOffset = job.slice(stepStart + 1).search(/^ - name: /m); + return job.slice( + stepStart, + nextStepOffset < 0 ? undefined : stepStart + 1 + nextStepOffset, + ); +} function backendStepIndex(stepName: string) { const backendJobStart = workflow.indexOf(' backend-tests:'); @@ -20,6 +75,122 @@ function backendStepIndex(stepName: string) { } describe('project CI workflow', () => { + it('keeps every job on the isolated preinstalled CI image boundary', () => { + expect(workflow.match(/^ runs-on: genarrative-ci$/gm)).toHaveLength(4); + expect(workflow).not.toContain('actions/checkout'); + expect(workflow).not.toContain('actions/setup-node'); + expect(workflow).not.toMatch(/^\s+run: .*\b(?:apt|rustup)\b/m); + + for (const jobName of jobNames) { + const job = jobSection(jobName); + const checkout = job.indexOf('genarrative-gitea-checkout'); + const validateImage = job.indexOf( + 'GENARRATIVE_GITEA_CI_CHECK_RUNTIME=1 bash scripts/check-gitea-ci-job-image.sh', + ); + const installDependencies = job.indexOf( + '- name: Install npm dependencies', + ); + + expect(checkout).toBeGreaterThanOrEqual(0); + expect(validateImage).toBeGreaterThan(checkout); + expect(installDependencies).toBeGreaterThan(validateImage); + } + + expect(imageDockerfile).toMatch( + /^ARG RUST_IMAGE=[^\s]+@sha256:[a-f0-9]{64}$/m, + ); + expect(imageDockerfile).toMatch( + /^ARG RUNNER_IMAGE=[^\s]+@sha256:[a-f0-9]{64}$/m, + ); + }); + + it('retries every root and AI game creator npm clean install as a bounded whole command', () => { + for (const jobName of jobNames) { + const install = stepSection(jobName, 'Install npm dependencies'); + expect(install).toContain('bash scripts/ci-npm-ci-with-retry.sh'); + expect(install).not.toContain('--prefix'); + } + + for (const jobName of ['frontend-tests', 'native-shell-tests'] as const) { + const install = stepSection( + jobName, + 'Install AI game creator dependencies', + ); + expect(install).toContain( + 'bash scripts/ci-npm-ci-with-retry.sh --prefix apps/ai-game-creator-shell', + ); + } + + expect(npmCiRetryScript).toContain( + 'max_attempts="${GENARRATIVE_CI_NPM_CI_ATTEMPTS:-3}"', + ); + expect(npmCiRetryScript).toContain( + 'base_delay_seconds="${GENARRATIVE_CI_NPM_CI_RETRY_DELAY_SECONDS:-5}"', + ); + expect(npmCiRetryScript).toContain( + 'for attempt in $(seq 1 "${max_attempts}"); do', + ); + expect(npmCiRetryScript).toContain('if npm ci "$@"; then'); + expect(npmCiRetryScript).toContain( + 'if [[ "${attempt}" -eq "${max_attempts}" ]]; then', + ); + }); + + it('builds and verifies image caches against both AI game creator locks', () => { + const npmManifest = 'apps/ai-game-creator-shell/package.json'; + const npmLock = 'apps/ai-game-creator-shell/package-lock.json'; + const rustManifest = 'apps/ai-game-creator-shell/src-tauri/Cargo.toml'; + const rustLock = 'apps/ai-game-creator-shell/src-tauri/Cargo.lock'; + + for (const [path, expectedCount] of [ + [npmManifest, 2], + [npmLock, 3], + [rustManifest, 2], + [rustLock, 3], + ] as const) { + expect(imageBuildScript.split(path)).toHaveLength(expectedCount + 1); + expect(imageDockerignore).toContain(`!${path}`); + } + + expect(imageBuildScript).toContain( + '--build-arg "AGC_NPM_LOCK_SHA256=${agc_npm_lock_sha256}"', + ); + expect(imageBuildScript).toContain( + '--build-arg "AGC_RUST_LOCK_SHA256=${agc_rust_lock_sha256}"', + ); + expect(imageDockerfile).toContain('ARG AGC_NPM_LOCK_SHA256'); + expect(imageDockerfile).toContain('ARG AGC_RUST_LOCK_SHA256'); + expect(imageDockerfile).toContain( + '--prefix /usr/local/share/genarrative-ci/agc-npm', + ); + expect( + imageDockerfile.match( + /--manifest-path \/tmp\/genarrative-cargo-cache\/apps\/ai-game-creator-shell\/src-tauri\/Cargo\.toml/g, + ), + ).toHaveLength(1); + expect(imageDockerfile).toContain( + 'cargo_fetch_with_retry /tmp/genarrative-cargo-cache/apps/ai-game-creator-shell/src-tauri/Cargo.toml', + ); + expect(imageDockerfile).toContain( + 'GENARRATIVE_GITEA_CI_AGC_NPM_LOCK_SHA256=${AGC_NPM_LOCK_SHA256}', + ); + expect(imageDockerfile).toContain( + 'GENARRATIVE_GITEA_CI_AGC_RUST_LOCK_SHA256=${AGC_RUST_LOCK_SHA256}', + ); + + expect(imageCheckScript).toContain(npmLock); + expect(imageCheckScript).toContain(rustLock); + expect(imageCheckScript).toContain( + '${GENARRATIVE_GITEA_CI_AGC_NPM_LOCK_SHA256:-}', + ); + expect(imageCheckScript).toContain( + '${GENARRATIVE_GITEA_CI_AGC_RUST_LOCK_SHA256:-}', + ); + expect(imageCheckScript).toContain( + '::warning title=CI dependency cache is partial::', + ); + }); + it('prepares locked server-rs dependencies before the first Cargo build gate', () => { const prepareDependencies = backendStepIndex( 'Prepare server-rs Rust dependencies',