feat(douyu): 整合自动化中心并修复重登队列

This commit is contained in:
yml
2026-11-22 21:17:37 +08:00
parent af05c54c3f
commit 94de1d167b
18 changed files with 854 additions and 1029 deletions
-349
View File
@@ -1,349 +0,0 @@
"""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",
# 「同一账号多久重检」= 7 天(就是界面上那个字段)
interval_seconds=7 * 24 * 3600,
batch_size=10,
concurrency=3,
enabled=False,
)
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",
hv_checked_days_ago: float | None = None,
hv_status: str | None = None,
) -> 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
)
if hv_checked_days_ago is not None:
acc.high_value_checked_at = datetime.now(UTC) - timedelta(
days=hv_checked_days_ago
)
if hv_status is not None:
acc.high_value_bairimeng_status = hv_status
acc.high_value_mengtu_status = hv_status
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_assigned_and_recent_login_lead_within_the_same_group(self):
"""没查过的仍整体优先;同一档里已分配的在前,再按最近登录。"""
stale = self.account("ovr-stale-assigned", queried_days_ago=30)
stale.assigned_to = self.admin.id
recent = self.account("ovr-recent")
recent.cookie_updated_at = datetime.now(UTC)
older = self.account("ovr-older")
older.cookie_updated_at = datetime.now(UTC) - timedelta(days=10)
assigned = self.account("ovr-assigned")
assigned.assigned_to = self.admin.id
assigned.cookie_updated_at = datetime.now(UTC) - timedelta(days=1)
self.session.commit()
names = [a.username for a in self.selected()]
assert names.index("ovr-assigned") < names.index("ovr-recent")
assert names.index("ovr-recent") < names.index("ovr-older")
assert names.index("ovr-older") < names.index("ovr-stale-assigned")
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_recheck_interval_comes_from_the_ui_field(self):
"""**用界面上那个「同一账号多久重检」** —— 与其它批量作业同一套口径。
之前这里读的是 `job.config["refresh_days"]`,于是界面上那个字段
对本作业是**死的**(改了没反应)—— 正是用户批评过的
「看起来能配、其实配的是不相干的东西」。
"""
self.job.interval_seconds = 100 * 24 * 3600
self.session.commit()
self.account("ovr-30d", queried_days_ago=30)
assert "ovr-30d" not in [a.username for a in self.selected()], (
"重检间隔 100 天时,30 天前查过的账号不该重新入选"
)
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 TestHighValueRefresh(_Base):
"""高价值结论有**自己独立的时钟** —— 资料刷新必须跟着它走。
缺口长这样:一键查询里商品接口单独失败(其它项都拿到了),
``overview_queried_at`` 被刷新了、``high_value_checked_at`` 却停在原地。
资料看起来很新,于是**再也不会被选中** —— 高价值状态永久停在旧结论。
"""
def test_stale_high_value_is_selected_even_if_overview_is_fresh(self):
"""资料 1 天前查过(很新),但高价值结论 30 天前 —— 必须重选。"""
self.account(
"ovr-hv-stale",
queried_days_ago=1,
hv_checked_days_ago=30,
hv_status="available",
)
assert "ovr-hv-stale" in [a.username for a in self.selected()]
def test_fresh_high_value_is_not_selected(self):
"""资料和高价值结论都新鲜 —— 不该被重选(避免重复请求)。"""
self.account(
"ovr-hv-fresh",
queried_days_ago=1,
hv_checked_days_ago=1,
hv_status="available",
)
assert "ovr-hv-fresh" not in [a.username for a in self.selected()]
def test_never_checked_high_value_is_not_selected_on_its_own(self):
"""从来没查到过高价值结论,**不**因此被重选。
这种情况要么是「账号根本还没查过资料」(本来就该被选中),
要么是**商品接口对这个账号稳定失败**(没配 manualID、账号没有宝典)。
后者若每轮都重选,同一批账号会反复撞墙,把真正有新欠账的挤掉。
「资料过期」那条已经兜住了它。
"""
self.account("ovr-hv-never", queried_days_ago=1)
assert "ovr-hv-never" not in [a.username for a in self.selected()]
def test_high_value_uses_the_same_ruler_as_the_ui_field(self):
"""高价值过期**用同一把尺子**(界面上那个「同一账号多久重检」)。
之前这里读的是 `job.config["refresh_days"]`,界面上那个字段对本作业
就是死的;同理,若高价值另设一个窗口,那个字段对本作业就变成半死的。
"""
self.job.interval_seconds = 100 * 24 * 3600
self.session.commit()
# 资料 30 天前(按 100 天不该重选),高价值结论 60 天前(同样不该)
self.account(
"ovr-hv-60d",
queried_days_ago=30,
hv_checked_days_ago=60,
hv_status="available",
)
assert "ovr-hv-60d" not in [a.username for a in self.selected()]
# 缩回 7 天 → 高价值 60 天前就过期了,要重选
self.job.interval_seconds = 7 * 24 * 3600
self.session.commit()
assert "ovr-hv-60d" in [a.username for a in self.selected()]
def test_high_value_refresh_still_requires_valid_ck(self):
"""高价值过期**不能**绕过 CK 有效性 —— CK 失效的查了也是白查。"""
self.account(
"ovr-hv-stale-invalid",
ck="invalid",
queried_days_ago=1,
hv_checked_days_ago=30,
hv_status="available",
)
assert "ovr-hv-stale-invalid" not in [a.username for a in self.selected()]
def test_invalidated_conclusion_is_rechecked_within_six_hours(self):
"""结论被**作废**(unknown)的账号,6 小时内要重查 —— 哪怕重检间隔是 7 天。
场景:CK 失效期间账号可能被别人手工登录并兑换(这些号共享给客服),
所以 CK 一判定失效就把结论作废。CK 恢复之后必须**尽快**重新确认 ——
否则运营会一直拿着一份「待确认」的结论,而它其实已经变了。
⚠️ 这里用的是「资料重检间隔」(7 天)之外的**另一个**窗口:
作废的结论等 7 天太久了,运营等不了。但也不能更密 ——
商城接口对某些账号稳定失败,那类账号会一直是 unknown,每轮都重选
就是反复撞墙,挤掉真正有新欠账的账号。
"""
# 资料 1 天前(很新,按 7 天不该重选),高价值结论是 10 小时前的 unknown
self.account(
"ovr-hv-invalidated",
queried_days_ago=1,
hv_checked_days_ago=10 / 24,
hv_status="unknown",
)
assert "ovr-hv-invalidated" in [a.username for a in self.selected()]
def test_invalidated_conclusion_is_not_rechecked_too_often(self):
"""作废的结论 6 小时内**不**重选 —— 不能反复撞同一个失败的接口。"""
self.account(
"ovr-hv-invalidated-fresh",
queried_days_ago=1,
hv_checked_days_ago=1 / 24, # 1 小时前
hv_status="unknown",
)
assert "ovr-hv-invalidated-fresh" not in [a.username for a in self.selected()]
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"
+1
View File
@@ -550,6 +550,7 @@ class TestAutomationApi(_Base):
assert payload["scheduler"]["running"] is False
keys = [item["key"] for item in payload["jobs"]]
assert ck_health.JOB_KEY in keys
assert "account_overview_refresh" not in keys
job = next(item for item in payload["jobs"] if item["key"] == ck_health.JOB_KEY)
assert job["enabled"] is False
assert job["interval_seconds"] == 6 * 3600
+38
View File
@@ -22,6 +22,7 @@ from datetime import UTC, datetime, timedelta
from tests.support.db import get_test_engine, make_test_session
from web.backend.models import (
Account,
AccountWorkItem,
AutomationJob,
LoginAttempt,
LoginQuotaSetting,
@@ -735,6 +736,43 @@ class TestSettlementVisibility(_Base):
checker.close()
stale.close()
def test_settlement_releases_claimed_account_without_attempt(self):
"""预算跳过等无尝试记录场景也必须结束 running 工作项。"""
account = self.make_account("acc-no-attempt")
now = datetime.now(UTC)
work = AccountWorkItem(
account_id=account.id,
work_type=ck_relogin.WORK_LOGIN_RELOGIN,
priority=30,
status="running",
due_at=now,
attempt_count=1,
max_attempts=20,
dedupe_key=f"login_relogin:{account.id}",
started_at=now,
)
self.session.add(work)
self.session.commit()
counters = ck_relogin.settle_fresh_results(
"batch-without-attempts",
now=now,
max_attempts=2,
expected_account_ids=[account.id],
)
checker = make_test_session(self.engine)
try:
settled_work = checker.get(AccountWorkItem, work.id)
settled_account = checker.get(Account, account.id)
assert counters["scanned"] == 1
assert counters["retry"] == 1
assert settled_work.status == "pending"
assert settled_work.due_at >= now + timedelta(minutes=59)
assert settled_account.relogin_next_at >= now + timedelta(minutes=59)
finally:
checker.close()
def test_settlement_marks_real_success_as_valid(self):
"""结算看到 success 就必须清零退避,不能写成失败。"""
account = self.make_account(
+1 -1
View File
@@ -42,7 +42,7 @@ TESTS_DIR = Path(__file__).resolve().parent
PROJECT_ROOT = TESTS_DIR.parent
VERSIONS_DIR = PROJECT_ROOT / "web" / "backend" / "migrations" / "versions"
HEAD_REVISION = "20261122_0116"
HEAD_REVISION = "20261122_0117"
# 迁移新增、但不声明在模型里的复合查询索引。
EXTRA_INDEXES = (