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
+33 -6
View File
@@ -22,6 +22,7 @@ import stat
import subprocess
import sys
import time
import traceback
import zipfile
from urllib.request import ProxyHandler, HTTPRedirectHandler, build_opener
@@ -81,6 +82,7 @@ def atomic(path, value):
def copy_atomic(source, target):
source, target = Path(source), Path(target)
target.parent.mkdir(parents=True, exist_ok=True)
temporary = target.with_name(target.name + '.update-' + secrets.token_hex(6))
try:
@@ -433,11 +435,8 @@ def backup_program(root, backup, release):
copy_atomic(root / 'manifest.json', backup / 'manifest.json')
# All private configuration (AI keys, deployment credentials and bootstrap
# 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)
for path in (backup / 'private').rglob('*'):
if path.is_file():
with path.open('rb') as stream:
os.fsync(stream.fileno())
# 复制时在写句柄上刷盘;Windows 不支持在只读句柄上执行 _commit。
shutil.copytree(root / '.runtime', backup / 'private', ignore=shutil.ignore_patterns('update', '*.lock', '*.log', 'service.json', 'stop-request.json'), symlinks=False, copy_function=copy_atomic)
atomic(backup / 'program.json', {'all': names, 'present': present})
@@ -497,6 +496,22 @@ def rollback(root, backup, database_backed_up):
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):
directory = root / '.runtime/update'
for _ in range(300):
@@ -512,6 +527,7 @@ def worker(launcher, token, root=ROOT):
with exclusive_update(root):
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
failure_stage = 'validateRelease'
try:
envelope = decode(download(LATEST))
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。')
write_status('downloading', '正在下载并校验更新。', root)
info = release['archive']
failure_stage = 'downloadArchive'
download(BASE + info['file'], archive, info['size'], info['sha256'])
failure_stage = 'unpackArchive'
unpack(archive, stage, release)
# Revalidate immediately before any service or disk mutation.
validate_release(envelope, 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})
failure_stage = 'stopService'
controller(root, 'stop', timeout=180)
stopped = True
failure_stage = 'backupProgram'
backup_program(root, backup, release)
program_saved = True
config = read(root / '.runtime/deployment.json')
failure_stage = 'startBackupDatabase'
database = controller(root, 'serve', True)
wait_ready(root, database, True)
failure_stage = 'backupDatabase'
with (backup / 'database.sql').open('xb') as output:
mysql(root, config, 'mysqldump', stdout=output)
output.flush()
@@ -560,14 +582,18 @@ def worker(launcher, token, root=ROOT):
'programInventorySha256': digest(backup / 'program.json')})
write_status('installing', '正在安装更新。', root)
changed = True
failure_stage = 'replaceProgram'
replace_program(root, stage, backup, release)
failure_stage = 'verifyProgram'
controller(root, 'verify', timeout=180)
write_status('restarting', '正在重新启动,请稍候。', root)
failure_stage = 'restartService'
process = controller(root, 'serve')
wait_ready(root, process)
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)
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。'
if stopped:
try:
@@ -579,7 +605,8 @@ def worker(launcher, token, root=ROOT):
wait_ready(root, process)
atomic(directory / 'journal.json', {'backup': backup.name, 'phase': 'rolledBack', 'databaseBackedUp': backed_up})
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。'
write_status('failed', message, root, available=False, finishedAt=datetime.now(timezone.utc).isoformat())
finally: