Merge pull request #268 from kevin9327/fix/pull-window-beijing

fix(ext-data): 定时拉取的时间窗口与落盘日期改用北京时间
This commit is contained in:
wshy
2026-09-08 21:06:36 +08:00
committed by GitHub
2 changed files with 113 additions and 3 deletions
+10 -3
View File
@@ -12,6 +12,7 @@ from typing import Any
import httpx import httpx
from app.market_time import cn_now, cn_today
from app.services.ext_data import ( from app.services.ext_data import (
ExtConfig, ExtConfig,
ExtConfigStore, ExtConfigStore,
@@ -43,14 +44,19 @@ def outbound_headers(user_headers: dict[str, str] | None = None) -> dict[str, st
def _in_time_window(start: str | None, end: str | None) -> bool: def _in_time_window(start: str | None, end: str | None) -> bool:
"""检查当前本地时间是否在每日时间窗口内。 """检查当前北京时间是否在每日时间窗口内。
start/end 为 "HH:MM" 格式。两者都为 None 时不限制(返回 True)。 start/end 为 "HH:MM" 格式。两者都为 None 时不限制(返回 True)。
支持跨午夜窗口(如 22:00-02:00)。 支持跨午夜窗口(如 22:00-02:00)。
用北京时间而不是本地时间: 这个窗口是照着 A 股交易时段设的, 而
market_time 模块开篇就写明「服务器/容器本地时区不可靠 (python:slim
镜像默认 UTC)」。UTC 容器里 9:30-15:00 的窗口实际落在北京 17:30-23:00,
每天都在收盘之后。
""" """
if not start or not end: if not start or not end:
return True return True
now = datetime.now().strftime("%H:%M") now = cn_now().strftime("%H:%M")
if start <= end: if start <= end:
return start <= now < end return start <= now < end
# 跨午夜: 如 22:00-02:00 # 跨午夜: 如 22:00-02:00
@@ -258,7 +264,8 @@ async def fetch_and_ingest(
Returns: Returns:
(rows_written, date_str) (rows_written, date_str)
""" """
day = target_date or date.today() # 同上: 落盘分区按北京日期, 否则 UTC 容器在北京时间 08:00 之前写的是前一天。
day = target_date or cn_today()
rows = await fetch_rows_for_date(config, day) rows = await fetch_rows_for_date(config, day)
if not rows: if not rows:
raise ValueError("提取到的行数为 0") raise ValueError("提取到的行数为 0")
+103
View File
@@ -0,0 +1,103 @@
"""扩展数据定时拉取的时间基准测试 — 窗口与落盘日期都按北京时间。
背景: `app/market_time.py` 开篇写明「服务器/容器本地时区不可靠 (python:slim
镜像默认 UTC), 交易时段判断、实时行情落盘日期等必须显式使用北京时间, 否则
Docker 部署时轮询窗口与真实交易时段完全错开」。
拉取窗口正是照着 A 股交易时段设的, 落盘分区正是行情日期, 两处却都用了裸的
`datetime.now()` / `date.today()`。UTC 容器里 09:30-15:00 的窗口实际落在北京
17:30-23:00, 每天都在收盘之后; 北京时间 08:00 之前落盘写的是前一天的分区。
判据不依赖跑测试的机器在哪个时区: 固定一个北京时刻, 断言窗口按它判断。
"""
from __future__ import annotations
from datetime import date, datetime
import pytest
from app.market_time import CN_TZ
from app.services import ext_pull
@pytest.fixture
def at_beijing(monkeypatch: pytest.MonkeyPatch):
"""把北京时间钉在给定时刻。"""
def freeze(moment: datetime):
monkeypatch.setattr(ext_pull, "cn_now", lambda: moment)
monkeypatch.setattr(ext_pull, "cn_today", lambda: moment.date())
return freeze
def test_window_follows_beijing_time_not_the_machine_clock(at_beijing) -> None:
# 北京 10:00, 正在早盘。
at_beijing(datetime(2026, 3, 2, 10, 0, tzinfo=CN_TZ))
assert ext_pull._in_time_window("09:30", "15:00") is True
# 北京 17:30 — UTC 容器上「本地 09:30」正是这个时刻, 修复前会误判为在窗口内。
at_beijing(datetime(2026, 3, 2, 17, 30, tzinfo=CN_TZ))
assert ext_pull._in_time_window("09:30", "15:00") is False
def test_window_across_midnight_still_works(at_beijing) -> None:
at_beijing(datetime(2026, 3, 2, 23, 0, tzinfo=CN_TZ))
assert ext_pull._in_time_window("22:00", "02:00") is True
at_beijing(datetime(2026, 3, 2, 1, 0, tzinfo=CN_TZ))
assert ext_pull._in_time_window("22:00", "02:00") is True
at_beijing(datetime(2026, 3, 2, 12, 0, tzinfo=CN_TZ))
assert ext_pull._in_time_window("22:00", "02:00") is False
def test_no_window_configured_is_still_unrestricted(at_beijing) -> None:
at_beijing(datetime(2026, 3, 2, 3, 0, tzinfo=CN_TZ))
assert ext_pull._in_time_window(None, None) is True
assert ext_pull._in_time_window("09:30", None) is True
assert ext_pull._in_time_window(None, "15:00") is True
async def test_default_landing_date_is_the_beijing_date(at_beijing, monkeypatch) -> None:
"""不传 target_date 时按北京日期落盘。
北京 00:30 对应 UTC 前一天 16:30, 裸 date.today() 在 UTC 容器上会把当天的
数据写进前一天的分区。
"""
at_beijing(datetime(2026, 3, 2, 0, 30, tzinfo=CN_TZ))
seen: dict[str, date] = {}
async def fake_fetch(config, target_date):
seen["fetched"] = target_date
return [{"symbol": "000001.SZ", "v": 1}]
def fake_write(rows, config, data_dir, snapshot_date):
seen["written"] = snapshot_date
return len(rows)
monkeypatch.setattr(ext_pull, "fetch_rows_for_date", fake_fetch)
monkeypatch.setattr(ext_pull, "rows_to_parquet", fake_write)
written, day_str = await ext_pull.fetch_and_ingest(object(), "/tmp")
assert written == 1
assert seen["fetched"] == date(2026, 3, 2)
assert seen["written"] == date(2026, 3, 2)
assert day_str == "2026-03-02"
async def test_an_explicit_target_date_still_wins(at_beijing, monkeypatch) -> None:
# 历史回补路径不受影响。
at_beijing(datetime(2026, 3, 2, 10, 0, tzinfo=CN_TZ))
seen: dict[str, date] = {}
async def fake_fetch(config, target_date):
seen["fetched"] = target_date
return [{"symbol": "000001.SZ", "v": 1}]
monkeypatch.setattr(ext_pull, "fetch_rows_for_date", fake_fetch)
monkeypatch.setattr(ext_pull, "rows_to_parquet", lambda *a, **k: 1)
_, day_str = await ext_pull.fetch_and_ingest(object(), "/tmp", date(2026, 1, 5))
assert seen["fetched"] == date(2026, 1, 5)
assert day_str == "2026-01-05"