order_site/deploy/backup-db.sh

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

# 数据库备份:postgres 容器内 pg_dump(custom 格式,自带压缩,MVCC 快照无需停机)
# → 本地暂存 → 上传阿里云 OSS;同时打包 apps/backend/data 下的平台 JSON 配置(排除日志与缓存)。
#
# 用法:
#   bash deploy/backup-db.sh             # 手动或 cron 调用
#   bash deploy/backup-db.sh --keep 14   # 本地保留最近 14 份(默认读 .env 的 DB_BACKUP_KEEP,再默认 7)
#
# 说明:
# - 上传由 backend 镜像内的 dist/db-backup.js 完成;OSS 凭证优先读 .env 的 DB_BACKUP_OSS_*,
#   未配置时复用 STORAGE_OSS_*。备份 key 形如 backups/postgres/order_site-<时间戳>.dump。
# - OSS 端的过期删除请配置桶生命周期规则,见 docs/数据库备份与恢复.md。
# - ubuntu-deploy.sh 在执行数据库迁移前会自动调用本脚本(--skip-backup 可跳过)。

PROJECT_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
ENV_FILE="${PROJECT_ROOT}/.env"
BACKUP_ROOT="${PROJECT_ROOT}/backups"
DB_DIR="${BACKUP_ROOT}/postgres"
CONFIG_DIR="${BACKUP_ROOT}/configs"
KEEP=""
BACKUP_PREFIX=""

log() { printf '\033[1;34m[backup]\033[0m %s\n' "$*"; }
warn() { printf '\033[1;33m[backup]\033[0m %s\n' "$*"; }
fail() { printf '\033[1;31m[backup]\033[0m %s\n' "$*" >&2; exit 1; }

parse_args() {
  while (($# > 0)); do
    case "$1" in
      --keep)
        [[ $# -ge 2 ]] || fail "--keep 需要一个正整数参数。"
        KEEP="$2"
        shift 2
        ;;
      -h|--help)
        sed -n '4,15p' "${BASH_SOURCE[0]}" | sed 's/^# \{0,1\}//'
        exit 0
        ;;
      *)
        fail "未知参数:$1。执行 bash deploy/backup-db.sh --help 查看用法。"
        ;;
    esac
  done

  if [[ -z "${KEEP}" ]]; then
    KEEP="${DB_BACKUP_KEEP:-$(read_env_value DB_BACKUP_KEEP || true)}"
  fi
  KEEP="${KEEP:-7}"
  [[ "${KEEP}" =~ ^[1-9][0-9]*$ ]] || fail "本地保留份数必须是正整数,当前值:${KEEP}"

  BACKUP_PREFIX="${DB_BACKUP_OSS_PREFIX:-$(read_env_value DB_BACKUP_OSS_PREFIX || true)}"
  BACKUP_PREFIX="${BACKUP_PREFIX%/}"
  BACKUP_PREFIX="${BACKUP_PREFIX:-backups/postgres}"
  [[ "${BACKUP_PREFIX}" != /* && "${BACKUP_PREFIX}" != *'..'* ]] || fail "DB_BACKUP_OSS_PREFIX 不能是绝对路径或包含 ..。"
}

read_env_value() {
  local key="$1"
  [[ -f "${ENV_FILE}" ]] || return 0
  awk -F= -v key="${key}" '
    $0 !~ /^[[:space:]]*#/ && $1 == key {
      value = substr($0, index($0, "=") + 1)
      gsub(/^[[:space:]]+|[[:space:]]+$/, "", value)
      gsub(/^"|"$/, "", value)
      gsub(/^'\''|'\''$/, "", value)
      print value
      exit
    }
  ' "${ENV_FILE}"
}

run_backup_tool() {
  # 用一次性容器而不是 exec,部署过程中 backend 未运行时也能上传。
  docker compose run --rm --no-deps -T backend node dist/db-backup.js "$@"
}

ensure_backup_tool() {
  if docker compose run --rm --no-deps -T backend test -f /app/dist/db-backup.js >/dev/null 2>&1; then
    return
  fi

  fail "当前 backend 镜像不包含 dist/db-backup.js,请先执行 docker compose build backend 重建镜像,再重新运行本脚本。"
}

run_pg_dump() {
  local database_url
  database_url="$(read_env_value DATABASE_URL)"
  if [[ -z "${database_url}" || "${database_url}" == *"@postgres:"* ]]; then
    docker compose exec -T postgres sh -c 'pg_dump -U "${POSTGRES_USER:-postgres}" -d "${POSTGRES_DB:-order_site}" -Fc'
  else
    log "DATABASE_URL 指向外部数据库,从 postgres 容器经网络直连导出"
    docker compose exec -T -e DB_DUMP_URL="${database_url}" postgres sh -c 'pg_dump "$DB_DUMP_URL" -Fc'
  fi
}

backup_database() {
  local stamp dump_file
  stamp="$(date +%Y%m%d-%H%M%S)"
  dump_file="${DB_DIR}/order_site-${stamp}.dump"

  log "导出数据库(pg_dump custom 格式,无需停机)"
  if ! run_pg_dump > "${dump_file}"; then
    rm -f "${dump_file}"
    fail "pg_dump 失败。可先用 docker compose logs postgres 排查数据库状态。"
  fi
  [[ -s "${dump_file}" ]] || fail "pg_dump 输出为空:${dump_file}"

  log "上传数据库备份到 OSS:${BACKUP_PREFIX}/$(basename "${dump_file}")"
  local dump_size
  dump_size="$(wc -c < "${dump_file}" | tr -d ' ')"
  if ! run_backup_tool upload --key "${BACKUP_PREFIX}/$(basename "${dump_file}")" --size "${dump_size}" < "${dump_file}"; then
    fail "备份上传 OSS 失败。数据库未受影响,本地已保留:${dump_file}"
  fi
  log "数据库备份完成:${dump_size} 字节"
}

backup_configs() {
  local stamp tar_file sample
  sample="$(
    find "${PROJECT_ROOT}/apps/backend/data" -type f \
      ! -path '*/logs/*' ! -path '*/cache/*' \
      ! -path '*/browser-sessions/*' ! -path '*/redeem-screenshots/*' 2>/dev/null | head -n 1
  )"
  if [[ -z "${sample}" ]]; then
    warn "apps/backend/data 下没有需要备份的配置文件,跳过配置打包"
    return
  fi

  stamp="$(date +%Y%m%d-%H%M%S)"
  tar_file="${CONFIG_DIR}/order_site-${stamp}.tar.gz"

  if ! tar -czf "${tar_file}" \
    --exclude='./logs' --exclude='./cache' \
    --exclude='./browser-sessions' --exclude='./redeem-screenshots' \
    -C "${PROJECT_ROOT}/apps/backend/data" .; then
    rm -f "${tar_file}"
    fail "打包 apps/backend/data 失败。"
  fi

  local config_prefix
  config_prefix="${BACKUP_PREFIX%/postgres}/configs"
  log "上传配置备份到 OSS:${config_prefix}/$(basename "${tar_file}")"
  local config_size
  config_size="$(wc -c < "${tar_file}" | tr -d ' ')"
  if ! run_backup_tool upload --key "${config_prefix}/$(basename "${tar_file}")" --kind app-configs --size "${config_size}" < "${tar_file}"; then
    fail "配置备份上传 OSS 失败。数据库备份已上传,本地配置备份已保留:${tar_file}"
  fi
  log "配置备份完成:${config_size} 字节"
}

prune_local() {
  local dir="$1" removed
  removed="$(cd "${dir}" && ls -1t order_site-*.dump order_site-*.tar.gz 2>/dev/null | tail -n +$((KEEP + 1)) || true)"
  if [[ -z "${removed}" ]]; then
    return
  fi
  while IFS= read -r file; do
    rm -f -- "${file}"
    log "清理本地旧备份:${file}(本地保留 ${KEEP} 份,OSS 仍由生命周期规则管理)"
  done <<<"${removed}"
}

main() {
  parse_args "$@"
  cd "${PROJECT_ROOT}"

  command -v docker >/dev/null 2>&1 || fail "未找到 docker。"
  docker compose version >/dev/null 2>&1 || fail "未找到 docker compose 插件。"
  if [[ -z "$(docker compose images backend -q 2>/dev/null)" ]]; then
    fail "backend 镜像不存在,请先完成部署:bash deploy/ubuntu-deploy.sh"
  fi

  ensure_backup_tool

  mkdir -p "${DB_DIR}" "${CONFIG_DIR}"
  backup_database
  backup_configs
  prune_local "${DB_DIR}"
  prune_local "${CONFIG_DIR}"
  log "备份完成"
}

main "$@"