#!/usr/bin/env bash
# runner-deadman —— CI runner 死信守望。CI 整条吊在一台 mac(mac-pangolin-2)上,
# mac 一关/掉线 CI 全停且无人知道。本脚本放在【另一台常在线主机】上,周期查 gitea
# Actions API 里该 runner 的 status,连续 offline 到阈值 → Telegram 告警;恢复也通知。
#
# 仿 deploy/bootstrap/monitor/deadman-watch.sh(同一发信/去抖范式),探活对象从
# 「TCP 端口」换成「gitea runner 在线状态」。
#
# 配置 /etc/pangolin-runner-deadman.env:
#   GITEA_URL=http://127.0.0.1:13000          # gitea 基址(经 relay 或直连)
#   GITEA_TOKEN=...                           # gitea 读写 key(Bitwarden「gitea 读写key」)
#   GITEA_REPO=wangjia/pangolin               # owner/repo
#   RUNNER_NAME=mac-pangolin-2                # 要守望的 runner 名
#   TELEGRAM_BOT_TOKEN=...  TELEGRAM_CHAT_ID=...
#   FAIL_THRESHOLD=2                          # 连续几次 offline 才告警(去抖)
#
# 在该常在线主机上每 3 分钟跑一次(cron,用户态即可):
#   */3 * * * * /usr/local/bin/runner-deadman >/dev/null 2>&1
# (mac 宿主可改用 launchd LaunchAgent,见 docs/ci-runner.md「韧性」节。)
set -uo pipefail

ENV_FILE=/etc/pangolin-runner-deadman.env
[ -r "$ENV_FILE" ] || { echo "缺少 $ENV_FILE" >&2; exit 1; }
# shellcheck disable=SC1090
. "$ENV_FILE"
: "${FAIL_THRESHOLD:=2}"
: "${GITEA_REPO:=wangjia/pangolin}"
: "${RUNNER_NAME:=mac-pangolin-2}"
STATE_DIR="${STATE_DIR:-$HOME/.local/state/pangolin-runner-deadman}"
mkdir -p "$STATE_DIR"
statef="${STATE_DIR}/${RUNNER_NAME}.fail"

send_tg() {
  local text="$1"
  [ -n "${TELEGRAM_BOT_TOKEN:-}" ] && [ -n "${TELEGRAM_CHAT_ID:-}" ] || { echo "$text"; return; }
  curl -fsS --max-time 15 \
    "https://api.telegram.org/bot${TELEGRAM_BOT_TOKEN}/sendMessage" \
    --data-urlencode "chat_id=${TELEGRAM_CHAT_ID}" \
    --data-urlencode "text=${text}" >/dev/null 2>&1 || echo "Telegram 发送失败" >&2
}

# 查 runner 状态。响应形如:
#   {"runners":[{"id":6,"name":"mac-pangolin-2","status":"online","busy":false,...}],...}
resp=$(curl -fsS --max-time 15 -H "Authorization: token ${GITEA_TOKEN}" \
  "${GITEA_URL%/}/api/v1/repos/${GITEA_REPO}/actions/runners" 2>/dev/null) || {
  # API 取不到(监控主机网络抖动 / relay 没起 / gitea 宕):不计入去抖,避免误报。
  echo "runner-deadman: gitea API 不可达,跳过本次(不改状态)" >&2
  exit 0
}

# 提取该 runner 的 status(优先 jq,缺 jq 退回 grep)。
if command -v jq >/dev/null 2>&1; then
  status=$(printf '%s' "$resp" | jq -r --arg n "$RUNNER_NAME" \
    '.runners[]? | select(.name==$n) | .status' 2>/dev/null)
else
  status=$(printf '%s' "$resp" \
    | grep -oE "\"name\":\"${RUNNER_NAME}\"[^}]*\"status\":\"[a-z]+\"" \
    | grep -oE '"status":"[a-z]+"' | tail -1 | cut -d'"' -f4)
fi

if [ "$status" = "online" ]; then
  if [ -f "$statef" ]; then
    send_tg "🟢 [CI runner ${RUNNER_NAME}] 恢复在线"
    rm -f "$statef"
  fi
else
  # status 为 offline / 空(runner 已从 gitea 掉线或不存在)。
  fails=$(( $(cat "$statef" 2>/dev/null || echo 0) + 1 ))
  echo "$fails" > "$statef"
  if [ "$fails" -eq "$FAIL_THRESHOLD" ]; then
    send_tg "🔴 [CI runner ${RUNNER_NAME}] 离线(status=${status:-缺失}),连续 ${fails} 次,CI 已停摆"
  fi
fi