完善数据库备份恢复链路

.env.server.example

@@ -57,6 +57,19 @@ STORAGE_OSS_ACCESS_KEY_ID=change-me-oss-access-key
 STORAGE_OSS_SECRET_ACCESS_KEY=change-me-oss-secret-key
 STORAGE_OSS_REGION=oss-cn-hangzhou
 
+# 数据库备份 → OSS(deploy/backup-db.sh 与 dist/db-backup.js 使用)。
+# 默认复用上面的 STORAGE_OSS_* 上传到同桶 backups/ 前缀,以下仅需按需调整。
+DB_BACKUP_OSS_PREFIX=backups/postgres
+# 服务器本地保留的备份份数(OSS 端过期请配置桶生命周期规则)
+DB_BACKUP_KEEP=7
+# 建议为备份单独创建 RAM 用户、只授予 backups/ 前缀权限(策略见 docs/数据库备份与恢复.md);
+# 使用独立桶/密钥时取消注释覆盖:
+# DB_BACKUP_OSS_ENDPOINT=https://oss-cn-hangzhou-internal.aliyuncs.com
+# DB_BACKUP_OSS_BUCKET=change-me-backup-bucket
+# DB_BACKUP_OSS_ACCESS_KEY_ID=change-me-backup-access-key
+# DB_BACKUP_OSS_SECRET_ACCESS_KEY=change-me-backup-secret-key
+# DB_BACKUP_OSS_REGION=oss-cn-hangzhou
+
 # Admin
 ADMIN_SESSION_SECRET=change-me-long-random-session-secret
 ADMIN_DEFAULT_USERS_JSON=[{"username":"admin","password":"change-me-admin-password","role":"admin"},{"username":"operator","password":"change-me-operator-password","role":"operator"}]

.gitignore

@@ -32,4 +32,7 @@ apps/backend/data/logs
 send_code/
 .reasonix/
 *.tsbuildinfo
-.zcode/
+.zcode/
+
+# 服务器本地备份暂存目录(deploy/backup-db.sh)
+/backups/

apps/backend/src/config/env-documentation.test.ts

@@ -17,6 +17,8 @@ const DEPLOYMENT_ONLY_ENV_NAMES = [
   'BACKEND_PORT',
   'CADDY_SITE_ADDR',
   'CHOKIDAR_USEPOLLING',
+  'DB_BACKUP_KEEP',
+  'DB_BACKUP_OSS_PREFIX',
   'KUASHOU_INDUSTRY_RATE_LIMIT_MAX',
   'MINIO_API_PORT',
   'MINIO_CONSOLE_PORT',

apps/backend/src/db-backup.test.ts

@@ -0,0 +1,202 @@
+import assert from 'node:assert/strict'
+import test from 'node:test'
+import { Readable } from 'node:stream'
+
+import {
+  manifestKeyFor,
+  parseBackupArgs,
+  pickLatestDumpKey,
+  resolveBackupStorageConfig,
+  runDownload,
+  runLatest,
+  runList,
+  runUpload,
+} from './db-backup.js'
+import type { StorageObjectSummary } from './services/file-storage/object-storage.js'
+
+test('resolveBackupStorageConfig 优先 DB_BACKUP_OSS_* 并按字段回退 STORAGE_OSS_*', () => {
+  const config = resolveBackupStorageConfig({
+    DB_BACKUP_OSS_ENDPOINT: 'https://backup.example.com',
+    STORAGE_OSS_ENDPOINT: 'https://storage.example.com',
+    STORAGE_OSS_BUCKET: 'order-site',
+    STORAGE_OSS_ACCESS_KEY_ID: 'storage-key',
+    DB_BACKUP_OSS_ACCESS_KEY_ID: 'backup-key',
+    STORAGE_OSS_SECRET_ACCESS_KEY: 'storage-secret',
+  })
+
+  assert.equal(config.endpoint, 'https://backup.example.com')
+  assert.equal(config.bucket, 'order-site')
+  assert.equal(config.accessKeyId, 'backup-key')
+  assert.equal(config.secretAccessKey, 'storage-secret')
+  assert.equal(config.region, 'oss-cn-hangzhou')
+})
+
+test('resolveBackupStorageConfig 全部未配置时返回空值由调用方校验', () => {
+  const config = resolveBackupStorageConfig({})
+  assert.equal(config.endpoint, '')
+  assert.equal(config.bucket, '')
+})
+
+test('parseBackupArgs 解析命令与参数并去掉前缀尾部斜杠', () => {
+  const args = parseBackupArgs([
+    'upload',
+    '--key',
+    'backups/postgres/order_site-20260824-030000.dump',
+    '--prefix',
+    'backups/postgres/',
+    '--kind',
+    'app-configs',
+    '--size',
+    '1234',
+  ])
+
+  assert.equal(args.command, 'upload')
+  assert.equal(args.key, 'backups/postgres/order_site-20260824-030000.dump')
+  assert.equal(args.prefix, 'backups/postgres')
+  assert.equal(args.kind, 'app-configs')
+  assert.equal(args.size, 1234)
+})
+
+test('parseBackupArgs 无参数时使用默认前缀与类型', () => {
+  const args = parseBackupArgs(['list'])
+  assert.equal(args.command, 'list')
+  assert.equal(args.prefix, 'backups/postgres')
+  assert.equal(args.kind, 'postgres-dump')
+  assert.equal(args.key, '')
+  assert.equal(args.allowUnverified, false)
+})
+
+test('parseBackupArgs 支持显式兼容未校验备份', () => {
+  assert.equal(parseBackupArgs(['latest', '--allow-unverified']).allowUnverified, true)
+})
+
+test('manifestKeyFor 在备份 key 后追加清单后缀', () => {
+  assert.equal(manifestKeyFor('backups/postgres/a.dump'), 'backups/postgres/a.dump.manifest.json')
+})
+
+test('pickLatestDumpKey 只认 dump 并取最近修改的对象', () => {
+  const key = pickLatestDumpKey([
+    {
+      key: 'backups/postgres/a.dump',
+      size: 1,
+      etag: '',
+      lastModified: new Date('2026-08-01T00:00:00Z'),
+    },
+    {
+      key: 'backups/postgres/a.dump.manifest.json',
+      size: 1,
+      etag: '',
+      lastModified: new Date('2026-08-23T00:00:00Z'),
+    },
+    {
+      key: 'backups/configs/a.tar.gz',
+      size: 1,
+      etag: '',
+      lastModified: new Date('2026-08-23T00:00:00Z'),
+    },
+    {
+      key: 'backups/postgres/b.dump',
+      size: 1,
+      etag: '',
+      lastModified: new Date('2026-08-22T00:00:00Z'),
+    },
+  ])
+
+  assert.equal(key, 'backups/postgres/b.dump')
+})
+
+test('pickLatestDumpKey 无备份时返回 null,无时间时按 key 兜底', () => {
+  assert.equal(pickLatestDumpKey([]), null)
+  assert.equal(
+    pickLatestDumpKey([
+      { key: 'backups/postgres/order_site-1.dump', size: 1, etag: '' },
+      { key: 'backups/postgres/order_site-2.dump', size: 1, etag: '' },
+    ]),
+    'backups/postgres/order_site-2.dump',
+  )
+})
+
+test('runUpload 支持流式输入并生成带长度和 sha256 的 manifest', async () => {
+  const uploads: Array<{ key: string; content: Buffer; contentLength?: number }> = []
+  const storage = {
+    putObject: async (input: {
+      key: string
+      content: Buffer | Readable
+      contentLength?: number
+    }) => {
+      const chunks: Buffer[] = []
+      if (Buffer.isBuffer(input.content)) {
+        chunks.push(input.content)
+      } else {
+        for await (const chunk of input.content) chunks.push(Buffer.from(chunk))
+      }
+      uploads.push({
+        key: input.key,
+        content: Buffer.concat(chunks),
+        contentLength: input.contentLength,
+      })
+    },
+  } as never
+
+  await runUpload(
+    storage,
+    parseBackupArgs([
+      'upload',
+      '--key',
+      'backups/configs/a.tar.gz',
+      '--kind',
+      'app-configs',
+      '--size',
+      '5',
+    ]),
+    Readable.from([Buffer.from('hello')]),
+  )
+
+  assert.equal(uploads.length, 2)
+  assert.equal(uploads[0]?.content.toString(), 'hello')
+  assert.equal(uploads[0]?.contentLength, 5)
+  assert.match(uploads[1]?.content.toString() || '', /"size": 5/)
+  assert.match(uploads[1]?.content.toString() || '', /sha256/)
+})
+
+test('runLatest 跳过没有 manifest 的新对象并选择最近的有效备份', async () => {
+  const objects: StorageObjectSummary[] = [
+    {
+      key: 'backups/postgres/new.dump',
+      size: 5,
+      etag: '',
+      lastModified: new Date('2026-08-23T03:00:00Z'),
+    },
+    {
+      key: 'backups/postgres/valid.dump',
+      size: 5,
+      etag: '',
+      lastModified: new Date('2026-08-22T03:00:00Z'),
+    },
+  ]
+  const storage = {
+    listObjects: async () => objects,
+    getObject: async (key: string) => {
+      if (key.endsWith('new.dump.manifest.json')) {
+        const error = Object.assign(new Error('missing'), { name: 'NoSuchKey' })
+        throw error
+      }
+      return {
+        reader: Readable.from([
+          JSON.stringify({
+            key: 'backups/postgres/valid.dump',
+            kind: 'postgres-dump',
+            size: 5,
+            sha256: 'a'.repeat(64),
+            createdAt: '2026-08-22T03:00:00Z',
+          }),
+        ]),
+      }
+    },
+  } as never
+
+  assert.equal(
+    await runLatest(storage, parseBackupArgs(['latest', '--prefix', 'backups/postgres'])),
+    'backups/postgres/valid.dump',
+  )
+})

apps/backend/src/db-backup.ts

@@ -0,0 +1,539 @@
+import crypto from 'node:crypto'
+import fs from 'node:fs'
+import fsPromises from 'node:fs/promises'
+import path from 'node:path'
+import process from 'node:process'
+import { Transform, type Readable } from 'node:stream'
+import { pipeline } from 'node:stream/promises'
+import { pathToFileURL } from 'node:url'
+
+import {
+  createConfiguredObjectStorage,
+  type ObjectStorage,
+  type StorageObjectSummary,
+} from './services/file-storage/object-storage.js'
+
+// 数据库备份传输工具:在 backend 容器内运行,负责备份文件与阿里云 OSS 之间的传输与校验。
+// 由 deploy/backup-db.sh 与 deploy/restore-db.sh 通过 docker compose 调用。
+// 日志一律走 stderr,stdout 只输出机器可读内容(下载的二进制、latest 的 key、list 的行),
+// 便于宿主机脚本重定向捕获。
+//
+// 用法:
+//   node dist/db-backup.js upload --key backups/postgres/xxx.dump [--kind postgres-dump|app-configs] [--size bytes]
+//     从 stdin 读取备份内容上传,并生成 <key>.manifest.json(size/sha256/迁移记录)。
+//   node dist/db-backup.js download --key backups/postgres/xxx.dump [--allow-unverified]
+//     下载到 stdout,并按清单校验 sha256 与大小。
+//   node dist/db-backup.js latest [--prefix backups/postgres] [--allow-unverified]
+//     在 stdout 打印最新 .dump 备份的 key(一行)。
+//   node dist/db-backup.js list [--prefix backups/postgres]
+//     按时间倒序列出 key、大小、时间。
+//
+// OSS 配置:DB_BACKUP_OSS_* 优先,未设置的字段回退 STORAGE_OSS_*。
+
+const USAGE = `用法:
+  node dist/db-backup.js upload --key <oss-key> [--kind postgres-dump|app-configs] [--size bytes]
+  node dist/db-backup.js download --key <oss-key> [--allow-unverified]
+  node dist/db-backup.js latest [--prefix backups/postgres] [--allow-unverified]
+  node dist/db-backup.js list [--prefix backups/postgres]
+
+环境变量:DB_BACKUP_OSS_{ENDPOINT,BUCKET,ACCESS_KEY_ID,SECRET_ACCESS_KEY,REGION} 优先,
+缺省字段复用 STORAGE_OSS_*;默认前缀 backups/postgres。`
+
+const DEFAULT_BACKUP_PREFIX = 'backups/postgres'
+
+export type BackupStorageConfig = {
+  endpoint: string
+  bucket: string
+  accessKeyId: string
+  secretAccessKey: string
+  region: string
+}
+
+export type BackupCliArgs = {
+  command: string
+  key: string
+  prefix: string
+  kind: string
+  allowUnverified: boolean
+  size?: number
+}
+
+type BackupManifest = {
+  key: string
+  kind: string
+  size: number
+  sha256: string
+  createdAt: string
+  database: string | null
+  serverVersion: string | null
+  migrations: string[] | null
+}
+
+function logInfo(message: string): void {
+  process.stderr.write(`[db-backup] ${message}\n`)
+}
+
+function fail(message: string): never {
+  process.stderr.write(`[db-backup] ${message}\n`)
+  process.exit(1)
+}
+
+export function resolveBackupStorageConfig(
+  env: Record<string, string | undefined>,
+): BackupStorageConfig {
+  return {
+    endpoint: readWithFallback(env, 'DB_BACKUP_OSS_ENDPOINT', 'STORAGE_OSS_ENDPOINT'),
+    bucket: readWithFallback(env, 'DB_BACKUP_OSS_BUCKET', 'STORAGE_OSS_BUCKET'),
+    accessKeyId: readWithFallback(env, 'DB_BACKUP_OSS_ACCESS_KEY_ID', 'STORAGE_OSS_ACCESS_KEY_ID'),
+    secretAccessKey: readWithFallback(
+      env,
+      'DB_BACKUP_OSS_SECRET_ACCESS_KEY',
+      'STORAGE_OSS_SECRET_ACCESS_KEY',
+    ),
+    region:
+      readWithFallback(env, 'DB_BACKUP_OSS_REGION', 'STORAGE_OSS_REGION') || 'oss-cn-hangzhou',
+  }
+}
+
+function readWithFallback(
+  env: Record<string, string | undefined>,
+  primary: string,
+  fallback: string,
+): string {
+  return String(env[primary] || '').trim() || String(env[fallback] || '').trim()
+}
+
+export function parseBackupArgs(argv: string[]): BackupCliArgs {
+  const args: BackupCliArgs = {
+    command: String(argv[0] || ''),
+    key: '',
+    prefix: DEFAULT_BACKUP_PREFIX,
+    kind: 'postgres-dump',
+    allowUnverified: false,
+  }
+
+  for (let index = 1; index < argv.length; index += 1) {
+    const arg = argv[index]
+    const value = argv[index + 1]
+    if (arg === '--allow-unverified') {
+      args.allowUnverified = true
+      continue
+    }
+    if (arg === '--size' && value !== undefined) {
+      const size = Number(value)
+      if (Number.isSafeInteger(size) && size >= 0) args.size = size
+      index += 1
+      continue
+    }
+    if ((arg === '--key' || arg === '--prefix' || arg === '--kind') && value !== undefined) {
+      if (arg === '--key') args.key = value
+      if (arg === '--prefix') args.prefix = value.replace(/\/+$/, '')
+      if (arg === '--kind') args.kind = value
+      index += 1
+    }
+  }
+
+  return args
+}
+
+export function manifestKeyFor(key: string): string {
+  return `${key}.manifest.json`
+}
+
+export function pickLatestDumpKey(objects: StorageObjectSummary[]): string | null {
+  let latest: { key: string; time: number } | null = null
+  for (const item of objects) {
+    if (!item.key.endsWith('.dump')) {
+      continue
+    }
+    const time = item.lastModified ? item.lastModified.getTime() : 0
+    if (!latest || time > latest.time || (time === latest.time && item.key > latest.key)) {
+      latest = { key: item.key, time }
+    }
+  }
+  return latest ? latest.key : null
+}
+
+function createBackupStorage(): ObjectStorage {
+  const config = resolveBackupStorageConfig(process.env)
+  const missing = (['endpoint', 'bucket', 'accessKeyId', 'secretAccessKey'] as const)
+    .filter((field) => !config[field])
+    .map((field) => `DB_BACKUP_OSS_${field.replace(/([A-Z])/g, '_$1').toUpperCase()}`)
+  if (missing.length) {
+    logInfo(`缺少 OSS 配置:${missing.join(', ')}(或对应的 STORAGE_OSS_*)`)
+    process.exit(2)
+  }
+
+  return createConfiguredObjectStorage(config)
+}
+
+export async function runUpload(
+  storage: ObjectStorage,
+  args: BackupCliArgs,
+  content: Buffer | Readable,
+): Promise<void> {
+  if (!args.key) {
+    fail(`upload 需要 --key。${USAGE}`)
+  }
+  if (args.size === 0) {
+    fail('输入为空,没有可上传的备份内容。')
+  }
+  const hash = crypto.createHash('sha256')
+  let size = 0
+  let uploadContent: Buffer | Readable = content
+  if (Buffer.isBuffer(content)) {
+    size = content.length
+    hash.update(content)
+  } else {
+    const hashingStream = new HashingStream(hash, (chunkSize) => {
+      size += chunkSize
+    })
+    content.pipe(hashingStream)
+    uploadContent = hashingStream
+  }
+  if (size === 0 && Buffer.isBuffer(content)) {
+    fail('输入为空,没有可上传的备份内容。')
+  }
+
+  const snapshot = args.kind === 'postgres-dump' ? await readDatabaseSnapshot() : null
+
+  await storage.putObject({
+    key: args.key,
+    content: uploadContent,
+    contentType: 'application/octet-stream',
+    ...(Buffer.isBuffer(content)
+      ? { contentLength: content.length }
+      : args.size === undefined
+        ? {}
+        : { contentLength: args.size }),
+  })
+  if (size === 0) {
+    fail('输入为空,没有可上传的备份内容。')
+  }
+  if (args.size !== undefined && args.size !== size) {
+    fail(`输入大小与 --size 不一致:期望 ${args.size} 字节,实际 ${size} 字节。`)
+  }
+  const sha256 = hash.digest('hex')
+
+  const manifest: BackupManifest = {
+    key: args.key,
+    kind: args.kind,
+    size,
+    sha256,
+    createdAt: new Date().toISOString(),
+    database: snapshot ? snapshot.database : null,
+    serverVersion: snapshot ? snapshot.serverVersion : null,
+    migrations: snapshot ? snapshot.migrations : null,
+  }
+  await storage.putObject({
+    key: manifestKeyFor(args.key),
+    content: Buffer.from(`${JSON.stringify(manifest, null, 2)}\n`),
+    contentType: 'application/json',
+  })
+
+  logInfo(
+    `上传完成:${args.key}(${formatSize(size)},sha256=${sha256.slice(0, 12)}…),清单 ${manifestKeyFor(args.key)}`,
+  )
+  if (args.kind === 'postgres-dump' && !snapshot) {
+    logInfo('未能读取数据库迁移记录(数据库暂不可达不影响备份本身,清单相应字段为空)。')
+  }
+}
+
+export async function runDownload(storage: ObjectStorage, args: BackupCliArgs): Promise<Buffer> {
+  if (!args.key) {
+    fail(`download 需要 --key。${USAGE}`)
+  }
+
+  const stored = await storage.getObject(args.key)
+  const hash = crypto.createHash('sha256')
+  const chunks: Buffer[] = []
+  for await (const chunk of stored.reader) {
+    const buffer = Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk)
+    hash.update(buffer)
+    chunks.push(buffer)
+  }
+  const content = Buffer.concat(chunks)
+  const sha256 = hash.digest('hex')
+
+  const manifest = await readManifest(storage, args.key)
+  if (!manifest) {
+    if (!args.allowUnverified) {
+      fail(
+        `缺少 ${manifestKeyFor(args.key)},拒绝使用未校验备份。需要兼容旧备份时显式传 --allow-unverified。`,
+      )
+    }
+    logInfo(`未找到 ${manifestKeyFor(args.key)},已显式允许跳过 sha256 校验。`)
+  } else {
+    validateManifest(manifest, args.key, content.length, sha256, args.kind)
+    logInfo(`sha256 校验通过:${args.key}(${formatSize(content.length)})`)
+  }
+
+  return content
+}
+
+export async function runLatest(storage: ObjectStorage, args: BackupCliArgs): Promise<string> {
+  const objects = (await storage.listObjects(`${args.prefix}/`))
+    .filter((item) => item.key.endsWith('.dump'))
+    .sort((left, right) => {
+      const leftTime = left.lastModified ? left.lastModified.getTime() : 0
+      const rightTime = right.lastModified ? right.lastModified.getTime() : 0
+      return rightTime - leftTime || right.key.localeCompare(left.key)
+    })
+
+  for (const item of objects) {
+    try {
+      const manifest = await readManifest(storage, item.key)
+      if (manifest) {
+        validateManifestMetadata(manifest, item.key, 'postgres-dump')
+        const migrations = manifest.migrations ? `,迁移 ${manifest.migrations.length} 个` : ''
+        logInfo(
+          `最新备份:${item.key}(${formatSize(manifest.size)},${manifest.createdAt}${migrations})`,
+        )
+        return item.key
+      }
+    } catch (error) {
+      logInfo(`跳过无效备份 ${item.key}:${error instanceof Error ? error.message : String(error)}`)
+    }
+    if (args.allowUnverified) {
+      logInfo(`最新备份:${item.key}(已显式允许无清单备份)。`)
+      return item.key
+    }
+  }
+
+  if (!objects.length) {
+    fail(`前缀 ${args.prefix}/ 下没有 .dump 备份。`)
+  }
+  fail(`前缀 ${args.prefix}/ 下没有带有效 manifest 的 .dump 备份。`)
+}
+
+export async function runList(storage: ObjectStorage, args: BackupCliArgs): Promise<string[]> {
+  const objects = (await storage.listObjects(`${args.prefix}/`))
+    .filter((item) => !item.key.endsWith('.manifest.json'))
+    .sort((left, right) => {
+      const leftTime = left.lastModified ? left.lastModified.getTime() : 0
+      const rightTime = right.lastModified ? right.lastModified.getTime() : 0
+      return rightTime - leftTime
+    })
+
+  const lines = objects.map((item) => {
+    const time = item.lastModified ? item.lastModified.toISOString() : '-'
+    return `${item.key}\t${item.size}\t${time}`
+  })
+  logInfo(`前缀 ${args.prefix}/ 下共 ${objects.length} 个备份。`)
+  return lines
+}
+
+async function readManifest(storage: ObjectStorage, key: string): Promise<BackupManifest | null> {
+  try {
+    const stored = await storage.getObject(manifestKeyFor(key))
+    const chunks: Buffer[] = []
+    for await (const chunk of stored.reader) {
+      chunks.push(Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk))
+    }
+    return JSON.parse(Buffer.concat(chunks).toString('utf8')) as BackupManifest
+  } catch (error) {
+    if (isMissingObjectError(error)) {
+      return null
+    }
+    throw error
+  }
+}
+
+function validateManifestMetadata(
+  manifest: BackupManifest,
+  key: string,
+  expectedKind: string,
+): void {
+  if (manifest.key !== key || manifest.kind !== expectedKind) {
+    throw new Error(`manifest 与备份 key/type 不匹配`)
+  }
+  if (!Number.isSafeInteger(manifest.size) || manifest.size <= 0) {
+    throw new Error(`manifest size 无效`)
+  }
+  if (!/^[a-f0-9]{64}$/i.test(manifest.sha256)) {
+    throw new Error(`manifest sha256 无效`)
+  }
+  if (!manifest.createdAt || Number.isNaN(Date.parse(manifest.createdAt))) {
+    throw new Error(`manifest createdAt 无效`)
+  }
+}
+
+function validateManifest(
+  manifest: BackupManifest,
+  key: string,
+  size: number,
+  sha256: string,
+  expectedKind: string,
+): void {
+  validateManifestMetadata(manifest, key, expectedKind)
+  if (manifest.sha256.toLowerCase() !== sha256.toLowerCase()) {
+    fail(`sha256 校验失败:${key} 内容与清单不一致,请勿使用该备份。`)
+  }
+  if (manifest.size !== size) {
+    fail(`大小校验失败:${key} 期望 ${manifest.size} 字节,实际 ${size} 字节。`)
+  }
+}
+
+function isMissingObjectError(error: unknown): boolean {
+  const candidate = error as {
+    name?: string
+    Code?: string
+    $metadata?: { httpStatusCode?: number }
+  }
+  return (
+    candidate?.$metadata?.httpStatusCode === 404 ||
+    ['NotFound', 'NoSuchKey', 'NoSuchBucket'].includes(
+      String(candidate?.name || candidate?.Code || ''),
+    )
+  )
+}
+
+class HashingStream extends Transform {
+  constructor(
+    private readonly hash: crypto.Hash,
+    private readonly onChunk: (size: number) => void,
+  ) {
+    super()
+  }
+
+  _transform(
+    chunk: Buffer,
+    _encoding: BufferEncoding,
+    callback: (error?: Error | null, data?: Buffer) => void,
+  ): void {
+    this.hash.update(chunk)
+    this.onChunk(chunk.length)
+    callback(null, chunk)
+  }
+}
+
+async function downloadToStdout(storage: ObjectStorage, args: BackupCliArgs): Promise<void> {
+  if (!args.key) {
+    fail(`download 需要 --key。${USAGE}`)
+  }
+
+  const stored = await storage.getObject(args.key)
+  const hash = crypto.createHash('sha256')
+  const tempPath = path.join('/tmp', `order-site-backup-${process.pid}-${Date.now()}.dump`)
+  let size = 0
+  try {
+    const output = fs.createWriteStream(tempPath, { mode: 0o600 })
+    for await (const chunk of stored.reader) {
+      const buffer = Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk)
+      hash.update(buffer)
+      size += buffer.length
+      if (!output.write(buffer)) {
+        await new Promise<void>((resolve, reject) => {
+          output.once('drain', resolve)
+          output.once('error', reject)
+        })
+      }
+    }
+    await new Promise<void>((resolve, reject) => {
+      output.once('close', resolve)
+      output.once('error', reject)
+      output.end()
+    })
+
+    const sha256 = hash.digest('hex')
+    const manifest = await readManifest(storage, args.key)
+    if (!manifest) {
+      if (!args.allowUnverified) {
+        fail(
+          `缺少 ${manifestKeyFor(args.key)},拒绝使用未校验备份。需要兼容旧备份时显式传 --allow-unverified。`,
+        )
+      }
+      logInfo(`未找到 ${manifestKeyFor(args.key)},已显式允许跳过 sha256 校验。`)
+    } else {
+      validateManifest(manifest, args.key, size, sha256, args.kind)
+      logInfo(`sha256 校验通过:${args.key}(${formatSize(size)})`)
+    }
+    await pipeline(fs.createReadStream(tempPath), process.stdout)
+  } finally {
+    await fsPromises.rm(tempPath, { force: true })
+  }
+}
+
+// 尽力而为地补充数据库快照信息;数据库不可达时返回 null,不影响备份传输。
+async function readDatabaseSnapshot(): Promise<{
+  database: string | null
+  serverVersion: string | null
+  migrations: string[] | null
+} | null> {
+  try {
+    const { runtimeConfig } = await import('./config/runtime.js')
+    const { query, closeDb } = await import('./db/client.js')
+    try {
+      let database: string | null = null
+      try {
+        const url = new URL(String(runtimeConfig.database?.url || ''))
+        database = decodeURIComponent(url.pathname.replace(/^\//, '')) || null
+      } catch {
+        database = null
+      }
+      const [versionResult, migrationsResult] = await Promise.all([
+        query<{ server_version: string }>('SHOW server_version'),
+        query<{ filename: string }>('SELECT filename FROM schema_migrations ORDER BY filename'),
+      ])
+      return {
+        database,
+        serverVersion: versionResult.rows[0]?.server_version || null,
+        migrations: migrationsResult.rows.map((row) => row.filename),
+      }
+    } finally {
+      await closeDb()
+    }
+  } catch (error) {
+    logInfo(`读取数据库快照失败:${error instanceof Error ? error.message : String(error)}`)
+    return null
+  }
+}
+
+function formatSize(size: number): string {
+  if (size >= 1024 * 1024) return `${(size / (1024 * 1024)).toFixed(2)} MB`
+  if (size >= 1024) return `${(size / 1024).toFixed(2)} KB`
+  return `${size} B`
+}
+
+async function main(): Promise<void> {
+  const args = parseBackupArgs(process.argv.slice(2))
+  if (!['upload', 'download', 'latest', 'list'].includes(args.command)) {
+    logInfo(USAGE)
+    process.exitCode = 2
+    return
+  }
+
+  const storage = createBackupStorage()
+
+  switch (args.command) {
+    case 'upload':
+      await runUpload(storage, args, process.stdin)
+      break
+    case 'download':
+      await downloadToStdout(storage, args)
+      break
+    case 'latest':
+      process.stdout.write(`${await runLatest(storage, args)}\n`)
+      break
+    case 'list':
+      for (const line of await runList(storage, args)) {
+        process.stdout.write(`${line}\n`)
+      }
+      break
+  }
+}
+
+function isInvokedAsScript(entryPath: string | undefined): boolean {
+  if (!entryPath) {
+    return false
+  }
+  try {
+    return import.meta.url === pathToFileURL(path.resolve(entryPath)).href
+  } catch {
+    return false
+  }
+}
+
+if (isInvokedAsScript(process.argv[1])) {
+  await main()
+}

apps/backend/src/services/file-storage/object-storage.ts

@@ -70,9 +70,10 @@ export class ObjectStorage {
 
   async putObject(input: {
     key: string
-    content: Buffer
+    content: Buffer | Readable
     contentType: string
     metadata?: Record<string, string>
+    contentLength?: number
   }): Promise<void> {
     await this.ensureBucketReady()
     await this.client.send(
@@ -81,6 +82,7 @@ export class ObjectStorage {
         Key: input.key,
         Body: input.content,
         ContentType: input.contentType,
+        ...(input.contentLength === undefined ? {} : { ContentLength: input.contentLength }),
         ...(input.metadata ? { Metadata: sanitizeObjectMetadata(input.metadata) } : {}),
       }),
     )
@@ -125,13 +127,17 @@ export class ObjectStorage {
     return stat
   }
 
-  async listObjects(): Promise<StorageObjectSummary[]> {
+  async listObjects(prefix?: string): Promise<StorageObjectSummary[]> {
     await this.ensureBucketReady()
     const objects: StorageObjectSummary[] = []
     let continuationToken: string | undefined
     do {
       const output = await this.client.send(
-        new ListObjectsV2Command({ Bucket: this.bucket, ContinuationToken: continuationToken }),
+        new ListObjectsV2Command({
+          Bucket: this.bucket,
+          ...(prefix ? { Prefix: prefix } : {}),
+          ContinuationToken: continuationToken,
+        }),
       )
       for (const item of output.Contents || []) {
         if (!item.Key) {
@@ -166,7 +172,9 @@ export class ObjectStorage {
             throw error
           }
           if (!this.pathStyle) {
-            throw new Error(`OSS bucket ${this.bucket} 不存在，请先在阿里云控制台创建`)
+            throw new Error(`OSS bucket ${this.bucket} 不存在，请先在阿里云控制台创建`, {
+              cause: error,
+            })
           }
           await this.client.send(new CreateBucketCommand({ Bucket: this.bucket }))
         }
@@ -197,6 +205,9 @@ export class StorageRouter {
   }
 
   async putObject(input: Parameters<ObjectStorage['putObject']>[0]): Promise<void> {
+    if (!Buffer.isBuffer(input.content) && this.mode === 'dual') {
+      throw new Error('dual 存储模式不支持同一可读流双写，请传入 Buffer')
+    }
     if (this.mode === 'minio') {
       await this.legacy.putObject(input)
       return

deploy/backup-db.sh

@@ -0,0 +1,175 @@
+#!/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
+    }
+  ' "${ENV_FILE}"
+}
+
+run_backup_tool() {
+  # 用一次性容器而不是 exec,部署过程中 backend 未运行时也能上传。
+  docker compose run --rm --no-deps -T backend node dist/db-backup.js "$@"
+}
+
+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 "$(docker compose images backend -q 2>/dev/null)" ]]; then
+    fail "backend 镜像不存在,请先完成部署:bash deploy/ubuntu-deploy.sh"
+  fi
+
+  mkdir -p "${DB_DIR}" "${CONFIG_DIR}"
+  backup_database
+  backup_configs
+  prune_local "${DB_DIR}"
+  prune_local "${CONFIG_DIR}"
+  log "备份完成"
+}
+
+main "$@"

deploy/restore-db.sh

@@ -0,0 +1,376 @@
+#!/usr/bin/env bash
+set -Eeuo pipefail
+
+# 数据库恢复与备份验证。
+#
+# 用法:
+#   bash deploy/restore-db.sh --list                       # 列出 OSS 中的数据库备份(倒序)
+#   bash deploy/restore-db.sh --verify                     # 用最新备份做恢复演练(临时容器,不影响线上库)
+#   bash deploy/restore-db.sh --verify --key backups/postgres/order_site-xxx.dump
+#   bash deploy/restore-db.sh                              # 交互确认后,恢复最新备份到线上库
+#   bash deploy/restore-db.sh --key backups/postgres/order_site-xxx.dump --yes
+#   bash deploy/restore-db.sh --restore-configs              # 同时恢复最新 data 配置备份
+#   bash deploy/restore-db.sh --allow-unverified             # 仅兼容历史无 manifest 备份
+#
+# 恢复流程:下载备份(校验 sha256)→ 停 backend/web → pg_restore --clean → 执行迁移 → 重启并等待健康检查。
+# 详见 docs/数据库备份与恢复.md。
+
+PROJECT_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+ENV_FILE="${PROJECT_ROOT}/.env"
+TMP_DIR="${PROJECT_ROOT}/backups/tmp"
+MODE="restore"
+KEY=""
+CONFIG_KEY=""
+RESTORE_CONFIGS=false
+ASSUME_YES="${RESTORE_CONFIRM:-}"
+ALLOW_UNVERIFIED=false
+BACKUP_PREFIX=""
+CONFIG_PREFIX=""
+RESTORE_ACTIVE=false
+RESTORE_SUCCEEDED=false
+BACKEND_WAS_RUNNING=false
+WEB_WAS_RUNNING=false
+RESTORE_TARGET=""
+CONFIG_TARGET=""
+VERIFY_CONTAINER=""
+
+log() { printf '\033[1;34m[restore]\033[0m %s\n' "$*"; }
+warn() { printf '\033[1;33m[restore]\033[0m %s\n' "$*"; }
+fail() { printf '\033[1;31m[restore]\033[0m %s\n' "$*" >&2; exit 1; }
+
+parse_args() {
+  while (($# > 0)); do
+    case "$1" in
+      --list|--verify)
+        [[ "${MODE}" == "restore" ]] || fail "--list 与 --verify 互斥。"
+        MODE="${1#--}"
+        shift
+        ;;
+      --key)
+        [[ $# -ge 2 ]] || fail "--key 需要一个备份 key 参数。"
+        KEY="$2"
+        shift 2
+        ;;
+      --config-key)
+        [[ $# -ge 2 ]] || fail "--config-key 需要一个配置备份 key 参数。"
+        CONFIG_KEY="$2"
+        RESTORE_CONFIGS=true
+        shift 2
+        ;;
+      --restore-configs)
+        RESTORE_CONFIGS=true
+        shift
+        ;;
+      --allow-unverified)
+        ALLOW_UNVERIFIED=true
+        shift
+        ;;
+      --yes)
+        ASSUME_YES=1
+        shift
+        ;;
+      -h|--help)
+        sed -n '4,13p' "${BASH_SOURCE[0]}" | sed 's/^# \{0,1\}//'
+        exit 0
+        ;;
+      *)
+        fail "未知参数:$1。执行 bash deploy/restore-db.sh --help 查看用法。"
+        ;;
+    esac
+  done
+}
+
+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
+    }
+  ' "${ENV_FILE}"
+}
+
+run_backup_tool() {
+  docker compose run --rm --no-deps -T backend node dist/db-backup.js "$@"
+}
+
+restore_cleanup() {
+  local exit_code=$?
+  if [[ "${RESTORE_ACTIVE}" == "true" && "${RESTORE_SUCCEEDED}" != "true" && "${exit_code}" -ne 0 ]]; then
+    warn "恢复失败,尝试恢复恢复前正在运行的业务服务。数据库可能处于部分恢复状态,请立即检查。"
+    if [[ "${BACKEND_WAS_RUNNING}" == "true" ]]; then
+      docker compose up -d --no-deps backend >/dev/null 2>&1 || warn "backend 自动重启失败。"
+    fi
+    if [[ "${WEB_WAS_RUNNING}" == "true" ]]; then
+      docker compose up -d --no-deps web >/dev/null 2>&1 || warn "web 自动重启失败。"
+    fi
+  fi
+  [[ -n "${RESTORE_TARGET}" ]] && rm -f -- "${RESTORE_TARGET}"
+  [[ -n "${CONFIG_TARGET}" ]] && rm -f -- "${CONFIG_TARGET}"
+  [[ -n "${VERIFY_CONTAINER}" ]] && docker rm -f -v "${VERIFY_CONTAINER}" >/dev/null 2>&1 || true
+  trap - EXIT
+  exit "${exit_code}"
+}
+
+is_external_database() {
+  local database_url
+  database_url="$(read_env_value DATABASE_URL)"
+  [[ -n "${database_url}" && "${database_url}" != *"@postgres:"* ]]
+}
+
+resolve_key() {
+  if [[ -n "${KEY}" ]]; then
+    printf '%s\n' "${KEY}"
+    return
+  fi
+  local latest
+  local latest_args=(latest --prefix "${BACKUP_PREFIX}")
+  [[ "${ALLOW_UNVERIFIED}" == "true" ]] && latest_args+=(--allow-unverified)
+  if ! latest="$(run_backup_tool "${latest_args[@]}")"; then
+    fail "未能获取最新备份 key。可先用 bash deploy/restore-db.sh --list 查看已有备份。"
+  fi
+  printf '%s\n' "${latest}"
+}
+
+resolve_config_key() {
+  if [[ -n "${CONFIG_KEY}" ]]; then
+    printf '%s\n' "${CONFIG_KEY}"
+    return
+  fi
+  local latest
+  latest="$(run_backup_tool list --prefix "${CONFIG_PREFIX}" | awk -F '\t' '$1 ~ /\.tar\.gz$/ { print $1; exit }' || true)"
+  [[ -n "${latest}" ]] || fail "未找到配置备份。可使用 --config-key 指定配置备份,或不启用 --restore-configs。"
+  printf '%s\n' "${latest}"
+}
+
+download_backup() {
+  local key="$1" target="$2" kind="${3:-postgres-dump}"
+  log "下载备份并校验 sha256:${key}"
+  local verification_args=(--key "${key}" --kind "${kind}")
+  [[ "${ALLOW_UNVERIFIED}" == "true" ]] && verification_args+=(--allow-unverified)
+  if ! run_backup_tool download "${verification_args[@]}" > "${target}"; then
+    rm -f "${target}"
+    fail "下载或校验失败:${key}"
+  fi
+  [[ -s "${target}" ]] || fail "下载内容为空:${key}"
+}
+
+run_pg_restore_stdin() {
+  local database_url
+  database_url="$(read_env_value DATABASE_URL)"
+  if [[ -z "${database_url}" || "${database_url}" == *"@postgres:"* ]]; then
+    docker compose exec -T postgres sh -c 'pg_restore -U "${POSTGRES_USER:-postgres}" -d "${POSTGRES_DB:-order_site}" --clean --if-exists --no-owner'
+  else
+    log "DATABASE_URL 指向外部数据库,从 postgres 容器经网络直连恢复"
+    docker compose exec -T -e DB_RESTORE_URL="${database_url}" postgres sh -c 'pg_restore --dbname="$DB_RESTORE_URL" --clean --if-exists --no-owner'
+  fi
+}
+
+service_is_running() {
+  local service="$1" container_id status
+  container_id="$(docker compose ps -q "${service}" 2>/dev/null || true)"
+  [[ -n "${container_id}" ]] || return 1
+  status="$(docker inspect -f '{{.State.Status}}' "${container_id}" 2>/dev/null || true)"
+  [[ "${status}" == "running" ]]
+}
+
+restore_configs() {
+  local source="$1" staging="${TMP_DIR}/config-extract-$$"
+  log "恢复配置备份到 apps/backend/data"
+  while IFS= read -r entry; do
+    case "${entry}" in
+      /*|../*|*/../*|..)
+        fail "配置备份包含不安全路径:${entry}"
+        ;;
+    esac
+  done < <(tar -tzf "${source}")
+  rm -rf -- "${staging}"
+  mkdir -p "${staging}"
+  if ! tar -xzf "${source}" --no-same-owner -C "${staging}"; then
+    rm -rf -- "${staging}"
+    fail "配置备份解包失败,未修改现有配置。"
+  fi
+  cp -a "${staging}/." "${PROJECT_ROOT}/apps/backend/data/"
+  rm -rf -- "${staging}"
+}
+
+restore_live_database() {
+  local key="$1" target="$2" config_target="${3:-}"
+
+  warn "即将用备份覆盖当前数据库:${key}"
+  warn "当前库中的订单、后台用户、审计日志等数据都会被备份内容替换。"
+  if [[ "${ASSUME_YES}" != "1" ]]; then
+    if [[ ! -t 0 ]]; then
+      fail "当前不是交互式终端。确认恢复请执行:RESTORE_CONFIRM=1 bash deploy/restore-db.sh --key ${key}"
+    fi
+    local answer
+    printf '请输入 RESTORE 确认恢复数据库:'
+    read -r answer
+    [[ "${answer}" == "RESTORE" ]] || fail "未确认恢复,已停止。"
+  fi
+
+  BACKEND_WAS_RUNNING=false
+  WEB_WAS_RUNNING=false
+  service_is_running backend && BACKEND_WAS_RUNNING=true
+  service_is_running web && WEB_WAS_RUNNING=true
+  RESTORE_ACTIVE=true
+  log "停止 backend/web,避免恢复期间产生写入"
+  docker compose stop backend web >/dev/null 2>&1 || true
+  if service_is_running backend || service_is_running web; then
+    fail "无法确认 backend/web 已停止,为避免恢复期间写入已中止。"
+  fi
+
+  log "pg_restore 恢复中(--clean --if-exists)"
+  if ! run_pg_restore_stdin < "${target}"; then
+    fail "pg_restore 失败。数据库处于中间状态,请根据日志处理;必要时用 --list 选择更早的备份重试。"
+  fi
+
+  log "执行数据库迁移(补齐备份之后新增的迁移)"
+  if ! docker compose run --rm --no-deps backend node dist/db/migrate.js; then
+    fail "迁移失败。请查看日志;数据已恢复到备份时点。"
+  fi
+
+  if [[ -n "${config_target}" ]]; then
+    restore_configs "${config_target}"
+  fi
+
+  log "重启 backend/web"
+  docker compose up -d --no-deps backend >/dev/null || fail "backend 启动失败。"
+  wait_backend_healthy
+  docker compose up -d --no-deps web >/dev/null || fail "web 启动失败。"
+
+  local health_url site_addr
+  site_addr="$(read_env_value CADDY_SITE_ADDR)"
+  health_url="${site_addr%/}/health/ready"
+  if curl -fsS "${health_url}" >/dev/null 2>&1; then
+    log "健康检查通过:${health_url}"
+  else
+    fail "健康检查未通过:${health_url},请检查 docker compose logs backend。"
+  fi
+  RESTORE_SUCCEEDED=true
+  log "恢复完成。建议抽查近期订单与后台登录确认数据正确。"
+}
+
+wait_backend_healthy() {
+  local attempt container_id status
+  for ((attempt = 1; attempt <= 60; attempt++)); do
+    container_id="$(docker compose ps -q backend 2>/dev/null || true)"
+    if [[ -n "${container_id}" ]]; then
+      status="$(docker inspect -f '{{if .State.Health}}{{.State.Health.Status}}{{end}}' "${container_id}" 2>/dev/null || true)"
+      if [[ "${status}" == "healthy" ]]; then
+        return
+      fi
+    fi
+    sleep 2
+  done
+  fail "等待 backend 健康超时(120 秒),请检查 docker compose logs backend。"
+}
+
+resolve_postgres_image() {
+  local container_id image
+  container_id="$(docker compose ps -q postgres 2>/dev/null || true)"
+  if [[ -n "${container_id}" ]]; then
+    image="$(docker inspect -f '{{.Config.Image}}' "${container_id}" 2>/dev/null || true)"
+    if [[ -n "${image}" ]]; then
+      printf '%s\n' "${image}"
+      return
+    fi
+  fi
+  image="$(docker compose images postgres -q 2>/dev/null | head -n 1 || true)"
+  [[ -n "${image}" ]] || fail "无法确定 postgres 镜像(线上 postgres 未运行)。"
+  printf '%s\n' "${image}"
+}
+
+verify_backup() {
+  local key="$1" target="$2" image cid attempt
+
+  image="$(resolve_postgres_image)"
+  log "使用镜像 ${image} 启动临时验证容器(不接入任何业务网络)"
+
+  docker rm -f order-site-backup-verify >/dev/null 2>&1 || true
+  cid="$(docker run -d --name order-site-backup-verify -e POSTGRES_PASSWORD=verify-only "${image}")"
+  VERIFY_CONTAINER="order-site-backup-verify"
+
+  for ((attempt = 1; attempt <= 30; attempt++)); do
+    if docker exec "${cid}" pg_isready -U postgres >/dev/null 2>&1; then
+      break
+    fi
+    sleep 1
+  done
+  docker exec "${cid}" pg_isready -U postgres >/dev/null 2>&1 || fail "临时验证容器启动超时。"
+
+  log "在临时容器中恢复备份"
+  docker cp "${target}" "${cid}:/tmp/verify.dump" >/dev/null
+  docker exec "${cid}" createdb -U postgres order_site >/dev/null
+  if ! docker exec "${cid}" pg_restore -U postgres -d order_site --no-owner /tmp/verify.dump; then
+    fail "备份无法恢复:pg_restore 报错,该备份不可用。"
+  fi
+
+  log "备份可恢复。已应用迁移:"
+  docker exec "${cid}" psql -U postgres -d order_site -At -c 'SELECT filename FROM schema_migrations ORDER BY filename'
+
+  log "各表行数:"
+  docker exec "${cid}" psql -U postgres -d order_site -At -F '=' \
+    -c "SELECT format('SELECT %L, count(*) FROM %I;', tablename) FROM pg_tables WHERE schemaname = 'public' ORDER BY tablename \\gexec"
+
+  log "验证通过,已清理临时容器。建议每月执行一次。"
+  docker rm -f -v "${VERIFY_CONTAINER}" >/dev/null 2>&1 || true
+  VERIFY_CONTAINER=""
+}
+
+list_backups() {
+  log "OSS 中的数据库备份(key/字节/时间):"
+  run_backup_tool list --prefix "${BACKUP_PREFIX}"
+}
+
+main() {
+  parse_args "$@"
+  cd "${PROJECT_ROOT}"
+
+  command -v docker >/dev/null 2>&1 || fail "未找到 docker。"
+  docker compose version >/dev/null 2>&1 || fail "未找到 docker compose 插件。"
+
+  BACKUP_PREFIX="${DB_BACKUP_OSS_PREFIX:-$(read_env_value DB_BACKUP_OSS_PREFIX || true)}"
+  BACKUP_PREFIX="${BACKUP_PREFIX%/}"
+  BACKUP_PREFIX="${BACKUP_PREFIX:-backups/postgres}"
+  [[ "${BACKUP_PREFIX}" != /* && "${BACKUP_PREFIX}" != *'..'* ]] || fail "DB_BACKUP_OSS_PREFIX 不能是绝对路径或包含 ..。"
+  CONFIG_PREFIX="${BACKUP_PREFIX%/postgres}/configs"
+
+  if [[ "${MODE}" == "list" ]]; then
+    list_backups
+    return
+  fi
+
+  mkdir -p "${TMP_DIR}"
+  local key target config_key
+  key="$(resolve_key)"
+  target="${TMP_DIR}/$(basename "${key}")"
+  RESTORE_TARGET="${target}"
+
+  download_backup "${key}" "${target}" postgres-dump
+
+  if [[ "${RESTORE_CONFIGS}" == "true" && "${MODE}" == "restore" ]]; then
+    config_key="$(resolve_config_key)"
+    CONFIG_TARGET="${TMP_DIR}/$(basename "${config_key}")"
+    download_backup "${config_key}" "${CONFIG_TARGET}" app-configs
+  fi
+
+  if [[ "${MODE}" == "verify" ]]; then
+    verify_backup "${key}" "${target}"
+    rm -f -- "${target}"
+    return
+  fi
+
+  if is_external_database; then
+    warn "DATABASE_URL 指向外部数据库:pg_restore 将经网络直连外部库并覆盖其数据,请确认目标正确。"
+  fi
+  restore_live_database "${key}" "${target}" "${CONFIG_TARGET:-}"
+}
+
+trap restore_cleanup EXIT
+main "$@"

deploy/ubuntu-deploy.sh

@@ -16,6 +16,8 @@ HEALTH_URL=""
 READY_TIMEOUT="${READY_TIMEOUT:-120}"
 RESET_DB=false
 RESET_DB_CONFIRM="${RESET_DB_CONFIRM:-}"
+SKIP_BACKUP="${SKIP_BACKUP:-false}"
+RESET_PRE_BACKUP_DONE=false
 
 log() {
   printf '\033[1;34m[deploy]\033[0m %s\n' "$*"
@@ -38,6 +40,7 @@ print_usage() {
 
 参数：
   --reset-db    删除 Compose PostgreSQL 数据卷 postgres_data，再按 migrations 目录全量重建（001 基线 + 全部增量）。
+  --skip-backup 跳过部署过程中的自动数据库备份（备份失败时确认为可接受后可用它紧急部署）。
   --yes         跳过 --reset-db 的交互确认，也可使用环境变量 RESET_DB_CONFIRM=1。
   -h, --help    显示帮助。
 
@@ -47,6 +50,8 @@ print_usage() {
   它不会删除 apps/backend/data 下的平台 JSON 配置，也不会删除 Caddy 证书卷。
 
   图片存储使用阿里云 OSS（.env 的 STORAGE_OSS_*），生产编排不含 MinIO。
+  每次部署在执行数据库迁移前会自动备份到 OSS（deploy/backup-db.sh）；
+  --reset-db 在删除数据卷前也会先抓一份最后备份。备份与恢复见 docs/数据库备份与恢复.md。
 EOF
 }
 
@@ -56,6 +61,9 @@ parse_args() {
       --reset-db)
         RESET_DB=true
         ;;
+      --skip-backup)
+        SKIP_BACKUP=true
+        ;;
       --yes)
         RESET_DB_CONFIRM=1
         ;;
@@ -340,12 +348,68 @@ restart_web() {
   fail "Web 启动失败，请根据上面的日志排查。"
 }
 
+run_scheduled_backup() {
+  local label="$1"
+
+  if [[ "${SKIP_BACKUP}" == "true" ]]; then
+    warn "已指定 --skip-backup，跳过${label}前的数据库备份。"
+    return
+  fi
+
+  log "${label}前自动备份数据库到 OSS"
+  if bash "${PROJECT_ROOT}/deploy/backup-db.sh"; then
+    log "${label}前备份完成"
+    return
+  fi
+
+  fail "${label}前备份失败，已停止部署。确认为可接受的失败后可用 --skip-backup 继续。"
+}
+
+backup_before_reset() {
+  if [[ "${RESET_DB}" != "true" ]]; then
+    return
+  fi
+  if [[ "${SKIP_BACKUP}" == "true" ]]; then
+    warn "已指定 --skip-backup，跳过删除数据卷前的最后一份备份。"
+    return
+  fi
+
+  local container_id status
+  container_id="$(docker compose ps -q postgres 2>/dev/null | head -n 1 || true)"
+  status=""
+  [[ -n "${container_id}" ]] && status="$(docker inspect -f '{{.State.Status}}' "${container_id}" 2>/dev/null || true)"
+  if [[ "${status}" != "running" ]]; then
+    local volume_name
+    volume_name="$(resolve_postgres_volume_name)"
+    if docker volume inspect "${volume_name}" >/dev/null 2>&1; then
+      warn "检测到已有 PostgreSQL 数据卷但容器未运行，先启动数据库以完成重置前备份。"
+      start_postgres
+    else
+      warn "未检测到已有 PostgreSQL 数据卷，跳过重置前备份（首次部署属正常现象）。"
+      return
+    fi
+  fi
+
+  run_scheduled_backup "重置数据库"
+  RESET_PRE_BACKUP_DONE=true
+}
+
+backup_before_migrate() {
+  if [[ "${RESET_PRE_BACKUP_DONE}" == "true" ]]; then
+    log "已在重置数据库前完成最后备份，跳过重置后的空库备份。"
+    return
+  fi
+  run_scheduled_backup "执行迁移"
+}
+
 deploy_stack() {
   check_compose_config
   pull_base_images
   build_images
+  backup_before_reset
   reset_database
   start_postgres
+  backup_before_migrate
   run_database_migrations
   restart_backend
   restart_web

docs/数据库备份与恢复.md

@@ -0,0 +1,145 @@
+# 数据库备份与恢复
+
+生产数据库(PostgreSQL,Compose 容器)每日备份到阿里云 OSS,与图片存储共用 OSS 基础设施,但使用独立的 `backups/` 前缀与(建议)独立的 RAM 凭证。
+
+## 备份内容与存放
+
+| 内容 | 方式 | OSS key 前缀 |
+| --- | --- | --- |
+| PostgreSQL 全库 | `pg_dump -Fc`(custom 格式,自带压缩,MVCC 快照无需停机) | `backups/postgres/` |
+| 平台 JSON 配置(`apps/backend/data`,排除日志/缓存) | `tar -czf` | `backups/configs/` |
+
+- 每个备份对象附带 `<key>.manifest.json` 清单:大小、sha256、创建时间、数据库版本、已应用迁移列表。没有有效清单的备份默认不会被 `latest` 选择,恢复也会拒绝;只有明确使用 `--allow-unverified` 才兼容历史备份。
+- 不备份的内容:图片(已在 OSS)、Caddy 证书(自动重新签发)、日志(`apps/backend/data/logs`)。
+- PostgreSQL dump 是数据库内部一致快照,但数据库与 OSS 不共享事务;备份时刻附近的新图片可能只存在一侧。恢复后应按 `file_assets.object_key` 抽查 OSS 对象,并为图片桶开启版本控制/保留策略。
+- 服务器本地在 `backups/` 目录保留最近 `DB_BACKUP_KEEP` 份(默认 7),OSS 端过期由桶生命周期规则管理。
+- 涉及脚本:`deploy/backup-db.sh`(备份)、`deploy/restore-db.sh`(恢复/演练)、后端镜像内 `dist/db-backup.js`(OSS 传输与校验)。
+
+## 首次开通
+
+以下步骤在阿里云控制台与服务器上各执行一次。
+
+### 1. RAM 授权(建议独立,可直接复用现有凭证)
+
+现有 `STORAGE_OSS_*` 的 RAM 用户拥有整个桶的权限,也能直接跑备份。但备份包含全部订单与用户数据,是最敏感的资产,建议单独创建一个 RAM 用户 `order-site-backup`,只授予 `backups/` 前缀,自定义权限策略:
+
+```json
+{
+  "Version": "1",
+  "Statement": [
+    {
+      "Effect": "Allow",
+      "Action": ["oss:PutObject", "oss:GetObject"],
+      "Resource": ["acs:oss:*:*:你的桶名/backups/*"]
+    },
+    {
+      "Effect": "Allow",
+      "Action": ["oss:HeadBucket"],
+      "Resource": ["acs:oss:*:*:你的桶名"]
+    },
+    {
+      "Effect": "Allow",
+      "Action": ["oss:ListObjects"],
+      "Resource": ["acs:oss:*:*:你的桶名"],
+      "Condition": {
+        "StringLike": { "oss:Prefix": ["backups/", "backups/*"] }
+      }
+    }
+  ]
+}
+```
+
+为该用户创建 AccessKey,填入服务器 `.env` 的 `DB_BACKUP_OSS_ACCESS_KEY_ID` / `DB_BACKUP_OSS_SECRET_ACCESS_KEY`(不填则回退 `STORAGE_OSS_*`)。若用独立备份桶,同时覆盖 `DB_BACKUP_OSS_ENDPOINT` / `DB_BACKUP_OSS_BUCKET` / `DB_BACKUP_OSS_REGION`。
+
+### 2. 桶生命周期规则(控制存储成本)
+
+在 OSS 控制台为目标桶添加生命周期规则:
+
+- 前缀 `backups/`:最后一次访问 90 天后删除(按需调整,数据库每日一份,90 份足够覆盖一个季度)。
+- 建议同时开启版本控制:误删备份时可找回旧版本;给非当前版本配置 7 天过期,避免版本堆积。
+
+### 3. 手动执行一次,验证链路
+
+```bash
+bash deploy/backup-db.sh
+bash deploy/restore-db.sh --list
+bash deploy/restore-db.sh --verify
+```
+
+`--verify` 会把最新带有效清单的备份恢复到一个隔离的临时 PostgreSQL 容器(不接入业务网络,不影响线上库),并打印迁移列表与各表行数。建议之后每月执行一次演练——没验证过的备份等于没有备份。
+
+### 4. 配置每日定时备份
+
+```bash
+crontab -e
+```
+
+```cron
+# 每天 03:00 备份数据库与配置到 OSS(目录需先存在,首次开通时脚本已创建)
+0 3 * * * cd /srv/order_site && bash deploy/backup-db.sh >> backups/backup.log 2>&1
+```
+
+把 `/srv/order_site` 换成实际项目目录。日志追加在 `backups/backup.log`,可用 logrotate 管理;备份失败时 cron 不通知,请定期查看该文件,或后续接入告警 webhook。
+
+## 日常运维
+
+- **部署自动备份**:`bash deploy/ubuntu-deploy.sh` 在每次执行数据库迁移前会自动备份一次;`--reset-db` 会在删除数据卷前抓最后一份旧库备份,并跳过重置后空库备份,避免空库被误选为 latest。备份失败会中止部署,确认为可接受后可用 `--skip-backup` 继续。
+- **手动备份**:`bash deploy/backup-db.sh`(可加 `--keep N` 覆盖本地保留份数)。
+- **查看备份列表**:`bash deploy/restore-db.sh --list`。
+- **恢复演练**:`bash deploy/restore-db.sh --verify`(可加 `--key` 指定备份)。
+- **兼容历史备份**:仅在确认对象来源可信时加 `--allow-unverified`,不要作为生产默认参数。
+
+## 恢复
+
+### 场景 A:整库恢复到最新备份(最常用)
+
+```bash
+bash deploy/restore-db.sh
+```
+
+流程:下载最新备份并强制校验 sha256 → 停 backend/web 并确认已停止 → `pg_restore --clean --if-exists` → 执行迁移补差 → 重启并等待健康检查。任一步失败会尝试恢复原先运行的服务,并明确报告数据库可能处于部分恢复状态。执行前需交互输入 `RESTORE` 确认(非交互终端用 `RESTORE_CONFIRM=1` 或 `--yes`)。
+
+如需同时恢复 `apps/backend/data` 配置:
+
+```bash
+bash deploy/restore-db.sh --restore-configs
+```
+
+也可以用 `--config-key` 指定配置归档。配置归档会先校验路径,再解包到 data 目录。
+
+### 场景 B:恢复到指定时间点
+
+```bash
+bash deploy/restore-db.sh --list
+bash deploy/restore-db.sh --key backups/postgres/order_site-20260823-030000.dump
+```
+
+### 场景 C:服务器全损(数据卷丢失/换机)
+
+在新服务器完成部署(`docs/部署启动与数据库重建.md`)后,把旧服务器 `.env`(含 OSS 凭证)恢复到新服务器,再执行场景 A。图片不受影响(在 OSS),平台 JSON 配置可从 `backups/configs/order_site-<时间戳>.tar.gz` 解回 `apps/backend/data/`。
+
+### 场景 D:外部数据库(DATABASE_URL 非 postgres 服务)
+
+`backup-db.sh` 与 `restore-db.sh` 会自动识别 `.env` 的 `DATABASE_URL`:指向外部库时,从 postgres 容器经网络直连执行 pg_dump/pg_restore,因此服务器仍需有可用的 postgres 镜像/容器作为客户端。恢复外部库前会额外告警确认目标。
+
+### 恢复后检查
+
+- 打开后台登录、近期订单列表,抽查数据到备份时间点为止。
+- `bash deploy/restore-db.sh --list` 确认下一次定时备份已恢复产生新对象。
+
+## 故障排查
+
+| 现象 | 处理 |
+| --- | --- |
+| `[db-backup] HeadBucket 403` | RAM 策略缺 `oss:HeadBucket`,按上文策略补齐 |
+| `ListObjects 403` | 策略缺 `oss:ListObjects` 或 `oss:Prefix` 条件与实际前缀不匹配 |
+| `PutObject 403` | `Resource` 里的桶名/前缀与 `.env` 的 `DB_BACKUP_OSS_BUCKET` / `DB_BACKUP_OSS_PREFIX` 不一致 |
+| `缺少 OSS 配置` | `.env` 里 `DB_BACKUP_OSS_*` 与 `STORAGE_OSS_*` 都为空;改完后备份脚本用的是镜像内程序,若同时改了代码需重新构建 |
+| 下载校验 `sha256 校验失败` | 备份对象被篡改或截断,换更早的备份重试 |
+| `pg_restore` 报版本错误 | 不会发生在本编排内(pg_dump/pg_restore 与库同镜像);仅外部库场景需保证客户端 ≥ 服务端大版本 |
+
+## 安全注意
+
+- 备份 RAM 凭证只进服务器 `.env`(已 gitignore),不要提交到仓库。
+- 备份桶/前缀保持私有,不要开公共读写;`backups/` 下的对象不要通过前端暴露。
+- 若备份桶与图片桶相同,确认图片访问逻辑只读图片前缀,不会枚举 `backups/`。