修复 CI 领取冲突导致网关永久锁定
Project CI / AI game creator shell Rust lane 1/2 (pull_request) Has been cancelled
Project CI / AI game creator shell Rust lane 2/2 (pull_request) Has been cancelled
Project CI / AI game creator shell Rust crates (pull_request) Has been cancelled
Project CI / Backend tests (pull_request) Has been cancelled
Project CI / Native shell tests (pull_request) Has been cancelled
Project CI / Frontend tests (pull_request) Has been cancelled
Project CI / Repository checks (pull_request) Has been cancelled
Project CI / AI game creator shell web tests (pull_request) Has been cancelled

识别 Gitea 已回滚的任务分配冲突,允许 runner 正常重试。
保留未知响应与未完成领取的保护,增加脱敏原因日志。
补充冲突重试及错误边界测试,同步部署文档与共享记忆。
This commit is contained in:
2026-10-07 09:20:04 +00:00
parent e6933762d2
commit d7ddf5fbc1
4 changed files with 89 additions and 5 deletions
@@ -112,6 +112,8 @@ Gitea 基础镜像通过专用 `genarrative-ci-images` Buildx builder 持久复
Gitea Rust 缓存自动维护由宿主 `genarrative-ci-cache.timer` 收集同一 master push run 六个 Rust job 的原生 V4 缓存产物,不重复执行 Cargo 预热。只传本轮新 key,命中对象只传使用时间;宿主与真实来源镜像对象合并、去重、按新近使用时间裁剪到 4 GiB,从无对象缓存基础镜像重新组装。源 run 不要求全绿,但取消、缺组、旧 attempt、未完成上传或混用来源镜像不得采用。网关暂停新 FetchTask、在途领取结束、持久化任务账本清空且内层活动容器为空才切换,不打断运行中的 CI。首次接入/升级网关需空闲窗口;Token 只需普通仓库 `write:repository`,不查管理员 API。候选装载后清理已收集 artifact,遗留项保留 7 天;真实 master CI 验证后才清理旧镜像,保留当前、一个回滚版、基础镜像及容器引用。部署入口见 `deploy/container/README.md`,合并代码不等于服务启用。
领取网关只对已核对 Gitea 1.26.4 事务回滚的精确 `CreateTaskForRunner: update run <id>: run has changed` Connect 错误允许 runner 重试,避免普通并发更新冲突永久阻断 CI;完整响应格式约束见 `deploy/container/README.md`。其它未知领取结果继续持久化保护,不自动清除 `uncertain`;恢复既有锁前需核实任务与在途请求收敛。容器日志只记录固定原因分类,不打印请求、响应或认证信息。
修改 Gitea workflow 的 job 显示名称、ID 或缓存导出组时,必须同步维护器的 `JOBS` / `RUST_JOB_IDS`;`test_gitea_cache_maintenance.py` 直接对照实际 workflow 检查全集和导出映射,避免自动刷新或镜像验收因名单漂移长期等待。维护器 `Api.request` 的 `method` 是必填关键字参数,GET 也必须显式指定,不根据 body 推断请求方法。
Gitea 缓存部署必须区分网络:runner 的 RPC 走 `gitea-runner-fetch-gate:8080`;内层 job 的 checkout/上传走映射到 `172.30.0.3` 的 `http://genarrative-station/git`;宿主专用 clone 走 `http://127.0.0.1:3003`。不要把 runner 可达的 `gitea:3000` 配给 job。内层 Docker 使用 `10.240.0.0/16`、每 job `/24` 的默认地址池,避开外层 `172.30/172.31` 网段;恢复领取前必须在真实 job 网络里验证 checkout 与 Gitea API,不能只验证 FetchTask。具体配置与遗留空网络处理见 `deploy/container/README.md`。