# -*- coding: utf-8 -*- """数据库备份、恢复与自动备份线程。 (自 auth_system.py 拆出,行为与原实现一致。) """ import math import os import shutil import sqlite3 import sys import threading import time from pathlib import Path from typing import Any import auth_system # 路径常量以包属性为唯一来源(动态读取,便于测试/运维覆盖) from .db import ( _DB_LOCK, get_conn, init_db, ) from .settings import ( get_setting, set_setting, ) # ============ 数据库备份 ============ # 备份命名专用锁:create_backup 不再持有全局 _DB_LOCK(不阻塞写请求), # 但同秒并发的两次备份仍需保证文件名唯一 _BACKUP_NAME_LOCK = threading.Lock() def create_backup() -> str: """创建备份(唯一文件名——同秒冲突自动追加序号,不再互相覆盖)。 WAL 模式下用独立只读连接走 SQLite 在线备份 API:拿到的是一致性快照, 且不持有 _DB_LOCK——原实现备份期间阻塞全站写请求,大库时会卡住所有用户。""" auth_system.BACKUP_DIR.mkdir(parents=True, exist_ok=True) with _BACKUP_NAME_LOCK: base = 'backup_' + time.strftime('%Y%m%d_%H%M%S') name = base + '.db' k = 1 while (auth_system.BACKUP_DIR / name).exists(): name = f'{base}-{k}.db' k += 1 # 占位创建目标文件(锁内),保证并发下同秒命名仍唯一 dest = sqlite3.connect(str(auth_system.BACKUP_DIR / name)) conn = get_conn() try: try: conn.backup(dest) finally: dest.close() # 任何备份产生后统一裁剪(手动/自动/上传同配额) try: prune_backups(20, 5) except Exception: pass return name except Exception: # 备份失败:清掉占位文件,不留半个坏备份 try: (auth_system.BACKUP_DIR / name).unlink(missing_ok=True) except OSError: pass raise finally: conn.close() def list_backups() -> list[dict[str, Any]]: """备份列表(按文件修改时间倒序——按名字排序会让 upload_ 永远排前,误删新自动备份)""" if not auth_system.BACKUP_DIR.exists(): return [] out = [] for p in auth_system.BACKUP_DIR.glob('*.db'): st = p.stat() out.append({'name': p.name, 'size': st.st_size, 'time': time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(st.st_mtime)), '_mtime': st.st_mtime}) out.sort(key=lambda x: x['_mtime'], reverse=True) for o in out: o.pop('_mtime', None) return out def backup_path(name: str) -> Path | None: name = Path(name).name # 防目录穿越 p = auth_system.BACKUP_DIR / name return p if p.exists() and p.suffix == '.db' else None def _backup_is_valid_db(src: Path) -> bool: """深度校验备份文件:SQLite 头 + 能真正打开 + 含 users 表,三者缺一不可""" try: with open(src, 'rb') as f: if f.read(16) != b'SQLite format 3\x00'[:16]: return False probe = sqlite3.connect(f'file:{src}?mode=ro', uri=True) try: tables = {r[0] for r in probe.execute("SELECT name FROM sqlite_master WHERE type='table'")} return 'users' in tables finally: probe.close() except Exception: return False # 恢复前静止钩子——由应用层注册(取消全部在途下载任务),避免 worker 按旧库账目结算 _before_restore_hook = None def register_before_restore_hook(fn) -> None: global _before_restore_hook _before_restore_hook = fn def restore_backup(name: str) -> None: """恢复备份(先验证、留回滚、后替换、失败自动回滚,源备份文件保留不移动) 恢复前先通过钩子取消全部在途任务(worker 不再引用旧库用户结算)""" src = backup_path(name) if not src: raise ValueError('备份文件不存在') if not _backup_is_valid_db(src): raise ValueError('不是有效的站点数据库备份(校验失败,未做任何修改)') if _before_restore_hook is not None: try: _before_restore_hook() except Exception: pass safety = auth_system.BACKUP_DIR / ('pre_restore_' + time.strftime('%Y%m%d_%H%M%S') + '.db') with _DB_LOCK: # 0) 先尽力清 WAL 并合并进主库(被占用导致 checkpoint 忙时明确失败,不假装成功); # 必须在留安全副本之前做——否则副本缺 WAL 里未合并的最近提交,回滚点不完整 conn = get_conn() try: row = conn.execute('PRAGMA wal_checkpoint(TRUNCATE)').fetchone() busy = bool(row and row[0] == 1) # 1 = busy finally: conn.close() if busy: raise RuntimeError('数据库当前有其他读写在途(checkpoint 忙),请稍后重试恢复') # 1) checkpoint 后留安全副本(包含全部已提交数据,可回滚) shutil.copy2(str(auth_system.DB_PATH), str(safety)) # 2) 源备份复制到临时文件(不动源备份本身) tmp = auth_system.DB_PATH.with_suffix('.db.restore_tmp') shutil.copy2(str(src), str(tmp)) # 3) 原子替换 + 立即校验 os.replace(str(tmp), str(auth_system.DB_PATH)) try: init_db() except Exception: # 替换后初始化失败:回滚安全副本,恢复原库 shutil.copy2(str(safety), str(auth_system.DB_PATH)) init_db() raise RuntimeError('恢复后数据库初始化失败,已自动回滚到恢复前状态') from None # 裁剪移出 _DB_LOCK——glob/unlink 的 I/O 不阻塞全站写 try: prune_backups(20, 5) except Exception: pass def delete_backup(name: str) -> bool: p = backup_path(name) # 允许删除 pre_restore_ 安全副本(配额裁剪已保证至少保留 # 最近 5 份;原实现一刀切拒删导致恢复频繁时无界累积且管理员手动也删不掉) if p: p.unlink() return True return False # 自动备份线程 _backup_thread_started = False _BACKUP_ERR_TS = {'ts': 0.0, 'n': 0} # 备份失败限频记录(失败不能零痕迹) def prune_backups(keep_normal: int = 20, keep_pre_restore: int = 5) -> None: """备份配额裁剪(从自动备份循环抽出为独立函数)—— 非pre_restore 保留最近 N 份、pre_restore_ 安全副本保留最近 M 份。 在 create_backup/上传备份/恢复备份后统一调用(原实现只在自动备份循环里 裁剪,手动备份/上传/恢复会让 data/backups 无界累积)。""" baks = [b for b in list_backups() if not b['name'].startswith('pre_restore_')] for b in baks[keep_normal:]: try: delete_backup(b['name']) except Exception: pass pre = [b for b in list_backups() if b['name'].startswith('pre_restore_')] for b in pre[keep_pre_restore:]: try: (auth_system.BACKUP_DIR / b['name']).unlink() except OSError: pass def _auto_backup_loop() -> None: while True: try: interval = float(get_setting('backup_interval_hours', '0') or 0) # 非有限值(NaN/inf)或负数一律视为关闭 if not math.isfinite(interval): interval = 0 set_setting('backup_interval_hours', '0') if interval > 0: last = float(get_setting('last_backup_ts', '0') or 0) if not math.isfinite(last): last = 0 if time.time() - last >= interval * 3600: create_backup() set_setting('last_backup_ts', str(time.time())) # 仅保留最近 20 份(按修改时间,不按文件名) # pre_restore_* 安全副本不计入 20 份配额,仅各自保留最近 5 份 prune_backups(20, 5) except Exception: # 定时备份失败绝不能静默(磁盘满/权限问题时 owner 以为有备份、 # 出事时没有)——限频告警,每 10 分钟最多一条(对齐监控线程口径) import traceback as _tb _now = time.time() _BACKUP_ERR_TS['n'] += 1 if _now - _BACKUP_ERR_TS['ts'] > 600: print(f'[警告] 自动备份失败(近 10 分钟累计 {_BACKUP_ERR_TS["n"]} 次): ' f'{_tb.format_exc()[-400:]}', file=sys.stderr) _BACKUP_ERR_TS['ts'] = _now _BACKUP_ERR_TS['n'] = 0 time.sleep(600) def start_auto_backup() -> None: global _backup_thread_started if _backup_thread_started: return t = threading.Thread(target=_auto_backup_loop, daemon=True) t.start() # 仅在真实启动成功后置位,失败可重试 _backup_thread_started = True