重写在线备份链路

backend/.env.prod.example

@@ -48,26 +48,23 @@ LOG_ENABLE_CONSOLE=false
 LOG_ENABLE_FILE=true
 LOG_RETAIN_DAYS=14
 
-# MinIO 容器初始化变量。迁移完成前仍保留，用于回退与校验。
-MINIO_ROOT_USER=change-minio-user
-MINIO_ROOT_PASSWORD=change-minio-password
-# 当前主对象存储。使用内置 MinIO 时，STORAGE_* 密钥必须和 MINIO_ROOT_* 保持一致。
-STORAGE_ENDPOINT=http://minio:9000
-STORAGE_BUCKET=hfb-sys
-STORAGE_ACCESS_KEY_ID=change-minio-user
-STORAGE_SECRET_ACCESS_KEY=change-minio-password
-STORAGE_REGION=
-STORAGE_BUCKET_LOOKUP=auto
-
-# OSS 迁移镜像端。迁移期间设置后，新上传文件会同时写入 MinIO 与 OSS。
+# 主对象存储：图片等数据已迁移到阿里云 OSS，MinIO 已从生产编排移除。
 # 杭州 ECS 应使用内网 Endpoint：https://oss-cn-hangzhou-internal.aliyuncs.com
 # OSS 的 S3 兼容访问使用 cn-hangzhou 区域和 DNS Bucket 寻址。
+STORAGE_ENDPOINT=https://oss-cn-hangzhou-internal.aliyuncs.com
+STORAGE_BUCKET=hfb-sys-assets
+STORAGE_ACCESS_KEY_ID=change-oss-access-key-id
+STORAGE_SECRET_ACCESS_KEY=change-oss-access-key-secret
+STORAGE_REGION=cn-hangzhou
+STORAGE_BUCKET_LOOKUP=dns
+
+# 历史镜像写入配置，迁移完成后保持为空。
 STORAGE_MIRROR_ENDPOINT=
-STORAGE_MIRROR_BUCKET=hfb-sys-assets
+STORAGE_MIRROR_BUCKET=
 STORAGE_MIRROR_ACCESS_KEY_ID=
 STORAGE_MIRROR_SECRET_ACCESS_KEY=
-STORAGE_MIRROR_REGION=cn-hangzhou
-STORAGE_MIRROR_BUCKET_LOOKUP=dns
+STORAGE_MIRROR_REGION=
+STORAGE_MIRROR_BUCKET_LOOKUP=
 
 # 生产环境建议接入真实短信服务；未配置时不要使用 mock 对外运营。
 SMS_PROVIDER=aliyun
@@ -102,3 +99,25 @@ FIELD_ENCRYPTION_LEGACY_KEY=
 EXTERNAL_UPLOAD_SECRET=
 # 可选：逗号分隔的 IP 或 CIDR 白名单，例如 203.0.113.10,10.0.0.0/8。
 EXTERNAL_UPLOAD_ALLOWED_IPS=
+
+# 在线备份（scripts/backup-online.sh / scripts/archive-binlog.sh）。所有项为生产必填。
+# 本地备份根目录，必须是独立数据盘上的 /data 子目录；包含加密全量备份与 binlog 状态。
+BACKUP_DIR=/data/backups
+# 隔离恢复演练的数据目录，必须与 BACKUP_DIR 不同；prepare 后的 MySQL 原始数据只放在这里。
+BACKUP_RESTORE_DIR=/data/restore
+# 备份专用 OSS Bucket（与业务 Bucket 分离），ossutil 使用。
+# 生成方式：ossutil config 交互配置，或环境变量 OSS_ACCESS_KEY_ID / OSS_ACCESS_KEY_SECRET。
+BACKUP_OSS_URI=oss://hfb-backup
+# 客户端加密密码短语（独立保管，与业务密钥一同纳入 KMS/密码管理）。
+# 生成方式：openssl rand -hex 32
+BACKUP_PASSPHRASE=change-to-random-backup-passphrase
+# OSS 服务端 KMS 加密 Key ID，备份脚本强制使用，不可留空。
+BACKUP_KMS_KEY_ID=change-kms-key-id
+# XtraBackup 专用 MySQL 账号，部署脚本会幂等创建并授予最小备份权限。
+BACKUP_MYSQL_USER=hfb_backup
+# 生成方式：openssl rand -hex 24。不要和 MySQL 应用账号或 root 密码复用。
+BACKUP_MYSQL_PASSWORD=change-backup-mysql-password
+# 本地保留策略（OSS 侧用 Bucket 生命周期规则管理，见脚本头部注释）。
+BACKUP_KEEP_DAILY=14
+BACKUP_KEEP_WEEKLY=8
+BACKUP_KEEP_MONTHLY=12

deploy/README.md

@@ -8,7 +8,7 @@ https://选号网域名        # CADDY_SHOW_DOMAIN（可选），hfb_show 静态
 https://monitor.主站域名  # Beszel 监控
 ```
 
-生产部署使用 Caddy 作为公网入口，只暴露宿主机 `80`、`443` 端口。Caddy 会自动申请和续签 HTTPS 证书，并直接托管前端静态资源（主站 `dist` 与可选的选号网 `show-dist` 在构建 Caddy 镜像时打入）；后端、MySQL、Redis、MinIO 均通过 Docker 内网通信。选号网与主站**共用同一 backend**，浏览器访问选号网域名下的 `/api/*` 由 Caddy 反代到本机 backend，**不会直连数据库**。
+生产部署使用 Caddy 作为公网入口，只暴露宿主机 `80`、`443` 端口。Caddy 会自动申请和续签 HTTPS 证书，并直接托管前端静态资源（主站 `dist` 与可选的选号网 `show-dist` 在构建 Caddy 镜像时打入）；后端、MySQL、Redis 均通过 Docker 内网通信，图片对象存储使用阿里云 OSS。选号网与主站**共用同一 backend**，浏览器访问选号网域名下的 `/api/*` 由 Caddy 反代到本机 backend，**不会直连数据库**。
 
 部署前请确认：
 
@@ -32,12 +32,16 @@ MYSQL_ROOT_PASSWORD
 MYSQL_PASSWORD
 MYSQL_DSN
 JWT_SECRET
-MINIO_ROOT_USER
-MINIO_ROOT_PASSWORD
 STORAGE_ACCESS_KEY_ID
 STORAGE_SECRET_ACCESS_KEY
 BACKEND_UID
 BACKEND_GID
+BACKUP_DIR
+BACKUP_RESTORE_DIR
+BACKUP_OSS_URI
+BACKUP_PASSPHRASE
+BACKUP_KMS_KEY_ID
+BACKUP_MYSQL_PASSWORD
 ```
 
 `BACKEND_UID`、`BACKEND_GID` 填写服务器业务用户的数值 ID，例如：
@@ -49,29 +53,68 @@ id -g yml
 
 后端容器会使用该 UID/GID 运行。部署脚本由 root 执行时会自动把已有日志修正为该属主；非 root 执行且发现旧的 root 日志时，会给出需要执行的 `chown` 命令后停止。
 
-### 迁移对象存储到阿里云 OSS
+### 对象存储：阿里云 OSS
 
-创建私有 Bucket 后，先保持 `STORAGE_*` 指向当前 MinIO，在 `backend/.env` 配置 OSS 镜像端：
+图片等业务文件存储在私有 OSS Bucket，`backend/.env` 的 `STORAGE_*` 直接指向 OSS：
 
 ```text
-STORAGE_MIRROR_ENDPOINT=https://oss-cn-hangzhou-internal.aliyuncs.com
-STORAGE_MIRROR_BUCKET=hfb-sys-assets
-STORAGE_MIRROR_ACCESS_KEY_ID=change-oss-access-key-id
-STORAGE_MIRROR_SECRET_ACCESS_KEY=change-oss-access-key-secret
-STORAGE_MIRROR_REGION=cn-hangzhou
-STORAGE_MIRROR_BUCKET_LOOKUP=dns
+STORAGE_ENDPOINT=https://oss-cn-hangzhou-internal.aliyuncs.com
+STORAGE_BUCKET=hfb-sys-assets
+STORAGE_ACCESS_KEY_ID=change-oss-access-key-id
+STORAGE_SECRET_ACCESS_KEY=change-oss-access-key-secret
+STORAGE_REGION=cn-hangzhou
+STORAGE_BUCKET_LOOKUP=dns
 ```
 
-RAM 凭证只授予该 Bucket 的列举、读取、写入对象权限，不要使用阿里云主账号 AccessKey。镜像端启用后，新上传文件会同步写入 MinIO 和 OSS；任一镜像写入失败时上传会失败，避免数据库引用未同步文件。
+RAM 凭证只授予该 Bucket 的列举、读取、写入对象权限，不要使用阿里云主账号 AccessKey。杭州 ECS 必须使用内网 Endpoint，避免跨公网访问。
 
-部署支持镜像配置的镜像后，执行以下命令同步历史对象；命令可重复执行，不会删除 MinIO 数据：
+历史 MinIO 数据已通过 `scripts/migrate-minio-to-oss.sh` 迁移完成，生产编排不再包含 MinIO；旧 MinIO volume 观察期结束后可删除。
+
+### 数据库在线备份与时间点恢复
+
+生产 MySQL 开启了 `ROW` 格式 binlog 和崩溃安全刷盘。备份期间业务持续读写：每日用 XtraBackup 做物理全量备份，每分钟轮转并归档已关闭的 binlog。备份对象先做客户端加密，再强制以 OSS KMS 加密上传；远端仅在 payload、checksum 和 manifest 都上传成功后写入 `complete.env`。
+
+部署脚本会创建或更新 `BACKUP_MYSQL_USER`，该用户只有 XtraBackup 所需的全局备份权限和业务库只读权限，不开放远程 root。
+
+备份 Bucket 必须独立于业务 Bucket，备份 RAM 身份仅授予该 Bucket 的列举、读写对象及指定 KMS Key 的加解密权限。推荐 Bucket 开启版本控制和不可变保留策略。
+
+部署机需安装 `ossutil`、`openssl` 和 `flock`，并使用与 cron 相同的业务用户完成 `ossutil` 凭证配置；部署脚本会在启动服务前检查这些依赖。
 
 ```bash
-./scripts/migrate-minio-to-oss.sh
-./scripts/migrate-minio-to-oss.sh --verify
+# 每日全量热备，本地留存并上传 OSS 备份桶
+./scripts/backup-online.sh full
+
+# binlog 分钟级归档，配合全量实现最近保留窗口内的时间点恢复
+./scripts/archive-binlog.sh archive
+
+# 校验本地状态对应的远端完整对象 / 清理过期本地备份
+./scripts/archive-binlog.sh verify
+./scripts/backup-online.sh status
+./scripts/backup-online.sh prune
 ```
 
-校验通过后，将 `STORAGE_*` 改为同一组 OSS 配置，再清空所有 `STORAGE_MIRROR_*` 配置并重新部署。杭州 ECS 必须使用 `oss-cn-hangzhou-internal.aliyuncs.com`，避免跨公网访问。MinIO 保留至少 30 天后再考虑下线。
+推荐 cron（以部署用户运行，并把日志写到受限目录）：
+
+```cron
+* * * * * cd /srv/hfb_sys && ./scripts/archive-binlog.sh archive >> /data/backups/archive-binlog.log 2>&1
+30 4 * * * cd /srv/hfb_sys && ./scripts/backup-online.sh full >> /data/backups/full-backup.log 2>&1
+15 5 * * * cd /srv/hfb_sys && ./scripts/backup-online.sh prune >> /data/backups/prune.log 2>&1
+10 5 * * 0 cd /srv/hfb_sys && ./scripts/archive-binlog.sh verify >> /data/backups/archive-verify.log 2>&1
+```
+
+OSS 生命周期建议：`mysql/full/daily/` 30 天、`weekly/` 60 天、`monthly/` 365 天、`mysql/binlog/` 至少 45 天。时间点恢复窗口受最早全量备份和 binlog 保留期共同限制。
+
+每周必须在隔离目录做一次恢复演练，绝不向生产容器回放：
+
+```bash
+# 从 status 输出选取一个带 complete.env 的全量备份 URI。
+./scripts/restore-online.sh prepare oss://hfb-backup/mysql/full/daily/<server_uuid>/<timestamp>/complete.env
+./scripts/restore-online.sh fetch-all /data/restore/<server_uuid>/<timestamp>
+./scripts/restore-online.sh start /data/restore/<server_uuid>/<timestamp>
+./scripts/restore-online.sh replay /data/restore/<server_uuid>/<timestamp> hfb-restore-<timestamp> '2026-08-16 20:00:00'
+```
+
+恢复容器使用 `--network none`，只可通过 `docker exec` 检查。演练结束后手动删除名为 `hfb-restore-*` 的恢复容器和 `/data/restore` 对应目录；生产容器不在恢复工具的可选目标中。
 
 ### 同机部署选号网（hfb_show，可选）
 
@@ -108,12 +151,11 @@ HFB_SHOW_DIR=../hfb_show
 
 脚本会自动完成：
 
-- 检查 `backend/.env` 是否仍指向 `127.0.0.1`、`localhost` 或占位值。
-- 使用内置 MinIO 时，检查 `STORAGE_ACCESS_KEY_ID` / `STORAGE_SECRET_ACCESS_KEY` 是否和 `MINIO_ROOT_USER` / `MINIO_ROOT_PASSWORD` 一致。
+- 检查 `backend/.env` 是否仍指向 `127.0.0.1`、`localhost`、占位值或已下线的 MinIO。
 - 构建并启动生产容器。
 - 使用配置的非 root UID/GID 运行后端，并检查日志目录属主。
 - 通过 Caddy 自动申请或续签 HTTPS 证书。
-- 等待 MySQL、Redis、MinIO 就绪。
+- 等待 MySQL、Redis 就绪。
 - 按顺序执行尚未应用的数据库迁移。
 - 重启后端并检查 `https://你的域名/api/health`。
 

deploy/docker-compose.prod.yml

@@ -75,30 +75,6 @@ services:
       timeout: 5s
       retries: 10
 
-  minio:
-    image: minio/minio:RELEASE.2025-09-07T16-13-09Z
-    restart: unless-stopped
-    command: server /data --console-address ":9001"
-    env_file:
-      - ../backend/.env
-    environment:
-      TZ: Asia/Shanghai
-    volumes:
-      - minio_data:/data
-    deploy:
-      resources:
-        limits:
-          memory: 512M
-          cpus: '0.5'
-        reservations:
-          memory: 256M
-          cpus: '0.25'
-    healthcheck:
-      test: ["CMD", "mc", "ready", "local"]
-      interval: 10s
-      timeout: 5s
-      retries: 10
-
   backend:
     build:
       context: ../backend
@@ -124,8 +100,6 @@ services:
         condition: service_healthy
       redis:
         condition: service_healthy
-      minio:
-        condition: service_healthy
     expose:
       - "8080"
 
@@ -189,7 +163,6 @@ volumes:
   caddy_config:
   mysql_data:
   redis_data:
-  minio_data:
   beszel_data:
   beszel_socket:
   beszel_agent_data:

deploy/mysql/my.cnf

@@ -8,3 +8,13 @@ default-character-set = utf8mb4
 character-set-server = utf8mb4
 collation-server = utf8mb4_unicode_ci
 init-connect = 'SET NAMES utf8mb4'
+
+# 时间点恢复基础：binlog 由 scripts/archive-binlog.sh 每分钟归档到 OSS 备份桶。
+# 本地按 7 天自动过期，归档与本地保留策略见 scripts/backup-online.sh。
+server-id = 1
+log_bin = mysql-bin
+binlog_format = ROW
+binlog_row_image = FULL
+binlog_expire_logs_seconds = 604800
+sync_binlog = 1
+innodb_flush_log_at_trx_commit = 1

scripts/archive-binlog.sh

@@ -0,0 +1,175 @@
+#!/usr/bin/env bash
+set -euo pipefail
+umask 077
+
+# 分钟级 binlog 归档。每次轮转后只归档已关闭文件；每个对象以
+# server_uuid、原始文件名和原始 sha256 命名，避免实例重建后的编号碰撞。
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=backup-common.sh
+source "${SCRIPT_DIR}/backup-common.sh"
+
+MYSQL_DATA_DIR='/var/lib/mysql'
+work_dir=''
+cleanup() {
+  if [[ -n "${work_dir}" && -d "${work_dir}" ]]; then
+    rm -rf -- "${work_dir}"
+  fi
+}
+trap cleanup EXIT
+
+require_archive_config() {
+  backup_require_env BACKUP_DIR >/dev/null
+  backup_require_env BACKUP_OSS_URI >/dev/null
+  backup_require_env BACKUP_PASSPHRASE >/dev/null
+  backup_require_env BACKUP_KMS_KEY_ID >/dev/null
+}
+
+closed_binlogs() {
+  local mysql_cid="$1"
+  docker exec "${mysql_cid}" sh -c \
+    'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot --protocol=socket -N -B -e "SHOW BINARY LOGS;"' \
+    2>/dev/null | awk 'NR > 1 { print previous } { previous = $1 } END { }'
+}
+
+state_last_index() {
+  local state_dir="$1"
+  local value
+  [[ -f "${state_dir}/last-archived.env" ]] || return
+  value="$(awk -F= '$1 == "binlog_index" { print $2; exit }' "${state_dir}/last-archived.env")"
+  [[ "${value}" =~ ^[0-9]+$ ]] && printf '%s' "${value}"
+}
+
+write_state() {
+  local state_dir="$1" filename="$2" index="$3" raw_sha="$4" cipher_sha="$5" remote_dir="$6"
+  {
+    printf 'filename=%s\n' "${filename}"
+    printf 'binlog_index=%s\n' "${index}"
+    printf 'raw_sha256=%s\n' "${raw_sha}"
+    printf 'cipher_sha256=%s\n' "${cipher_sha}"
+    printf 'remote_dir=%s\n' "${remote_dir}"
+  } > "${state_dir}/${filename}.${raw_sha}.state"
+  cp "${state_dir}/${filename}.${raw_sha}.state" "${state_dir}/last-archived.env"
+}
+
+archive_one() {
+  local mysql_cid="$1" state_dir="$2" server_uuid="$3" oss_uri="$4" filename="$5"
+  local source_path source_sha raw_file raw_sha cipher_file cipher_sha cipher_size binlog_index object_id remote_dir
+  source_path="${MYSQL_DATA_DIR}/${filename}"
+  if ! docker exec "${mysql_cid}" test -f "${source_path}" >/dev/null 2>&1; then
+    backup_error "待归档 binlog 已从 MySQL 数据目录消失：${filename}；PITR 链路已断裂"
+    exit 1
+  fi
+  raw_file="${work_dir}/${filename}"
+  docker cp "${mysql_cid}:${source_path}" "${raw_file}"
+  raw_sha="$(backup_sha256 "${raw_file}")"
+  source_sha="$(docker exec "${mysql_cid}" sha256sum "${source_path}" | awk '{print $1}')"
+  [[ "${source_sha}" == "${raw_sha}" ]] || {
+    backup_error "binlog 拷贝校验失败：${filename}"
+    exit 1
+  }
+  if compgen -G "${state_dir}/${filename}.${raw_sha}.state" >/dev/null; then
+    return
+  fi
+  binlog_index="$(backup_binlog_index "${filename}")"
+  object_id="${filename}.${raw_sha}"
+  remote_dir="${oss_uri}/mysql/binlog/${server_uuid}/${object_id}"
+  cipher_file="${work_dir}/${object_id}.enc"
+  backup_encrypt_file "${raw_file}" "${cipher_file}"
+  cipher_sha="$(backup_sha256 "${cipher_file}")"
+  cipher_size="$(backup_file_size "${cipher_file}")"
+  printf '%s  %s\n' "${cipher_sha}" 'binlog.enc' > "${work_dir}/${object_id}.enc.sha256"
+  {
+    printf 'format_version=1\n'
+    printf 'server_uuid=%s\n' "${server_uuid}"
+    printf 'source_binlog=%s\n' "${filename}"
+    printf 'binlog_index=%s\n' "${binlog_index}"
+    printf 'raw_sha256=%s\n' "${raw_sha}"
+    printf 'payload=binlog.enc\n'
+    printf 'payload_sha256=%s\n' "${cipher_sha}"
+    printf 'payload_size_bytes=%s\n' "${cipher_size}"
+    printf 'created_at=%s\n' "$(date -Iseconds)"
+  } > "${work_dir}/${object_id}.complete.env"
+
+  backup_oss_upload_kms "${cipher_file}" "${remote_dir}/binlog.enc"
+  backup_oss_upload_kms "${work_dir}/${object_id}.enc.sha256" "${remote_dir}/binlog.enc.sha256"
+  # complete.env 最后上传，恢复程序只使用有该文件的 binlog。
+  backup_oss_upload_kms "${work_dir}/${object_id}.complete.env" "${remote_dir}/complete.env"
+  write_state "${state_dir}" "${filename}" "${binlog_index}" "${raw_sha}" "${cipher_sha}" "${remote_dir}"
+  rm -f -- "${raw_file}" "${cipher_file}" "${work_dir}/${object_id}.enc.sha256" "${work_dir}/${object_id}.complete.env"
+}
+
+archive() {
+  backup_require_cmd docker
+  backup_require_cmd ossutil
+  backup_require_cmd openssl
+  require_archive_config
+  local backup_dir oss_uri mysql_cid server_uuid state_dir first_binlog last_index first_index expected_index filename index
+  backup_dir="$(backup_require_env BACKUP_DIR)"
+  backup_validate_dir "${backup_dir}"
+  oss_uri="$(backup_require_env BACKUP_OSS_URI)"
+  backup_validate_oss_uri "${oss_uri}"
+  backup_lock_or_exit "${backup_dir}/.binlog.lock"
+  mysql_cid="$(backup_mysql_container_id)"
+  server_uuid="$(backup_server_uuid)"
+  state_dir="${backup_dir}/binlog-state/${server_uuid}"
+  work_dir="${backup_dir}/.tmp-binlog-$$"
+  mkdir -p "${state_dir}" "${work_dir}"
+
+  docker exec "${mysql_cid}" sh -c \
+    'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot --protocol=socket -e "FLUSH BINARY LOGS;"' >/dev/null
+  first_binlog="$(docker exec "${mysql_cid}" sh -c \
+    'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot --protocol=socket -N -B -e "SHOW BINARY LOGS;"' | head -1 | awk '{print $1}')"
+  [[ -n "${first_binlog}" ]] || { backup_error '无法读取 binlog 列表'; exit 1; }
+  last_index="$(state_last_index "${state_dir}" || true)"
+  first_index="$(backup_binlog_index "${first_binlog}")"
+  if [[ -n "${last_index}" && "${first_index}" -gt $((last_index + 1)) ]]; then
+    backup_error "发现未归档 binlog 缺口：已归档至序号 ${last_index}，当前最早为 ${first_index}"
+    exit 1
+  fi
+
+  expected_index="${last_index}"
+  while IFS= read -r filename; do
+    [[ -n "${filename}" ]] || continue
+    index="$(backup_binlog_index "${filename}")"
+    if [[ -n "${expected_index}" && "${index}" -le "${expected_index}" ]]; then
+      continue
+    fi
+    if [[ -n "${expected_index}" && "${index}" -ne $((expected_index + 1)) ]]; then
+      backup_error "发现未归档 binlog 缺口：期望序号 $((expected_index + 1))，实际为 ${index}"
+      exit 1
+    fi
+    archive_one "${mysql_cid}" "${state_dir}" "${server_uuid}" "${oss_uri}" "${filename}"
+    expected_index="${index}"
+  done < <(closed_binlogs "${mysql_cid}")
+  backup_log "binlog 归档完成，实例 ${server_uuid}"
+}
+
+verify() {
+  backup_require_cmd ossutil
+  require_archive_config
+  local backup_dir server_uuid state_dir state_file remote_dir
+  backup_dir="$(backup_require_env BACKUP_DIR)"
+  backup_validate_dir "${backup_dir}"
+  backup_lock_or_exit "${backup_dir}/.binlog.lock"
+  server_uuid="$(backup_server_uuid)"
+  state_dir="${backup_dir}/binlog-state/${server_uuid}"
+  [[ -d "${state_dir}" ]] || { backup_error '尚未产生 binlog 归档状态'; exit 1; }
+  while IFS= read -r -d '' state_file; do
+    remote_dir="$(awk -F= '$1 == "remote_dir" { print substr($0, index($0, "=") + 1); exit }' "${state_file}")"
+    [[ -n "${remote_dir}" ]] || { backup_error "归档状态不完整：${state_file}"; exit 1; }
+    ossutil stat "${remote_dir}/binlog.enc" >/dev/null
+    ossutil stat "${remote_dir}/binlog.enc.sha256" >/dev/null
+    ossutil stat "${remote_dir}/complete.env" >/dev/null
+  done < <(find "${state_dir}" -maxdepth 1 -name '*.state' -print0)
+  backup_log '全部本地 binlog 归档状态均有远端完整对象'
+}
+
+case "${1:-archive}" in
+  archive) archive ;;
+  verify) verify ;;
+  *)
+    printf '用法: %s {archive|verify}\n' "$0" >&2
+    exit 1
+    ;;
+esac

scripts/backup-common.sh

@@ -0,0 +1,193 @@
+#!/usr/bin/env bash
+
+# 在线备份脚本共用函数。仅供 scripts/backup-online.sh、archive-binlog.sh 和
+# restore-online.sh source，不要单独执行。
+
+BACKUP_ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+BACKUP_COMPOSE_FILE="${BACKUP_ROOT_DIR}/deploy/docker-compose.prod.yml"
+BACKUP_ENV_FILE="${BACKUP_ROOT_DIR}/backend/.env"
+BACKUP_XTRABACKUP_IMAGE="${XTRABACKUP_IMAGE:-percona/percona-xtrabackup:8.4}"
+
+backup_log() {
+  printf '[backup] %s\n' "$*" >&2
+}
+
+backup_error() {
+  printf '[backup] ERROR: %s\n' "$*" >&2
+}
+
+backup_require_cmd() {
+  if ! command -v "$1" >/dev/null 2>&1; then
+    backup_error "缺少命令：$1"
+    exit 1
+  fi
+}
+
+backup_env_value() {
+  local key="$1"
+  awk -F= -v key="${key}" '$1 == key { sub(/^[^=]*=/, ""); print; exit }' "${BACKUP_ENV_FILE}"
+}
+
+backup_require_env() {
+  local key="$1"
+  local value
+  value="$(backup_env_value "${key}")"
+  if [[ -z "${value}" ]]; then
+    backup_error "${BACKUP_ENV_FILE} 缺少或未填写 ${key}"
+    exit 1
+  fi
+  printf '%s' "${value}"
+}
+
+backup_compose() {
+  docker compose --env-file "${BACKUP_ENV_FILE}" -f "${BACKUP_COMPOSE_FILE}" "$@"
+}
+
+backup_mysql_container_id() {
+  local cid
+  cid="$(backup_compose ps -q mysql 2>/dev/null || true)"
+  if [[ -z "${cid}" ]]; then
+    backup_error "MySQL 容器未运行"
+    exit 1
+  fi
+  printf '%s' "${cid}"
+}
+
+backup_mysql_network() {
+  local network
+  network="$(docker inspect -f '{{range $k, $v := .NetworkSettings.Networks}}{{$k}}{{end}}' "$1")"
+  if [[ -z "${network}" ]]; then
+    backup_error "无法读取 MySQL Docker 网络"
+    exit 1
+  fi
+  printf '%s' "${network}"
+}
+
+backup_mysql_root_query() {
+  backup_compose exec -T mysql sh -c \
+    'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot --protocol=socket -N -B "$MYSQL_DATABASE" -e "$1"' \
+    sh "$1"
+}
+
+backup_validate_dir() {
+  local dir="$1"
+  if [[ "${dir}" != /data/* || "${dir}" == "/data" ]]; then
+    backup_error "备份目录必须是 /data 之下的具体绝对路径，当前值：${dir}"
+    exit 1
+  fi
+}
+
+backup_lock_or_exit() {
+  local path="$1"
+  local wait_seconds="${2:-${BACKUP_LOCK_WAIT_SECONDS:-0}}"
+  backup_require_cmd flock
+  mkdir -p "$(dirname "${path}")"
+  exec 9>"${path}"
+  if [[ ! "${wait_seconds}" =~ ^[0-9]+$ ]]; then
+    backup_error "锁等待时间必须是非负整数：${wait_seconds}"
+    exit 1
+  fi
+  if ! flock -w "${wait_seconds}" 9; then
+    backup_error "已有同类备份任务在执行，拒绝并发运行"
+    exit 1
+  fi
+}
+
+backup_server_uuid() {
+  local value
+  value="$(backup_mysql_root_query 'SELECT @@server_uuid;' 2>/dev/null || true)"
+  if [[ ! "${value}" =~ ^[0-9A-Fa-f-]{36}$ ]]; then
+    backup_error "无法读取 MySQL server_uuid"
+    exit 1
+  fi
+  printf '%s' "${value,,}"
+}
+
+backup_binlog_index() {
+  local filename="$1"
+  if [[ "${filename}" =~ ^mysql-bin\.([0-9]{6,})$ ]]; then
+    printf '%d' "$((10#${BASH_REMATCH[1]}))"
+    return
+  fi
+  backup_error "不支持的 binlog 文件名：${filename}"
+  exit 1
+}
+
+backup_sha256() {
+  if command -v sha256sum >/dev/null 2>&1; then
+    sha256sum "$1" | awk '{print $1}'
+    return
+  fi
+  shasum -a 256 "$1" | awk '{print $1}'
+}
+
+backup_file_size() {
+  if [[ "$(uname)" == 'Darwin' ]]; then
+    stat -f%z "$1"
+  else
+    stat -c%s "$1"
+  fi
+}
+
+backup_encrypt_file() {
+  local source="$1"
+  local destination="$2"
+  local passphrase
+  passphrase="$(backup_require_env BACKUP_PASSPHRASE)"
+  BACKUP_PASSPHRASE="${passphrase}" openssl enc -aes-256-cbc -pbkdf2 -iter 600000 -salt \
+    -pass env:BACKUP_PASSPHRASE \
+    -in "${source}" \
+    -out "${destination}"
+  unset passphrase
+}
+
+backup_decrypt_file() {
+  local source="$1"
+  local destination="$2"
+  local passphrase
+  passphrase="$(backup_require_env BACKUP_PASSPHRASE)"
+  BACKUP_PASSPHRASE="${passphrase}" openssl enc -d -aes-256-cbc -pbkdf2 -iter 600000 \
+    -pass env:BACKUP_PASSPHRASE \
+    -in "${source}" \
+    -out "${destination}"
+  unset passphrase
+}
+
+backup_oss_upload_kms() {
+  local source="$1"
+  local destination="$2"
+  local kms_key_id
+  kms_key_id="$(backup_require_env BACKUP_KMS_KEY_ID)"
+  ossutil cp "${source}" "${destination}" \
+    --encryption KMS --kms-master-key-id "${kms_key_id}"
+}
+
+backup_validate_oss_uri() {
+  local uri="$1"
+  if [[ ! "${uri}" =~ ^oss://[A-Za-z0-9][A-Za-z0-9._-]*$ ]]; then
+    backup_error "BACKUP_OSS_URI 必须是 oss://bucket 格式，当前值：${uri}"
+    exit 1
+  fi
+}
+
+backup_oss_download() {
+  local source="$1"
+  local destination="$2"
+  ossutil cp "${source}" "${destination}"
+}
+
+backup_write_complete_marker() {
+  local path="$1"
+  local payload_name="$2"
+  local payload_sha256="$3"
+  local payload_size="$4"
+  local server_uuid="$5"
+  {
+    printf 'format_version=1\n'
+    printf 'created_at=%s\n' "$(date -Iseconds)"
+    printf 'server_uuid=%s\n' "${server_uuid}"
+    printf 'payload=%s\n' "${payload_name}"
+    printf 'payload_sha256=%s\n' "${payload_sha256}"
+    printf 'payload_size_bytes=%s\n' "${payload_size}"
+  } > "${path}"
+}

scripts/backup-online.sh

@@ -0,0 +1,206 @@
+#!/usr/bin/env bash
+set -euo pipefail
+umask 077
+
+# MySQL 在线全量物理备份。XtraBackup 在业务读写期间运行，备份只在
+# 上传 payload、manifest、checksum 后最后写入 complete.env，恢复程序只接受
+# 带 complete.env 的备份集。
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=backup-common.sh
+source "${SCRIPT_DIR}/backup-common.sh"
+
+cleanup_dir=''
+cleanup() {
+  if [[ -n "${cleanup_dir}" && -d "${cleanup_dir}" ]]; then
+    rm -rf -- "${cleanup_dir}"
+  fi
+}
+trap cleanup EXIT
+
+require_backup_config() {
+  backup_require_env BACKUP_DIR >/dev/null
+  backup_require_env BACKUP_OSS_URI >/dev/null
+  backup_require_env BACKUP_PASSPHRASE >/dev/null
+  backup_require_env BACKUP_KMS_KEY_ID >/dev/null
+  backup_require_env BACKUP_MYSQL_USER >/dev/null
+  backup_require_env BACKUP_MYSQL_PASSWORD >/dev/null
+}
+
+upload_backup_set() {
+  local local_dir="$1"
+  local level="$2"
+  local ts="$3"
+  local server_uuid="$4"
+  local oss_uri="$5"
+  local remote_dir="${oss_uri}/mysql/full/${level}/${server_uuid}/${ts}"
+
+  backup_log "上传 ${level} 全量备份到 ${remote_dir}"
+  backup_oss_upload_kms "${local_dir}/mysql-full.tgz.enc" "${remote_dir}/mysql-full.tgz.enc"
+  backup_oss_upload_kms "${local_dir}/manifest.env" "${remote_dir}/manifest.env"
+  backup_oss_upload_kms "${local_dir}/mysql-full.tgz.enc.sha256" "${remote_dir}/mysql-full.tgz.enc.sha256"
+  # complete.env 必须最后上传，作为远端原子完成标识。
+  backup_oss_upload_kms "${local_dir}/complete.env" "${remote_dir}/complete.env"
+}
+
+full_backup() {
+  backup_require_cmd docker
+  backup_require_cmd openssl
+  backup_require_cmd ossutil
+  require_backup_config
+
+  local backup_dir oss_uri mysql_cid network server_uuid backup_user backup_password ts tmp_dir target_dir
+  backup_dir="$(backup_require_env BACKUP_DIR)"
+  backup_validate_dir "${backup_dir}"
+  oss_uri="$(backup_require_env BACKUP_OSS_URI)"
+  backup_validate_oss_uri "${oss_uri}"
+  backup_lock_or_exit "${backup_dir}/.full.lock"
+  mysql_cid="$(backup_mysql_container_id)"
+  network="$(backup_mysql_network "${mysql_cid}")"
+  server_uuid="$(backup_server_uuid)"
+  backup_user="$(backup_require_env BACKUP_MYSQL_USER)"
+  backup_password="$(backup_require_env BACKUP_MYSQL_PASSWORD)"
+  ts="$(date +%Y%m%d_%H%M%S)"
+  tmp_dir="${backup_dir}/.tmp-full-${ts}-$$"
+  target_dir="${backup_dir}/full/daily/${ts}"
+  cleanup_dir="${tmp_dir}"
+  mkdir -p "${tmp_dir}/xtrabackup"
+
+  cat > "${tmp_dir}/xtrabackup-client.cnf" <<EOF
+[client]
+user=${backup_user}
+password=${backup_password}
+host=mysql
+EOF
+  chmod 600 "${tmp_dir}/xtrabackup-client.cnf"
+
+  backup_log "开始全量热备，实例 ${server_uuid}"
+  docker run --rm \
+    --network "${network}" \
+    --volumes-from "${mysql_cid}:ro" \
+    -v "${tmp_dir}/xtrabackup:/backup" \
+    -v "${tmp_dir}/xtrabackup-client.cnf:/etc/xtrabackup-client.cnf:ro" \
+    "${BACKUP_XTRABACKUP_IMAGE}" \
+    xtrabackup \
+    --defaults-extra-file=/etc/xtrabackup-client.cnf \
+    --backup \
+    --target-dir=/backup/data
+  rm -f "${tmp_dir}/xtrabackup-client.cnf"
+  unset backup_password
+
+  local binlog_file binlog_pos
+  binlog_file="$(awk -F$'\t' 'NR == 1 { print $1 }' "${tmp_dir}/xtrabackup/data/xtrabackup_binlog_info")"
+  binlog_pos="$(awk -F$'\t' 'NR == 1 { print $2 }' "${tmp_dir}/xtrabackup/data/xtrabackup_binlog_info")"
+  if [[ -z "${binlog_file}" || ! "${binlog_pos}" =~ ^[0-9]+$ ]]; then
+    backup_error "未读取到有效 xtrabackup_binlog_info"
+    exit 1
+  fi
+
+  {
+    printf 'format_version=1\n'
+    printf 'created_at=%s\n' "$(date -Iseconds)"
+    printf 'host=%s\n' "$(hostname)"
+    printf 'server_uuid=%s\n' "${server_uuid}"
+    printf 'mysql_version=%s\n' "$(backup_mysql_root_query 'SELECT VERSION();')"
+    printf 'goose_version=%s\n' "$(backup_mysql_root_query 'SELECT COALESCE(MAX(version_id), 0) FROM goose_db_version;')"
+    printf 'binlog_file=%s\n' "${binlog_file}"
+    printf 'binlog_pos=%s\n' "${binlog_pos}"
+    printf 'git_commit=%s\n' "$(git -C "${BACKUP_ROOT_DIR}" rev-parse HEAD 2>/dev/null || true)"
+  } > "${tmp_dir}/manifest.env"
+
+  # 关闭包含 XtraBackup 位点的当前 binlog 并归档。全量备份只有在该文件
+  # 已有远端 complete 标识后才会被标记完成。
+  BACKUP_LOCK_WAIT_SECONDS=120 "${SCRIPT_DIR}/archive-binlog.sh" archive
+  if ! find "${backup_dir}/binlog-state/${server_uuid}" -maxdepth 1 -type f \
+    -name "${binlog_file}.*.state" -print -quit | grep -q .; then
+    backup_error "XtraBackup 位点 ${binlog_file}:${binlog_pos} 尚未完成 binlog 归档"
+    exit 1
+  fi
+
+  tar czf "${tmp_dir}/mysql-full.tgz" \
+    -C "${tmp_dir}/xtrabackup/data" . \
+    -C "${tmp_dir}" manifest.env
+  backup_encrypt_file "${tmp_dir}/mysql-full.tgz" "${tmp_dir}/mysql-full.tgz.enc"
+  rm -f "${tmp_dir}/mysql-full.tgz"
+
+  local payload_sha payload_size
+  payload_sha="$(backup_sha256 "${tmp_dir}/mysql-full.tgz.enc")"
+  payload_size="$(backup_file_size "${tmp_dir}/mysql-full.tgz.enc")"
+  printf '%s  %s\n' "${payload_sha}" 'mysql-full.tgz.enc' > "${tmp_dir}/mysql-full.tgz.enc.sha256"
+  backup_write_complete_marker "${tmp_dir}/complete.env" 'mysql-full.tgz.enc' "${payload_sha}" "${payload_size}" "${server_uuid}"
+
+  mkdir -p "${target_dir}"
+  mv "${tmp_dir}/mysql-full.tgz.enc" "${target_dir}/"
+  mv "${tmp_dir}/mysql-full.tgz.enc.sha256" "${target_dir}/"
+  mv "${tmp_dir}/manifest.env" "${target_dir}/"
+  mv "${tmp_dir}/complete.env" "${target_dir}/"
+  rm -rf "${tmp_dir}/xtrabackup"
+  cleanup_dir=''
+  rmdir "${tmp_dir}"
+
+  upload_backup_set "${target_dir}" daily "${ts}" "${server_uuid}" "${oss_uri}"
+  if [[ "$(date +%u)" == '7' ]]; then
+    mkdir -p "${backup_dir}/full/weekly/${ts}"
+    ln "${target_dir}/mysql-full.tgz.enc" "${backup_dir}/full/weekly/${ts}/mysql-full.tgz.enc"
+    cp "${target_dir}/mysql-full.tgz.enc.sha256" "${target_dir}/manifest.env" "${target_dir}/complete.env" "${backup_dir}/full/weekly/${ts}/"
+    upload_backup_set "${target_dir}" weekly "${ts}" "${server_uuid}" "${oss_uri}"
+  fi
+  if [[ "$(date +%d)" == '01' ]]; then
+    mkdir -p "${backup_dir}/full/monthly/${ts}"
+    ln "${target_dir}/mysql-full.tgz.enc" "${backup_dir}/full/monthly/${ts}/mysql-full.tgz.enc"
+    cp "${target_dir}/mysql-full.tgz.enc.sha256" "${target_dir}/manifest.env" "${target_dir}/complete.env" "${backup_dir}/full/monthly/${ts}/"
+    upload_backup_set "${target_dir}" monthly "${ts}" "${server_uuid}" "${oss_uri}"
+  fi
+  backup_log "全量热备完成：${target_dir}，位点 ${binlog_file}:${binlog_pos}"
+}
+
+prune_level() {
+  local base="$1"
+  local days="$2"
+  [[ -d "${base}" ]] || return
+  find "${base}" -mindepth 1 -maxdepth 1 -type d -mtime "+${days}" -print0 |
+    while IFS= read -r -d '' dir; do
+      backup_log "清理过期本地备份：${dir}"
+      rm -rf -- "${dir}"
+    done
+}
+
+prune_backups() {
+  local backup_dir daily weekly monthly
+  backup_dir="$(backup_require_env BACKUP_DIR)"
+  backup_validate_dir "${backup_dir}"
+  backup_lock_or_exit "${backup_dir}/.full.lock"
+  daily="$(backup_env_value BACKUP_KEEP_DAILY)"; daily="${daily:-14}"
+  weekly="$(backup_env_value BACKUP_KEEP_WEEKLY)"; weekly="${weekly:-8}"
+  monthly="$(backup_env_value BACKUP_KEEP_MONTHLY)"; monthly="${monthly:-12}"
+  [[ "${daily}" =~ ^[0-9]+$ && "${weekly}" =~ ^[0-9]+$ && "${monthly}" =~ ^[0-9]+$ ]] || {
+    backup_error 'BACKUP_KEEP_* 必须是非负整数'
+    exit 1
+  }
+  prune_level "${backup_dir}/full/daily" "${daily}"
+  prune_level "${backup_dir}/full/weekly" "$((weekly * 7))"
+  prune_level "${backup_dir}/full/monthly" "$((monthly * 31))"
+}
+
+status() {
+  local backup_dir oss_uri
+  backup_dir="$(backup_require_env BACKUP_DIR)"
+  oss_uri="$(backup_require_env BACKUP_OSS_URI)"
+  backup_validate_oss_uri "${oss_uri}"
+  backup_log '本地最近完整全量备份：'
+  find "${backup_dir}/full" -name complete.env -print 2>/dev/null | sort | tail -8 || true
+  backup_log 'OSS 最近完整全量备份：'
+  ossutil ls "${oss_uri}/mysql/full/" 2>/dev/null | grep 'complete.env' | tail -8 || true
+  backup_log 'OSS 最近 binlog 完成标识：'
+  ossutil ls "${oss_uri}/mysql/binlog/" -r 2>/dev/null | grep 'complete.env' | tail -8 || true
+}
+
+case "${1:-}" in
+  full) full_backup ;;
+  prune) prune_backups ;;
+  status) status ;;
+  *)
+    printf '用法: %s {full|prune|status}\n' "$0" >&2
+    exit 1
+    ;;
+esac

scripts/backup-prod.sh

@@ -1,533 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-COMPOSE_FILE="${ROOT_DIR}/deploy/docker-compose.prod.yml"
-BACKEND_ENV="${ROOT_DIR}/backend/.env"
-BACKUP_ROOT="${BACKUP_ROOT:-${ROOT_DIR}/backups}"
-HELPER_IMAGE="${BACKUP_HELPER_IMAGE:-redis:7.4-alpine}"
-if [[ "${BACKUP_ROOT}" != /* ]]; then
-  BACKUP_ROOT="${ROOT_DIR}/${BACKUP_ROOT}"
-fi
-
-RED='\033[0;31m'
-GREEN='\033[0;32m'
-YELLOW='\033[1;33m'
-BLUE='\033[1;34m'
-NC='\033[0m'
-
-log() {
-  printf "${BLUE}[backup]${NC} %s\n" "$*" >&2
-}
-
-log_success() {
-  printf "${GREEN}[backup]${NC} %s\n" "$*" >&2
-}
-
-log_warn() {
-  printf "${YELLOW}[backup]${NC} %s\n" "$*" >&2
-}
-
-log_error() {
-  printf "${RED}[backup]${NC} %s\n" "$*" >&2
-}
-
-need_cmd() {
-  if ! command -v "$1" >/dev/null 2>&1; then
-    log_error "缺少命令：$1"
-    exit 1
-  fi
-}
-
-absolute_dir() {
-  local dir="$1"
-  (
-    cd "${dir}"
-    pwd -P
-  )
-}
-
-show_help() {
-  cat << EOF
-用法: ./scripts/backup-prod.sh <命令> [选项]
-
-命令:
-  backup                  创建完整生产备份
-  restore <备份目录>       从备份目录恢复生产数据（需要 --yes）
-  list                    列出本机备份目录
-
-选项:
-  --yes                   确认执行破坏性恢复，只用于 restore
-  --no-stop               backup 时不主动停止 backend/caddy，适合只读低风险窗口（不推荐）
-  -h, --help              显示帮助
-
-环境变量:
-  BACKUP_ROOT=/path       备份根目录，默认 ./backups
-  BACKUP_HELPER_IMAGE=img 用于打包/还原 Docker volume 的镜像，默认 redis:7.4-alpine
-
-备份内容:
-  - MySQL 逻辑备份 mysql.sql
-  - MinIO volume: minio_data.tgz
-  - Redis volume: redis_data.tgz
-  - Caddy 证书/配置 volume: caddy_data.tgz, caddy_config.tgz
-  - backend/.env、deploy 配置、backend/logs
-  - 当前 Git commit、Docker Compose 状态、goose 迁移状态
-
-示例:
-  ./scripts/backup-prod.sh backup
-  ./scripts/backup-prod.sh list
-  ./scripts/backup-prod.sh restore backups/20260618_153000 --yes
-EOF
-}
-
-compose() {
-  local app_env caddy_domain caddy_email admin_base_path
-  app_env="$(env_value APP_ENV)"
-  caddy_domain="${CADDY_DOMAIN:-$(env_value CADDY_DOMAIN)}"
-  caddy_email="${CADDY_EMAIL:-$(env_value CADDY_EMAIL)}"
-  admin_base_path="${VITE_ADMIN_BASE_PATH:-$(env_value VITE_ADMIN_BASE_PATH)}"
-
-  if [[ "${app_env}" == "production" && (-z "${caddy_domain}" || -z "${caddy_email}") ]]; then
-    log_error "APP_ENV=production 时必须配置 CADDY_DOMAIN 和 CADDY_EMAIL"
-    exit 1
-  fi
-
-  CADDY_DOMAIN="${caddy_domain:-backup.localhost}" \
-    CADDY_EMAIL="${caddy_email:-backup@example.com}" \
-    VITE_ADMIN_BASE_PATH="${admin_base_path:-/manage-7x9k2p}" \
-    docker compose --env-file "${BACKEND_ENV}" -f "${COMPOSE_FILE}" "$@"
-}
-
-require_env_file() {
-  if [[ ! -f "${BACKEND_ENV}" ]]; then
-    log_error "未找到 backend/.env，无法读取生产环境变量"
-    exit 1
-  fi
-}
-
-env_value() {
-  local key="$1"
-  awk -F= -v key="${key}" '$1 == key { sub(/^[^=]*=/, ""); print; exit }' "${BACKEND_ENV}"
-}
-
-require_env() {
-  local key="$1"
-  local value
-  value="$(env_value "${key}")"
-  if [[ -z "${value}" ]]; then
-    log_error "backend/.env 缺少或未填写 ${key}"
-    exit 1
-  fi
-  printf "%s" "${value}"
-}
-
-container_id() {
-  local service="$1"
-  local cid
-  cid="$(compose ps -a -q "${service}" 2>/dev/null || true)"
-  if [[ -z "${cid}" ]]; then
-    log "创建 ${service} 容器以挂载 volume..."
-    compose create "${service}" >/dev/null
-    cid="$(compose ps -a -q "${service}" 2>/dev/null || true)"
-  fi
-  if [[ -z "${cid}" ]]; then
-    log_error "无法找到 ${service} 容器"
-    exit 1
-  fi
-  printf "%s" "${cid}"
-}
-
-service_was_running() {
-  local service="$1"
-  local cid status
-  cid="$(compose ps -q "${service}" 2>/dev/null || true)"
-  if [[ -z "${cid}" ]]; then
-    return 1
-  fi
-  status="$(docker inspect -f '{{.State.Status}}' "${cid}" 2>/dev/null || true)"
-  [[ "${status}" == "running" ]]
-}
-
-start_services() {
-  if [[ "$#" -eq 0 ]]; then
-    return
-  fi
-  log "启动服务：$*"
-  compose up -d "$@"
-}
-
-stop_services() {
-  if [[ "$#" -eq 0 ]]; then
-    return
-  fi
-  log "停止服务：$*"
-  compose stop "$@"
-}
-
-start_recorded_services() {
-  if [[ "$#" -eq 0 ]]; then
-    return
-  fi
-  start_services "$@"
-}
-
-wait_service_ready() {
-  local service="$1"
-  local label="$2"
-  local timeout="${3:-120}"
-  local attempt cid status
-
-  log "等待 ${label} 就绪..."
-  for ((attempt = 1; attempt <= timeout; attempt++)); do
-    cid="$(compose ps -q "${service}" 2>/dev/null || true)"
-    if [[ -n "${cid}" ]]; then
-      status="$(docker inspect -f '{{if .State.Health}}{{.State.Health.Status}}{{else}}{{.State.Status}}{{end}}' "${cid}" 2>/dev/null || true)"
-      if [[ "${status}" == "healthy" || "${status}" == "running" ]]; then
-        log_success "${label} 已就绪"
-        return
-      fi
-    fi
-    sleep 1
-  done
-
-  log_error "${label} 启动超时"
-  compose logs --tail=80 "${service}" || true
-  exit 1
-}
-
-remember_running() {
-  local service="$1"
-  local result_array_name="$2"
-  if service_was_running "${service}"; then
-    eval "${result_array_name}+=(\"${service}\")"
-  fi
-}
-
-tar_from_service_volume() {
-  local service="$1"
-  local source_dir="$2"
-  local backup_dir="$3"
-  local output_name="$4"
-  local cid
-  cid="$(container_id "${service}")"
-  log "备份 ${service}:${source_dir} -> ${output_name}"
-  docker run --rm \
-    --volumes-from "${cid}:ro" \
-    -v "${backup_dir}:/backup" \
-    "${HELPER_IMAGE}" \
-    sh -c "cd '${source_dir}' && tar czf '/backup/${output_name}' ."
-}
-
-restore_service_volume() {
-  local service="$1"
-  local target_dir="$2"
-  local backup_dir="$3"
-  local input_name="$4"
-  local cid
-  if [[ ! -f "${backup_dir}/${input_name}" ]]; then
-    log_warn "跳过 ${service}:${target_dir}，备份文件不存在：${input_name}"
-    return
-  fi
-  cid="$(container_id "${service}")"
-  log "还原 ${input_name} -> ${service}:${target_dir}"
-  docker run --rm \
-    --volumes-from "${cid}" \
-    -v "${backup_dir}:/backup:ro" \
-    "${HELPER_IMAGE}" \
-    sh -c "rm -rf '${target_dir}'/* '${target_dir}'/.[!.]* '${target_dir}'/..?* 2>/dev/null || true; tar xzf '/backup/${input_name}' -C '${target_dir}'"
-}
-
-copy_if_exists() {
-  local source="$1"
-  local target="$2"
-  if [[ -e "${source}" ]]; then
-    cp -a "${source}" "${target}"
-  fi
-}
-
-write_manifest() {
-  local backup_dir="$1"
-  {
-    printf "created_at=%s\n" "$(date -Iseconds)"
-    printf "host=%s\n" "$(hostname)"
-    printf "git_commit=%s\n" "$(git -C "${ROOT_DIR}" rev-parse HEAD 2>/dev/null || true)"
-    printf "git_branch=%s\n" "$(git -C "${ROOT_DIR}" branch --show-current 2>/dev/null || true)"
-    printf "mysql_database=%s\n" "$(require_env MYSQL_DATABASE)"
-    printf "storage_bucket=%s\n" "$(require_env STORAGE_BUCKET)"
-  } > "${backup_dir}/manifest.env"
-
-  git -C "${ROOT_DIR}" log --oneline -5 > "${backup_dir}/git-log.txt" 2>/dev/null || true
-  compose ps > "${backup_dir}/docker-compose-ps.txt" 2>/dev/null || true
-  compose config > "${backup_dir}/docker-compose.config.rendered.yml" 2>/dev/null || true
-}
-
-backup_mysql() {
-  local backup_dir="$1"
-  log "备份 MySQL -> mysql.sql"
-  compose exec -T mysql sh -c \
-    'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysqldump -uroot --single-transaction --routines --triggers --events --hex-blob --default-character-set=utf8mb4 "$MYSQL_DATABASE"' \
-    > "${backup_dir}/mysql.sql"
-  compose exec -T mysql sh -c 'mysql --version' > "${backup_dir}/mysql-version.txt" 2>/dev/null || true
-}
-
-backup_goose_status() {
-  local backup_dir="$1"
-  log "记录 goose 迁移状态"
-  compose run --rm --no-deps backend /app/goose -dir /app/migrations mysql "$(goose_dsn "$(require_env MYSQL_DSN)")" status \
-    > "${backup_dir}/goose-status.txt" 2>&1 || true
-}
-
-goose_dsn() {
-  local dsn="$1"
-  if [[ "${dsn}" != *"multiStatements="* ]]; then
-    if [[ "${dsn}" == *"?"* ]]; then
-      dsn="${dsn}&multiStatements=true"
-    else
-      dsn="${dsn}?multiStatements=true"
-    fi
-  fi
-  printf "%s" "${dsn}"
-}
-
-backup_files() {
-  local backup_dir="$1"
-  log "备份配置文件"
-  mkdir -p "${backup_dir}/config"
-  copy_if_exists "${BACKEND_ENV}" "${backup_dir}/config/backend.env"
-  copy_if_exists "${COMPOSE_FILE}" "${backup_dir}/config/docker-compose.prod.yml"
-  copy_if_exists "${ROOT_DIR}/deploy/caddy/Caddyfile" "${backup_dir}/config/Caddyfile"
-  copy_if_exists "${ROOT_DIR}/deploy/mysql/my.cnf" "${backup_dir}/config/mysql.my.cnf"
-
-  if [[ -d "${ROOT_DIR}/backend/logs" ]]; then
-    log "备份后端日志 -> backend_logs.tgz"
-    tar czf "${backup_dir}/backend_logs.tgz" -C "${ROOT_DIR}/backend" logs
-  fi
-}
-
-restore_files() {
-  local backup_dir="$1"
-  local safety_dir="${BACKUP_ROOT}/pre-restore-files-$(date +%Y%m%d_%H%M%S)"
-  mkdir -p "${safety_dir}"
-
-  log "恢复配置文件（当前文件会先备份到 ${safety_dir}）"
-  copy_if_exists "${BACKEND_ENV}" "${safety_dir}/backend.env"
-  copy_if_exists "${COMPOSE_FILE}" "${safety_dir}/docker-compose.prod.yml"
-  copy_if_exists "${ROOT_DIR}/deploy/caddy/Caddyfile" "${safety_dir}/Caddyfile"
-  copy_if_exists "${ROOT_DIR}/deploy/mysql/my.cnf" "${safety_dir}/mysql.my.cnf"
-
-  copy_if_exists "${backup_dir}/config/backend.env" "${BACKEND_ENV}"
-  copy_if_exists "${backup_dir}/config/docker-compose.prod.yml" "${COMPOSE_FILE}"
-  copy_if_exists "${backup_dir}/config/Caddyfile" "${ROOT_DIR}/deploy/caddy/Caddyfile"
-  copy_if_exists "${backup_dir}/config/mysql.my.cnf" "${ROOT_DIR}/deploy/mysql/my.cnf"
-
-  if [[ -f "${backup_dir}/backend_logs.tgz" ]]; then
-    log "恢复后端日志"
-    mkdir -p "${ROOT_DIR}/backend"
-    rm -rf "${ROOT_DIR}/backend/logs"
-    tar xzf "${backup_dir}/backend_logs.tgz" -C "${ROOT_DIR}/backend"
-  fi
-}
-
-prepare_restore_env() {
-  local backup_dir="$1"
-  if [[ ! -f "${BACKEND_ENV}" ]]; then
-    if [[ ! -f "${backup_dir}/config/backend.env" ]]; then
-      log_error "当前缺少 backend/.env，备份中也没有 config/backend.env，无法恢复"
-      exit 1
-    fi
-    log_warn "当前缺少 backend/.env，先从备份恢复环境变量文件"
-    mkdir -p "$(dirname "${BACKEND_ENV}")"
-    cp -a "${backup_dir}/config/backend.env" "${BACKEND_ENV}"
-  fi
-  if [[ ! -f "${COMPOSE_FILE}" && -f "${backup_dir}/config/docker-compose.prod.yml" ]]; then
-    log_warn "当前缺少 deploy/docker-compose.prod.yml，先从备份恢复 Compose 文件"
-    mkdir -p "$(dirname "${COMPOSE_FILE}")"
-    cp -a "${backup_dir}/config/docker-compose.prod.yml" "${COMPOSE_FILE}"
-  fi
-  require_env_file
-}
-
-restore_mysql() {
-  local backup_dir="$1"
-  local database
-  database="$(require_env MYSQL_DATABASE)"
-  if [[ ! "${database}" =~ ^[A-Za-z0-9_]+$ ]]; then
-    log_error "MYSQL_DATABASE 只能包含字母、数字和下划线，当前值：${database}"
-    exit 1
-  fi
-  if [[ ! -f "${backup_dir}/mysql.sql" ]]; then
-    log_error "缺少 MySQL 备份文件：${backup_dir}/mysql.sql"
-    exit 1
-  fi
-
-  start_services mysql
-  wait_service_ready mysql "MySQL"
-  log "重建数据库 ${database}"
-  compose exec -T mysql sh -c \
-    'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot -e "DROP DATABASE IF EXISTS \`$MYSQL_DATABASE\`; CREATE DATABASE \`$MYSQL_DATABASE\` CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci;"'
-  log "导入 MySQL 备份"
-  compose exec -T mysql sh -c 'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot "$MYSQL_DATABASE"' < "${backup_dir}/mysql.sql"
-}
-
-backup_all() {
-  require_env_file
-  need_cmd docker
-  local no_stop="$1"
-  local backup_dir="${BACKUP_ROOT}/$(date +%Y%m%d_%H%M%S)"
-  local restart_services=()
-  local backup_ok=0
-
-  backup_failure_cleanup() {
-    local exit_code=$?
-    trap - ERR
-    if [[ "${no_stop}" == "0" && "${backup_ok}" != "1" ]]; then
-      log_warn "备份失败，尝试恢复备份前运行的服务"
-      start_recorded_services "${restart_services[@]+"${restart_services[@]}"}" || true
-    fi
-    exit "${exit_code}"
-  }
-  trap backup_failure_cleanup ERR
-
-  umask 077
-  mkdir -p "${backup_dir}"
-  chmod 700 "${backup_dir}"
-  log "备份目录：${backup_dir}"
-
-  if [[ "${no_stop}" == "0" ]]; then
-    remember_running caddy restart_services
-    remember_running backend restart_services
-    stop_services caddy backend
-  else
-    log_warn "已开启 --no-stop，备份期间仍可能有写入，建议只在低风险窗口使用"
-  fi
-
-  start_services mysql
-  wait_service_ready mysql "MySQL"
-  backup_mysql "${backup_dir}"
-  backup_goose_status "${backup_dir}"
-
-  if [[ "${no_stop}" == "0" ]]; then
-    remember_running minio restart_services
-    remember_running redis restart_services
-    stop_services minio redis caddy
-  fi
-
-  tar_from_service_volume minio /data "${backup_dir}" minio_data.tgz
-  tar_from_service_volume redis /data "${backup_dir}" redis_data.tgz
-  tar_from_service_volume caddy /data "${backup_dir}" caddy_data.tgz
-  tar_from_service_volume caddy /config "${backup_dir}" caddy_config.tgz
-  backup_files "${backup_dir}"
-  write_manifest "${backup_dir}"
-
-  if [[ "${no_stop}" == "0" ]]; then
-    start_recorded_services "${restart_services[@]+"${restart_services[@]}"}"
-  fi
-
-  backup_ok=1
-  trap - ERR
-  log_success "备份完成：${backup_dir}"
-  log_warn "备份包含生产密钥，请离线加密保存，不要提交到 Git。"
-}
-
-restore_all() {
-  need_cmd docker
-  local backup_dir="$1"
-  local yes="$2"
-  local restore_ok=0
-
-  if [[ -z "${backup_dir}" || ! -d "${backup_dir}" ]]; then
-    log_error "备份目录不存在：${backup_dir}"
-    exit 1
-  fi
-  if [[ "${yes}" != "1" ]]; then
-    log_error "restore 会覆盖当前生产数据库、MinIO、Redis、Caddy volume 和配置文件。确认执行请追加 --yes"
-    exit 1
-  fi
-  backup_dir="$(absolute_dir "${backup_dir}")"
-
-  restore_failure_cleanup() {
-    local exit_code=$?
-    trap - ERR
-    if [[ "${restore_ok}" != "1" ]]; then
-      log_warn "恢复失败，尝试启动当前配置下的服务，便于继续排查"
-      compose up -d || true
-    fi
-    exit "${exit_code}"
-  }
-  trap restore_failure_cleanup ERR
-
-  prepare_restore_env "${backup_dir}"
-  log_warn "即将从 ${backup_dir} 恢复，当前生产数据会被覆盖。"
-  stop_services backend caddy minio redis
-  restore_files "${backup_dir}"
-  restore_mysql "${backup_dir}"
-
-  restore_service_volume minio /data "${backup_dir}" minio_data.tgz
-  restore_service_volume redis /data "${backup_dir}" redis_data.tgz
-  restore_service_volume caddy /data "${backup_dir}" caddy_data.tgz
-  restore_service_volume caddy /config "${backup_dir}" caddy_config.tgz
-
-  log "启动生产服务"
-  compose up -d
-  restore_ok=1
-  trap - ERR
-  log_success "恢复完成。建议立刻执行健康检查并抽查图片/二维码/订单数据。"
-}
-
-list_backups() {
-  if [[ ! -d "${BACKUP_ROOT}" ]]; then
-    log_warn "备份目录不存在：${BACKUP_ROOT}"
-    return
-  fi
-  find "${BACKUP_ROOT}" -mindepth 2 -maxdepth 2 -type f -name manifest.env -exec dirname {} \; | sort
-}
-
-main() {
-  local command="${1:-}"
-  local no_stop=0
-  local yes=0
-  local restore_dir=""
-
-  if [[ -z "${command}" || "${command}" == "-h" || "${command}" == "--help" ]]; then
-    show_help
-    exit 0
-  fi
-  shift || true
-
-  case "${command}" in
-    backup)
-      while [[ $# -gt 0 ]]; do
-        case "$1" in
-          --no-stop) no_stop=1; shift ;;
-          -h|--help) show_help; exit 0 ;;
-          *) log_error "未知选项：$1"; show_help; exit 1 ;;
-        esac
-      done
-      backup_all "${no_stop}"
-      ;;
-    restore)
-      restore_dir="${1:-}"
-      if [[ $# -gt 0 ]]; then
-        shift
-      fi
-      while [[ $# -gt 0 ]]; do
-        case "$1" in
-          --yes) yes=1; shift ;;
-          -h|--help) show_help; exit 0 ;;
-          *) log_error "未知选项：$1"; show_help; exit 1 ;;
-        esac
-      done
-      restore_all "${restore_dir}" "${yes}"
-      ;;
-    list)
-      list_backups
-      ;;
-    *)
-      log_error "未知命令：${command}"
-      show_help
-      exit 1
-      ;;
-  esac
-}
-
-main "$@"

scripts/deploy-prod.sh

@@ -146,23 +146,28 @@ validate_env() {
     exit 1
   fi
 
-  local app_env caddy_domain caddy_email mysql_dsn redis_addr storage_endpoint backend_uid backend_gid
-  local minio_root_user minio_root_password storage_access_key_id storage_secret_access_key
+  local app_env caddy_domain caddy_email mysql_dsn redis_addr storage_endpoint backend_uid backend_gid backup_dir backup_restore_dir backup_oss_uri backup_user
   app_env="$(require_env APP_ENV)"
   caddy_domain="$(require_env CADDY_DOMAIN)"
   caddy_email="$(require_env CADDY_EMAIL)"
   mysql_dsn="$(require_env MYSQL_DSN)"
   redis_addr="$(require_env REDIS_ADDR)"
   storage_endpoint="$(require_env STORAGE_ENDPOINT)"
-  require_env MYSQL_DATABASE >/dev/null
+  local mysql_database
+  mysql_database="$(require_env MYSQL_DATABASE)"
   require_env MYSQL_ROOT_PASSWORD >/dev/null
   require_env MYSQL_USER >/dev/null
   require_env MYSQL_PASSWORD >/dev/null
   require_env JWT_SECRET >/dev/null
-  minio_root_user="$(require_env MINIO_ROOT_USER)"
-  minio_root_password="$(require_env MINIO_ROOT_PASSWORD)"
-  storage_access_key_id="$(require_env STORAGE_ACCESS_KEY_ID)"
-  storage_secret_access_key="$(require_env STORAGE_SECRET_ACCESS_KEY)"
+  require_env STORAGE_ACCESS_KEY_ID >/dev/null
+  require_env STORAGE_SECRET_ACCESS_KEY >/dev/null
+  backup_dir="$(require_env BACKUP_DIR)"
+  backup_restore_dir="$(require_env BACKUP_RESTORE_DIR)"
+  backup_oss_uri="$(require_env BACKUP_OSS_URI)"
+  require_env BACKUP_PASSPHRASE >/dev/null
+  require_env BACKUP_KMS_KEY_ID >/dev/null
+  backup_user="$(require_env BACKUP_MYSQL_USER)"
+  require_env BACKUP_MYSQL_PASSWORD >/dev/null
   backend_uid="$(require_env BACKEND_UID)"
   backend_gid="$(require_env BACKEND_GID)"
 
@@ -197,14 +202,36 @@ validate_env() {
     exit 1
   fi
   if [[ "${storage_endpoint}" == *"127.0.0.1"* || "${storage_endpoint}" == *"localhost"* ]]; then
-    log_error "STORAGE_ENDPOINT 仍指向本机，请改为 http://minio:9000"
+    log_error "STORAGE_ENDPOINT 仍指向本机，请改为阿里云 OSS Endpoint（如 https://oss-cn-hangzhou.aliyuncs.com）"
     exit 1
   fi
-  if [[ "${storage_endpoint}" == "http://minio:9000" || "${storage_endpoint}" == "minio:9000" ]]; then
-    if [[ "${storage_access_key_id}" != "${minio_root_user}" || "${storage_secret_access_key}" != "${minio_root_password}" ]]; then
-      log_error "使用内置 MinIO 时，STORAGE_ACCESS_KEY_ID/SECRET_ACCESS_KEY 必须和 MINIO_ROOT_USER/ROOT_PASSWORD 保持一致"
-      exit 1
-    fi
+  if [[ "${storage_endpoint}" == *"minio"* ]]; then
+    log_error "STORAGE_ENDPOINT 仍指向已下线的 MinIO，请改为阿里云 OSS Endpoint"
+    exit 1
+  fi
+  if [[ ! "${storage_endpoint}" =~ ^https://oss-[A-Za-z0-9-]+\.aliyuncs\.com/?$ ]]; then
+    log_error "STORAGE_ENDPOINT 必须是阿里云 OSS HTTPS Endpoint，当前值：${storage_endpoint}"
+    exit 1
+  fi
+  if [[ "${backup_dir}" != /data/* || "${backup_dir}" == "/data" ]]; then
+    log_error "BACKUP_DIR 必须是 /data 之下的具体绝对路径，当前值：${backup_dir}"
+    exit 1
+  fi
+  if [[ "${backup_restore_dir}" != /data/* || "${backup_restore_dir}" == "/data" || "${backup_restore_dir}" == "${backup_dir}" ]]; then
+    log_error "BACKUP_RESTORE_DIR 必须是与 BACKUP_DIR 不同的 /data 子目录，当前值：${backup_restore_dir}"
+    exit 1
+  fi
+  if [[ ! "${backup_oss_uri}" =~ ^oss://[A-Za-z0-9][A-Za-z0-9._-]*$ ]]; then
+    log_error "BACKUP_OSS_URI 必须是 oss://bucket 格式，当前值：${backup_oss_uri}"
+    exit 1
+  fi
+  if [[ ! "${backup_user}" =~ ^[A-Za-z0-9_]{1,32}$ ]]; then
+    log_error "BACKUP_MYSQL_USER 只能包含字母、数字和下划线，长度不超过 32"
+    exit 1
+  fi
+  if [[ ! "${mysql_database}" =~ ^[A-Za-z0-9_]+$ ]]; then
+    log_error "MYSQL_DATABASE 只能包含字母、数字和下划线，当前值：${mysql_database}"
+    exit 1
   fi
   if awk -F= '/^[A-Z0-9_]+=/{ if ($0 ~ /=(change-|保持你的|你的)/) found=1 } END { exit found ? 0 : 1 }' "${BACKEND_ENV}"; then
     log_error "backend/.env 里还有占位值，请先替换后再部署"
@@ -407,6 +434,34 @@ mysql_root_exec() {
   compose exec -T mysql sh -c 'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot "$@"' sh "$@"
 }
 
+sql_string_literal() {
+  local value="$1"
+  value="${value//\\/\\\\}"
+  value="${value//\'/\'\'}"
+  printf '%s' "${value}"
+}
+
+ensure_mysql_backup_user() {
+  local database backup_user backup_password escaped_password
+  database="$(require_env MYSQL_DATABASE)"
+  backup_user="$(require_env BACKUP_MYSQL_USER)"
+  backup_password="$(require_env BACKUP_MYSQL_PASSWORD)"
+  escaped_password="$(sql_string_literal "${backup_password}")"
+  unset backup_password
+
+  log "配置 MySQL 在线备份账号 ${backup_user}"
+  printf "CREATE USER IF NOT EXISTS \`%s\`@'%%' IDENTIFIED BY '%s';\n" "${backup_user}" "${escaped_password}" |
+    compose exec -T mysql sh -c 'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot --protocol=socket'
+  printf "ALTER USER \`%s\`@'%%' IDENTIFIED BY '%s';\n" "${backup_user}" "${escaped_password}" |
+    compose exec -T mysql sh -c 'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot --protocol=socket'
+  printf "GRANT BACKUP_ADMIN, PROCESS, RELOAD, LOCK TABLES, REPLICATION CLIENT ON *.* TO \`%s\`@'%%';\n" "${backup_user}" |
+    compose exec -T mysql sh -c 'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot --protocol=socket'
+  printf "GRANT SELECT ON performance_schema.log_status TO \`%s\`@'%%';\n" "${backup_user}" |
+    compose exec -T mysql sh -c 'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot --protocol=socket'
+  printf "GRANT SELECT ON \`%s\`.* TO \`%s\`@'%%';\n" "${database}" "${backup_user}" |
+    compose exec -T mysql sh -c 'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot --protocol=socket'
+}
+
 goose_dsn() {
   local dsn="$1"
   if [[ "${dsn}" != *"multiStatements="* ]]; then
@@ -492,6 +547,9 @@ main() {
   cd "${ROOT_DIR}"
   need_cmd docker
   need_cmd curl
+  need_cmd openssl
+  need_cmd ossutil
+  need_cmd flock
   validate_env
   export_caddy_env
   prepare_log_dir
@@ -534,8 +592,8 @@ main() {
 
   wait_service_healthy mysql "MySQL"
   wait_service_healthy redis "Redis"
-  wait_service_healthy minio "MinIO"
   wait_service_healthy caddy "Caddy"
+  ensure_mysql_backup_user
   reset_database
   run_migrations
 

scripts/restore-online.sh

@@ -0,0 +1,248 @@
+#!/usr/bin/env bash
+set -euo pipefail
+umask 077
+
+# 只用于隔离恢复演练或灾难恢复准备。它从不触碰生产 mysql 容器：prepare
+# 在 BACKUP_RESTORE_DIR 下生成数据目录，start 仅启动 --network none 的临时容器。
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+# shellcheck source=backup-common.sh
+source "${SCRIPT_DIR}/backup-common.sh"
+
+field_value() {
+  local file="$1" key="$2"
+  awk -F= -v key="${key}" '$1 == key { print substr($0, index($0, "=") + 1); exit }' "${file}"
+}
+
+require_restore_dir() {
+  local dir
+  dir="$(backup_require_env BACKUP_RESTORE_DIR)"
+  backup_validate_dir "${dir}"
+  printf '%s' "${dir}"
+}
+
+validate_sha256() {
+  [[ "$1" =~ ^[0-9a-f]{64}$ ]] || { backup_error "非法 SHA-256：$1"; exit 1; }
+}
+
+validate_server_uuid() {
+  [[ "$1" =~ ^[0-9a-f-]{36}$ ]] || { backup_error "非法 server_uuid：$1"; exit 1; }
+}
+
+verify_downloaded_payload() {
+  local payload="$1" checksum_file="$2" complete_file="$3"
+  local expected actual expected_size actual_size
+  expected="$(field_value "${complete_file}" payload_sha256)"
+  expected_size="$(field_value "${complete_file}" payload_size_bytes)"
+  validate_sha256 "${expected}"
+  [[ "${expected_size}" =~ ^[0-9]+$ ]] || { backup_error '完成标识中的文件大小不合法'; exit 1; }
+  actual="$(backup_sha256 "${payload}")"
+  actual_size="$(backup_file_size "${payload}")"
+  [[ "${expected}" == "${actual}" && "${expected_size}" == "${actual_size}" ]] || {
+    backup_error "下载文件校验失败：${payload}"
+    exit 1
+  }
+  grep -Fqx "${expected}  $(basename "${payload}")" "${checksum_file}" || {
+    backup_error "校验文件内容不匹配：${checksum_file}"
+    exit 1
+  }
+}
+
+prepare_full() {
+  backup_require_cmd docker
+  backup_require_cmd ossutil
+  backup_require_cmd openssl
+  local complete_uri="$1"
+  [[ "${complete_uri}" == oss://*/complete.env ]] || {
+    backup_error 'prepare 参数必须是完整全量备份的 complete.env OSS URI'
+    exit 1
+  }
+  local restore_dir remote_dir ts server_uuid work_dir download_dir
+  restore_dir="$(require_restore_dir)"
+  remote_dir="${complete_uri%/complete.env}"
+  ts="${remote_dir##*/}"
+  server_uuid="$(basename "$(dirname "${remote_dir}")")"
+  validate_server_uuid "${server_uuid}"
+  work_dir="${restore_dir}/${server_uuid}/${ts}"
+  download_dir="${work_dir}/.download"
+  [[ ! -e "${work_dir}" ]] || { backup_error "恢复目录已存在：${work_dir}"; exit 1; }
+  mkdir -p "${download_dir}" "${work_dir}/data"
+  trap 'rm -rf -- "${download_dir}"' EXIT
+
+  backup_oss_download "${complete_uri}" "${download_dir}/complete.env"
+  [[ "$(field_value "${download_dir}/complete.env" payload)" == 'mysql-full.tgz.enc' ]] || {
+    backup_error '不是受支持的全量备份完成标识'
+    exit 1
+  }
+  [[ "$(field_value "${download_dir}/complete.env" server_uuid)" == "${server_uuid}" ]] || {
+    backup_error '备份路径与完成标识的 server_uuid 不一致'
+    exit 1
+  }
+  backup_oss_download "${remote_dir}/mysql-full.tgz.enc" "${download_dir}/mysql-full.tgz.enc"
+  backup_oss_download "${remote_dir}/mysql-full.tgz.enc.sha256" "${download_dir}/mysql-full.tgz.enc.sha256"
+  verify_downloaded_payload "${download_dir}/mysql-full.tgz.enc" "${download_dir}/mysql-full.tgz.enc.sha256" "${download_dir}/complete.env"
+  backup_decrypt_file "${download_dir}/mysql-full.tgz.enc" "${download_dir}/mysql-full.tgz"
+  tar xzf "${download_dir}/mysql-full.tgz" -C "${work_dir}/data"
+  local manifest_uuid binlog_file binlog_pos
+  manifest_uuid="$(field_value "${work_dir}/data/manifest.env" server_uuid)"
+  binlog_file="$(field_value "${work_dir}/data/manifest.env" binlog_file)"
+  binlog_pos="$(field_value "${work_dir}/data/manifest.env" binlog_pos)"
+  [[ "${manifest_uuid}" == "${server_uuid}" ]] || { backup_error '归档内 manifest 与完成标识不一致'; exit 1; }
+  backup_binlog_index "${binlog_file}" >/dev/null
+  [[ "${binlog_pos}" =~ ^[0-9]+$ ]] || { backup_error 'manifest 中 binlog 位点非法'; exit 1; }
+  docker run --rm -v "${work_dir}/data:/restore" "${BACKUP_XTRABACKUP_IMAGE}" \
+    xtrabackup --prepare --target-dir=/restore
+  {
+    printf 'server_uuid=%s\n' "${server_uuid}"
+    printf 'binlog_file=%s\n' "${binlog_file}"
+    printf 'binlog_pos=%s\n' "${binlog_pos}"
+    printf 'source_complete_uri=%s\n' "${complete_uri}"
+    printf 'prepared_at=%s\n' "$(date -Iseconds)"
+  } > "${work_dir}/prepared.env"
+  rm -rf -- "${download_dir}"
+  trap - EXIT
+  backup_log "全量备份已 prepare：${work_dir}"
+  backup_log "下一步执行：$0 fetch-all ${work_dir}"
+}
+
+fetch_binlog() {
+  backup_require_cmd docker
+  backup_require_cmd ossutil
+  backup_require_cmd openssl
+  local prepared_dir="$1" complete_uri="$2"
+  [[ -f "${prepared_dir}/prepared.env" ]] || { backup_error "未 prepare 的恢复目录：${prepared_dir}"; exit 1; }
+  [[ "${complete_uri}" == oss://*/complete.env ]] || { backup_error 'binlog 参数必须是 complete.env OSS URI'; exit 1; }
+  local remote_dir temp_dir complete source_file raw_sha payload server_uuid expected_uuid target_base
+  remote_dir="${complete_uri%/complete.env}"
+  temp_dir="${prepared_dir}/.binlog-download-$$"
+  mkdir -p "${temp_dir}" "${prepared_dir}/binlogs"
+  trap 'rm -rf -- "${temp_dir}"' EXIT
+  backup_oss_download "${complete_uri}" "${temp_dir}/complete.env"
+  source_file="$(field_value "${temp_dir}/complete.env" source_binlog)"
+  raw_sha="$(field_value "${temp_dir}/complete.env" raw_sha256)"
+  payload="$(field_value "${temp_dir}/complete.env" payload)"
+  server_uuid="$(field_value "${temp_dir}/complete.env" server_uuid)"
+  expected_uuid="$(field_value "${prepared_dir}/prepared.env" server_uuid)"
+  backup_binlog_index "${source_file}" >/dev/null
+  validate_sha256 "${raw_sha}"
+  validate_server_uuid "${server_uuid}"
+  [[ "${payload}" == 'binlog.enc' && "${server_uuid}" == "${expected_uuid}" ]] || {
+    backup_error 'binlog 完成标识与恢复目标不匹配'
+    exit 1
+  }
+  target_base="${prepared_dir}/binlogs/${source_file}.${raw_sha}"
+  if [[ -f "${target_base}.meta.env" && -f "${target_base}" ]]; then
+    rm -rf -- "${temp_dir}"
+    trap - EXIT
+    return
+  fi
+  backup_oss_download "${remote_dir}/binlog.enc" "${temp_dir}/binlog.enc"
+  backup_oss_download "${remote_dir}/binlog.enc.sha256" "${temp_dir}/binlog.enc.sha256"
+  verify_downloaded_payload "${temp_dir}/binlog.enc" "${temp_dir}/binlog.enc.sha256" "${temp_dir}/complete.env"
+  backup_decrypt_file "${temp_dir}/binlog.enc" "${target_base}"
+  docker run --rm -v "${prepared_dir}/binlogs:/binlogs:ro" mysql:8.4 \
+    mysqlbinlog --verify-binlog-checksum "/binlogs/$(basename "${target_base}")" >/dev/null
+  mv "${temp_dir}/complete.env" "${target_base}.meta.env"
+  rm -rf -- "${temp_dir}"
+  trap - EXIT
+}
+
+fetch_all() {
+  backup_require_cmd ossutil
+  local prepared_dir="$1" server_uuid oss_uri complete_uri
+  [[ -f "${prepared_dir}/prepared.env" ]] || { backup_error "未 prepare 的恢复目录：${prepared_dir}"; exit 1; }
+  server_uuid="$(field_value "${prepared_dir}/prepared.env" server_uuid)"
+  oss_uri="$(backup_require_env BACKUP_OSS_URI)"
+  backup_validate_oss_uri "${oss_uri}"
+  while IFS= read -r complete_uri; do
+    [[ -n "${complete_uri}" ]] || continue
+    fetch_binlog "${prepared_dir}" "${complete_uri}"
+  done < <(ossutil ls "${oss_uri}/mysql/binlog/${server_uuid}/" -r | awk '$NF ~ /\/complete\.env$/ { print $NF }')
+  backup_log "已下载实例 ${server_uuid} 的全部完整 binlog"
+}
+
+start_isolated_mysql() {
+  backup_require_cmd docker
+  local prepared_dir="$1" name="${2:-}"
+  [[ -f "${prepared_dir}/prepared.env" ]] || { backup_error "未 prepare 的恢复目录：${prepared_dir}"; exit 1; }
+  if [[ -z "${name}" ]]; then
+    name="hfb-restore-$(basename "${prepared_dir}")"
+  fi
+  [[ "${name}" =~ ^hfb-restore-[A-Za-z0-9_-]+$ ]] || { backup_error '恢复容器名必须以 hfb-restore- 开头'; exit 1; }
+  [[ -z "$(docker ps -aq -f "name=^/${name}$")" ]] || { backup_error "恢复容器已存在：${name}"; exit 1; }
+  docker run -d --name "${name}" --network none \
+    -e "MYSQL_ROOT_PASSWORD=$(backup_require_env MYSQL_ROOT_PASSWORD)" \
+    -v "${prepared_dir}/data:/var/lib/mysql" mysql:8.4 \
+    --skip-log-bin --skip-networking=1 >/dev/null
+  local attempt
+  for ((attempt = 1; attempt <= 60; attempt++)); do
+    if docker exec "${name}" sh -c 'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysqladmin -uroot --protocol=socket ping' >/dev/null 2>&1; then
+      backup_log "隔离 MySQL 已启动：${name}"
+      return
+    fi
+    sleep 1
+  done
+  backup_error "隔离 MySQL 启动失败，请执行 docker logs ${name}"
+  exit 1
+}
+
+replay_binlogs() {
+  backup_require_cmd docker
+  local prepared_dir="$1" target_container="$2" stop_datetime="${3:-}"
+  [[ -f "${prepared_dir}/prepared.env" ]] || { backup_error "未 prepare 的恢复目录：${prepared_dir}"; exit 1; }
+  local production_cid
+  production_cid="$(backup_mysql_container_id)"
+  [[ "${target_container}" != "${production_cid}" ]] || { backup_error '禁止向生产 MySQL 回放 binlog'; exit 1; }
+  docker inspect "${target_container}" >/dev/null
+  [[ -d "${prepared_dir}/binlogs" ]] || { backup_error '尚未下载 binlog'; exit 1; }
+  local start_file start_pos metadata index source previous_index found=0
+  start_file="$(field_value "${prepared_dir}/prepared.env" binlog_file)"
+  start_pos="$(field_value "${prepared_dir}/prepared.env" binlog_pos)"
+  mapfile -t metadata < <(
+    for metadata in "${prepared_dir}"/binlogs/*.meta.env; do
+      [[ -f "${metadata}" ]] || continue
+      index="$(field_value "${metadata}" binlog_index)"
+      source="$(field_value "${metadata}" source_binlog)"
+      printf '%010d\t%s\t%s\n' "${index}" "${source}" "${metadata}"
+    done | sort -n
+  )
+  [[ "${#metadata[@]}" -gt 0 ]] || { backup_error '没有可回放的 binlog'; exit 1; }
+  local binlog_paths=()
+  previous_index=''
+  local row path
+  for row in "${metadata[@]}"; do
+    IFS=$'\t' read -r index source metadata <<< "${row}"
+    if [[ "${source}" == "${start_file}" ]]; then
+      found=1
+    fi
+    [[ "${found}" == '1' ]] || continue
+    if [[ -n "${previous_index}" && "${index}" -ne $((previous_index + 1)) ]]; then
+      backup_error "恢复 binlog 不连续：${previous_index} -> ${index}"
+      exit 1
+    fi
+    path="/binlogs/$(basename "${metadata%.meta.env}")"
+    binlog_paths+=("${path}")
+    previous_index="${index}"
+  done
+  [[ "${found}" == '1' ]] || { backup_error "缺少全量备份起始 binlog：${start_file}"; exit 1; }
+  local mysqlbinlog_args=(mysqlbinlog "--start-position=${start_pos}")
+  if [[ -n "${stop_datetime}" ]]; then
+    mysqlbinlog_args+=("--stop-datetime=${stop_datetime}")
+  fi
+  mysqlbinlog_args+=("${binlog_paths[@]}")
+  docker run --rm -i -v "${prepared_dir}/binlogs:/binlogs:ro" mysql:8.4 "${mysqlbinlog_args[@]}" |
+    docker exec -i "${target_container}" sh -c 'MYSQL_PWD="$MYSQL_ROOT_PASSWORD" mysql -uroot --protocol=socket'
+  backup_log "binlog 已回放至隔离 MySQL：${target_container}"
+}
+
+case "${1:-}" in
+  prepare) [[ $# -eq 2 ]] || { backup_error '用法：prepare <full-complete-uri>'; exit 1; }; prepare_full "$2" ;;
+  fetch-all) [[ $# -eq 2 ]] || { backup_error '用法：fetch-all <prepared-dir>'; exit 1; }; fetch_all "$2" ;;
+  fetch-binlog) [[ $# -eq 3 ]] || { backup_error '用法：fetch-binlog <prepared-dir> <binlog-complete-uri>'; exit 1; }; fetch_binlog "$2" "$3" ;;
+  start) [[ $# -le 3 && $# -ge 2 ]] || { backup_error '用法：start <prepared-dir> [container-name]'; exit 1; }; start_isolated_mysql "$2" "${3:-}" ;;
+  replay) [[ $# -ge 3 && $# -le 4 ]] || { backup_error '用法：replay <prepared-dir> <container> [YYYY-MM-DD HH:MM:SS]'; exit 1; }; replay_binlogs "$2" "$3" "${4:-}" ;;
+  *)
+    printf '用法: %s {prepare|fetch-all|fetch-binlog|start|replay} ...\n' "$0" >&2
+    exit 1
+    ;;
+esac