memind/scripts/rollback-portal-runtime-prod.sh

#!/usr/bin/env bash
set -euo pipefail

ROOT="$(cd "$(dirname "$0")/.." && pwd)"
HOST="${STUDIO_HOST:-58.38.22.103}"
REMOTE_ROOT="${STUDIO_REMOTE_ROOT:-/Users/john/Project}"
APP_DIR="${REMOTE_ROOT}/Memind"
BACKUP_DIR="${REMOTE_ROOT}/backups/memind"
ARCHIVE_DIR="${REMOTE_ROOT}/archives"
HEALTH_URL="${STUDIO_HEALTH_URL:-http://127.0.0.1:8081/api/status}"
PORTAL_LABEL="cn.tkmind.memind-portal"
LAUNCHD_GUI="gui/$(id -u)"
AUTO_YES=0
RELEASE_ID=""

usage() {
  cat <<'EOF'
用法:
  bash scripts/rollback-portal-runtime-prod.sh <RELEASE_ID> [--yes]
  bash scripts/rollback-portal-runtime-prod.sh --list [--limit N]

说明:
  将 103 Portal 回退到指定 release 发布前的 live 目录。

  RELEASE_ID 形如 20260701-075736-16828a5（与 release-portal-runtime-prod.sh 输出一致）。

  回退优先级:
    1. /Users/john/Project/archives/Memind-source-before-<RELEASE_ID>
    2. /Users/john/Project/backups/memind/memind-full-<RELEASE_ID>-before.tar.gz

  当前失败的 live 目录会移入 archive，便于事后排查。

示例:
  bash scripts/rollback-portal-runtime-prod.sh --list
  bash scripts/rollback-portal-runtime-prod.sh 20260701-075736-16828a5 --yes
EOF
}

while [[ $# -gt 0 ]]; do
  case "$1" in
    --yes|-y) AUTO_YES=1 ;;
    --list)
      ssh -o BatchMode=yes "${HOST}" "bash -s" <<'REMOTE'
set -euo pipefail
echo "=== 当前 live manifest ==="
if [[ -f /Users/john/Project/Memind/.release-manifest.txt ]]; then
  cat /Users/john/Project/Memind/.release-manifest.txt
else
  echo "(无 manifest)"
fi
echo
echo "=== 可回退 archive（最近 10 条）==="
ls -1dt /Users/john/Project/archives/Memind-source-before-* 2>/dev/null | head -10 || true
echo
echo "=== 可回退 full backup（最近 10 条）==="
ls -1dt /Users/john/Project/backups/memind/memind-full-*-before.tar.gz 2>/dev/null | head -10 || true
REMOTE
      exit 0
      ;;
    -h|--help)
      usage
      exit 0
      ;;
    *)
      if [[ -z "${RELEASE_ID}" ]]; then
        RELEASE_ID="$1"
      else
        echo "未知参数: $1" >&2
        usage >&2
        exit 1
      fi
      ;;
  esac
  shift
done

if [[ -z "${RELEASE_ID}" ]]; then
  usage >&2
  exit 1
fi

say() {
  printf '\n[%s] %s\n' "$(date +%H:%M:%S)" "$*"
}

say "103 回退预检查"
ssh -o BatchMode=yes -o ConnectTimeout=15 "${HOST}" "echo rollback-ssh-ok" >/dev/null

ARCHIVED_SOURCE="${ARCHIVE_DIR}/Memind-source-before-${RELEASE_ID}"
FULL_BACKUP="${BACKUP_DIR}/memind-full-${RELEASE_ID}-before.tar.gz"

ssh -o BatchMode=yes "${HOST}" "test -d '${ARCHIVED_SOURCE}' || test -f '${FULL_BACKUP}'" || {
  echo "找不到 release ${RELEASE_ID} 的回退源：" >&2
  echo "  ${ARCHIVED_SOURCE}" >&2
  echo "  ${FULL_BACKUP}" >&2
  echo "可先运行: bash scripts/rollback-portal-runtime-prod.sh --list" >&2
  exit 1
}

if [[ "${AUTO_YES}" -ne 1 ]]; then
  say "回退确认"
  echo "目标主机: ${HOST}"
  echo "回退 release: ${RELEASE_ID}"
  echo "当前 live: ${APP_DIR}"
  ssh -o BatchMode=yes "${HOST}" "
    if [[ -d '${ARCHIVED_SOURCE}' ]]; then echo 'source=archived_source'; fi
    if [[ -f '${FULL_BACKUP}' ]]; then echo 'source=full_backup'; fi
    if [[ -f '${APP_DIR}/.release-manifest.txt' ]]; then echo '--- current manifest ---'; cat '${APP_DIR}/.release-manifest.txt'; fi
  "
  read -r -p "确认回退到 ${RELEASE_ID} 发布前版本? [y/N] " confirm </dev/tty
  [[ "${confirm}" =~ ^[Yy]$ ]] || exit 0
fi

say "在 103 执行回退"
ssh -o BatchMode=yes "${HOST}" \
  "RELEASE_ID='${RELEASE_ID}' APP_DIR='${APP_DIR}' ARCHIVED_SOURCE='${ARCHIVED_SOURCE}' FULL_BACKUP='${FULL_BACKUP}' ARCHIVE_DIR='${ARCHIVE_DIR}' HEALTH_URL='${HEALTH_URL}' PORTAL_LABEL='${PORTAL_LABEL}' LAUNCHD_GUI='${LAUNCHD_GUI}' /bin/bash" <<'REMOTE_SCRIPT'
set -euo pipefail

ROLLBACK_TS="$(date +%Y%m%d-%H%M%S)"
FAILED_LIVE="${ARCHIVE_DIR}/Memind-failed-after-${RELEASE_ID}-${ROLLBACK_TS}"
RESTORE_DIR="${APP_DIR}.rollback-${ROLLBACK_TS}"

say() {
  printf '\n[%s] %s\n' "$(date +%H:%M:%S)" "$*"
}

say "停止 Portal"
launchctl bootout "${LAUNCHD_GUI}/${PORTAL_LABEL}" >/dev/null 2>&1 || true
lsof -tiTCP:8081 -sTCP:LISTEN 2>/dev/null | xargs kill 2>/dev/null || true
sleep 2
lsof -tiTCP:8081 -sTCP:LISTEN 2>/dev/null | xargs kill -9 2>/dev/null || true

say "准备回退目录"
rm -rf "${RESTORE_DIR}"
mkdir -p "${RESTORE_DIR}"

if [[ -d "${ARCHIVED_SOURCE}" ]]; then
  say "从 archived_source 恢复: ${ARCHIVED_SOURCE}"
  cp -a "${ARCHIVED_SOURCE}/." "${RESTORE_DIR}/"
elif [[ -f "${FULL_BACKUP}" ]]; then
  say "从 full backup 恢复: ${FULL_BACKUP}"
  tar -xzf "${FULL_BACKUP}" -C "${RESTORE_DIR}"
else
  echo "rollback source missing" >&2
  exit 1
fi

say "归档当前失败 live 目录"
if [[ -d "${APP_DIR}" ]]; then
  rm -rf "${FAILED_LIVE}"
  mv "${APP_DIR}" "${FAILED_LIVE}"
fi
mv "${RESTORE_DIR}" "${APP_DIR}"

cat > "${APP_DIR}/.rollback-manifest.txt" <<EOF
rollback_at=${ROLLBACK_TS}
rolled_back_release=${RELEASE_ID}
failed_live=${FAILED_LIVE}
restored_from=$([[ -d "${ARCHIVED_SOURCE}" ]] && echo archived_source || echo full_backup)
EOF

say "重启 Portal"
if [[ -x "${APP_DIR}/scripts/run-memind-portal-prod.sh" ]]; then
  cat > "${HOME}/Library/LaunchAgents/${PORTAL_LABEL}.plist" <<EOF
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
  <key>Label</key>
  <string>${PORTAL_LABEL}</string>
  <key>ProgramArguments</key>
  <array>
    <string>${APP_DIR}/scripts/run-memind-portal-prod.sh</string>
  </array>
  <key>WorkingDirectory</key>
  <string>${APP_DIR}</string>
  <key>EnvironmentVariables</key>
  <dict>
    <key>PATH</key>
    <string>/opt/homebrew/bin:/opt/homebrew/opt/node@24/bin:/usr/local/bin:/usr/bin:/bin</string>
  </dict>
  <key>RunAtLoad</key>
  <true/>
  <key>KeepAlive</key>
  <true/>
  <key>ThrottleInterval</key>
  <integer>10</integer>
  <key>StandardOutPath</key>
  <string>${HOME}/Library/Logs/memind-portal.log</string>
  <key>StandardErrorPath</key>
  <string>${HOME}/Library/Logs/memind-portal.log</string>
</dict>
</plist>
EOF
  launchctl bootstrap "${LAUNCHD_GUI}" "${HOME}/Library/LaunchAgents/${PORTAL_LABEL}.plist" >/dev/null 2>&1 || true
  launchctl enable "${LAUNCHD_GUI}/${PORTAL_LABEL}" >/dev/null 2>&1 || true
  launchctl kickstart -k "${LAUNCHD_GUI}/${PORTAL_LABEL}" >/dev/null 2>&1 || \
    nohup "${APP_DIR}/scripts/run-memind-portal-prod.sh" >> "${HOME}/Library/Logs/memind-portal.log" 2>&1 &
else
  nohup /opt/homebrew/opt/node@24/bin/node "${APP_DIR}/server.mjs" >> "${HOME}/Library/Logs/memind-portal.log" 2>&1 &
fi

say "健康检查"
for _ in $(seq 1 60); do
  portal_code="$(curl -s -o /dev/null -w '%{http_code}' "${HEALTH_URL}" || true)"
  if [[ "${portal_code}" == "200" ]]; then
    break
  fi
  sleep 2
done
portal_code="$(curl -s -o /dev/null -w '%{http_code}' "${HEALTH_URL}" || true)"
if [[ "${portal_code}" != "200" ]]; then
  echo "rollback health check failed: portal=${portal_code}" >&2
  exit 1
fi

say "回退完成"
printf 'live_dir=%s\n' "${APP_DIR}"
printf 'failed_live=%s\n' "${FAILED_LIVE}"
printf 'rolled_back_release=%s\n' "${RELEASE_ID}"
REMOTE_SCRIPT

say "103 回退后状态"
ssh -o BatchMode=yes "${HOST}" "curl -s '${HEALTH_URL}' && echo && cat '${APP_DIR}/.rollback-manifest.txt' 2>/dev/null || true"