bookworm-smart-assistant/scripts/deploy-transactional.sh

#!/usr/bin/env bash
# ═══════════════════════════════════════════════════════════════════
# B7: 事务性部署脚本模板 (Transactional Deploy)
# 用途: 在 ECS 上执行零停机部署，支持原子回滚
#
# 安全设计:
#   - 蓝绿部署: 新版本部署到 staging 目录，切换符号链接激活
#   - Gold Release: 维护已知良好版本标记，回滚始终回到 gold
#   - Migration 检测: 含 DB migration 时禁止自动回滚
#   - nohup 安全: SSH 断开不中断部署
#   - 全程审计日志
#
# 用法 (由 deploy-wrapper.sh 白名单调用):
#   bash deploy-transactional.sh deploy
#   bash deploy-transactional.sh rollback <commit-hash>
#   bash deploy-transactional.sh status
#   bash deploy-transactional.sh health
# ═══════════════════════════════════════════════════════════════════
set -euo pipefail

# ─── 配置 ─────────────────────────────────────────────────────
APP_NAME="${APP_NAME:-bookworm-web}"
APP_BASE="/var/www/${APP_NAME}"
APP_CURRENT="${APP_BASE}/current"       # 符号链接 → 当前活跃版本
APP_RELEASES="${APP_BASE}/releases"     # 版本目录
APP_SHARED="${APP_BASE}/shared"         # 共享资源 (.env, uploads, logs)
APP_REPO="${APP_BASE}/repo"             # Git 仓库
GOLD_FILE="${APP_BASE}/gold-release.txt"
DEPLOY_LOG="/var/log/${APP_NAME}-deploy.log"
MAX_RELEASES=5                          # 保留最近 N 个版本
HEALTH_URL="${HEALTH_URL:-http://localhost:3000/api/health}"
HEALTH_TIMEOUT=30                       # 健康检查超时秒数

# ─── 工具函数 ─────────────────────────────────────────────────
ts() { date '+%Y-%m-%d %H:%M:%S'; }
log() { echo "[$(ts)] $*" | tee -a "${DEPLOY_LOG}"; }
die() { log "FATAL: $*"; exit 1; }

# ─── 部署命令 ─────────────────────────────────────────────────

cmd_deploy() {
  log "═══ 开始部署 ${APP_NAME} ═══"

  # 1. 拉取最新代码
  log "Step 1/7: 拉取代码..."
  cd "${APP_REPO}"
  git fetch origin main --prune
  git reset --hard origin/main
  local COMMIT
  COMMIT=$(git rev-parse HEAD)
  local SHORT_COMMIT="${COMMIT:0:8}"
  log "目标 commit: ${SHORT_COMMIT}"

  # 2. 检测是否包含 migration
  local HAS_MIGRATION=false
  if git diff HEAD~1 --name-only 2>/dev/null | grep -qi 'migration\|migrate\|schema'; then
    HAS_MIGRATION=true
    log "WARNING: 检测到数据库 migration，回滚将受限"
  fi

  # 3. 创建新版本目录 (蓝绿)
  local RELEASE_DIR="${APP_RELEASES}/$(date +%Y%m%d_%H%M%S)_${SHORT_COMMIT}"
  log "Step 2/7: 创建版本目录 ${RELEASE_DIR}..."
  mkdir -p "${RELEASE_DIR}"
  rsync -a --exclude=node_modules --exclude=.git "${APP_REPO}/" "${RELEASE_DIR}/"

  # 4. 安装依赖 + 构建
  log "Step 3/7: 安装依赖..."
  cd "${RELEASE_DIR}"
  if [ -f "pnpm-lock.yaml" ]; then
    pnpm install --frozen-lockfile --prod 2>&1 | tail -5
  elif [ -f "package-lock.json" ]; then
    npm ci --production 2>&1 | tail -5
  fi

  log "Step 4/7: 构建..."
  if grep -q '"build"' package.json 2>/dev/null; then
    pnpm build 2>&1 | tail -10 || npm run build 2>&1 | tail -10
  fi

  # 5. 链接共享资源
  log "Step 5/7: 链接共享资源..."
  [ -f "${APP_SHARED}/.env" ] && ln -sf "${APP_SHARED}/.env" "${RELEASE_DIR}/.env"
  [ -d "${APP_SHARED}/uploads" ] && ln -sf "${APP_SHARED}/uploads" "${RELEASE_DIR}/uploads"

  # 6. 数据库 migration (如果有)
  if [ "${HAS_MIGRATION}" = true ]; then
    log "Step 5.5/7: 执行数据库 migration..."
    if [ -f "${RELEASE_DIR}/prisma/schema.prisma" ]; then
      cd "${RELEASE_DIR}" && npx prisma migrate deploy 2>&1 | tail -5
    fi
    # migration 成功后标记，用于判断是否可回滚
    echo "MIGRATION_APPLIED=${COMMIT}" >> "${RELEASE_DIR}/.deploy-meta"
  fi

  # 7. 原子切换符号链接
  log "Step 6/7: 切换活跃版本..."
  local TMP_LINK="${APP_CURRENT}.new"
  ln -sfn "${RELEASE_DIR}" "${TMP_LINK}"
  mv -Tf "${TMP_LINK}" "${APP_CURRENT}"
  log "符号链接已切换: current → ${RELEASE_DIR}"

  # 8. 重启应用
  log "Step 7/7: 重启应用..."
  if command -v pm2 &>/dev/null; then
    cd "${APP_CURRENT}" && pm2 reload ecosystem.config.js --update-env 2>&1 || \
    pm2 reload all 2>&1
  fi

  # 9. 健康检查
  log "等待应用启动..."
  local healthy=false
  for i in $(seq 1 ${HEALTH_TIMEOUT}); do
    if curl -sf --max-time 5 "${HEALTH_URL}" >/dev/null 2>&1; then
      healthy=true
      break
    fi
    sleep 1
  done

  if [ "${healthy}" = true ]; then
    log "健康检查通过！"
    # 标记为 gold release
    echo "${COMMIT}" > "${GOLD_FILE}"
    log "已标记 gold release: ${SHORT_COMMIT}"

    # 清理旧版本 (保留最近 N 个)
    cleanup_old_releases
    log "═══ 部署成功 (${SHORT_COMMIT}) ═══"
    echo "DEPLOY_SUCCESS"
  else
    log "ERROR: 健康检查超时 (${HEALTH_TIMEOUT}s)"

    # 回滚 (仅非 migration 场景)
    if [ "${HAS_MIGRATION}" = true ]; then
      log "ERROR: 含 migration，无法自动回滚。请手动处理。"
      echo "DEPLOY_FAIL_MIGRATION_NO_ROLLBACK"
      exit 1
    fi

    # 自动回滚到前一个版本
    log "正在自动回滚..."
    cmd_rollback_to_previous
    echo "DEPLOY_FAIL_ROLLED_BACK"
    exit 1
  fi
}

# ─── 回滚命令 ─────────────────────────────────────────────────

cmd_rollback() {
  local target_commit="${1:-}"

  if [ -z "${target_commit}" ]; then
    # 无指定 commit → 回滚到 gold release
    if [ -f "${GOLD_FILE}" ]; then
      target_commit=$(cat "${GOLD_FILE}")
      log "回滚到 gold release: ${target_commit:0:8}"
    else
      die "无 gold release 记录，无法回滚"
    fi
  fi

  # 校验 commit hash 格式
  if [[ ! "${target_commit}" =~ ^[0-9a-f]{40}$ ]]; then
    die "无效的 commit hash: ${target_commit}"
  fi

  # 查找对应的版本目录
  local target_dir=""
  for dir in "${APP_RELEASES}"/*/; do
    if [[ "${dir}" == *"${target_commit:0:8}"* ]]; then
      target_dir="${dir%/}"
      break
    fi
  done

  if [ -n "${target_dir}" ] && [ -d "${target_dir}" ]; then
    # 检查目标版本是否含 migration
    if [ -f "${target_dir}/.deploy-meta" ] && grep -q "MIGRATION_APPLIED" "${target_dir}/.deploy-meta"; then
      log "WARNING: 目标版本含已执行的 migration，回滚可能导致 schema 不一致"
    fi

    # 切换符号链接
    local TMP_LINK="${APP_CURRENT}.new"
    ln -sfn "${target_dir}" "${TMP_LINK}"
    mv -Tf "${TMP_LINK}" "${APP_CURRENT}"
    pm2 reload all 2>&1 || true
    log "已回滚到: ${target_dir}"
    echo "ROLLBACK_SUCCESS"
  else
    # 版本目录不存在，从 git 恢复
    log "版本目录不存在，从 git checkout 恢复..."
    cd "${APP_REPO}" && git checkout "${target_commit}" -- .
    cmd_deploy
  fi
}

cmd_rollback_to_previous() {
  # 获取当前版本之前的最新版本目录
  local current_target
  current_target=$(readlink -f "${APP_CURRENT}" 2>/dev/null || echo "")
  local prev_dir=""

  for dir in $(ls -dt "${APP_RELEASES}"/*/); do
    dir="${dir%/}"
    if [ "${dir}" != "${current_target}" ]; then
      prev_dir="${dir}"
      break
    fi
  done

  if [ -n "${prev_dir}" ]; then
    log "回滚到前一版本: ${prev_dir}"
    local TMP_LINK="${APP_CURRENT}.new"
    ln -sfn "${prev_dir}" "${TMP_LINK}"
    mv -Tf "${TMP_LINK}" "${APP_CURRENT}"
    pm2 reload all 2>&1 || true
    echo "ROLLBACK_SUCCESS"
  elif [ -f "${GOLD_FILE}" ]; then
    log "无前一版本，回滚到 gold release"
    cmd_rollback "$(cat "${GOLD_FILE}")"
  else
    die "无可用的回滚目标"
  fi
}

# ─── 状态/健康命令 ────────────────────────────────────────────

cmd_status() {
  echo "=== ${APP_NAME} 部署状态 ==="
  echo "当前版本: $(readlink -f "${APP_CURRENT}" 2>/dev/null || echo 'N/A')"
  echo "Gold release: $(cat "${GOLD_FILE}" 2>/dev/null | head -c 8 || echo 'N/A')"
  echo "可用版本:"
  ls -dt "${APP_RELEASES}"/*/ 2>/dev/null | head -${MAX_RELEASES} | while read -r d; do
    echo "  - $(basename "${d}")"
  done
  echo "PM2 状态:"
  pm2 status 2>&1 || echo "  pm2 不可用"
}

cmd_health() {
  local code
  code=$(curl -sf -o /dev/null -w "%{http_code}" --max-time 10 "${HEALTH_URL}" 2>/dev/null || echo "000")
  if [ "${code}" = "200" ]; then
    echo "HEALTH_OK (${HEALTH_URL} → ${code})"
  else
    echo "HEALTH_FAIL (${HEALTH_URL} → ${code})"
    exit 1
  fi
}

# ─── 清理旧版本 ──────────────────────────────────────────────

cleanup_old_releases() {
  local count
  count=$(ls -d "${APP_RELEASES}"/*/ 2>/dev/null | wc -l)
  if [ "${count}" -gt "${MAX_RELEASES}" ]; then
    local to_remove=$((count - MAX_RELEASES))
    log "清理 ${to_remove} 个旧版本..."
    ls -dt "${APP_RELEASES}"/*/ | tail -${to_remove} | while read -r d; do
      # 不删除 gold release 对应的版本
      local gold_commit
      gold_commit=$(cat "${GOLD_FILE}" 2>/dev/null || echo "")
      if [ -n "${gold_commit}" ] && [[ "${d}" == *"${gold_commit:0:8}"* ]]; then
        log "保留 gold release 版本: $(basename "${d}")"
        continue
      fi
      log "删除旧版本: $(basename "${d}")"
      rm -rf "${d}"
    done
  fi
}

# ─── 入口 ─────────────────────────────────────────────────────

case "${1:-}" in
  deploy)       cmd_deploy ;;
  rollback)     cmd_rollback "${2:-}" ;;
  status)       cmd_status ;;
  health)       cmd_health ;;
  *)
    echo "用法: $0 {deploy|rollback [commit]|status|health}"
    exit 1
    ;;
esac