修复:旧浏览器样式与安全操作编号兼容,复用已完成的 Windows 数据库恢复

This commit is contained in:
2026-11-05 14:08:09 +08:00
parent c2410c4ce1
commit b208f93945
18 changed files with 477 additions and 32 deletions
+112 -18
View File
@@ -121,13 +121,98 @@ def rebind_sources(query, manifest, archive):
raise RuntimeError('来源文件校验失败,恢复未切换。')
query('UPDATE `' + entry['table'] + '` SET `' + entry['column'] + '`=' + sql_text(str(target))
+ ' WHERE id=' + sql_text(entry['id']) + ' AND `' + entry['column'] + '`=' + sql_text(entry['oldPath']) + ';')
actual = query('SELECT `' + entry['column'] + '` FROM `' + entry['table'] + '` WHERE id=' + sql_text(entry['id']) + ';').strip()
if actual != str(target):
actual = query('SELECT HEX(`' + entry['column'] + '`) FROM `' + entry['table'] + '` WHERE id=' + sql_text(entry['id']) + ';').strip().lower()
# 按 UTF-8 十六进制核对,避免 mysql --batch 转义 Windows 反斜杠。
if actual != str(target).encode('utf-8').hex():
raise RuntimeError('来源文件路径重绑定失败。')
counts[entry['table']] += 1
return counts
def restored_source_paths(query, manifest):
"""批量读取原始路径;HEX 避免 Windows 路径受到客户端转义影响。"""
allowed = {('source_revision', 'archive_path'), ('source_candidate', 'source_path'),
('source_candidate', 'archive_path'), ('source_upload_draft', 'source_path')}
result = {}
pairs = {(entry['table'], entry['column']) for entry in manifest['pathBindings']}
if not pairs.issubset(allowed):
raise RuntimeError('来源路径绑定类型无效。')
for table, column in sorted(pairs):
values = {}
for line in query('SELECT id,HEX(`' + column + '`) FROM `' + table + '` WHERE `' + column + '` IS NOT NULL;').splitlines():
identity, encoded = line.split('\t', 1)
values[identity] = bytes.fromhex(encoded).decode('utf-8')
for entry in manifest['pathBindings']:
if (entry['table'], entry['column']) == (table, column):
if entry['id'] not in values:
raise RuntimeError('恢复库缺少已登记的来源引用。')
result[(table, column, entry['id'])] = values[entry['id']]
return result
def reusable_archive(manifest, paths):
"""只复用本部署目录此前恢复生成、且全部文件通过校验的归档。"""
parent = (control.DATA / 'migrations').resolve()
attempts = set()
for entry in manifest['pathBindings']:
actual = paths[(entry['table'], entry['column'], entry['id'])]
if actual == entry['oldPath']:
continue
relative = relative_name(entry['relative'])
path = Path(actual)
if not path.is_absolute() or not path.resolve().is_relative_to(parent):
raise RuntimeError('恢复库来源路径不属于本目录。')
parts = path.relative_to(parent).parts
if (len(parts) != len(relative.parts) + 2 or not re.fullmatch(r'[a-f0-9]{16}', parts[0])
or parts[1] != 'archive' or tuple(parts[2:]) != relative.parts):
raise RuntimeError('恢复库来源路径与原迁移包不一致。')
attempts.add(parts[0])
if len(attempts) != 1:
raise RuntimeError('无法确定上次恢复的来源归档,未重新导入。')
archive = parent / next(iter(attempts)) / 'archive'
observed = {path.relative_to(archive).as_posix() for path in archive.rglob('*') if path.is_file()}
if observed != set(manifest['archiveFiles']):
raise RuntimeError('上次恢复的来源文件数量不一致。')
for name, expected in manifest['archiveFiles'].items():
target = archive.joinpath(*relative_name(name).parts)
if (target.is_symlink() or not target.resolve().is_relative_to(archive.resolve())
or target.stat().st_size != expected['bytes'] or sha256(target) != expected['sha256']):
raise RuntimeError('上次恢复的来源文件未通过校验。')
return archive
def select_reusable_import(value, manifest):
"""复核完整导入及来源文件后复用暂存库;绝不回退为重新导入。"""
names = run_query(value, 'SHOW DATABASES;').splitlines()
candidates = [name for name in names if re.fullmatch(r'quality_restore_[a-f0-9]{12,32}', name)
and name != value.get('mysqlDatabase')]
verified = []
for name in candidates:
query = lambda statement: run_query(value, statement, name)
try:
paths = restored_source_paths(query, manifest)
archive = reusable_archive(manifest, paths)
counts = table_counts(query)
if counts != manifest['tableCounts']:
continue
users = [json.loads(line) for line in query("SELECT JSON_OBJECT('id',id,'username',username,'passwordHashSha256',SHA2(password_hash,256),'active',active) FROM app_user ORDER BY id;").splitlines()]
if sorted(users, key=lambda row: row['id']) != sorted(manifest['sourceUsers'], key=lambda row: row['id']):
continue
if query_evidence(query) != manifest['queryEvidence']:
continue
# 已确认原路径或同一受控归档路径,后续仍以当前值作为条件更新。
bound = {**manifest, 'pathBindings': [{**entry, 'oldPath': paths[(entry['table'], entry['column'], entry['id'])]}
for entry in manifest['pathBindings']]}
verified.append((archive.stat().st_mtime_ns, name, archive, counts, bound))
except (RuntimeError, ValueError, OSError):
continue
if not verified:
raise RuntimeError('未找到通过完整校验的已导入恢复库。已有数据库保留,本次未重新导入;请保留 runtimedata/logs 供核查。')
_, database, archive, counts, bound = max(verified, key=lambda item: item[0])
print('已核验并复用上次导入完成的数据库,跳过数据库重新导入。', flush=True)
return database, archive, counts, bound
def relocate_native(stage, source_root):
source_root = PurePosixPath(source_root)
def relocated(value):
@@ -247,8 +332,11 @@ def restore_locked(check_runtime):
raise RuntimeError('迁移包文件不完整,已停止恢复。')
if manifest.get('modelConfigSha256') and sha256(PAYLOAD / 'model-config.json') != manifest['modelConfigSha256']:
raise RuntimeError('模型配置校验失败。')
if shutil.disk_usage(ROOT).free < manifest['minimumFreeBytes']:
raise RuntimeError('可用磁盘空间不足,请至少预留 ' + str((manifest['minimumFreeBytes'] + 1024**3 - 1) // 1024**3) + ' GB。')
reuse_only = os.environ.get('QUALITY_MIGRATION_REUSE_ONLY') == '1'
required_free = (sum(item['bytes'] for item in manifest['nativeHistoryFiles'].values()) + 128 * 1024**2
if reuse_only else manifest['minimumFreeBytes'])
if shutil.disk_usage(ROOT).free < required_free:
raise RuntimeError('可用磁盘空间不足,请至少预留 ' + str((required_free + 1024**3 - 1) // 1024**3) + ' GB。')
attempt = uuid.uuid4().hex[:16]
stage = control.PRIVATE / ('migration-stage-' + attempt)
stage.mkdir(mode=0o700)
@@ -256,7 +344,8 @@ def restore_locked(check_runtime):
archive.parent.mkdir(parents=True, mode=0o700)
native = stage / 'nao-users'
print('正在恢复来源文件和分析历史……', flush=True)
extract_checked(PAYLOAD / 'archive.tar.gz', archive, manifest['archiveFiles'])
if not reuse_only:
extract_checked(PAYLOAD / 'archive.tar.gz', archive, manifest['archiveFiles'])
extract_checked(PAYLOAD / 'native-history.tar.gz', native, manifest['nativeHistoryFiles'])
native_count = relocate_native(native, manifest['sourceProjectRoot'])
if native_count != manifest['nativeHistoryDatabases']:
@@ -271,18 +360,23 @@ def restore_locked(check_runtime):
database = 'quality_restore_' + uuid.uuid4().hex
try:
start_database(value)
run_query(value, 'CREATE DATABASE `' + database + '` CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci;')
restore_dump(value, database, PAYLOAD / 'database.sql.gz')
query = lambda statement: run_query(value, statement, database)
print('正在核对全部数据表、原账号和正式统计结果……', flush=True)
actual_counts = table_counts(query)
if actual_counts != manifest['tableCounts']:
raise RuntimeError('恢复后的数据表行数不一致,未切换正式使用库。')
actual_users = [json.loads(line) for line in query("SELECT JSON_OBJECT('id',id,'username',username,'passwordHashSha256',SHA2(password_hash,256),'active',active) FROM app_user ORDER BY id;").splitlines()]
if sorted(actual_users, key=lambda row: row['id']) != sorted(manifest['sourceUsers'], key=lambda row: row['id']):
raise RuntimeError('原用户账号校验失败,未切换使用库。')
if query_evidence(query) != manifest['queryEvidence']:
raise RuntimeError('正式版本和月度统计结果不一致,未切换使用库。')
if reuse_only:
print('正在核验上次已导入的数据库和来源文件,不会重新导入……', flush=True)
database, archive, actual_counts, manifest = select_reusable_import(value, manifest)
query = lambda statement: run_query(value, statement, database)
else:
run_query(value, 'CREATE DATABASE `' + database + '` CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci;')
restore_dump(value, database, PAYLOAD / 'database.sql.gz')
query = lambda statement: run_query(value, statement, database)
print('正在核对全部数据表、原账号和正式统计结果……', flush=True)
actual_counts = table_counts(query)
if actual_counts != manifest['tableCounts']:
raise RuntimeError('恢复后的数据表行数不一致,未切换正式使用库。')
actual_users = [json.loads(line) for line in query("SELECT JSON_OBJECT('id',id,'username',username,'passwordHashSha256',SHA2(password_hash,256),'active',active) FROM app_user ORDER BY id;").splitlines()]
if sorted(actual_users, key=lambda row: row['id']) != sorted(manifest['sourceUsers'], key=lambda row: row['id']):
raise RuntimeError('原用户账号校验失败,未切换使用库。')
if query_evidence(query) != manifest['queryEvidence']:
raise RuntimeError('正式版本和月度统计结果不一致,未切换使用库。')
rebound = rebind_sources(query, manifest, archive)
run_query(value, 'GRANT ALL PRIVILEGES ON `' + database + "`.* TO 'quality_app'@'127.0.0.1';")
run_query(value, 'FLUSH TABLES;')
@@ -304,7 +398,7 @@ def restore_locked(check_runtime):
'totalRows': sum(actual_counts.values()), 'sourceReferencesRebound': rebound,
'modelProfilesMigrated': manifest['modelProfilesMigrated'], 'modelProfilesExcluded': manifest['modelProfilesExcluded'],
'nativeHistoryDatabases': native_count, 'originalDatabasesPreserved': True,
'archiveDirectory': new_config['archiveDirectory']}
'archiveDirectory': new_config['archiveDirectory'], 'reusedCompletedImport': reuse_only}
plan = {'attempt': attempt, 'previousConfig': value, 'newConfig': new_config, 'privateSwitches': switches}
commit_plan(plan, receipt)
print('迁移恢复完成。请使用原来的账号和密码登录;MiMo 配置未迁移。', flush=True)