feat(meituan-cps): 经纬度→城市离线反查 + rec/销量最高按城市过滤 (#116)

## 主要功能
新增离线「经纬度 → 美团 cityId」反查,让 `rec`(智能推荐)与 `top-sales`(销量最高)从离线库只返**同城券**(此前会混返异地券)。

- `app/utils/geo.py` + `app/utils/meituan_city.py`:坐标 → 美团 `city_id`(reverse_geocoder 离线反查,零网络)。
- `feed?tab=rec` / `/top-sales`:按 `city_id` 过滤;解析不出 / 老客户端不带坐标 → 降级返空。
- `top-sales` 与 `rec` 一致置空库内距离(相对城市默认点、对用户无意义)。

---------

Co-authored-by: guke <guke@autohome.com.cn>
Reviewed-on: #116
Co-authored-by: guke <guke@wonderable.ai>
Co-committed-by: guke <guke@wonderable.ai>
This commit was merged in pull request #116.
This commit is contained in:
2026-07-05 09:31:53 +08:00
committed by marco
parent 3eb439b5ae
commit 7b6756f936
13 changed files with 1501 additions and 5 deletions
+170
View File
@@ -0,0 +1,170 @@
"""把 meituan_coupon 的线上采样 TSV 灌进本地 SQLite。
用途:本地开发/调试时,把线上 `meituan_coupon` 表的采样数据(tests/meituan_coupon_data.tsv)
灌进 dev 库(默认 `./data/app.db`),免得每次都实时打美团接口。
TSV 说明:
- 制表符分隔,每行一条记录,列顺序与线上 PostgreSQL 物理列一致
(image_size / image_type 是后加的迁移,排在最后两列 —— 与本地 SQLite 一致)。
- 空字段 = NULL(文件里没有 `\\N` 标记)。
- 个别记录的文本/JSON 字段内含换行,会把一条逻辑行拆成多物理行 —— 按“累计到 26 列”重组。
- 文件尾部可能有一条被导出截断的残行(列数不足 / raw JSON 不完整),直接跳过。
datetime 三列(first_seen/last_seen/updated_at)去掉尾部时区偏移(`+08`),
存成 SQLAlchemy 在 SQLite 上用的朴素格式 `YYYY-MM-DD HH:MM:SS.ffffff`,保证 ORM 能读回。
用法:
python scripts/load_meituan_coupon_tsv.py # 默认 TSV + .env 里的库
python scripts/load_meituan_coupon_tsv.py path/to.tsv # 指定 TSV
DATABASE_URL=sqlite:///./data/app.db python scripts/load_meituan_coupon_tsv.py
"""
from __future__ import annotations
import json
import os
import re
import sqlite3
import sys
from pathlib import Path
_PROJECT_ROOT = Path(__file__).resolve().parent.parent
# 列顺序 = TSV 字段顺序 = 本地 SQLite 物理列顺序
COLS = [
"id", "source", "platform", "biz_line", "city_id", "product_view_sign",
"sku_view_id", "name", "brand_name", "sell_price_cents", "original_price_cents",
"head_url", "sale_volume", "sale_volume_num", "commission_percent",
"commission_amount_cents", "poi_name", "available_poi_num", "delivery_distance_m",
"dedup_key", "raw", "first_seen", "last_seen", "updated_at", "image_size", "image_type",
]
NCOL = len(COLS)
# 按列做类型转换(空串 -> None)。未列出的列 = 原样字符串(source/city_id/... 等 NOT NULL 文本)。
_INT_COLS = {0, 2, 3, 9, 10, 13, 15, 17, 24} # id, platform, biz_line, prices, ...
_FLOAT_COLS = {14, 18} # commission_percent, delivery_distance_m
_NULLABLE_STR_COLS = {6, 7, 8, 11, 12, 16, 25} # sku_view_id, name, brand_name, ...
_DT_COLS = {21, 22, 23} # first_seen, last_seen, updated_at
_RAW_COL = 20
_TZ_SUFFIX = re.compile(r"[+-]\d{2}(:?\d{2})?$") # 尾部时区偏移 +08 / +08:00 / +0800
def _resolve_sqlite_path() -> Path:
"""从 DATABASE_URL(env 或 .env)解析出 SQLite 文件路径。只支持 sqlite://。"""
url = os.environ.get("DATABASE_URL", "")
if not url:
env = _PROJECT_ROOT / ".env"
if env.exists():
for line in env.read_text(encoding="utf-8").splitlines():
if line.strip().startswith("DATABASE_URL="):
url = line.split("=", 1)[1].strip()
break
if not url:
url = "sqlite:///./data/app.db"
if not url.startswith("sqlite:"):
sys.exit(f"仅支持 sqlite:// 库,当前 DATABASE_URL={url!r}")
rest = url.split("sqlite:///", 1)[1] if "sqlite:///" in url else url.split("sqlite://", 1)[1]
p = Path(rest)
if not p.is_absolute():
p = (_PROJECT_ROOT / rest).resolve()
return p
def _reconstruct_rows(text: str) -> tuple[list[list[str]], int]:
"""把文件文本重组成一条条 26 列的逻辑行。返回 (rows, skipped)。
单个字段内含换行 -> 一条逻辑行被拆成多物理行:累计字段,拆点用 \\n 重新拼回,
直到凑满 26 列。列数溢出(内嵌 TAB / 错位)或文件尾残行 -> 跳过并计数。
"""
lines = text.split("\n")
while lines and lines[-1] == "":
lines.pop()
rows: list[list[str]] = []
skipped = 0
buf: list[str] = []
for raw_line in lines:
parts = raw_line.split("\t")
if not buf:
buf = parts
else:
buf[-1] += "\n" + parts[0] # 拼回被换行拆开的字段
buf.extend(parts[1:])
if len(buf) == NCOL:
rows.append(buf)
buf = []
elif len(buf) > NCOL: # 溢出:数据异常,丢弃这段重新开始
print(f" [skip] 列数溢出({len(buf)}>{NCOL}),field0={buf[0][:20]!r}")
skipped += 1
buf = []
if buf: # 文件尾被截断的残行
print(f" [skip] 尾部残行不足 {NCOL} 列(实 {len(buf)} 列),field0={buf[0][:20]!r}")
skipped += 1
return rows, skipped
def _convert(row: list[str]) -> tuple | None:
"""按列类型转换一行;非法(必填 int 为空 / raw 非 JSON)返回 None。"""
out: list = []
for i, v in enumerate(row):
if i in _DT_COLS:
out.append(_TZ_SUFFIX.sub("", v))
continue
if i == _RAW_COL:
try:
json.loads(v)
except Exception as e:
print(f" [skip] id={row[0]} raw 非法 JSON: {e}")
return None
out.append(v)
continue
if i in _INT_COLS:
out.append(int(v) if v != "" else None)
elif i in _FLOAT_COLS:
out.append(float(v) if v != "" else None)
elif i in _NULLABLE_STR_COLS:
out.append(v if v != "" else None)
else: # 必填文本列,原样
out.append(v)
return tuple(out)
def main() -> None:
tsv = Path(sys.argv[1]) if len(sys.argv) > 1 else _PROJECT_ROOT / "tests" / "meituan_coupon_data.tsv"
if not tsv.is_absolute():
tsv = (_PROJECT_ROOT / tsv).resolve()
db = _resolve_sqlite_path()
print(f"TSV: {tsv}")
print(f"DB : {db}")
if not tsv.exists():
sys.exit(f"TSV 不存在: {tsv}")
if not db.exists():
sys.exit(f"SQLite 库不存在: {db}(先跑 alembic upgrade head 建表)")
rows, skipped = _reconstruct_rows(tsv.read_text(encoding="utf-8"))
print(f"重组逻辑行: {len(rows)} 跳过(残/异常): {skipped}")
records = []
bad = 0
for r in rows:
rec = _convert(r)
if rec is None:
bad += 1
continue
records.append(rec)
print(f"可入库: {len(records)} 转换失败: {bad}")
placeholders = ",".join(["?"] * NCOL)
sql = f"INSERT OR REPLACE INTO meituan_coupon ({','.join(COLS)}) VALUES ({placeholders})"
con = sqlite3.connect(str(db))
try:
before = con.execute("SELECT count(*) FROM meituan_coupon").fetchone()[0]
con.executemany(sql, records)
con.commit()
after = con.execute("SELECT count(*) FROM meituan_coupon").fetchone()[0]
finally:
con.close()
print(f"入库前 {before} 行 -> 入库后 {after} 行(本次 {len(records)} 条)")
if __name__ == "__main__":
main()
+189
View File
@@ -0,0 +1,189 @@
"""给指定用户灌一批 mock 比价记录(开发 / 真机走查用)。
为什么:首页「上次比价」横幅(4 分钟新鲜窗口)、比价记录页、「我的」省钱战绩卡都读
`comparison_record`。dev 库里这个用户没数据时这些 UI 是空的,本脚本灌一批真实感的外卖
比价记录,方便端上走查。
用法(项目根、已 pip install -e . 的环境):
python scripts/mock_compare_records.py # 默认 user 昵称后缀 DlSKoc5S7, 10 条
python scripts/mock_compare_records.py --user DlSKoc5S7 # 按 昵称后缀/用户名/邀请码/手机号
python scripts/mock_compare_records.py --user 5 # 直接按 user.id
python scripts/mock_compare_records.py --user 5 --count 6
要点:
- 幂等:trace_id 固定为 mock-<标识>-NN,重跑覆盖同号记录(不会越灌越多)。
- 第 1 条 created_at = 跑脚本当下(4 分钟内),其余铺在近 7 天;重跑会刷新时间戳
(所以重跑一次即可让「上次比价」横幅重新进入 4 分钟窗口)。
- best_* / saved / status 由真实的 repositories.comparison.upsert_record 从 comparison_results 派生。
"""
from __future__ import annotations
import argparse
from datetime import datetime, timedelta
from sqlalchemy import or_, select
from app.core.rewards import CN_TZ
from app.db.session import SessionLocal
from app.models.user import User
from app.repositories.comparison import upsert_record
from app.schemas.compare_record import ComparisonRecordIn
# 平台名 → id(展示用;横幅按 platform_name 渲染,id 仅留痕)
_PLATFORM_ID = {"美团": "meituan", "京东": "jd", "淘宝闪购": "taobao", "饿了么": "ele"}
# 10 单真实感外卖订单:source=美团,targets=(平台, 到手价/元, 优惠来源名|None, 红包额/元)。
# failed=True 表示目标平台没采到价(targets 价置 None → 派生 status=failed)。
_CATALOG = [
dict(store="瑞幸咖啡(国贸店)", items=[("生椰拿铁(大杯)", 2)],
source=("美团", 39.8),
targets=[("淘宝闪购", 29.9, "平台红包", 6.0), ("京东", 33.0, None, 0)]),
dict(store="茶百道(大悦城店)", items=[("杨梅冰萃(大杯)", 1), ("脆波波奶茶", 1)],
source=("美团", 31.0),
targets=[("京东", 24.5, "百亿补贴", 4.5), ("饿了么", 27.9, None, 0)]),
dict(store="麦当劳(西单店)", items=[("板烧鸡腿堡套餐", 1)],
source=("美团", 32.0),
targets=[("京东", 25.9, "神券", 3.0), ("淘宝闪购", 28.0, None, 0)]),
dict(store="肯德基(朝阳大悦城店)", items=[("疯狂星期四全家桶", 1)],
source=("美团", 89.0),
targets=[("京东", 79.9, "百亿补贴", 9.0)]),
dict(store="张亮麻辣烫(双井店)", items=[("自选麻辣烫", 1)],
source=("美团", 35.0),
targets=[("京东", None, None, 0), ("淘宝闪购", None, None, 0)],
failed=True, fail_reason="京东外卖、淘宝闪购均未找到该店"),
dict(store="必胜客(王府井店)", items=[("超级至尊比萨(9寸)", 1), ("香辣鸡翅", 1)],
source=("美团", 78.0),
targets=[("淘宝闪购", 62.0, "跨店满减", 10.0), ("京东", 69.0, None, 0)]),
dict(store="蜜雪冰城(中关村店)", items=[("多肉葡萄", 2), ("冰鲜柠檬水", 1)],
source=("美团", 21.0),
targets=[("京东", 16.5, None, 0)]),
dict(store="海底捞外送(三里屯店)", items=[("番茄锅底", 1), ("鲜毛肚", 2), ("虾滑", 1)],
source=("美团", 168.0),
targets=[("京东", 155.0, "大额神券", 13.0), ("饿了么", 162.0, None, 0)]),
dict(store="华莱士(回龙观店)", items=[("全鸡汉堡套餐", 1)],
source=("美团", 26.0),
targets=[("京东", None, None, 0)],
failed=True, fail_reason="该商品在京东外卖未上架"),
dict(store="星巴克(国贸店)", items=[("燕麦拿铁(大杯)", 2)],
source=("美团", 66.0),
targets=[("饿了么", 58.0, "会员券", 8.0), ("京东", 61.0, None, 0)]),
]
# 第 2..N 条的 created_at 相对当下的回退量(都 > 4 分钟,确保只有第 1 条落在新鲜窗口)。
_OFFSETS = [
timedelta(minutes=25), timedelta(hours=2, minutes=10), timedelta(hours=6),
timedelta(days=1, hours=3), timedelta(days=1, hours=20), timedelta(days=2, hours=9),
timedelta(days=3, hours=14), timedelta(days=5, hours=7), timedelta(days=7, hours=2),
]
def _resolve_user(db, ident: str) -> User | None:
"""按 id(纯数字)/ 用户名 / 邀请码 / 手机号 / 昵称(或昵称去掉「用户」前缀的后缀)解析用户。"""
if ident.isdigit():
u = db.get(User, int(ident))
if u is not None:
return u
return db.execute(
select(User).where(
or_(
User.username == ident,
User.invite_code == ident,
User.phone == ident,
User.nickname == ident,
User.nickname == f"用户{ident}",
User.nickname.like(f"%{ident}"),
)
)
).scalars().first()
def _build_results(order: dict) -> tuple[list[dict], str]:
"""造 comparison_results + information 文案;返回 (results, information)。"""
src_name, src_price = order["source"]
failed = order.get("failed", False)
rows: list[dict] = [dict(
platform_id=_PLATFORM_ID.get(src_name), platform_name=src_name,
price=src_price, is_source=True, status="success",
)]
for pname, price, cname, csaved in order["targets"]:
row = dict(
platform_id=_PLATFORM_ID.get(pname), platform_name=pname,
price=(None if failed else price), is_source=False,
status=("store_not_found" if failed else "success"),
)
if not failed and csaved:
row["coupon_saved"] = csaved
row["coupon_name"] = cname
rows.append(row)
# rank:有价的按升序 1..N;无价的不排
priced = sorted((r for r in rows if r["price"] is not None), key=lambda r: r["price"])
for i, r in enumerate(priced, start=1):
r["rank"] = i
if failed:
return rows, order.get("fail_reason", "目标平台未找到该商品")
best = priced[0]
saved = src_price - best["price"]
if best["is_source"]:
info = f"{src_name}已是最低价,本单未比出更低"
else:
info = f"已在{best['platform_name']}比出最低价 ¥{best['price']:.2f}(比{src_name}省 ¥{saved:.2f})"
return rows, info
def main() -> None:
parser = argparse.ArgumentParser(description="给用户灌一批 mock 比价记录")
parser.add_argument("--user", default="DlSKoc5S7", help="user.id / 用户名 / 邀请码 / 手机号 / 昵称后缀")
parser.add_argument("--count", type=int, default=10, help="条数(1..10),默认 10")
args = parser.parse_args()
count = max(1, min(args.count, len(_CATALOG)))
now = datetime.now(CN_TZ).replace(tzinfo=None)
db = SessionLocal()
try:
user = _resolve_user(db, args.user)
if user is None:
raise SystemExit(f"❌ 找不到用户:{args.user}(试试 --user <id>)")
print(f"→ 用户 id={user.id} nickname={user.nickname} phone={user.phone}")
ok = failed = 0
for idx in range(count):
order = _CATALOG[idx]
src_name, src_price = order["source"]
results, info = _build_results(order)
payload = ComparisonRecordIn(
trace_id=f"mock-{args.user}-{idx:02d}",
business_type="food",
device_id="mock-seed",
store_name=order["store"],
source_platform_id=_PLATFORM_ID.get(src_name),
source_platform_name=src_name,
source_price=src_price,
items=[{"name": n, "qty": q} for n, q in order["items"]],
comparison_results=results,
total_dish_count=sum(q for _, q in order["items"]),
information=info,
app_version="mock",
)
rec = upsert_record(db, user_id=user.id, payload=payload)
# created_at:第 1 条 = 当下(4 分钟新鲜窗口);其余铺近 7 天。重跑刷新时间戳。
rec.created_at = now if idx == 0 else now - _OFFSETS[idx - 1]
db.commit()
if rec.status == "success":
ok += 1
else:
failed += 1
tag = "🆕now" if idx == 0 else f"-{_OFFSETS[idx - 1]}"
print(f" [{rec.status:7}] id={rec.id} {order['store']} best={rec.best_platform_name} "
f"¥{(rec.best_price_cents or 0)/100:.2f} saved=¥{(rec.saved_amount_cents or 0)/100:.2f} {tag}")
print(f"✅ 已为 user id={user.id} 写入 {count} 条(success {ok} / failed {failed});"
f"第 1 条 created_at=当下,4 分钟内进 app 可见「上次比价」横幅。")
finally:
db.close()
if __name__ == "__main__":
main()