修复:Windows 更新备份使用写句柄刷盘,保留失败阶段诊断

This commit is contained in:
2026-11-05 15:06:16 +08:00
parent 4843768694
commit 2315c83067
5 changed files with 87 additions and 11 deletions
+6
View File
@@ -58,3 +58,9 @@ python3 deploy/offline/publish_update.py \
旧版检查会逐个探测完整运行环境清单的磁盘路径,慢磁盘上可能超过接口等待时间。新版仅探测会替换、删除或备份的程序路径;完整清单语法、重复项、运行环境指纹、公钥及签名验证仍保留。 旧版检查会逐个探测完整运行环境清单的磁盘路径,慢磁盘上可能超过接口等待时间。新版仅探测会替换、删除或备份的程序路径;完整清单语法、重复项、运行环境指纹、公钥及签名验证仍保留。
已有部署若无法通过界面更新,将生成的“修复更新检查.cmd”放入原 QualityCenter 文件夹双击,再回到“任务与监控 → 系统更新”检查更新。修复只接受已知程序摘要,备份后原子替换更新检查程序及其清单摘要;保留版本、数据、配置、缓存修复和信任公钥,不重启服务、不导入数据库。更新或恢复任务运行期间拒绝修复。 已有部署若无法通过界面更新,将生成的“修复更新检查.cmd”放入原 QualityCenter 文件夹双击,再回到“任务与监控 → 系统更新”检查更新。修复只接受已知程序摘要,备份后原子替换更新检查程序及其清单摘要;保留版本、数据、配置、缓存修复和信任公钥,不重启服务、不导入数据库。更新或恢复任务运行期间拒绝修复。
### Windows 更新备份刷盘修复(2026-11-05)
Windows 的刷盘操作要求写句柄。程序及私有配置备份统一使用写入、flush、fsync 后再原子替换,移除复制后以只读句柄刷盘的步骤。备份失败仍停止更新并保留旧程序与数据库;不降低备份持久化要求。
“修复Windows更新.cmd”适配原始更新器和已安装检查超时修复的更新器,保留原清单、缓存修复和版本信息。更新失败诊断写入 `.runtime/update/failure.json`,只包含阶段、异常类型、系统错误码和源码位置,不含业务正文、凭据或异常参数;回退异常单独记录在 `rollback-failure.json`。
File diff suppressed because one or more lines are too long
+31
View File
@@ -232,6 +232,37 @@ class UpdateTests(unittest.TestCase):
update.worker(987654, 'a' * 32, self.root) update.worker(987654, 'a' * 32, self.root)
return calls, restores, queries return calls, restores, queries
def test_windows_flush_requires_write_handle_through_complete_update(self):
import os
if os.name != 'nt':
import fcntl
real_fsync = os.fsync
flushed = []
def windows_fsync(fd):
mode = fcntl.fcntl(fd, fcntl.F_GETFL) & os.O_ACCMODE if os.name != 'nt' else os.O_RDWR
if mode == os.O_RDONLY:
raise OSError(9, 'Windows _commit requires writable file descriptor')
flushed.append(fd)
return real_fsync(fd)
with patch.object(update.os, 'fsync', side_effect=windows_fsync):
self.worker_harness()
state = update.read(self.root / '.runtime/update/status.json')
self.assertEqual('completed', state['state'], state['message'])
self.assertTrue(flushed)
backup = next((self.root / '.runtime/update').glob('backup-*'))
self.assertEqual(self.config, update.read(backup / 'private/deployment.json'))
self.assertEqual(b'{"apiKey":"keep-private"}', (backup / 'private/analysis-api-settings.json').read_bytes())
def test_failure_receipt_records_stage_and_codes_without_private_details(self):
error = OSError(5, 'privatePasswordValue must never be logged')
update.record_failure(self.root, 'backupProgram', error)
receipt = update.read(self.root / '.runtime/update/failure.json')
self.assertEqual('backupProgram', receipt['stage'])
self.assertEqual(5, receipt['errno'])
self.assertNotIn('privatePasswordValue', json.dumps(receipt))
with patch.object(update, 'atomic', side_effect=OSError('disk full')):
update.record_failure(self.root, 'backupProgram', error)
def test_complete_update_orchestration_with_isolated_database_backup(self): def test_complete_update_orchestration_with_isolated_database_backup(self):
calls, restores, queries = self.worker_harness() calls, restores, queries = self.worker_harness()
self.assertEqual([('stop', False), ('serve', True), ('stop', False), ('verify', False), ('serve', False)], calls) self.assertEqual([('stop', False), ('serve', True), ('stop', False), ('verify', False), ('serve', False)], calls)
@@ -109,6 +109,17 @@ class UpdateCheckRepairTest(unittest.TestCase):
self.assertEqual((self.root / fix.SCRIPT).read_bytes(), self.original) self.assertEqual((self.root / fix.SCRIPT).read_bytes(), self.original)
path.unlink() path.unlink()
def test_previous_check_repair_version_can_upgrade_without_losing_manifest_fields(self):
prior = b'previous optimized checker fixture'
digest = hashlib.sha256(prior).hexdigest()
(self.root / fix.SCRIPT).write_bytes(prior)
value = json.loads(self.manifest_path.read_text())
value['files'][0]['sha256'] = digest
self.manifest_path.write_text(json.dumps(value))
with patch.object(fix, 'PREVIOUS_SHA', digest):
self.assertTrue(fix.repair(self.root)['updateCheckRepaired'])
self.assertEqual(fix.PATCHED_SCRIPT, (self.root / fix.SCRIPT).read_bytes())
def test_embedded_program_matches_reviewed_source(self): def test_embedded_program_matches_reviewed_source(self):
self.assertEqual(fix.PATCHED_SCRIPT, (HERE / 'update_manager.py').read_bytes()) self.assertEqual(fix.PATCHED_SCRIPT, (HERE / 'update_manager.py').read_bytes())
+33 -6
View File
@@ -22,6 +22,7 @@ import stat
import subprocess import subprocess
import sys import sys
import time import time
import traceback
import zipfile import zipfile
from urllib.request import ProxyHandler, HTTPRedirectHandler, build_opener from urllib.request import ProxyHandler, HTTPRedirectHandler, build_opener
@@ -81,6 +82,7 @@ def atomic(path, value):
def copy_atomic(source, target): def copy_atomic(source, target):
source, target = Path(source), Path(target)
target.parent.mkdir(parents=True, exist_ok=True) target.parent.mkdir(parents=True, exist_ok=True)
temporary = target.with_name(target.name + '.update-' + secrets.token_hex(6)) temporary = target.with_name(target.name + '.update-' + secrets.token_hex(6))
try: try:
@@ -433,11 +435,8 @@ def backup_program(root, backup, release):
copy_atomic(root / 'manifest.json', backup / 'manifest.json') copy_atomic(root / 'manifest.json', backup / 'manifest.json')
# All private configuration (AI keys, deployment credentials and bootstrap # All private configuration (AI keys, deployment credentials and bootstrap
# flags) is backed up locally, but never becomes a release payload. # flags) is backed up locally, but never becomes a release payload.
shutil.copytree(root / '.runtime', backup / 'private', ignore=shutil.ignore_patterns('update', '*.lock', '*.log', 'service.json', 'stop-request.json'), symlinks=False) # 复制时在写句柄上刷盘;Windows 不支持在只读句柄上执行 _commit。
for path in (backup / 'private').rglob('*'): shutil.copytree(root / '.runtime', backup / 'private', ignore=shutil.ignore_patterns('update', '*.lock', '*.log', 'service.json', 'stop-request.json'), symlinks=False, copy_function=copy_atomic)
if path.is_file():
with path.open('rb') as stream:
os.fsync(stream.fileno())
atomic(backup / 'program.json', {'all': names, 'present': present}) atomic(backup / 'program.json', {'all': names, 'present': present})
@@ -497,6 +496,22 @@ def rollback(root, backup, database_backed_up):
wait_ready(root, process) wait_ready(root, process)
def record_failure(root, stage, error, name='failure.json'):
# 仅记录阶段、异常类型、系统错误码和代码位置,不记录正文、凭据或异常参数。
receipt = {'stage': stage, 'errorType': type(error).__name__,
'occurredAt': datetime.now(timezone.utc).isoformat(),
'frames': [{'file': Path(frame.f_code.co_filename).name, 'function': frame.f_code.co_name, 'line': line}
for frame, line in traceback.walk_tb(error.__traceback__)]}
for key in ('errno', 'winerror'):
value = getattr(error, key, None)
if type(value) is int:
receipt[key] = value
try:
atomic(root / '.runtime/update' / name, receipt)
except Exception:
pass # 诊断写入失败不能阻止原有回退流程。
def worker(launcher, token, root=ROOT): def worker(launcher, token, root=ROOT):
directory = root / '.runtime/update' directory = root / '.runtime/update'
for _ in range(300): for _ in range(300):
@@ -512,6 +527,7 @@ def worker(launcher, token, root=ROOT):
with exclusive_update(root): with exclusive_update(root):
backup = directory / ('backup-' + datetime.now(timezone.utc).strftime('%Y%m%dT%H%M%SZ') + '-' + secrets.token_hex(4)) backup = directory / ('backup-' + datetime.now(timezone.utc).strftime('%Y%m%dT%H%M%SZ') + '-' + secrets.token_hex(4))
stopped = backed_up = program_saved = changed = False stopped = backed_up = program_saved = changed = False
failure_stage = 'validateRelease'
try: try:
envelope = decode(download(LATEST)) envelope = decode(download(LATEST))
release = validate_release(envelope, root) release = validate_release(envelope, root)
@@ -529,19 +545,25 @@ def worker(launcher, token, root=ROOT):
raise UpdateError('磁盘剩余空间不足,请释放空间后重试。') raise UpdateError('磁盘剩余空间不足,请释放空间后重试。')
write_status('downloading', '正在下载并校验更新。', root) write_status('downloading', '正在下载并校验更新。', root)
info = release['archive'] info = release['archive']
failure_stage = 'downloadArchive'
download(BASE + info['file'], archive, info['size'], info['sha256']) download(BASE + info['file'], archive, info['size'], info['sha256'])
failure_stage = 'unpackArchive'
unpack(archive, stage, release) unpack(archive, stage, release)
# Revalidate immediately before any service or disk mutation. # Revalidate immediately before any service or disk mutation.
validate_release(envelope, root) validate_release(envelope, root)
write_status('backingUp', '正在备份,服务即将重新启动。', root) write_status('backingUp', '正在备份,服务即将重新启动。', root)
atomic(directory / 'journal.json', {'backup': backup.name, 'phase': 'stopping', 'databaseBackedUp': False}) atomic(directory / 'journal.json', {'backup': backup.name, 'phase': 'stopping', 'databaseBackedUp': False})
failure_stage = 'stopService'
controller(root, 'stop', timeout=180) controller(root, 'stop', timeout=180)
stopped = True stopped = True
failure_stage = 'backupProgram'
backup_program(root, backup, release) backup_program(root, backup, release)
program_saved = True program_saved = True
config = read(root / '.runtime/deployment.json') config = read(root / '.runtime/deployment.json')
failure_stage = 'startBackupDatabase'
database = controller(root, 'serve', True) database = controller(root, 'serve', True)
wait_ready(root, database, True) wait_ready(root, database, True)
failure_stage = 'backupDatabase'
with (backup / 'database.sql').open('xb') as output: with (backup / 'database.sql').open('xb') as output:
mysql(root, config, 'mysqldump', stdout=output) mysql(root, config, 'mysqldump', stdout=output)
output.flush() output.flush()
@@ -560,14 +582,18 @@ def worker(launcher, token, root=ROOT):
'programInventorySha256': digest(backup / 'program.json')}) 'programInventorySha256': digest(backup / 'program.json')})
write_status('installing', '正在安装更新。', root) write_status('installing', '正在安装更新。', root)
changed = True changed = True
failure_stage = 'replaceProgram'
replace_program(root, stage, backup, release) replace_program(root, stage, backup, release)
failure_stage = 'verifyProgram'
controller(root, 'verify', timeout=180) controller(root, 'verify', timeout=180)
write_status('restarting', '正在重新启动,请稍候。', root) write_status('restarting', '正在重新启动,请稍候。', root)
failure_stage = 'restartService'
process = controller(root, 'serve') process = controller(root, 'serve')
wait_ready(root, process) wait_ready(root, process)
atomic(directory / 'journal.json', {'backup': backup.name, 'phase': 'complete', 'databaseBackedUp': True}) atomic(directory / 'journal.json', {'backup': backup.name, 'phase': 'complete', 'databaseBackedUp': True})
write_status('completed', '更新完成。', root, version=release['version'], finishedAt=datetime.now(timezone.utc).isoformat(), available=False) write_status('completed', '更新完成。', root, version=release['version'], finishedAt=datetime.now(timezone.utc).isoformat(), available=False)
except Exception as error: except Exception as error:
record_failure(root, failure_stage, error)
message = str(error) if isinstance(error, UpdateError) else '更新未完成,请查看本机更新日志。' message = str(error) if isinstance(error, UpdateError) else '更新未完成,请查看本机更新日志。'
if stopped: if stopped:
try: try:
@@ -579,7 +605,8 @@ def worker(launcher, token, root=ROOT):
wait_ready(root, process) wait_ready(root, process)
atomic(directory / 'journal.json', {'backup': backup.name, 'phase': 'rolledBack', 'databaseBackedUp': backed_up}) atomic(directory / 'journal.json', {'backup': backup.name, 'phase': 'rolledBack', 'databaseBackedUp': backed_up})
message += ' 已恢复原版本。' message += ' 已恢复原版本。'
except Exception: except Exception as rollback_error:
record_failure(root, 'rollback', rollback_error, 'rollback-failure.json')
message += ' 自动恢复未完成,请联系管理员使用本机更新备份恢复。' message += ' 自动恢复未完成,请联系管理员使用本机更新备份恢复。'
write_status('failed', message, root, available=False, finishedAt=datetime.now(timezone.utc).isoformat()) write_status('failed', message, root, available=False, finishedAt=datetime.now(timezone.utc).isoformat())
finally: finally: