e7b37d7719
ci-pangolin / Lint — shellcheck (pull_request) Successful in 9s
ci-pangolin / OpenAPI Sync Check (pull_request) Successful in 23s
ci-pangolin / Redline Scan — 脱敏 (UI 文案) (pull_request) Successful in 5s
ci-pangolin / Flutter — analyze + test (pull_request) Failing after 37s
ci-pangolin / Portable SQL — 可移植性 (mysql/sqlite) (pull_request) Successful in 6s
ci-pangolin / Codegen Drift — token 生成物未漂移 (pull_request) Successful in 4s
ci-pangolin / Go — build + test (pull_request) Successful in 19s
ci-pangolin / E2E Smoke — L4 进程级端到端 (pull_request) Successful in 17s
ci-pangolin / Go — integration (mysql/redis testcontainers) (pull_request) Successful in 4m16s
ci-pangolin / Golden — 视觉回归 (components + auth) (pull_request) Successful in 15s
ci-pangolin / Lint — shellcheck (push) Successful in 5s
ci-pangolin / OpenAPI Sync Check (push) Successful in 23s
ci-pangolin / Redline Scan — 脱敏 (UI 文案) (push) Successful in 6s
ci-pangolin / Flutter — analyze + test (push) Failing after 15s
ci-pangolin / Portable SQL — 可移植性 (mysql/sqlite) (push) Successful in 5s
ci-pangolin / Codegen Drift — token 生成物未漂移 (push) Successful in 5s
ci-pangolin / Go — build + test (push) Successful in 7s
ci-pangolin / E2E Smoke — L4 进程级端到端 (push) Successful in 19s
ci-pangolin / Go — integration (mysql/redis testcontainers) (push) Failing after 4m29s
ci-pangolin / Golden — 视觉回归 (components + auth) (push) Successful in 13s
CI 整条吊在一台 mac,关机/掉线即全停且无人知道。新增
deploy/runner/runner-deadman.sh(仿 bootstrap/monitor/deadman-watch.sh):
放另一台常在线主机,每 3 分钟查 gitea Actions API 里 mac-pangolin-2 的 status,
连续 offline 到阈值 → Telegram 告警,恢复也通知;API 取不到时跳过不误报。
响应字段已对照真 gitea API 核验({"runners":[{"name":...,"status":"online"}]}),
shellcheck 干净,实跑验证 online→不告警、缺失→进告警分支。
docs/ci-runner.md 增「韧性」节(告警部署 + cron/launchd)并标注持久化激活仍待
本机手动跑(pkill/launchctl,Claude 跑不了);验证段 job 数 9→10。
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
97 lines
4.8 KiB
Markdown
97 lines
4.8 KiB
Markdown
# Pangolin CI Runner 记录
|
||
|
||
> gitea Actions self-hosted runner 设置记录。CI 工作流见 `.gitea/workflows/ci.yml`(9 个 job)。
|
||
|
||
## 决定:用 1 个 runner
|
||
|
||
| 项 | 值 |
|
||
|---|---|
|
||
| 名称 | **mac-pangolin-2**(gitea id 6) |
|
||
| 级别 | **项目级**(绑 `wangjia/pangolin`,只服务本仓库) |
|
||
| 机器 | mac 本机(`/Users/wangjia/bin/act_runner` v0.6.1) |
|
||
| 执行模式 | **host**(非容器) |
|
||
| label | `nas`(即 `runs-on: nas`) |
|
||
| 连接 | 经 jiu 的 TCP relay `127.0.0.1:13000` → NAS gitea `192.168.3.200:3000` |
|
||
|
||
### 为什么是 host 模式 + mac
|
||
- ci.yml 的 9 个 job **全是 docker-in-docker**(每个 job 内部 `docker run …` 拉 golang/flutter/redis 等镜像)。container 模式(像 jiu 的 `ubuntu-latest:docker://catthehacker`)在非特权容器里跑不了 DinD,所以必须 **host 模式**。
|
||
- mac 的 Docker Desktop 能跑 Linux 容器,host 模式下 `docker run <linux-image>` 可用。
|
||
- 理想归宿其实是 **NAS(Linux)host runner**(给 NAS 现有 runner 加 `nas:host` label),mac 只是当前落点。
|
||
|
||
### 为什么要 relay(13000)
|
||
mac 上的 Shadowrocket 网络扩展会拦截 act_runner 这个 Go 二进制的流量,直连 NAS gitea 失败。jiu 已有一个常驻 python TCP relay(`~/bin/tcp_relay.py`,`127.0.0.1:13000 → 192.168.3.200:3000`),act_runner 连本地 relay 绕过。pangolin runner **复用**这个 relay。
|
||
|
||
## 配置文件:`~/.act_runner_pangolin2_config.yaml`
|
||
```yaml
|
||
log:
|
||
level: info
|
||
runner:
|
||
file: /Users/wangjia/.act_runner_pangolin2 # 注册凭证(.runner),勿入 git
|
||
capacity: 1
|
||
timeout: 3h
|
||
labels:
|
||
- "nas:host"
|
||
cache:
|
||
enabled: true
|
||
host:
|
||
workdir_parent: /Users/wangjia/.cache/act-pangolin2
|
||
```
|
||
|
||
## 持久化(launchd 开机自启)
|
||
仿 jiu 的 `com.jiu.act-runner`。文件在仓库 `deploy/runner/`:
|
||
- `act_runner_pangolin_start.sh` —— 启动脚本(等 jiu 的 relay 13000 就绪 → 起 daemon → 退出自重启)。
|
||
- `com.pangolin.act-runner.plist` —— launchd(RunAtLoad + KeepAlive)。
|
||
|
||
**部署步骤**(一次性):
|
||
```bash
|
||
# 1. 文件就位
|
||
cp deploy/runner/act_runner_pangolin_start.sh ~/bin/ && chmod +x ~/bin/act_runner_pangolin_start.sh
|
||
cp deploy/runner/com.pangolin.act-runner.plist ~/Library/LaunchAgents/
|
||
# 2. 停掉临时 daemon(否则和 launchd 起的抢同一注册)
|
||
pkill -f act_runner_pangolin2_config.yaml
|
||
# 3. 加载常驻服务(开机自启 + 崩溃自重启)
|
||
launchctl load ~/Library/LaunchAgents/com.pangolin.act-runner.plist
|
||
```
|
||
之后 mac 重启 / 退出登录都会自动拉起。日志 `/tmp/act_runner_pangolin2.log`。
|
||
卸载:`launchctl unload ~/Library/LaunchAgents/com.pangolin.act-runner.plist`。
|
||
|
||
> 依赖 jiu 的 relay(13000)常驻:jiu 的 `com.jiu.act-runner` 在 → relay 在 → 本 runner 能连 NAS gitea。
|
||
|
||
> ⚠️ **持久化激活状态**:上面三条一次性命令含 `pkill`/`launchctl`,需在 mac 本机手动执行
|
||
> (Claude 跑不了)。未激活前 runner 是临时 daemon,mac 重启即停。激活后用下面的离线告警兜底。
|
||
|
||
## 韧性:runner 离线告警(单点兜底)
|
||
|
||
CI 整条吊在这一台 mac 上,mac 关机/掉线 → CI 全停且**无人知道**。脚本
|
||
`deploy/runner/runner-deadman.sh`(仿 `bootstrap/monitor/deadman-watch.sh`)放在**另一台
|
||
常在线主机**上,每 3 分钟查 gitea Actions API 里 `mac-pangolin-2` 的 `status`,连续
|
||
offline 到阈值 → Telegram 告警,恢复也通知;API 取不到时跳过不误报。
|
||
|
||
**部署**(在常在线主机,如 jiu 的 mac / NAS):
|
||
```bash
|
||
cp deploy/runner/runner-deadman.sh /usr/local/bin/ && chmod +x /usr/local/bin/runner-deadman
|
||
# /etc/pangolin-runner-deadman.env(不入 git):
|
||
# GITEA_URL=http://127.0.0.1:13000 GITEA_TOKEN=<Bitwarden「gitea 读写key」>
|
||
# GITEA_REPO=wangjia/pangolin RUNNER_NAME=mac-pangolin-2
|
||
# TELEGRAM_BOT_TOKEN=... TELEGRAM_CHAT_ID=... FAIL_THRESHOLD=2
|
||
# cron(每 3 分钟):
|
||
echo '*/3 * * * * /usr/local/bin/runner-deadman >/dev/null 2>&1' | crontab -
|
||
```
|
||
mac 宿主可改 launchd LaunchAgent(`StartInterval` 180 + 同一脚本),仿
|
||
`com.pangolin.act-runner.plist`。响应字段已对照真 API 核验:
|
||
`{"runners":[{"name":"mac-pangolin-2","status":"online",...}]}`。
|
||
|
||
## 已删除:mac-pangolin(第一次注册的,id 5)
|
||
第一次注册的 runner(label `nas`,config `~/.act_runner_pangolin_config.yaml`)。job 调试时已从 **gitea 删除注册**(API DELETE,204)。本机残留待清:
|
||
```bash
|
||
# orphan daemon 进程(拿不到任务、空转报 "unregistered runner",无害但占进程)
|
||
pkill -f 'act_runner_pangolin_config.yaml' # 只匹配 orphan,不碰 -2 和 jiu
|
||
# orphan 文件
|
||
rm -f ~/.act_runner_pangolin ~/.act_runner_pangolin_config.yaml
|
||
rm -rf ~/.cache/act-pangolin
|
||
```
|
||
|
||
## 验证
|
||
- gitea → 仓库 Settings → Actions → Runners 应只见 **mac-pangolin-2**(online)。
|
||
- 向 `main` 开 PR 触发 10 个 job(后增覆盖率闸 + go-integration);首次全绿见 run #298(commit `65bd7c3`)。
|