yml2213
fd578683b6
## 为什么不能一次性跑全量
生产约 **8,000 个**「CK 有效但没查过」的账号。一次性发起 ≈ **8 万次斗鱼请求**，
很容易触发风控 —— 而风控一旦触发是**所有账号一起受影响**。试水 200 个时
0.65 秒/账号看着很香（全量 86 分钟），但那正是危险的地方：**跑得越快越容易被封**。
## 新作业 `account_overview_refresh`
照「自动巡检」（`ck_health`）那套做，并复用它的两道保护。
| 项 | 默认 | 说明 |
| --- | --- | --- |
| `interval_seconds` | 30 分钟 | 多久一轮 |
| `batch_size` | 30 | 每轮查多少（约 20 秒跑完）|
| `concurrency` | 3 | 并发 |
| `refresh_days` | 7（config 可调）| 查过的多久后再查 |
| `enabled` | **False** | 默认关，由运维手动开 |
按默认值：8,000 个约 **3 天**补完，全程不打满斗鱼接口。
## 两类账号**分开**（用户明确要求）
- **没查过的**（`overview_queried_at IS NULL`）**优先**
- 查过的只在超过 `refresh_days` 后重新入选
否则"周期性刷新老账号"会永远挤掉"从没查过的新账号"。
MySQL 没有 `NULLS FIRST`，用 `ORDER BY overview_queried_at IS NULL DESC` 等价实现。
## 只查有效 CK
`ck_check_status == 'valid'`。失效的查了也是白查（只会多一条失败记录），
交给「CK 检测与重登」处理。另外沿用 `cookie_operable_filter()`（有 CK 且无在途尝试）、
排除人工状态（停用/归档）与「不参与自动化」的账号。
## 风控熔断（沿用巡检口径）
一轮里失败率 > 50%（且样本 ≥ 10）就**自己把自己关掉**，交给人确认。
消息里写明可能原因，否则运维只看到"失败率高"会一头雾水：
失败率过高，已自动关闭作业（大面积失败通常是**被风控**或接口抖动，请先确认再重开）
样本 < 10 不熔断 —— 1 个失败就是 100%，比例没意义。
## 查询记录
每轮都建真实的 `douyu_tasks` 行（payload 标 `source=scheduled_refresh`），
「哪个账号、什么时候、查出来什么」都有据可查，也能在任务列表里
区分「定时巡检」与「人工点查询」。账号上的 `overview_queried_at`
则是"下次该不该再查"的依据。
## 验证
- **镜像库实测**：`本轮查 3 个：成功 3 / 失败 0（剩 7784 个待查）`
- 本地 `TZ=Asia/Shanghai` **632 passed / 6 skipped**；ruff 全绿
- 新增 11 条用例：选择口径（只查有效 CK、没查过优先、refresh_days 可调、
人工状态排除、batch_size 限流）+ 熔断（高失败率关闭、正常不关、小样本不熔断）
+ 一条回归：计数不能走带 LIMIT 的查询（作业第一版就是这么挂的）
211 lines
8.1 KiB
Python
211 lines
8.1 KiB
Python
"""定时补全账号资料的作业（`account_overview_refresh`）。
|
|
|
|
**为什么不能一次性跑全量**：生产约 8,000 个待查账号，一次性发起 ≈ 8 万次斗鱼请求，
|
|
很容易触发风控 —— 而风控的代价是所有账号一起受影响。所以做成定时分批，
|
|
并复用「自动巡检」的两道保护（`cookie_operable_filter` + 错误率熔断）。
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from datetime import UTC, datetime, timedelta
|
|
|
|
from tests.support.db import get_test_engine, make_test_session
|
|
from web.backend.models import Account, AutomationJob, User
|
|
from web.backend.services.automation.jobs import account_overview_refresh as job
|
|
|
|
|
|
class _Base:
|
|
def setup_method(self):
|
|
self.engine = get_test_engine()
|
|
self.session = make_test_session(self.engine)
|
|
self.admin = User(username="ovr-admin", password_hash="h", role="super_admin")
|
|
self.session.add(self.admin)
|
|
self.session.flush()
|
|
self.job = AutomationJob(
|
|
key=job.JOB_KEY,
|
|
title="账号资料定时补全",
|
|
interval_seconds=1800,
|
|
batch_size=10,
|
|
concurrency=3,
|
|
enabled=False,
|
|
config={"refresh_days": 7},
|
|
)
|
|
self.session.add(self.job)
|
|
self.session.commit()
|
|
|
|
def teardown_method(self):
|
|
self.session.close()
|
|
|
|
def account(
|
|
self,
|
|
name: str,
|
|
*,
|
|
ck: str = "valid",
|
|
queried_days_ago: int | None = None,
|
|
auto_managed: bool = True,
|
|
status: str = "active",
|
|
) -> Account:
|
|
acc = Account(
|
|
username=name,
|
|
password="p",
|
|
email=f"{name}@e.com",
|
|
email_password="ep",
|
|
cookie="acf_uid=1; acf_ccn=c",
|
|
ck_check_status=ck,
|
|
auto_managed=auto_managed,
|
|
status=status,
|
|
)
|
|
if queried_days_ago is not None:
|
|
acc.overview_queried_at = datetime.now(UTC) - timedelta(
|
|
days=queried_days_ago
|
|
)
|
|
self.session.add(acc)
|
|
self.session.commit()
|
|
return acc
|
|
|
|
def selected(self) -> list[Account]:
|
|
return job.build_select_query(self.session, self.job, datetime.now(UTC)).all()
|
|
|
|
|
|
class TestSelection(_Base):
|
|
def test_only_valid_ck_is_selected(self):
|
|
"""只查有效 CK —— 失效的查了也是白查，只会多一条失败记录。"""
|
|
self.account("ovr-valid", ck="valid")
|
|
self.account("ovr-invalid", ck="invalid")
|
|
self.account("ovr-error", ck="error")
|
|
names = [a.username for a in self.selected()]
|
|
assert "ovr-valid" in names
|
|
assert "ovr-invalid" not in names
|
|
assert "ovr-error" not in names
|
|
|
|
def test_never_queried_comes_first(self):
|
|
"""**没查过的优先** —— 否则周期性刷新会永远挤掉从没查过的新账号。"""
|
|
self.account("ovr-old", queried_days_ago=30)
|
|
self.account("ovr-never")
|
|
assert self.selected()[0].username == "ovr-never"
|
|
|
|
def test_recently_queried_is_skipped(self):
|
|
"""刚查过的要跳过（`refresh_days` 之内不再查）。"""
|
|
self.account("ovr-fresh", queried_days_ago=1)
|
|
names = [a.username for a in self.selected()]
|
|
assert "ovr-fresh" not in names
|
|
|
|
def test_stale_is_selected_again(self):
|
|
self.account("ovr-stale", queried_days_ago=30)
|
|
assert "ovr-stale" in [a.username for a in self.selected()]
|
|
|
|
def test_refresh_days_is_configurable(self):
|
|
self.job.config = {"refresh_days": 100}
|
|
self.session.commit()
|
|
self.account("ovr-30d", queried_days_ago=30)
|
|
assert "ovr-30d" not in [a.username for a in self.selected()]
|
|
|
|
def test_manual_and_unmanaged_accounts_are_excluded(self):
|
|
"""人工状态（停用/归档）与「不参与自动化」的账号不该被自动碰。"""
|
|
self.account("ovr-disabled", status="disabled")
|
|
self.account("ovr-archived", status="archived")
|
|
self.account("ovr-unmanaged", auto_managed=False)
|
|
names = [a.username for a in self.selected()]
|
|
for bad in ("ovr-disabled", "ovr-archived", "ovr-unmanaged"):
|
|
assert bad not in names
|
|
|
|
def test_batch_size_limits_the_round(self):
|
|
for i in range(5):
|
|
self.account(f"ovr-batch-{i}")
|
|
self.job.batch_size = 2
|
|
self.session.commit()
|
|
assert len(self.selected()) == 2
|
|
|
|
def test_remaining_count_works_with_limit_applied(self):
|
|
"""回归：计数不能走 `build_select_query`。
|
|
|
|
那个查询已经 `LIMIT` 了，再 `order_by(None)` 会被 SQLAlchemy 拒绝：
|
|
"order_by() being called on a Query which already has LIMIT"。
|
|
（真踩过 —— 作业第一版就是这么挂的。）
|
|
"""
|
|
for i in range(4):
|
|
self.account(f"ovr-count-{i}")
|
|
self.job.batch_size = 1
|
|
self.session.commit()
|
|
assert job.count_remaining(self.session, self.job, datetime.now(UTC)) >= 4
|
|
|
|
|
|
class TestCircuitBreaker(_Base):
|
|
"""失败率过高就**自己把自己关掉**（风控保护）。
|
|
|
|
与「自动巡检」同口径。大面积失败通常意味着被风控或斗鱼接口抖动 ——
|
|
这时候继续跑只会把更多账号写坏，停手交给人确认才是对的。
|
|
"""
|
|
|
|
def _run_with(self, total: int, failed: int):
|
|
"""打桩执行器，只关心失败率熔断的逻辑。"""
|
|
from web.backend.models import DouyuTask
|
|
|
|
for i in range(total):
|
|
self.account(f"ovr-cb-{i}")
|
|
|
|
def fake_create(db, ids, task_type, scope, created_by, payload=None):
|
|
batch_id = "cb-batch"
|
|
for account_id in ids:
|
|
self.session.add(
|
|
DouyuTask(
|
|
batch_id=batch_id,
|
|
account_id=account_id,
|
|
task_type=task_type,
|
|
handbook_scope=scope,
|
|
status="planned",
|
|
created_by=created_by,
|
|
)
|
|
)
|
|
self.session.commit()
|
|
return batch_id, len(ids)
|
|
|
|
session = self.session # 闭包捕获：下面 _FakeRunner 里的 self 是 runner
|
|
|
|
class _FakeRunner:
|
|
def __init__(self, *a, **k):
|
|
pass
|
|
|
|
def run(self):
|
|
# 把前 `failed` 个标成失败，其余成功
|
|
rows = (
|
|
session.query(DouyuTask)
|
|
.filter(DouyuTask.batch_id == "cb-batch")
|
|
.order_by(DouyuTask.id.asc())
|
|
.all()
|
|
)
|
|
for index, row in enumerate(rows):
|
|
row.status = "failed" if index < failed else "success"
|
|
session.commit()
|
|
|
|
import web.backend.services.automation.jobs.account_overview_refresh as mod
|
|
|
|
original_create = mod.create_douyu_planned_tasks
|
|
original_runner = mod.DouyuBatchRunner
|
|
mod.create_douyu_planned_tasks = fake_create
|
|
mod.DouyuBatchRunner = _FakeRunner
|
|
try:
|
|
return job.run_account_overview_refresh(self.session, self.job)
|
|
finally:
|
|
mod.create_douyu_planned_tasks = original_create
|
|
mod.DouyuBatchRunner = original_runner
|
|
|
|
def test_high_failure_ratio_disables_the_job(self):
|
|
out = self._run_with(total=20, failed=15) # 75% 失败
|
|
self.session.refresh(self.job)
|
|
assert out["paused"] is True
|
|
assert self.job.enabled is False, "失败率过高应自动关闭作业"
|
|
assert "风控" in out["message"]
|
|
|
|
def test_normal_failure_ratio_keeps_job_running(self):
|
|
out = self._run_with(total=20, failed=1) # 5% 失败（真实的 CK 失效率）
|
|
self.session.refresh(self.job)
|
|
assert out["paused"] is False
|
|
assert "风控" not in out["message"]
|
|
|
|
def test_small_sample_does_not_trigger_breaker(self):
|
|
"""样本太少时比例没有意义（1 个失败就是 100%），不该熔断。"""
|
|
out = self._run_with(total=3, failed=3)
|
|
self.session.refresh(self.job)
|
|
assert out["paused"] is False, "样本不足时不该熔断"
|