Merge branch 'main' of https://gitea.shaguabijia.com/WonderableAI/shaguabijia-app-server into feat/openobserve-api-metrics
# Conflicts: # app/main.py
This commit is contained in:
guke
@@ -0,0 +1,830 @@
|
||||
# 逐次比价/领券广告收益 Implementation Plan
|
||||
|
||||
> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
|
||||
|
||||
**Goal:** 让 admin「领券数据」和「比价记录」两个看板的 table 每一行显示这一次领券/比价产生的广告收益(预估元)。
|
||||
|
||||
**Architecture:** 客户端在信息流(Draw)展示上报 eCPM 时带上本场 `trace_id`,后端落到 `ad_ecpm_record.trace_id`(新列 + 索引)。两个看板在分页后,对**当前页**的 trace_id 批量聚合一次 `ad_ecpm_record` 的展示收益(单条收益 = min(eCPM元,¥500)/1000,与广告收益报表同口径),挂到每行。查询是按 trace_id 索引的单条聚合,与已上线的 `_ad_coins_by_trace`(比价记录页「比价赚N金币」)同一性能剖面。
|
||||
|
||||
**Tech Stack:** 后端 FastAPI + SQLAlchemy 2.0 + Alembic;客户端 Android(Kotlin/OkHttp);admin 前端 Next.js + React + Ant Design(shaguabijia-admin-web)。
|
||||
|
||||
---
|
||||
|
||||
## 背景与约束(执行前必读)
|
||||
|
||||
- **收益 ≠ 金币**。本功能查的是「我们赚的广告收益」(数据源 `ad_ecpm_record`,客户端自报 eCPM 折算的预估),不是发给用户的金币(那是 `ad_feed_reward_record`,已有 `trace_id`)。
|
||||
- **只能到「场景 + 单次」粒度**。`trace_id` 由客户端在比价(comparisonTraceId)/领券(sessionTraceId)全流程保持不变。激励视频、福利页、旧客户端不带 trace_id → 该列为 NULL,历史数据无法回填,只对升级后新数据生效。
|
||||
- **收益口径**(与 `app/admin/repositories/ad_revenue.py:165-167` 完全一致):
|
||||
单条展示收益(元) = `min(parse_ecpm_yuan(ecpm_raw), AD_ECPM_MAX_FEN/100) / 1000`
|
||||
其中 `AD_ECPM_MAX_FEN = 50000`(分)= ¥500 CPM 封顶,`parse_ecpm_yuan(x) = parse_ecpm_fen(x)/100`。
|
||||
- **性能前提**:`ad_ecpm_record` 是全库写入量最大的表。必须有 `ix_ad_ecpm_record_trace_id` 索引(本计划 Task A2 建),且只对**当前页**的 trace_id 聚合,绝不对整个日期区间聚合。
|
||||
- **跨仓库**:本计划涉及三个仓库,路径前缀:
|
||||
- 后端 `e:\project\shaguabijia-app-server`(相对路径即以此为根)
|
||||
- 客户端 `E:\project\shaguabijia-app-android`
|
||||
- admin 前端 `e:\project\shaguabijia-admin-web`
|
||||
|
||||
---
|
||||
|
||||
## File Structure
|
||||
|
||||
### 后端(shaguabijia-app-server)
|
||||
- Modify `app/models/ad_ecpm.py` — `AdEcpmRecord` 加 `trace_id` 列(索引)
|
||||
- Create `alembic/versions/ad_ecpm_trace_id.py` — 加列 + 索引,并收敛当前双 head
|
||||
- Modify `app/schemas/ad.py` — `EcpmReportIn` 加 `trace_id` 字段
|
||||
- Modify `app/repositories/ad_ecpm.py` — `create_ecpm_record` 持久化 `trace_id`;新增 `revenue_yuan_by_trace` 聚合器
|
||||
- Modify `app/api/v1/ad.py` — `ecpm_report` 透传 `trace_id`
|
||||
- Modify `app/admin/schemas/coupon_data.py` — `CouponDataRow` 加 `ad_revenue_yuan`
|
||||
- Modify `app/admin/repositories/coupon_data.py` — 逐页补 `ad_revenue_yuan`
|
||||
- Modify `app/admin/schemas/comparison.py` — `AdminComparisonListItem` 加 `ad_revenue_yuan`
|
||||
- Modify `app/admin/repositories/queries.py` — `list_comparison_records` 逐页补 `ad_revenue_yuan`
|
||||
- Create `tests/test_ad_ecpm_trace_revenue.py` — 聚合器 + 落库单测
|
||||
- Create `tests/test_board_ad_revenue.py` — 两个看板收益列单测
|
||||
|
||||
> `app/models/__init__.py` **不需改**:`AdEcpmRecord` 已注册,只是加列。
|
||||
|
||||
### 客户端(shaguabijia-app-android)
|
||||
- Modify `app/src/main/java/com/jishisongfu/shaguabijia/agent/network/ApiClient.kt` — `reportAdImpression` 加 `traceId` 参数
|
||||
- Modify `app/src/main/java/com/jishisongfu/shaguabijia/agent/service/ad/CompareAdController.kt` — 比价展示上报带 `traceId`
|
||||
- Modify `app/src/main/java/com/jishisongfu/shaguabijia/service/CouponForegroundService.kt` — 领券展示上报带 `traceId`
|
||||
|
||||
### admin 前端(shaguabijia-admin-web)
|
||||
- Modify `src/lib/types.ts` — `ComparisonRecordListItem` 加 `ad_revenue_yuan`
|
||||
- Modify `src/app/(main)/comparison-records/page.tsx` — 加「广告收益」列
|
||||
- Modify `src/app/(main)/coupon-data/page.tsx` — `CouponDataRow` 加字段 + 加「广告收益」列
|
||||
|
||||
---
|
||||
|
||||
## Phase A — 后端数据打通(落 trace_id + 收益聚合器)
|
||||
|
||||
### Task A1: `AdEcpmRecord` 加 `trace_id` 列
|
||||
|
||||
**Files:**
|
||||
- Modify: `app/models/ad_ecpm.py`
|
||||
|
||||
- [ ] **Step 1: 加列**
|
||||
|
||||
在 `app/models/ad_ecpm.py` 中,找到 `feed_scene` 这一行:
|
||||
|
||||
```python
|
||||
feed_scene: Mapped[str | None] = mapped_column(String(16), nullable=True)
|
||||
```
|
||||
|
||||
在其**下方**插入:
|
||||
|
||||
```python
|
||||
# 本次比价/领券 trace_id(信息流场景客户端带上):把这条展示收益归属到对应比价/领券记录。
|
||||
# 领券数据 / 比价记录看板按 trace_id 聚合"本次广告收益"。激励视频/福利/旧客户端 = NULL。
|
||||
trace_id: Mapped[str | None] = mapped_column(String(64), index=True, nullable=True)
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 提交**
|
||||
|
||||
```bash
|
||||
git add app/models/ad_ecpm.py
|
||||
git commit -m "feat(ad-ecpm): add trace_id column to AdEcpmRecord model"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### Task A2: 迁移 — 加列 + 索引,并收敛双 head
|
||||
|
||||
**Files:**
|
||||
- Create: `alembic/versions/ad_ecpm_trace_id.py`
|
||||
|
||||
> ⚠️ 当前 `alembic heads` 有**两个 head**:`11c44afbea58`(selfstat 表)与 `merge_pages_override_coupon_slot`(#126+领券合并)。本迁移用元组 `down_revision` 把二者收敛成单 head,同时加列,让 `alembic upgrade head`(单数,run.sh 用)恢复正常。
|
||||
|
||||
- [ ] **Step 1: 确认当前 heads 未漂移**
|
||||
|
||||
Run: `alembic heads`
|
||||
Expected: 恰好两行 —
|
||||
```
|
||||
11c44afbea58 (head)
|
||||
merge_pages_override_coupon_slot (head)
|
||||
```
|
||||
若不同(他人已合并/新增),把下面 `down_revision` 改成此刻实际的 head 列表。
|
||||
|
||||
- [ ] **Step 2: 建迁移文件**
|
||||
|
||||
Create `alembic/versions/ad_ecpm_trace_id.py`:
|
||||
|
||||
```python
|
||||
"""ad_ecpm_record.trace_id(展示收益归属到比价/领券 trace)+ 收敛双 head
|
||||
|
||||
信息流(Draw)展示 eCPM 上报时带上本场比价/领券 trace_id,落此列;领券数据 / 比价记录看板
|
||||
按 trace_id 聚合"本次广告收益"。激励视频/福利/旧客户端为 NULL。
|
||||
|
||||
顺带把当前两个 head(11c44afbea58 selfstat 表 + merge_pages_override_coupon_slot)收敛成
|
||||
单 head,让 `alembic upgrade head`(单数,部署/run.sh 用)恢复正常。
|
||||
|
||||
Revision ID: ad_ecpm_trace_id
|
||||
Revises: 11c44afbea58, merge_pages_override_coupon_slot
|
||||
Create Date: 2026-07-10
|
||||
"""
|
||||
from typing import Sequence, Union
|
||||
|
||||
from alembic import op
|
||||
import sqlalchemy as sa
|
||||
|
||||
|
||||
revision: str = "ad_ecpm_trace_id"
|
||||
down_revision: Union[str, Sequence[str], None] = (
|
||||
"11c44afbea58",
|
||||
"merge_pages_override_coupon_slot",
|
||||
)
|
||||
branch_labels: Union[str, Sequence[str], None] = None
|
||||
depends_on: Union[str, Sequence[str], None] = None
|
||||
|
||||
|
||||
def upgrade() -> None:
|
||||
# SQLite 下 ADD COLUMN(可空)与 CREATE INDEX 均原生支持,无需 batch_alter_table
|
||||
# (同 ad_feed_reward_trace_id 迁移)。
|
||||
op.add_column(
|
||||
"ad_ecpm_record",
|
||||
sa.Column("trace_id", sa.String(length=64), nullable=True),
|
||||
)
|
||||
op.create_index(
|
||||
op.f("ix_ad_ecpm_record_trace_id"),
|
||||
"ad_ecpm_record",
|
||||
["trace_id"],
|
||||
unique=False,
|
||||
)
|
||||
|
||||
|
||||
def downgrade() -> None:
|
||||
op.drop_index(op.f("ix_ad_ecpm_record_trace_id"), table_name="ad_ecpm_record")
|
||||
op.drop_column("ad_ecpm_record", "trace_id")
|
||||
```
|
||||
|
||||
- [ ] **Step 3: 应用迁移**
|
||||
|
||||
Run: `alembic upgrade head`
|
||||
Expected: 无报错(不再报 "multiple heads")。
|
||||
|
||||
- [ ] **Step 4: 验证单 head + 列存在**
|
||||
|
||||
Run: `alembic heads`
|
||||
Expected: 只有一行 `ad_ecpm_trace_id (head)`。
|
||||
|
||||
Run: `python -c "from sqlalchemy import inspect; from app.db.session import engine; print([c['name'] for c in inspect(engine).get_columns('ad_ecpm_record')])"`
|
||||
Expected: 输出的列名列表包含 `trace_id`。
|
||||
|
||||
- [ ] **Step 5: 提交**
|
||||
|
||||
```bash
|
||||
git add alembic/versions/ad_ecpm_trace_id.py
|
||||
git commit -m "feat(migration): add ad_ecpm_record.trace_id + index, converge heads"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### Task A3: `EcpmReportIn` 加 `trace_id` 字段
|
||||
|
||||
**Files:**
|
||||
- Modify: `app/schemas/ad.py`
|
||||
|
||||
- [ ] **Step 1: 加字段**
|
||||
|
||||
在 `app/schemas/ad.py` 的 `EcpmReportIn` 里,找到 `feed_scene` 字段定义(以 `feed_scene: str | None = Field(` 开头的那段)。在该字段**之后**插入:
|
||||
|
||||
```python
|
||||
trace_id: str | None = Field(
|
||||
None,
|
||||
max_length=64,
|
||||
description="本次比价/领券 trace_id(信息流场景带上):把这条展示收益归属到对应比价/领券,"
|
||||
"供领券数据/比价记录看板聚合本场广告收益;激励视频/福利为空",
|
||||
)
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 提交**
|
||||
|
||||
```bash
|
||||
git add app/schemas/ad.py
|
||||
git commit -m "feat(ad-schema): EcpmReportIn accepts trace_id"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### Task A4: `create_ecpm_record` 持久化 trace_id + 新增 `revenue_yuan_by_trace`
|
||||
|
||||
**Files:**
|
||||
- Modify: `app/repositories/ad_ecpm.py`
|
||||
- Test: `tests/test_ad_ecpm_trace_revenue.py`
|
||||
|
||||
- [ ] **Step 1: 写失败测试**
|
||||
|
||||
Create `tests/test_ad_ecpm_trace_revenue.py`:
|
||||
|
||||
```python
|
||||
"""ad_ecpm_record.trace_id 落库 + 按 trace 聚合广告收益(元)。"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import UTC, datetime
|
||||
|
||||
from sqlalchemy import delete
|
||||
|
||||
from app.db.session import SessionLocal
|
||||
from app.models.ad_ecpm import AdEcpmRecord
|
||||
from app.repositories import ad_ecpm as crud_ecpm
|
||||
|
||||
|
||||
def _ecpm(trace_id: str, ecpm_raw: str, session_id: str) -> AdEcpmRecord:
|
||||
"""构造一条 Draw 展示 eCPM(不 commit;ad_session_id 全局唯一,须各不相同)。"""
|
||||
return AdEcpmRecord(
|
||||
user_id=1,
|
||||
ad_type="draw",
|
||||
feed_scene="comparison",
|
||||
ad_session_id=session_id,
|
||||
ecpm_raw=ecpm_raw,
|
||||
trace_id=trace_id,
|
||||
report_date="2020-01-02",
|
||||
created_at=datetime(2020, 1, 2, tzinfo=UTC),
|
||||
)
|
||||
|
||||
|
||||
def test_revenue_yuan_by_trace_sums_and_clamps() -> None:
|
||||
"""同一 trace 多条展示求和;收益=min(eCPM元,¥500)/1000;无展示的 trace 不出现。
|
||||
|
||||
ecpm 200 分→2.0 元/千次→0.002 元/次;300 分→0.003;合计 0.005。
|
||||
"""
|
||||
db = SessionLocal()
|
||||
try:
|
||||
db.add_all([
|
||||
_ecpm("t1", "200", "sess-t1-a"),
|
||||
_ecpm("t1", "300", "sess-t1-b"),
|
||||
_ecpm("t2", "0", "sess-t2-a"),
|
||||
])
|
||||
db.flush()
|
||||
rev = crud_ecpm.revenue_yuan_by_trace(db, ["t1", "t2", "t3"])
|
||||
assert rev["t1"] == 0.005
|
||||
assert rev.get("t2", 0.0) == 0.0
|
||||
assert "t3" not in rev # 无展示的 trace 不出现在结果里
|
||||
finally:
|
||||
db.rollback()
|
||||
db.close()
|
||||
|
||||
|
||||
def test_revenue_yuan_by_trace_empty() -> None:
|
||||
"""空 trace 列表直接返回 {}(避免 IN () 非法)。"""
|
||||
db = SessionLocal()
|
||||
try:
|
||||
assert crud_ecpm.revenue_yuan_by_trace(db, []) == {}
|
||||
finally:
|
||||
db.close()
|
||||
|
||||
|
||||
def test_create_ecpm_record_persists_trace_id() -> None:
|
||||
"""create_ecpm_record 落 trace_id。"""
|
||||
db = SessionLocal()
|
||||
try:
|
||||
rec = crud_ecpm.create_ecpm_record(
|
||||
db, 1, ad_type="draw", ecpm_raw="150",
|
||||
ad_session_id="sess-trace-persist", feed_scene="coupon",
|
||||
trace_id="trace-xyz",
|
||||
)
|
||||
assert rec.trace_id == "trace-xyz"
|
||||
finally:
|
||||
db.execute(delete(AdEcpmRecord).where(AdEcpmRecord.ad_session_id == "sess-trace-persist"))
|
||||
db.commit()
|
||||
db.close()
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 跑测试确认失败**
|
||||
|
||||
Run: `pytest tests/test_ad_ecpm_trace_revenue.py -q`
|
||||
Expected: FAIL — `create_ecpm_record` 无 `trace_id` 参数(TypeError)/ `revenue_yuan_by_trace` 不存在(AttributeError)。
|
||||
|
||||
- [ ] **Step 3: 实现**
|
||||
|
||||
在 `app/repositories/ad_ecpm.py`:
|
||||
|
||||
(a) 顶部 import 区加(与现有 `from app.core.rewards import cn_today` 并列):
|
||||
|
||||
```python
|
||||
from app.core import rewards
|
||||
```
|
||||
|
||||
(b) `create_ecpm_record` 的签名里,在 `feed_scene: str | None = None,` 之后加一行参数:
|
||||
|
||||
```python
|
||||
trace_id: str | None = None,
|
||||
```
|
||||
|
||||
(c) 同函数体内构造 `AdEcpmRecord(...)` 处,在 `feed_scene=feed_scene,` 之后加一行:
|
||||
|
||||
```python
|
||||
trace_id=trace_id,
|
||||
```
|
||||
|
||||
(d) 文件末尾新增聚合器:
|
||||
|
||||
```python
|
||||
def revenue_yuan_by_trace(db: Session, trace_ids: list[str]) -> dict[str, float]:
|
||||
"""各 trace_id 的广告预估收益(元):按 trace_id 聚合 ad_ecpm_record 的展示收益。
|
||||
|
||||
单条展示收益 = min(eCPM元, AD_ECPM_MAX_FEN/100) / 1000(与 admin 广告收益报表同口径)。
|
||||
ecpm_raw 是字符串且需逐条钳顶,故取回后 Python 求和(行数=本页各 trace 的展示条数,很小)。
|
||||
trace_id 仅信息流(比价/领券)场景客户端带,激励视频/旧数据为 NULL,按 trace_id 过滤天然只算对应场景。
|
||||
只喂**当前页**的 trace_id(≤ 一页条数);空集合直接返回(避免 IN () 非法)。
|
||||
"""
|
||||
if not trace_ids:
|
||||
return {}
|
||||
rows = db.execute(
|
||||
select(AdEcpmRecord.trace_id, AdEcpmRecord.ecpm_raw).where(
|
||||
AdEcpmRecord.trace_id.in_(trace_ids),
|
||||
)
|
||||
).all()
|
||||
cap_yuan = rewards.AD_ECPM_MAX_FEN / 100.0
|
||||
out: dict[str, float] = {}
|
||||
for tid, ecpm_raw in rows:
|
||||
if not tid:
|
||||
continue
|
||||
out[tid] = out.get(tid, 0.0) + min(rewards.parse_ecpm_yuan(ecpm_raw), cap_yuan) / 1000.0
|
||||
return {tid: round(v, 6) for tid, v in out.items()}
|
||||
```
|
||||
|
||||
- [ ] **Step 4: 跑测试确认通过**
|
||||
|
||||
Run: `pytest tests/test_ad_ecpm_trace_revenue.py -q`
|
||||
Expected: PASS(3 passed)。
|
||||
|
||||
- [ ] **Step 5: 提交**
|
||||
|
||||
```bash
|
||||
git add app/repositories/ad_ecpm.py tests/test_ad_ecpm_trace_revenue.py
|
||||
git commit -m "feat(ad-ecpm): persist trace_id + revenue_yuan_by_trace aggregator"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### Task A5: `ecpm_report` 端点透传 trace_id
|
||||
|
||||
**Files:**
|
||||
- Modify: `app/api/v1/ad.py`
|
||||
|
||||
- [ ] **Step 1: 透传字段**
|
||||
|
||||
在 `app/api/v1/ad.py` 的 `ecpm_report` 函数里,找到 `crud_ecpm.create_ecpm_record(` 调用,在 `feed_scene=payload.feed_scene,` 之后加一行:
|
||||
|
||||
```python
|
||||
trace_id=payload.trace_id,
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 冒烟验证(手动,可选)**
|
||||
|
||||
启动后端(`./run.sh`),用一个有效用户 JWT 调:
|
||||
|
||||
Run:
|
||||
```bash
|
||||
curl -s -X POST http://127.0.0.1:8770/api/v1/ad/ecpm-report \
|
||||
-H "Authorization: Bearer <USER_JWT>" -H "Content-Type: application/json" \
|
||||
-d '{"ad_type":"draw","ecpm":"200","ad_session_id":"smoke-sess-1","feed_scene":"comparison","trace_id":"smoke-trace-1"}'
|
||||
```
|
||||
Expected: `{"ok":true}`;库里 `ad_ecpm_record` 出现一条 `trace_id='smoke-trace-1'` 的记录。
|
||||
|
||||
> 端点逻辑是纯透传,已由 A4 的 schema/repo 测试覆盖;此步仅人工确认接线。
|
||||
|
||||
- [ ] **Step 3: 提交**
|
||||
|
||||
```bash
|
||||
git add app/api/v1/ad.py
|
||||
git commit -m "feat(ad-api): ecpm-report forwards trace_id to record"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Phase B — 后端两个看板补收益列
|
||||
|
||||
### Task B1: 领券数据看板逐行补 `ad_revenue_yuan`
|
||||
|
||||
**Files:**
|
||||
- Modify: `app/admin/schemas/coupon_data.py`
|
||||
- Modify: `app/admin/repositories/coupon_data.py`
|
||||
- Test: `tests/test_board_ad_revenue.py`
|
||||
|
||||
- [ ] **Step 1: 写失败测试**
|
||||
|
||||
Create `tests/test_board_ad_revenue.py`:
|
||||
|
||||
```python
|
||||
"""两个看板逐行「本次广告收益」(元):按 trace_id 聚合 ad_ecpm_record。"""
|
||||
from __future__ import annotations
|
||||
|
||||
from datetime import UTC, date, datetime
|
||||
|
||||
from sqlalchemy import delete
|
||||
|
||||
from app.admin.repositories import queries
|
||||
from app.admin.repositories.coupon_data import coupon_data_report
|
||||
from app.db.session import SessionLocal
|
||||
from app.models.ad_ecpm import AdEcpmRecord
|
||||
from app.models.comparison import ComparisonRecord
|
||||
from app.models.coupon_state import CouponSession
|
||||
|
||||
|
||||
def _ecpm(trace_id: str, ecpm_raw: str, session_id: str, scene: str) -> AdEcpmRecord:
|
||||
return AdEcpmRecord(
|
||||
user_id=1, ad_type="draw", feed_scene=scene, ad_session_id=session_id,
|
||||
ecpm_raw=ecpm_raw, trace_id=trace_id, report_date="2020-01-02",
|
||||
created_at=datetime(2020, 1, 2, tzinfo=UTC),
|
||||
)
|
||||
|
||||
|
||||
def test_coupon_data_report_includes_ad_revenue() -> None:
|
||||
"""领券看板明细行带本次广告收益;200+300 分 → 0.005 元。"""
|
||||
db = SessionLocal()
|
||||
try:
|
||||
db.add(CouponSession(
|
||||
trace_id="rev-cp-1", device_id="d1", status="completed", app_env="prod",
|
||||
platforms=["meituan-waimai"], platform_success=["meituan-waimai"],
|
||||
started_at=datetime(2020, 1, 2, tzinfo=UTC), started_date=date(2020, 1, 2),
|
||||
))
|
||||
db.add_all([
|
||||
_ecpm("rev-cp-1", "200", "cp-sess-a", "coupon"),
|
||||
_ecpm("rev-cp-1", "300", "cp-sess-b", "coupon"),
|
||||
])
|
||||
db.flush()
|
||||
res = coupon_data_report(db, date_from="2020-01-02", date_to="2020-01-02", app_env="prod")
|
||||
row = next(r for r in res["items"] if r["trace_id"] == "rev-cp-1")
|
||||
assert row["ad_revenue_yuan"] == 0.005
|
||||
finally:
|
||||
db.rollback()
|
||||
db.close()
|
||||
|
||||
|
||||
def test_comparison_list_includes_ad_revenue() -> None:
|
||||
"""比价记录列表项带本次广告收益;200 分 → 0.002 元。
|
||||
|
||||
用独有 user_id 过滤,确保本行必落在第一页(避免共享测试库里同 user 记录多、分页把它挤掉)。
|
||||
"""
|
||||
db = SessionLocal()
|
||||
try:
|
||||
db.add(ComparisonRecord(
|
||||
trace_id="rev-cmp-1", user_id=987654, status="success", business_type="food",
|
||||
created_at=datetime(2020, 1, 2, tzinfo=UTC),
|
||||
))
|
||||
db.add(_ecpm("rev-cmp-1", "200", "cmp-sess-a", "comparison"))
|
||||
db.commit()
|
||||
items, _next, _total = queries.list_comparison_records(db, user_id=987654)
|
||||
row = next(it for it in items if it.trace_id == "rev-cmp-1")
|
||||
assert row.ad_revenue_yuan == 0.002
|
||||
finally:
|
||||
db.execute(delete(AdEcpmRecord).where(AdEcpmRecord.trace_id.in_(["rev-cmp-1"])))
|
||||
db.execute(delete(ComparisonRecord).where(ComparisonRecord.trace_id == "rev-cmp-1"))
|
||||
db.commit()
|
||||
db.close()
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 跑测试确认失败**
|
||||
|
||||
Run: `pytest tests/test_board_ad_revenue.py -q`
|
||||
Expected: FAIL — `coupon_data_report` 明细行无 `ad_revenue_yuan` 键(KeyError);`ComparisonRecord` 无 `ad_revenue_yuan`(AttributeError)。
|
||||
|
||||
- [ ] **Step 3: 领券 schema 加字段**
|
||||
|
||||
在 `app/admin/schemas/coupon_data.py` 的 `CouponDataRow` 里,`trace_url` 字段**之后**加:
|
||||
|
||||
```python
|
||||
ad_revenue_yuan: float = Field(
|
||||
0.0, description="本次领券看的信息流广告预估收益(元);按 trace_id 聚合 ad_ecpm_record"
|
||||
)
|
||||
```
|
||||
|
||||
- [ ] **Step 4: 领券 repo 逐页补收益**
|
||||
|
||||
在 `app/admin/repositories/coupon_data.py`:
|
||||
|
||||
(a) import 区(现有 `from app.models.user import User` 附近)加:
|
||||
|
||||
```python
|
||||
from app.repositories import ad_ecpm as crud_ecpm
|
||||
```
|
||||
|
||||
(b) `_session_to_row` 签名改为(加末位参数):
|
||||
|
||||
```python
|
||||
def _session_to_row(r, phone: str | None = None, nickname: str | None = None, ad_revenue_yuan: float = 0.0) -> dict:
|
||||
```
|
||||
|
||||
并在其返回的 dict 里,`"trace_url": r.trace_url,` 之后加一行:
|
||||
|
||||
```python
|
||||
"ad_revenue_yuan": ad_revenue_yuan,
|
||||
```
|
||||
|
||||
(c) 在 `coupon_data_report` 里,找到构造明细的这段:
|
||||
|
||||
```python
|
||||
items = []
|
||||
for r in page:
|
||||
phone, nickname = user_map.get(r.user_id, (None, None)) if r.user_id is not None else (None, None)
|
||||
items.append(_session_to_row(r, phone, nickname))
|
||||
```
|
||||
|
||||
替换为(新增 `rev_map` + 传入):
|
||||
|
||||
```python
|
||||
rev_map = crud_ecpm.revenue_yuan_by_trace(db, [r.trace_id for r in page])
|
||||
items = []
|
||||
for r in page:
|
||||
phone, nickname = user_map.get(r.user_id, (None, None)) if r.user_id is not None else (None, None)
|
||||
items.append(_session_to_row(r, phone, nickname, ad_revenue_yuan=rev_map.get(r.trace_id, 0.0)))
|
||||
```
|
||||
|
||||
> `coupon_user_records`(手机号抽屉)仍走 `_session_to_row(r)`,`ad_revenue_yuan` 取默认 0.0——抽屉不展示收益列,无需补;字段有默认值故 schema 校验不受影响。
|
||||
|
||||
- [ ] **Step 5: 跑领券用例确认通过**
|
||||
|
||||
Run: `pytest tests/test_board_ad_revenue.py::test_coupon_data_report_includes_ad_revenue -q`
|
||||
Expected: PASS。
|
||||
|
||||
- [ ] **Step 6: 提交**
|
||||
|
||||
```bash
|
||||
git add app/admin/schemas/coupon_data.py app/admin/repositories/coupon_data.py tests/test_board_ad_revenue.py
|
||||
git commit -m "feat(admin-coupon-data): per-session ad revenue column"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### Task B2: 比价记录看板逐行补 `ad_revenue_yuan`
|
||||
|
||||
**Files:**
|
||||
- Modify: `app/admin/schemas/comparison.py`
|
||||
- Modify: `app/admin/repositories/queries.py`
|
||||
|
||||
- [ ] **Step 1: 比价 schema 加字段**
|
||||
|
||||
在 `app/admin/schemas/comparison.py` 的 `AdminComparisonListItem` 里,`created_at: datetime` **之前**加:
|
||||
|
||||
```python
|
||||
ad_revenue_yuan: float = 0.0 # 本次比价看的信息流广告预估收益(元),queries 瞬态挂 ORM 实例上
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 比价 repo 逐页补收益**
|
||||
|
||||
在 `app/admin/repositories/queries.py`:
|
||||
|
||||
(a) import 区加:
|
||||
|
||||
```python
|
||||
from app.repositories import ad_ecpm
|
||||
```
|
||||
|
||||
(b) 在 `list_comparison_records` 里,找到:
|
||||
|
||||
```python
|
||||
_attach_user_info(db, items)
|
||||
return items, next_cursor, total
|
||||
```
|
||||
|
||||
替换为:
|
||||
|
||||
```python
|
||||
_attach_user_info(db, items)
|
||||
# 「本次比价看广告的预估收益」:按本页 trace_id 一次性聚合(同 _attach_user_info 逐页范式)。
|
||||
# ad_revenue_yuan 非 ORM 列,仅瞬态挂实例上供 AdminComparisonListItem(from_attributes)读出。
|
||||
rev = ad_ecpm.revenue_yuan_by_trace(db, [it.trace_id for it in items])
|
||||
for it in items:
|
||||
it.ad_revenue_yuan = rev.get(it.trace_id, 0.0)
|
||||
return items, next_cursor, total
|
||||
```
|
||||
|
||||
- [ ] **Step 3: 跑比价用例确认通过**
|
||||
|
||||
Run: `pytest tests/test_board_ad_revenue.py::test_comparison_list_includes_ad_revenue -q`
|
||||
Expected: PASS。
|
||||
|
||||
- [ ] **Step 4: 跑全量后端测试(确认无回归)**
|
||||
|
||||
Run: `pytest -q`
|
||||
Expected: 全绿(新增用例通过,原有用例不受影响)。
|
||||
|
||||
- [ ] **Step 5: Lint**
|
||||
|
||||
Run: `ruff check app/ tests/`
|
||||
Expected: 无新增告警。
|
||||
|
||||
- [ ] **Step 6: 提交**
|
||||
|
||||
```bash
|
||||
git add app/admin/schemas/comparison.py app/admin/repositories/queries.py
|
||||
git commit -m "feat(admin-comparison): per-comparison ad revenue column"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Phase C — 客户端上报带 trace_id(Android)
|
||||
|
||||
> 三处都在 `E:\project\shaguabijia-app-android`。改动极小:eCPM 上报点的 trace_id 已在作用域内(比价是 `showAd(traceId)` 参数,领券是 `sessionTraceId` 类字段),只是当前没往上带。发奖(`reportFeedReward`)已在带 traceId,可作参照。
|
||||
|
||||
### Task C1: `ApiClient.reportAdImpression` 加 `traceId` 参数
|
||||
|
||||
**Files:**
|
||||
- Modify: `app/src/main/java/com/jishisongfu/shaguabijia/agent/network/ApiClient.kt`
|
||||
|
||||
- [ ] **Step 1: 加参数**
|
||||
|
||||
找到 `reportAdImpression` 的参数列表,末尾 `feedScene: String? = null,` 之后加一行:
|
||||
|
||||
```kotlin
|
||||
traceId: String? = null,
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 写进 payload**
|
||||
|
||||
同函数内,找到:
|
||||
|
||||
```kotlin
|
||||
if (!feedScene.isNullOrBlank()) payload.put("feed_scene", feedScene)
|
||||
```
|
||||
|
||||
在其**下方**加一行:
|
||||
|
||||
```kotlin
|
||||
if (!traceId.isNullOrBlank()) payload.put("trace_id", traceId)
|
||||
```
|
||||
|
||||
- [ ] **Step 3: 提交**
|
||||
|
||||
```bash
|
||||
git add app/src/main/java/com/jishisongfu/shaguabijia/agent/network/ApiClient.kt
|
||||
git commit -m "feat(ad-report): reportAdImpression carries trace_id"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### Task C2: 比价展示上报带 traceId
|
||||
|
||||
**Files:**
|
||||
- Modify: `app/src/main/java/com/jishisongfu/shaguabijia/agent/service/ad/CompareAdController.kt`
|
||||
|
||||
- [ ] **Step 1: 传 traceId**
|
||||
|
||||
在 `showAd(traceId: String)` 内的 `onAdImpression` 回调里,找到 `apiClient.reportAdImpression(` 调用,其中 `feedScene = "comparison",` 之后加一行(`traceId` 即 `showAd` 的入参):
|
||||
|
||||
```kotlin
|
||||
traceId = traceId, // 本场比价 trace → 展示收益归属到该次比价
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 提交**
|
||||
|
||||
```bash
|
||||
git add app/src/main/java/com/jishisongfu/shaguabijia/agent/service/ad/CompareAdController.kt
|
||||
git commit -m "feat(compare-ad): report comparison impression with trace_id"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### Task C3: 领券展示上报带 traceId
|
||||
|
||||
**Files:**
|
||||
- Modify: `app/src/main/java/com/jishisongfu/shaguabijia/service/CouponForegroundService.kt`
|
||||
|
||||
- [ ] **Step 1: 传 traceId**
|
||||
|
||||
在 `onAdImpression` 回调里,找到 `apiClient.reportAdImpression(` 调用,其中 `feedScene = "coupon",` 之后加一行(`sessionTraceId` 为本类字段,整场领券不变):
|
||||
|
||||
```kotlin
|
||||
traceId = sessionTraceId, // 本场领券 trace → 展示收益归属到该次领券
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 编译验证**
|
||||
|
||||
Run(在 `E:\project\shaguabijia-app-android`): `./gradlew :app:compileDebugKotlin`
|
||||
Expected: BUILD SUCCESSFUL。
|
||||
|
||||
- [ ] **Step 3: 提交**
|
||||
|
||||
```bash
|
||||
git add app/src/main/java/com/jishisongfu/shaguabijia/service/CouponForegroundService.kt
|
||||
git commit -m "feat(coupon-ad): report coupon impression with trace_id"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## Phase D — admin 前端加「广告收益」列
|
||||
|
||||
> 两个页面都在 `e:\project\shaguabijia-admin-web`。收益值单位是**元**(小数,单次很小如 ¥0.0050),用 `.toFixed(4)` 展示。
|
||||
|
||||
### Task D1: 比价记录页加列
|
||||
|
||||
**Files:**
|
||||
- Modify: `src/lib/types.ts`
|
||||
- Modify: `src/app/(main)/comparison-records/page.tsx`
|
||||
|
||||
- [ ] **Step 1: 类型加字段**
|
||||
|
||||
在 `src/lib/types.ts` 的 `ComparisonRecordListItem` 接口里,`created_at: string;` **之前**加:
|
||||
|
||||
```typescript
|
||||
ad_revenue_yuan: number; // 本次比价看的信息流广告预估收益(元)
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 表格加列**
|
||||
|
||||
在 `src/app/(main)/comparison-records/page.tsx` 的 `columns` 数组里,找到「省」这一列(以 `title: '省',` 开头的对象),在其**之后**插入一列:
|
||||
|
||||
```tsx
|
||||
{
|
||||
title: '广告收益',
|
||||
key: 'ad_revenue',
|
||||
width: 96,
|
||||
align: 'right',
|
||||
render: (_, r) =>
|
||||
r.ad_revenue_yuan > 0 ? (
|
||||
<span style={{ color: '#3f8600' }}>¥{r.ad_revenue_yuan.toFixed(4)}</span>
|
||||
) : (
|
||||
<span style={{ color: '#ccc' }}>-</span>
|
||||
),
|
||||
},
|
||||
```
|
||||
|
||||
- [ ] **Step 3: 加宽横向滚动**
|
||||
|
||||
同文件找到比价记录主表的 `scroll={{ x: 1820 }}`,改为:
|
||||
|
||||
```tsx
|
||||
scroll={{ x: 1920 }}
|
||||
```
|
||||
|
||||
- [ ] **Step 4: 提交**
|
||||
|
||||
```bash
|
||||
git add src/lib/types.ts "src/app/(main)/comparison-records/page.tsx"
|
||||
git commit -m "feat(admin-web): ad revenue column in comparison records table"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
### Task D2: 领券数据页加列
|
||||
|
||||
**Files:**
|
||||
- Modify: `src/app/(main)/coupon-data/page.tsx`
|
||||
|
||||
- [ ] **Step 1: 接口加字段**
|
||||
|
||||
在 `src/app/(main)/coupon-data/page.tsx` 的 `interface CouponDataRow` 里,`trace_url: string | null;` **之后**加:
|
||||
|
||||
```typescript
|
||||
ad_revenue_yuan: number; // 本次领券看的信息流广告预估收益(元)
|
||||
```
|
||||
|
||||
- [ ] **Step 2: 加金额格式化辅助**
|
||||
|
||||
在 `fmtPct` 定义(以 `const fmtPct =` 开头)之后加:
|
||||
|
||||
```typescript
|
||||
// 元(小数)→ "¥0.0050"(空/≤0 显示 -)。单次广告收益很小,保留 4 位。
|
||||
const fmtYuan = (v: number | null | undefined): string =>
|
||||
v == null || v <= 0 ? '-' : `¥${v.toFixed(4)}`;
|
||||
```
|
||||
|
||||
- [ ] **Step 3: 主表加列**
|
||||
|
||||
找到主表 `columns`(`CouponDataPage` 组件内的 `const columns: ColumnsType<CouponDataRow> = [`)。在「耗时」列(`dataIndex: 'elapsed_ms'` 的对象)**之后**插入:
|
||||
|
||||
```tsx
|
||||
{
|
||||
title: '广告收益',
|
||||
dataIndex: 'ad_revenue_yuan',
|
||||
width: 100,
|
||||
align: 'right',
|
||||
render: (v: number) => fmtYuan(v),
|
||||
},
|
||||
```
|
||||
|
||||
- [ ] **Step 4: 加宽横向滚动**
|
||||
|
||||
找到主表 `scroll={{ x: 1450 }}`,改为:
|
||||
|
||||
```tsx
|
||||
scroll={{ x: 1560 }}
|
||||
```
|
||||
|
||||
- [ ] **Step 5: 前端类型检查 / 构建**
|
||||
|
||||
Run(在 `e:\project\shaguabijia-admin-web`): `npm run build`
|
||||
Expected: 构建成功、无 TS 类型错误。
|
||||
|
||||
- [ ] **Step 6: 提交**
|
||||
|
||||
```bash
|
||||
git add "src/app/(main)/coupon-data/page.tsx"
|
||||
git commit -m "feat(admin-web): ad revenue column in coupon data table"
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 端到端验证(全部任务完成后)
|
||||
|
||||
- [ ] **后端**:`pytest -q` 全绿;`alembic heads` 只有 `ad_ecpm_trace_id` 单 head。
|
||||
- [ ] **客户端**:装 debug 包,跑一次比价 + 一次领券(要有信息流广告展示);后端库 `ad_ecpm_record` 出现带 `trace_id`、`feed_scene in (comparison, coupon)` 的记录。
|
||||
- [ ] **看板**:admin 打开「比价记录」「领券数据」两页,新「广告收益」列对刚才那两次显示 > ¥0 的金额;旧数据(无 trace_id 上报)显示 `-`。
|
||||
- [ ] **口径核对**:任取一行,手动核对 `ad_ecpm_record` 中该 `trace_id` 的各条 `ecpm_raw`,按 `Σ min(ecpm/100, 500)/1000` 算出的值与页面一致。
|
||||
|
||||
---
|
||||
|
||||
## 回滚
|
||||
|
||||
- 前端/客户端:回退对应 commit 即可(纯展示/上报,无副作用)。
|
||||
- 后端:`alembic downgrade -1` 撤 `ad_ecpm_trace_id`(会拆回两个 head——与本计划实施前状态一致);看板收益列在无 trace_id 列时会因查询报错,故 downgrade 迁移前需先回退 Phase A/B 的 commit。正常情况不需回滚。
|
||||
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,296 @@
|
||||
# 15 天不活跃自动清零（金币 + 现金）设计
|
||||
|
||||
- **日期**：2026-07-16
|
||||
- **状态**：Draft — 待评审
|
||||
- **所属**：app-server（`app/`），含一处 admin 侧重构 + 一项 Android 端埋点依赖
|
||||
- **一句话**：连续 15 天不活跃的用户，自动清零其金币与现金；清零前按可配置节奏预警；全过程留审计以备纠纷排查。
|
||||
|
||||
---
|
||||
|
||||
## 1. 背景与目标
|
||||
|
||||
运营需要对**长期不活跃**用户的钱包余额做清理。两条硬性要求：
|
||||
|
||||
1. **可审计**：记录清零原因与**清零前的三桶余额**，便于后续排查与处理客户纠纷。
|
||||
2. **临清预警**：在临近清零前推送信息告知用户"因账号不活跃，账户里的 xx 金币和 xx 现金将被清零"。
|
||||
|
||||
### 非目标（本期不做）
|
||||
|
||||
- 不做真实推送通道（极光 JPush / 短信）的对接 —— 仅做**可插拔通知器 + 日志占位**，接口预留、后续无缝替换。
|
||||
- 不改动提现（`WithdrawOrder`）流程。
|
||||
- 不新增 `User.last_active_at` 列、不改鉴权热路径。
|
||||
|
||||
---
|
||||
|
||||
## 2. 需求
|
||||
|
||||
| # | 需求 | 落地 |
|
||||
|---|---|---|
|
||||
| R1 | 连续 15 天不活跃 → 清零金币 + 现金 | 每日 worker 扫描 + 逐用户事务清零（§6） |
|
||||
| R2 | 记录清零原因 + 清零前余额 | `inactivity_reset_log` 审计表 + 3 条钱包流水（§5、§7） |
|
||||
| R3 | 临清前预警"xx 金币 xx 现金将清零" | 阶段 A 预警 + `inactivity_notification_log`（§6、§7） |
|
||||
| R4 | 活跃口径与"用户管理"一致 | 抽共享模块 `activity.py`，admin 与 worker 共用（§4、§12） |
|
||||
| R5 | 预警时机完全可配置 | `INACTIVITY_*` 配置项（§8） |
|
||||
|
||||
---
|
||||
|
||||
## 3. 决策记录（来自评审问答）
|
||||
|
||||
| 决策点 | 结论 | 理由 |
|
||||
|---|---|---|
|
||||
| **活跃口径** | 与"用户管理"一致：`max(首页可见 show/home, 比价, 领券)`，**不含 last_login_at**；无任何信号时以 `created_at` 为非空基线 | 比价可从**浮窗**触发、不进首页；`last_login_at` 只在登录/换绑动作更新（re-login 也算），代表不了"在用 App"，故彻底排除 |
|
||||
| **"进首页"信号落地** | **方案 A：前端上报 `home_view` 埋点**（复用 `/analytics/events`），非新接口 | 三个活跃信号统一为同类埋点事件；零新接口零新列；与 admin 口径天然一致。B（鉴权接口 + 列）"更权威"的优势是假的——比价/领券仍是端上报事件，最弱环决定整体可信度 |
|
||||
| **清零范围** | **金币 + 折算现金**（**邀请现金不清**——产品红线，仅快照入审计） | 对应"账户里的金币和现金"；邀请奖励金与金币现金物理隔离、不可累加，见 `wallet.CoinAccount` 注释 |
|
||||
| **预警推送** | **可插拔通知器 + 日志占位**（v1），后续接 JPush/短信 | 现状无真实推送能力；先把清零主流程 + 审计做扎实，不阻塞 |
|
||||
| **预警时机** | **完全可配置**（提前天数列表 + 次数 + 执行点 + 通道） | R5 |
|
||||
| **触发方式** | **进程内每日 worker**，仿 `daily_exchange_worker` | 与项目最新模式一致，无需外部 cron |
|
||||
| **admin 共享口径** | 共享模块 + **重构 admin 改用它** | 单一真源，永不漂移（R4） |
|
||||
|
||||
### 已知取舍（可接受）
|
||||
|
||||
- analytics 的 `user_id` 是**端上报、未鉴权**（可伪造）。但伪造只能"保自己活跃、避免被清"，无收益，且正是本功能要防的行为，风险良性。活跃时间的非空基线由服务端权威的 `User.created_at` 提供（见 §4），不再依赖 `last_login_at`。与"用户管理"口径一致。
|
||||
|
||||
---
|
||||
|
||||
## 4. 活跃口径与共享模块 `app/repositories/activity.py`（新建）
|
||||
|
||||
活跃口径的**唯一真源**。app 侧模块，admin 可 import（`app.main` 不 import `app.admin`，反向允许）。
|
||||
|
||||
### 口径
|
||||
|
||||
```
|
||||
last_active = max(
|
||||
User.created_at, # 注册基线(恒非空;re-login 不推进,只有真实使用才推进)
|
||||
max AnalyticsEvent.created_at WHERE event IN ACTIVE_EVENTS,
|
||||
max CouponPromptEngagement.created_at WHERE engage_type == "claim_started",
|
||||
)
|
||||
不活跃判定：按北京自然日、0 点对齐（非从末次活跃时刻滚动 15×24h）
|
||||
last_active_date = 北京(last_active).date() # 末次活跃的北京日，记为「第 1 日」
|
||||
清零边界 = 北京 00:00 of (last_active_date + RESET_DAYS 天) =「第 (RESET_DAYS+1) 日 0 点」 # 15 → 第16日0点
|
||||
应清零 ⟺ (cn_today() − last_active_date).days ≥ RESET_DAYS
|
||||
⟺ last_active < cutoff， cutoff = 北京 00:00 of (cn_today() − (RESET_DAYS−1)) # 供 SQL 比较
|
||||
inactive_days = (cn_today() − last_active_date).days # 清零当日恰 = RESET_DAYS
|
||||
例：末次活跃 1/1 → 1/16 00:00(第16日0点)清零，当日 inactive_days=15；1/15 及之前不清
|
||||
```
|
||||
|
||||
### 模块内容
|
||||
|
||||
- 常量：
|
||||
- **首页可见活跃信号已定名：`event=show` + `page=home`**（前端确认，原占位 `home_view`；下文出现的 `home_view` 均指此信号）。活跃行为过滤见 `activity.active_event_condition()`：首页可见 ∪ 比价 `real_compare_start` ∪ 领券 `real_coupon_start`；`ACTIVE_EVENTS` 仅含后两个纯 event 名（首页可见是 event+page 组合、单列）。
|
||||
- `ACTIVE_ENGAGE_TYPE = "claim_started"`
|
||||
- `last_active_subqueries(db)` —— 复刻现 admin `queries._last_active_parts()`：两个按 `user_id` 的 `GROUP BY max(created_at)` 聚合子查询。
|
||||
- `last_active_expr(base_col, ev_sub, eng_sub, dialect)` —— 生成 `greatest`/`max`（PG `func.greatest`／SQLite `func.max`）；子聚合缺失时 `coalesce(子聚合, User.created_at)` 兜底（注册基线恒非空，**替代原 last_login_at**）。
|
||||
- `_norm_utc()` —— 沿用现 admin 的 naive→UTC 归一（SQLite naive / PG aware 混算保护）。
|
||||
- `reset_cutoff(reset_days)` / `warn_cutoff(reset_days, k)` —— 生成**北京 0 点对齐**的边界 datetime（见口径）：`reset_cutoff = 北京 00:00 of (cn_today() − (reset_days−1))`，供下面查询按 `last_active < cutoff` 比较。
|
||||
- `select_inactive_users(db, *, cutoff, with_balance=True)` —— **worker 专用**：join `CoinAccount`，筛 `last_active < cutoff`（cutoff = 北京 0 点对齐边界，见口径）且（`coin_balance>0 OR cash_balance_cents>0`；**邀请现金不清、不计入候选**），返回 `(user, account, last_active, inactive_days)`。
|
||||
- `select_warn_targets(db, *, reset_days, warn_days_before)` —— **worker 专用**：返回 `(user, account, last_active, inactive_days, stage)` 元组——各"提前天数"窗口内、有余额、本 streak 未推过档 `stage` 的用户（去重结合 `notification_log`，逻辑见 §9）。
|
||||
|
||||
> **参考现状**：现口径散落在 `app/admin/repositories/queries.py:38,91-124,199-204`（`_ACTIVE_EVENTS`/`_last_active_parts`/`greatest`）与 `app/admin/repositories/stats.py:51-52,138-146`（`COMPARE_START_EVENT`/`COUPON_START_EVENT`/活跃用户集）。这些改为从 `activity.py` 导入（§12）。
|
||||
|
||||
---
|
||||
|
||||
## 5. 数据模型（2 张新表，不动 `User`）
|
||||
|
||||
两表均登记进 `app/models/__init__.py`；一个 Alembic 迁移建两表（`render_as_batch`，SQLite 兼容）。
|
||||
|
||||
### ① `inactivity_reset_log` —— 清零审计（R2）
|
||||
|
||||
仿 `app/models/phone_rebind_log.py` 的简单审计表风格。
|
||||
|
||||
| 字段 | 类型 | 说明 |
|
||||
|---|---|---|
|
||||
| `id` | int PK autoincrement | |
|
||||
| `user_id` | int, index, not null | |
|
||||
| `coin_balance_before` | int, not null | 清零前金币 |
|
||||
| `cash_balance_cents_before` | int, not null | 清零前折算现金（分） |
|
||||
| `invite_cash_balance_cents_before` | int, not null | 清零前邀请现金（分） |
|
||||
| `last_active_at` | DateTime(tz), nullable | 判定时的最近活跃时间 |
|
||||
| `inactive_days` | int, not null | 判定时不活跃天数 |
|
||||
| `reason` | String(32), not null | 如 `"inactive_15d"` |
|
||||
| `reset_at` | DateTime(tz), server_default now(), index, not null | 清零时刻 |
|
||||
|
||||
### ② `inactivity_notification_log` —— 预警记录 + 去重 + 占位 outbox（R3）
|
||||
|
||||
| 字段 | 类型 | 说明 |
|
||||
|---|---|---|
|
||||
| `id` | int PK autoincrement | |
|
||||
| `user_id` | int, index, not null | |
|
||||
| `stage` | int, not null | 提前天数档（如 7 / 2） |
|
||||
| `inactive_days` | int, not null | 推送时不活跃天数 |
|
||||
| `coin_balance` | int, not null | 推送快照：告知用户的金币数 |
|
||||
| `cash_balance_cents` | int, not null | 推送快照：折算现金 |
|
||||
| `invite_cash_balance_cents` | int, not null | 推送快照：邀请现金 |
|
||||
| `channel` | String(16), not null | `"log"` / `"jpush"` / `"sms"` |
|
||||
| `status` | String(16), not null | `"placeholder"` / `"sent"` / `"failed"` |
|
||||
| `created_at` | DateTime(tz), server_default now(), index, not null | 去重锚点（见 §9） |
|
||||
|
||||
> 备注：不新增 `User.last_active_at` 列，不改 `get_current_user`。活跃时间由 §4 口径**实时计算**。
|
||||
|
||||
---
|
||||
|
||||
## 6. 清零 worker `app/core/inactivity_reset_worker.py`（新建）
|
||||
|
||||
**完全仿 [`app/core/daily_exchange_worker.py`](../../../app/core/daily_exchange_worker.py)**：App 启动自带 asyncio 任务，文件锁（`data/inactivity_reset.lock`）防同机多进程并发。**worker 常驻**；`INACTIVITY_RESET_ENABLED` 只决定是否**真清**：false（默认）= 只记审计名单、不动钱（dry-run），true = 真清。
|
||||
|
||||
### 调度
|
||||
|
||||
- 每 `INACTIVITY_RESET_CHECK_INTERVAL_SEC` 秒醒一次；`last_run: date` 守卫**北京日**，保证每日只跑一轮。
|
||||
- 仅当 `cn_today() != last_run` 且当前北京小时 `>= INACTIVITY_RESET_RUN_HOUR` 时执行（启动补跑同 daily_exchange 语义）。
|
||||
- **清零资格边界 = 第 16 日 0 点（北京，见 §4），与 worker 执行点解耦**：worker 于当日 `RUN_HOUR`（默认 3 点）跑，把已过边界者一并清；若要严格 0 点触发可置 `RUN_HOUR=0`，但注意与 `daily_auto_exchange` 的 0 点任务错峰。
|
||||
- lifespan 里 `start_inactivity_reset_worker()` / `stop_...`（仿 `start_daily_exchange_worker` 在 `app/main.py` 的接线）。
|
||||
|
||||
### 一轮 `run_once(db)` 两阶段（同一次运行、各自逐用户独立 commit）
|
||||
|
||||
**阶段 A — 预警**
|
||||
```
|
||||
for user, acc, last_active, inactive_days, stage in activity.select_warn_targets(...):
|
||||
notifier.send_inactivity_warning(user, balances=snapshot(acc), stage=stage, days_until_reset=RESET_DAYS-inactive_days)
|
||||
db.add(InactivityNotificationLog(..., channel=notifier.channel, status=notifier.last_status))
|
||||
db.commit() # 逐条独立
|
||||
```
|
||||
|
||||
**阶段 B — 清零**（`biz_type="inactivity_reset"`）
|
||||
```
|
||||
for user, acc, last_active, inactive_days in activity.select_inactive_users(db, cutoff=activity.reset_cutoff(RESET_DAYS)): # 北京 00:00 of (今天−(RESET_DAYS−1))
|
||||
try:
|
||||
acc = wallet.get_or_create_account(db, user.id, commit=False, lock=True) # 行锁
|
||||
before = (acc.coin_balance, acc.cash_balance_cents, acc.invite_cash_balance_cents)
|
||||
if acc.coin_balance == 0 and acc.cash_balance_cents == 0: continue # 邀请现金不清,不算可清余额
|
||||
log = InactivityResetLog(user_id=user.id, coin_balance_before=before[0],
|
||||
cash_balance_cents_before=before[1], invite_cash_balance_cents_before=before[2], # 邀请现金仅快照
|
||||
last_active_at=last_active, inactive_days=inactive_days, reason=f"inactive_{RESET_DAYS}d")
|
||||
db.add(log); db.flush() # 拿 log.id 作 ref_id 交叉链接
|
||||
if acc.coin_balance: wallet.grant_coins(db, user.id, -acc.coin_balance, biz_type="inactivity_reset", ref_id=str(log.id), remark="15天不活跃清零")
|
||||
if acc.cash_balance_cents: wallet.grant_cash(db, user.id, -acc.cash_balance_cents, biz_type="inactivity_reset", ref_id=str(log.id), remark="15天不活跃清零")
|
||||
# 邀请现金(invite_cash_balance_cents)不清:产品红线、两本账物理隔离,仅快照记入审计。
|
||||
db.commit()
|
||||
except SQLAlchemyError:
|
||||
db.rollback(); stats["failed"] += 1
|
||||
```
|
||||
|
||||
- `grant_*` 负数出账、`balance_after=0`、写**两条**流水（金币 + 折算现金；**邀请现金不清**）；`grant_coins` 负数**不**动 `total_coin_earned`（历史累计保留）。
|
||||
- 逐用户独立 commit：一个失败不影响其余。返回 `stats = {warned, warn_skipped, warn_failed, scanned, cleared, failed}` 并 `logger.info`。**预警逐用户 try/except 隔离、且预警整段异常也绝不阻塞清零**（清零是不可逆资金操作，不能被通知故障拖住）。
|
||||
|
||||
---
|
||||
|
||||
## 7. 预警与可插拔通知器
|
||||
|
||||
`app/integrations/notifier.py` 定义协议（外部投递属 integrations 层）：
|
||||
|
||||
```python
|
||||
class InactivityNotifier(Protocol):
|
||||
channel: str # "log" / "jpush" / "sms"
|
||||
last_status: str # "placeholder" / "sent" / "failed"
|
||||
def send_inactivity_warning(self, user, *, balances, stage, days_until_reset) -> None: ...
|
||||
```
|
||||
|
||||
- **v1 `LogNotifier`**（`channel="log"`）：`logger.warning("[inactivity-warn] user=%s coin=%s cash=%s invite=%s T-%s", ...)`，`last_status="placeholder"`。参照 `heartbeat_monitor_worker` 先例（"本期先不接推送，用终端打印代替"）。
|
||||
- 未来 `JPushNotifier` / `SmsNotifier`：实现同协议即可替换，worker 不改。
|
||||
- 选择：`INACTIVITY_NOTIFY_CHANNEL` → 工厂返回对应实现（未配到真实实现时回退 `LogNotifier`）。
|
||||
- 预警文案数据来自快照 `balances`，满足 R3"告知 xx 金币 xx 现金"。
|
||||
|
||||
---
|
||||
|
||||
## 8. 配置项（`app/core/config.py`）
|
||||
|
||||
```
|
||||
INACTIVITY_RESET_ENABLED = False # false(默认)=只记审计名单(dry-run,不动钱);true=真清
|
||||
INACTIVITY_RESET_DAYS = 15 # 不活跃阈值（天）
|
||||
INACTIVITY_WARN_DAYS_BEFORE = "7,2" # 清零前几天各推一次；空串=不推。逗号分隔，降序解析
|
||||
INACTIVITY_RESET_RUN_HOUR = 3 # 北京时间每日执行点（0-23）
|
||||
INACTIVITY_NOTIFY_CHANNEL = "log" # log(占位) / jpush / sms
|
||||
INACTIVITY_RESET_CHECK_INTERVAL_SEC = 1800 # worker 唤醒间隔（可复用现有间隔常量）
|
||||
```
|
||||
|
||||
- 清零范围（三桶）固定为常量，不做配置。
|
||||
- `INACTIVITY_WARN_DAYS_BEFORE` 语义（`inactive_days` 为北京自然日，见 §4）：档位 `k` ⟹ 当 `inactive_days >= RESET_DAYS-k` 且 `< RESET_DAYS` 且本 streak 未推过档 `k` 时预警，即在北京日 `last_active_date + (RESET_DAYS−k)` 触发（漏跑某天时补发最紧急未推档，§9）。
|
||||
- `INACTIVITY_RESET_RUN_HOUR` 只决定 worker 每日执行点，**不改变**"第 16 日 0 点"这一资格边界（§4/§6）。
|
||||
|
||||
---
|
||||
|
||||
## 9. 幂等与重新活跃
|
||||
|
||||
- **重新活跃自动退出**：`inactive_days` 由 §4 口径**实时算**。用户一有 `home_view`/比价/领券（**登录本身不算**），`last_active` 前移，自动移出预警与清零队列。**无需**显式"重置标记"。
|
||||
- **预警去重**：`inactivity_notification_log` 中存在 `stage==k 且 created_at > last_active` 的行 ⟹ 本 streak 已推过档 `k`，不重推。用户回归后 `last_active` 前移，旧预警行自然"失效"，开启新 streak。
|
||||
- **清零幂等**：阶段 B 只处理三桶非全 0 者；清完 = 0，次日不再匹配。worker 重启 / 多次唤醒 / 补跑均安全，不产生重复清零或重复流水。
|
||||
- **稳健补发**：worker 漏跑数天后，某用户可能同时满足多档；只补发**最紧急的未推档**（最小 `k`），避免一次刷屏。
|
||||
|
||||
---
|
||||
|
||||
## 10. 边界与安全
|
||||
|
||||
| 场景 | 处理 |
|
||||
|---|---|
|
||||
| 新用户 | `created_at` 作活跃基线（恒非空）→ 注册即"第 1 日活跃"；注册后连续 15 天无 home_view/比价/领券 才清 |
|
||||
| 在途提现 | 提现申请时现金已扣入 `WithdrawOrder`，当前余额已不含在途；只清当前余额、不动提现单。提现失败退款到已清账户 = 用户的钱，正常 |
|
||||
| 与 `daily_auto_exchange` 并存 | 各自逐用户幂等；金币多已日结折现金，三桶全清正好覆盖 |
|
||||
| 时区/日界 | 统一北京（`rewards.cn_today()`/`CN_TZ`）；**清零/预警按北京自然日 0 点对齐**（末次活跃记为第 1 日 → 第 16 日 0 点清零，见 §4），非滚动 24h；流水 `created_at` 沿用北京 wall-clock naive |
|
||||
| 误清防护 | worker 常驻默认 **dry-run**（`ENABLED=false` 只记审计名单、不动钱）；看准名单再置 `true` 真清（§13） |
|
||||
|
||||
---
|
||||
|
||||
## 11. 前端依赖：`home_view` 埋点（跨仓 — Android）
|
||||
|
||||
- **Android 端**（`shaguabijia-app-android`）需在**首页可见**（`onResume`/Tab 切入）时，向现有 `POST /api/v1/analytics/events` 批量上报里加一条 `event=<首页可见事件名>`（名称明天加埋点时定，暂记 `"home_view"`） 的事件，**携带登录后的 `user_id`**。
|
||||
- 客户端按会话/前台去重即可（服务端只取 `max(created_at)`，多报无害）。
|
||||
- **上线顺序依赖**：`home_view` 全量覆盖前，"进首页"信号缺失，只有比价/领券能推进活跃、其余落到 `created_at` 基线（"只开首页不操作"且注册满 15 天的用户会被误清）—— 故**开真清（`ENABLED=true`）必须待 `home_view` 铺满后再开**（§13）；dry-run 只记名单不动钱、可先开着看。
|
||||
|
||||
---
|
||||
|
||||
## 12. admin 重构范围与影响（R4）
|
||||
|
||||
- `app/admin/repositories/queries.py`：删本地 `_ACTIVE_EVENTS`/`_last_active_parts()`，改用 `activity.py` 的常量与子查询构造；`list_users` 的 `greatest(...)` 排序/筛选、`_attach_last_active` 均改走共享构造器。
|
||||
- `app/admin/repositories/stats.py`：`COMPARE_START_EVENT`/`COUPON_START_EVENT`/活跃用户集（`:138-146`）改用共享常量与口径。
|
||||
- **行为变化（预期内、需产品知会）**：admin 的"最近活跃 / DAU"口径变化——**移除 `last_login_at`（登录不再计为活跃）、以 `created_at` 为基线、纳入 `home_view`**。net：`home_view` 铺满后更准（真正把"开首页"算进活跃）；铺满前"只登录不操作"的用户活跃度会下降。
|
||||
- **回归底线**：现有 admin 用户列表 / stats 测试按新口径**更新预期**（last_login_at 移除 + created_at 基线 + home_view 纳入）；非活跃口径部分行为不变。
|
||||
|
||||
---
|
||||
|
||||
## 13. 灰度与上线顺序（安全优先）
|
||||
|
||||
1. **后端先行**：合入共享模块 + 两表 + worker + 通知器，`INACTIVITY_RESET_ENABLED=False`；活跃口径以 `created_at` 为非空基线、**不含 last_login_at**。
|
||||
2. **Android 发版**：上报 `home_view`；观察 analytics 覆盖率。
|
||||
3. **dry-run 灰度（默认即是）**：`INACTIVITY_RESET_ENABLED=False` 时 worker 常驻只写审计名单（`reason=inactive_Nd_dryrun`）、不动钱、不预警；核对名单准确。
|
||||
4. **开真清**：确认无误后置 `INACTIVITY_RESET_ENABLED=True`（转为真清 + 预警）。
|
||||
5. **收尾/监控**：持续观察 `home_view` 覆盖率与预警/清零名单；发现"活跃却被判不活跃"的漏报即回查埋点覆盖（口径已不含 last_login_at，登录不再兜底）。
|
||||
|
||||
---
|
||||
|
||||
## 14. 测试计划
|
||||
|
||||
- **活跃口径（共享模块）**：`home_view`/比价/领券 各单独命中都算活跃；**纯登录不算**；无信号用户以 `created_at` 计；`max` 取最新；naive/aware 混算不崩。
|
||||
- **admin 回归**：用户列表 / stats 按新口径更新预期（移除 last_login_at + created_at 基线 + home_view）。
|
||||
- **不活跃判定**：`last_active` 分别 `<15d / =15d / >15d` × 有/无余额 的命中矩阵。
|
||||
- **清零**：三桶归零；`inactivity_reset_log` 清前值正确；三条流水 `biz_type=inactivity_reset`、`balance_after=0`、`ref_id=log.id`；`total_coin_earned` 不变。
|
||||
- **预警**：命中窗口调 notifier + 写 `notification_log`；同 streak 不重推；回归后 `last_active` 前移可再次预警；漏跑补发最紧急档。
|
||||
- **worker**：常驻；`ENABLED=false` 走 dry-run（只记审计名单、不清、不预警）；文件锁互斥；逐用户失败隔离（一个抛错不影响其余，`failed` 计数）；重复跑幂等。
|
||||
- **配置**：`INACTIVITY_WARN_DAYS_BEFORE` 解析（含空串=不推）；`RESET_DAYS`/`RUN_HOUR` 生效。
|
||||
- 沿用 `tests/conftest.py`（临时 SQLite、`RATE_LIMIT_ENABLED=false`）；外部通知 monkeypatch。
|
||||
|
||||
---
|
||||
|
||||
## 15. 未来工作
|
||||
|
||||
- 接真实 `JPushNotifier`（需用户级 `registration_id` 覆盖 + JPush push API）/ `SmsNotifier`。
|
||||
- 如需 admin 后台可视化：不活跃/预警/清零名单与历史查询接口。
|
||||
- 如量级增长导致每日 join 扫描变慢：再考虑物化 `last_active_at`（当前每日一次可接受）。
|
||||
|
||||
---
|
||||
|
||||
## 附：涉及文件清单
|
||||
|
||||
**新增**
|
||||
- `app/repositories/activity.py` — 活跃口径唯一真源
|
||||
- `app/models/inactivity_reset_log.py` — 审计表
|
||||
- `app/models/inactivity_notification_log.py` — 预警/占位表
|
||||
- `app/core/inactivity_reset_worker.py` — 每日 worker（仿 daily_exchange_worker）
|
||||
- `app/integrations/notifier.py` — 通知器协议 + `LogNotifier`（真实 JPush/短信后续同层扩展）
|
||||
- `alembic/versions/<...>_add_inactivity_tables.py` — 建两表迁移
|
||||
- `docs/database/inactivity_reset_log.md` / `inactivity_notification_log.md` — 表字典（随实现补）
|
||||
- 对应 `tests/test_inactivity_reset.py`
|
||||
|
||||
**改动**
|
||||
- `app/models/__init__.py` — 注册两模型
|
||||
- `app/core/config.py` — `INACTIVITY_*` 配置
|
||||
- `app/main.py` — lifespan 接线 start/stop worker
|
||||
- `app/admin/repositories/queries.py`、`stats.py` — 改用 `activity.py`（§12）
|
||||
Reference in New Issue
Block a user