fix(review): 处理全量审查剩余项(M4-M8/M11/M12 + 21 条 low)

后端
- asset_service: delete_asset 先清 subdomains/site_certs 子表(SQLite 不强制
  FK 且 assets.id 无 AUTOINCREMENT,删最大 id 后新建会复用 rowid 导致孤儿误挂);
  metrics.db 清理改 best-effort,不再让已成功的删除返回 500
- provider_service: slug 唯一约束竞态补 IntegrityError 兜底(与 account_service 一致)
- account_service/sync_service: raise HTTPException 补 from e,保留原始异常链
- sync_service: 删死代码 _load_config;新增 BalanceUnsupported,平台无余额接口时
  计入 skipped 而不是静默当作失败;到期日写库前归一
- ai: 删掉 OpenAI 早已下线的 billing 余额接口实现,能力探测如实报「不支持」
- adapters/base: _overridden 改走 MRO,中间层基类实现的能力不再被误判为未实现
- core/timeutils: 删死代码 utcnow_iso(注释声称 Z 后缀,实际输出 +00:00)
- main: lifespan 退出时 await 被 cancel 的清理任务,避免 pending task 被销毁
- ssl_service: 写 site_certs/subdomains 前校验 asset_id 存在,杜绝悬空引用;
  到期阈值贯穿 _judge_status/_apply_probe/check_one/check_all_site_certs
- routers/credentials、accounts: 明文密码 GET 补 Cache-Control: no-store
- routers/stats: /stats/expiring 的 Query 描述与实际语义对齐(不设下界,含已过期)

Agent
- collector: 解析 sshd_config 的 Include(含 sshd_config.d/*.conf,按 sshd 语义
  「首次出现的值生效」),未显式配置的关键字按编译默认值判定,消除假 pass;
  防火墙检测改用 iptables -S 统计 -A 规则行并读 INPUT 策略(-L -n 恒有链头输出)
- reporter: 4xx(除 429)不再重试直接退出,429 优先按 Retry-After 退避

安全与部署
- core/security: require_agent_key 改为对外 fail-closed——tailnet/本机放行,
  外部来源未配 AGENT_KEY 时一律 401(此前全放行,而服务默认监听 0.0.0.0,
  公网上任何人可伪造监控指标与安全检查结果)
- main: 启动时把「未配 API_KEY/AGENT_KEY 则外部来源不可用」打进日志
- run.sh: HOST/PORT 可用环境变量覆盖,并说明默认 0.0.0.0 的权衡
- deploy/setup.sh: 新装时自动生成随机 API_KEY/AGENT_KEY 并提示同步到各 VPS;
  注册 vps-cert-check.timer
- deploy/vps-cert-check.{service,timer}: 新增每日 09:30 的站点证书到期探测
  (此前 check_site_certs.py 没有任何 unit,证书提醒不会自动跑)
- deploy/update.sh: sync_units 对 timer 用 enable --now(新增 timer 只 restart
  不会建开机自启软链),主服务 unit 变更后额外 restart;pip 安装失败不再被吞
- scripts/check_site_certs.py: threshold 参数真正生效并打印
- scripts/make_migration_bundle.py: 缺 assets.db 直接 abort;umask 077 + chmod 600
  (包内含 .env 的 MASTER_KEY 与 escrow,不能默认 644 落在 /tmp)
- scripts/restore_migration_bundle.py: 校验之外补必备文件预检,缺文件在触碰
  目标机状态之前就 abort

前端
- store: 拆分列表口径 assets 与统计口径 allAssets(无筛选时同源,不重复请求);
  删除已无消费者的 expiring/loadExpiring(每次 loadAll 白发一个请求)
- dashboard/monitor/providers/subscriptions/modals: 统计与计数改用 allAssets,
  资产页的筛选(含统计卡片点击跳转带的筛选)不再让其他视图的数字缩水
- modals: 账号同步后补 loadOverview;settings: 导入后改 loadAll
- api: countdownText 支持 unit/empty 选项,monitor/domains 各自的实现改为委托,
  assets/domains/subscriptions 的到期列不再直出「(-3天)」
- servers: cpu/mem/disk 百分比统一判空,采样缺项不再渲染「null%」
- assets: 同步 AI 余额的结果提示区分 skipped(平台无余额接口)与 failed
- modals: 修正凭据库头注释——2FA 动态码是选填 + 可事后补验,与实现一致

测试
- 新增 tests/test_agent_collector.py(8 条)、tests/test_security.py(11 条)
- 全量 113 passed;改动 JS 均通过 node --check,deploy 脚本通过 bash -n
- 浏览器实测:筛选态下总览/监控中心/平台页/付费资产的统计仍为全量口径,
  到期文案与 null 百分比渲染正确,console 零报错
This commit is contained in:
gouki
2026-09-23 10:15:52 +00:00
parent 61372f7ceb
commit f09c0f6e4b
37 changed files with 712 additions and 142 deletions
+4 -4
View File
@@ -113,13 +113,13 @@ def create_account(session: Session, data: AccountCreate) -> AccountRead:
try:
session.add(account)
session.commit()
except IntegrityError:
except IntegrityError as e:
# 并发创建同名账号时唯一约束兜底:检查与提交之间存在竞态窗口
session.rollback()
raise HTTPException(
status_code=status.HTTP_400_BAD_REQUEST,
detail=f"该平台下账号已存在:{name}",
)
) from e
session.refresh(account)
return _to_read(account, _asset_counts(session), _cred_of(session, account))
@@ -150,13 +150,13 @@ def update_account(session: Session, account_id: int, data: AccountUpdate) -> Ac
try:
session.add(account)
session.commit()
except IntegrityError:
except IntegrityError as e:
# 重命名撞上已有账号时唯一约束兜底,与 _check_name_taken 存在竞态窗口
session.rollback()
raise HTTPException(
status_code=status.HTTP_400_BAD_REQUEST,
detail=f"该平台下账号已存在:{account.name}",
)
) from e
session.refresh(account)
return _to_read(account, _asset_counts(session), _cred_of(session, account))
+27 -2
View File
@@ -299,17 +299,42 @@ def update_asset(session: Session, asset_id: int, data: AssetUpdate) -> AssetRea
def delete_asset(session: Session, asset_id: int) -> None:
"""删除资产及其详情,并清理 metrics.db 中的关联监控数据(避免孤儿数据)"""
"""删除资产及其详情/子表,并尽力清理 metrics.db 中的关联监控数据(避免孤儿数据)"""
asset = session.get(Asset, asset_id)
if not asset:
raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="资产不存在")
detail = _get_detail(session, asset)
if detail is not None:
session.delete(detail)
_delete_child_rows(session, asset_id)
session.delete(asset)
session.commit()
logger.info("删除资产 id=%s name=%s", asset_id, asset.name)
_cleanup_metrics_for_asset(asset_id)
# metrics.db 是独立库(可能正在被 Agent 写入而短暂锁定),
# 清理失败不应让已经成功的删除返回 500,否则前端会误报「删除失败」
try:
_cleanup_metrics_for_asset(asset_id)
except Exception as e: # noqa: BLE001
logger.warning(
"清理资产监控数据失败 asset_id=%s:%s(资产已删除,监控数据可后续手动清理)",
asset_id,
e,
)
def _delete_child_rows(session: Session, asset_id: int) -> None:
"""删除 assets.db 中引用该资产的子表行(subdomains / site_certs)
必须显式清理:SQLite 默认不强制外键(PRAGMA foreign_keys=0),残留行会变成孤儿;
而 assets.id 是 INTEGER PRIMARY KEY(无 AUTOINCREMENT),删掉最大 id 后
新建资产必然复用该 id,孤儿行会被误挂到不相干的新资产上。
"""
from sqlmodel import delete
from app.models.ssl import SiteCert, Subdomain
for model in (Subdomain, SiteCert):
session.exec(delete(model).where(model.asset_id == asset_id))
def _cleanup_metrics_for_asset(asset_id: int) -> None:
+20 -4
View File
@@ -3,6 +3,7 @@
from typing import List, Optional
from fastapi import HTTPException, status
from sqlalchemy.exc import IntegrityError
from sqlmodel import Session, select
from app.core import crypto
@@ -47,8 +48,16 @@ def create_provider(session: Session, data: ProviderCreate) -> ProviderRead:
provider_data = data.model_dump(exclude={"api_config"})
provider = Provider(**provider_data)
provider.api_config_encrypted = crypto.encrypt(data.api_config)
session.add(provider)
session.commit()
try:
session.add(provider)
session.commit()
except IntegrityError as e:
# 上面的 exists 检查与 commit 之间存在竞态窗口,靠 slug 唯一约束兜底,
# 否则并发建同名平台会直接把 IntegrityError 抛成 500
session.rollback()
raise HTTPException(
status_code=status.HTTP_400_BAD_REQUEST, detail=f"slug 已存在:{data.slug}"
) from e
session.refresh(provider)
return _to_read(provider)
@@ -64,8 +73,15 @@ def update_provider(
setattr(provider, key, value)
if data.api_config is not None:
provider.api_config_encrypted = crypto.encrypt(data.api_config)
session.add(provider)
session.commit()
try:
session.add(provider)
session.commit()
except IntegrityError as e:
# 改 slug 撞上已有平台时同样靠唯一约束兜底
session.rollback()
raise HTTPException(
status_code=status.HTTP_400_BAD_REQUEST, detail=f"slug 已存在:{provider.slug}"
) from e
session.refresh(provider)
return _to_read(provider)
+23 -8
View File
@@ -29,13 +29,13 @@ def days_until(d: Optional[date]) -> Optional[int]:
return (d - date.today()).days
def _judge_status(days: Optional[int]) -> str:
def _judge_status(days: Optional[int], threshold: int = EXPIRING_THRESHOLD) -> str:
"""按剩余天数定级:expired / expiring / valid"""
if days is None:
return "unknown"
if days < 0:
return "expired"
if days <= EXPIRING_THRESHOLD:
if days <= threshold:
return "expiring"
return "valid"
@@ -115,7 +115,20 @@ def list_subdomains(session: Session, asset_id: Optional[int] = None) -> list:
]
def _require_asset(session: Session, asset_id: Optional[int]) -> None:
"""校验关联资产存在
SQLite 默认不强制外键,写一个不存在的 asset_id 不会报错,
但列表页 join 不出资产名,删除资产时也清不到这些悬空行。
"""
if asset_id is None:
return
if not session.get(Asset, asset_id):
raise ValueError(f"关联资产不存在(id={asset_id})")
def create_subdomain(session: Session, data: dict) -> Subdomain:
_require_asset(session, data["asset_id"])
sub = Subdomain(
asset_id=data["asset_id"],
host=data["host"].strip().lower(),
@@ -157,7 +170,7 @@ def delete_subdomain(session: Session, sub_id: int) -> None:
# ---------------- 站点证书监控 ----------------
def _apply_probe(cert: SiteCert, info: dict) -> SiteCert:
def _apply_probe(cert: SiteCert, info: dict, threshold: int = EXPIRING_THRESHOLD) -> SiteCert:
"""把探测结果写入模型并定级"""
cert.subject_cn = info["subject_cn"]
cert.issuer = info["issuer"]
@@ -165,16 +178,16 @@ def _apply_probe(cert: SiteCert, info: dict) -> SiteCert:
cert.valid_to = info["valid_to"]
cert.fingerprint = info["fingerprint"]
cert.error = None
cert.status = _judge_status(days_until(info["valid_to"]))
cert.status = _judge_status(days_until(info["valid_to"]), threshold)
cert.last_checked_at = utcnow()
return cert
def check_one(session: Session, cert: SiteCert) -> SiteCert:
def check_one(session: Session, cert: SiteCert, threshold: int = EXPIRING_THRESHOLD) -> SiteCert:
"""重新探测单条证书记录(失败则标记 error,保留旧到期信息)"""
try:
info = probe_site_cert(cert.hostname, cert.port)
_apply_probe(cert, info)
_apply_probe(cert, info, threshold)
except Exception as e: # noqa: BLE001
cert.status = "error"
cert.error = str(e)[:200]
@@ -187,6 +200,7 @@ def check_one(session: Session, cert: SiteCert) -> SiteCert:
def create_site_cert(session: Session, hostname: str, port: int = 443, asset_id: Optional[int] = None) -> SiteCert:
"""创建探测目标并立即探测一次;hostname+port 已存在则复用并刷新"""
_require_asset(session, asset_id)
hostname = hostname.strip().lower()
existing = session.exec(
select(SiteCert).where(SiteCert.hostname == hostname, SiteCert.port == port)
@@ -223,12 +237,13 @@ def delete_site_cert(session: Session, cert_id: int) -> None:
session.commit()
def check_all_site_certs(session: Session) -> dict:
def check_all_site_certs(session: Session, threshold: int = EXPIRING_THRESHOLD) -> dict:
"""全量刷新所有站点证书,返回统计与异常清单
探测为纯网络 IO(单次最长 8s),用线程池并发探测后统一写库:
串行时 N 个站点最坏耗时 N×8s,并发后接近单站点耗时;
写库集中在主线程一次 commit(原来逐条 commit 产生 N 次事务)。
threshold:「即将到期」的天数阈值,默认 EXPIRING_THRESHOLD。
"""
from concurrent.futures import ThreadPoolExecutor
@@ -244,7 +259,7 @@ def check_all_site_certs(session: Session) -> dict:
for future in futures:
cert = futures[future]
try:
_apply_probe(cert, future.result())
_apply_probe(cert, future.result(), threshold)
except Exception as e: # noqa: BLE001
cert.status = "error"
cert.error = str(e)[:200]
+24 -17
View File
@@ -36,17 +36,6 @@ from app.models.provider import Provider
_VALID_STATUS = {s.value for s in AssetStatus}
def _load_config(provider: Provider) -> dict:
"""(已弃用)解密平台级 API 配置,仅为兼容历史数据保留"""
plain = crypto.decrypt(provider.api_config_encrypted)
if not plain:
return {}
try:
return json.loads(plain)
except (json.JSONDecodeError, TypeError):
return {}
def _load_account_config(account: Account) -> dict:
"""解密账号的 API 配置 JSON"""
plain = crypto.decrypt(account.api_config_encrypted)
@@ -334,6 +323,20 @@ _AI_SDK_MAP = {
}
class BalanceUnsupported(HTTPException):
"""该平台没有可用的余额查询接口
与「查询失败」区分开:sync_all_ai_balances 把它计入 skipped 而不是 failed,
否则每次「同步 AI 余额」都会把这类平台报成错误。
"""
def __init__(self, sdk_type: str):
super().__init__(
status_code=status.HTTP_400_BAD_REQUEST,
detail=f"{sdk_type} 暂不支持余额查询",
)
def refresh_ai_balance(session: Session, asset_id: int) -> dict:
"""刷新 AI 账号余额:解密 api_key → 适配器 get_account → 更新余额"""
asset = session.get(Asset, asset_id)
@@ -367,19 +370,18 @@ def refresh_ai_balance(session: Session, asset_id: int) -> dict:
)
adapter = registry.get_adapter(sdk_type, {"api_key": api_key})
if not adapter.capabilities().get("get_account"):
raise BalanceUnsupported(sdk_type)
try:
acc = adapter.get_account()
except NotImplementedError:
raise HTTPException(
status_code=status.HTTP_400_BAD_REQUEST,
detail=f"{sdk_type} 暂不支持余额查询",
)
except NotImplementedError as e:
raise BalanceUnsupported(sdk_type) from e
except HTTPException:
raise
except Exception as e: # noqa: BLE001
raise HTTPException(
status_code=status.HTTP_502_BAD_GATEWAY, detail=f"查询余额失败:{e}"
)
) from e
if acc.balance is not None:
ai.balance = acc.balance
@@ -417,6 +419,7 @@ def sync_all_ai_balances(session: Session, max_workers: int = 5) -> dict:
success = 0
failed = 0
skipped = 0
errors = []
with ThreadPoolExecutor(max_workers=max_workers) as pool:
futures = {pool.submit(_refresh_one, aid): name for aid, name in asset_ids}
@@ -425,6 +428,9 @@ def sync_all_ai_balances(session: Session, max_workers: int = 5) -> dict:
try:
future.result()
success += 1
except BalanceUnsupported:
# 平台本身没余额接口,不算失败(否则每次都报一堆红)
skipped += 1
except HTTPException as e:
failed += 1
errors.append(f"{name}: {e.detail}")
@@ -435,6 +441,7 @@ def sync_all_ai_balances(session: Session, max_workers: int = 5) -> dict:
"total": len(asset_ids),
"success": success,
"failed": failed,
"skipped": skipped,
"errors": errors,
"synced_at": utcnow().isoformat(),
}