fix(risk): 规则扫描加跨进程锁——手工触发与定时扫描此前可以同时跑
docs/25 P2 最后一项。核实后分清了两层，报告没区分： - **定时扫描是安全的**：risk_scan_scheduler.py 已有 MySQL 连接级咨询锁 （GET_LOCK，锁名 jr_risk_scan_schedule），跨进程互斥。 - **HTTP 端点不安全**：POST /api/v1/risk/alerts/scan → RiskScanService.scan() 只用了 **进程内** asyncio.Lock。多 Web worker、或 Worker 与 API 同时运行时形同虚设。 而扫描的幂等只有应用层的 _exists 查重 —— fin_risk_alert 的 trigger_rule_codes 是 JSON 数组，**无法建唯一索引兜底**（同一交易可命中多条规则，唯一键本应是"交易+规则"，而规则 埋在 JSON 里）。所以两条路径并发时会同时查不到、同时插入，产生重复预警。 **改动**： 1. 把 mysql_scan_lock 与锁名移到 pp/infrastructure/db.py —— 端点与调度器**必须共用 同一把锁**，放在基础设施层两个入口才都能引用（service 不该反向依赖 worker）。 调度器改为从那里 import。 2. **端点层加锁**（controllers/risk.py 的 scan 端点）：取不到锁就抛 RiskScanBusyError （与 service 内部那把进程内锁用同一错误类型与文案）。 **为什么不加在 RiskScanService.scan() 内部**：GET_LOCK 是**连接级**的，而调度器已经在 它自己的 session 上持锁；被两个入口共用的服务方法若再取同一把锁，取锁的连接并不是持锁的 那一个、必然返回 0 —— 会**把定时扫描自己挡死**。所以锁加在入口层，每个入口只取一次。 **实测**： - 无人持锁时扫描 → **200**「规则扫描完成」 - 本进程先取得跨进程锁后再调端点 → **409**「规则扫描正在执行，请稍后重试」 （同一进程内不同 session 也互斥，说明它是连接级的，正是跨进程所需） - 释放后再调 → **200**，恢复正常 顺带第 4 次遇到 409 复用错误码 RUN_NOT_CANCELLABLE，语义不符；属 P3 待处理项。 ruff / mypy(136 文件) / 639 unit+contract 全绿。
This commit is contained in:
@@ -23,13 +23,14 @@ from app.api.schemas.risk import (
|
||||
RiskNotificationPageQuery,
|
||||
)
|
||||
from app.core.contracts import RequestContext
|
||||
from app.infrastructure.db import mysql_scan_lock
|
||||
from app.service.risk_action_service import RiskActionService
|
||||
from app.service.risk_daily_report_mail_service import RiskDailyReportMailService
|
||||
from app.service.risk_daily_report_service import RiskDailyReportService
|
||||
from app.service.risk_evidence_archive_service import RiskEvidenceArchiveService
|
||||
from app.service.risk_notification_service import RiskNotificationService
|
||||
from app.service.risk_query_service import RiskQueryService
|
||||
from app.service.risk_scan_service import RiskScanService
|
||||
from app.service.risk_scan_service import RiskScanBusyError, RiskScanService
|
||||
|
||||
router = APIRouter(
|
||||
prefix="/api/v1/risk",
|
||||
@@ -62,7 +63,14 @@ async def scan_risk_alerts(
|
||||
context: RequestContext = Depends(build_request_context), # noqa: B008
|
||||
session: AsyncSession = Depends(get_session), # noqa: B008
|
||||
) -> dict[str, object]:
|
||||
data = await RiskScanService(session).scan(context)
|
||||
# 手工触发的扫描必须与定时扫描互斥，否则两条路径会同时查不到重复、同时插入。
|
||||
# 锁加在**入口层**而不是 `RiskScanService.scan()` 内部：`GET_LOCK` 是连接级的，
|
||||
# 而调度器已在它自己的 session 上持锁 —— 被两个入口共用的服务方法若再取同一把锁，
|
||||
# 取锁的连接不是持锁的那一个、必然失败，会**把定时扫描自己挡死**。
|
||||
async with mysql_scan_lock() as acquired:
|
||||
if not acquired:
|
||||
raise RiskScanBusyError("规则扫描正在执行，请稍后重试")
|
||||
data = await RiskScanService(session).scan(context)
|
||||
return _envelope(data, context)
|
||||
|
||||
|
||||
|
||||
@@ -1,3 +1,7 @@
|
||||
from collections.abc import AsyncIterator
|
||||
from contextlib import asynccontextmanager
|
||||
|
||||
from sqlalchemy import text
|
||||
from sqlalchemy.engine import make_url
|
||||
from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker, create_async_engine
|
||||
|
||||
@@ -29,3 +33,36 @@ if "init_command" not in _url.query:
|
||||
|
||||
engine = create_async_engine(_url, pool_pre_ping=True)
|
||||
SessionFactory = async_sessionmaker(engine, class_=AsyncSession, expire_on_commit=False)
|
||||
|
||||
# 规则扫描的**跨进程**咨询锁名。用 MySQL 连接级 `GET_LOCK` 而不是进程内 `asyncio.Lock`：
|
||||
# 后者只在单个进程内互斥，多 Web worker、或 Worker 与 API 同时运行时形同虚设。
|
||||
SCAN_LOCK_NAME = "jr_risk_scan_schedule"
|
||||
|
||||
|
||||
@asynccontextmanager
|
||||
async def mysql_scan_lock() -> AsyncIterator[bool]:
|
||||
"""约束规则扫描的跨进程并发；yield 是否取得锁。
|
||||
|
||||
**为什么两个入口都必须用它**：定时扫描（`worker/risk_scan_scheduler.py`）与手工触发的
|
||||
HTTP 端点（`POST /api/v1/risk/alerts/scan`）最终都调 `RiskScanService.scan()`，
|
||||
而扫描的幂等只有应用层的 `_exists` 查重 —— 两条路径并发时会同时查不到、同时插入，
|
||||
产生重复预警。锁要**共用同一个名字**才能互相排斥。
|
||||
|
||||
**为什么不放在 `scan()` 内部**：`GET_LOCK` 是**连接级**的。调度器已在自己的 session 上
|
||||
持锁，再新建 session 去调 `scan()`；若 `scan()` 又去取同一把锁，取锁的连接并不是持锁的
|
||||
那一个，`GET_LOCK` 返回 0 —— **调度器会把自己挡死**。所以锁加在**入口层**
|
||||
（调度器 / 端点各取一次），而不是被两个入口共用的服务方法里。
|
||||
|
||||
`GET_LOCK(name, 0)` 的 0 表示不等待、立即返回：取不到就跳过本轮，而不是排队。
|
||||
"""
|
||||
async with SessionFactory() as session:
|
||||
acquired = bool(
|
||||
await session.scalar(text("SELECT GET_LOCK(:name, 0)"), {"name": SCAN_LOCK_NAME})
|
||||
)
|
||||
try:
|
||||
yield acquired
|
||||
finally:
|
||||
if acquired:
|
||||
await session.scalar(
|
||||
text("SELECT RELEASE_LOCK(:name)"), {"name": SCAN_LOCK_NAME}
|
||||
)
|
||||
|
||||
@@ -9,17 +9,15 @@ from __future__ import annotations
|
||||
import argparse
|
||||
import asyncio
|
||||
import logging
|
||||
from collections.abc import AsyncIterator, Awaitable, Callable
|
||||
from contextlib import AbstractAsyncContextManager, asynccontextmanager
|
||||
from collections.abc import Awaitable, Callable
|
||||
from contextlib import AbstractAsyncContextManager
|
||||
from datetime import UTC, datetime, timedelta
|
||||
from typing import Any
|
||||
from uuid import uuid4
|
||||
|
||||
from sqlalchemy import text
|
||||
|
||||
from app.core.config import get_settings
|
||||
from app.core.contracts import RequestContext
|
||||
from app.infrastructure.db import SessionFactory, engine
|
||||
from app.infrastructure.db import SessionFactory, engine, mysql_scan_lock
|
||||
from app.model.audit import InteractionAudit
|
||||
from app.service.risk_scan_schedule_config import (
|
||||
RiskScanScheduleConfig,
|
||||
@@ -27,7 +25,6 @@ from app.service.risk_scan_schedule_config import (
|
||||
)
|
||||
from app.service.risk_scan_service import RiskScanService
|
||||
|
||||
SCAN_LOCK_NAME = "jr_risk_scan_schedule"
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
ConfigLoader = Callable[[], Awaitable[RiskScanScheduleConfig]]
|
||||
@@ -35,25 +32,8 @@ ScanExecutor = Callable[[], Awaitable[dict[str, int | str]]]
|
||||
AuditWriter = Callable[[str, dict[str, Any]], Awaitable[None]]
|
||||
LockFactory = Callable[[], AbstractAsyncContextManager[bool]]
|
||||
|
||||
|
||||
@asynccontextmanager
|
||||
async def mysql_scan_lock() -> AsyncIterator[bool]:
|
||||
"""使用 MySQL 连接级咨询锁约束跨进程并发。"""
|
||||
async with SessionFactory() as session:
|
||||
acquired = bool(
|
||||
await session.scalar(
|
||||
text("SELECT GET_LOCK(:name, 0)"),
|
||||
{"name": SCAN_LOCK_NAME},
|
||||
)
|
||||
)
|
||||
try:
|
||||
yield acquired
|
||||
finally:
|
||||
if acquired:
|
||||
await session.scalar(
|
||||
text("SELECT RELEASE_LOCK(:name)"),
|
||||
{"name": SCAN_LOCK_NAME},
|
||||
)
|
||||
# 跨进程扫描锁已移到 `app/infrastructure/db.py`：端点与调度器**必须共用同一把锁**，
|
||||
# 放在基础设施层两个入口才都能引用（service 不该反向依赖 worker）。
|
||||
|
||||
|
||||
async def default_scan_executor() -> dict[str, int | str]:
|
||||
|
||||
Reference in New Issue
Block a user