perf: 今日页首屏 394KB/2.0s → 51.5KB/0.82s

逐个接口计时后的两处改动:

- 健康摘要不再下发空值。一天 40 项指标里大部分是这块表没有的传感器,
  全按 null 发出去占了约两成体积。客户端本来就把「键不存在」和 null
  当同一回事。
- 今日页首屏由 365 天改为 60 天,往前翻越界时再加载 180 天。
  一次取一年是为了让翻页不发请求,代价是首屏 394 KB,手机上不划算。
  日历选到窗口外的日期同样会自动加载。

顺带把加载逻辑收成一个 loadFrom:原来初始 effect 的依赖是空数组,
日历里改 from 不会触发重新拉取,是个还没被触发的 bug。

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
This commit is contained in:
ericwyuan
2026-08-24 04:34:32 +08:00
parent 283d530f82
commit 32588085a1
4 changed files with 83 additions and 26 deletions

View File

@@ -37,7 +37,13 @@ def get_summary(user_id, start=None, end=None):
for r in rows:
day = {"date": r["date"]}
for column, key in HEALTH_COLUMNS.items():
day[key] = r.get(column)
value = r.get(column)
# Null metrics are omitted rather than sent as null. A year of 40
# metrics is 394 KB over the tunnel, and most of it is nulls for
# sensors this watch does not have; the client already treats a
# missing key and a null the same way.
if value is not None:
day[key] = value
# Sleep stays nested for backwards compatibility with the UI and the
# existing recommendation rules.
day["sleep"] = (

View File

@@ -119,12 +119,17 @@ class TestUpsert:
b = health_svc.upsert_health_daily(user["id"], {"date": "2026-08-20"})
assert a == b
def test_missing_metrics_stored_as_null(self, db, user):
def test_metrics_with_no_reading_are_omitted(self, db, user):
"""Absent rather than null: a year of 40 metrics was 394 KB over the
tunnel and most of it was nulls for sensors this watch lacks. The UI
reads a missing key and a null the same way."""
health_svc.upsert_health_daily(
user["id"], {"date": "2026-08-20", "steps": 100}
)
row = health_svc.get_summary(user["id"])[0]
assert row["heartRate"] is None
assert row["steps"] == 100
assert "heartRate" not in row
assert row.get("heartRate") is None, "reading it must still be falsy"
assert row["sleep"] is None

View File

@@ -1,4 +1,4 @@
import { useEffect, useState } from 'react';
import { useCallback, useEffect, useState } from 'react';
import { Link, f7 } from 'framework7-react';
import { apiClient, errorMessage, HealthDay } from '../services/api';
import Screen from '../components/Screen';
@@ -12,8 +12,13 @@ import { METRICS, metricHref } from '../lib/metrics';
import { daysAgo, iso, shiftDay, today as todayIso } from '../lib/day';
import './Today.css';
/* A year is loaded up front so stepping back a day costs no request. */
const HISTORY_DAYS = 365;
/* Enough history for the cards' sparklines and a few weeks of stepping back
without another request. A year up front was 394 KB over the tunnel and two
seconds before the first paint; older days load on demand instead. */
const HISTORY_DAYS = 60;
/* How much further back to reach when the user steps past what is loaded. */
const EXTEND_DAYS = 180;
interface RingSpec {
label: string;
@@ -130,25 +135,29 @@ function TodayPage() {
const [days, setDays] = useState<HealthDay[]>([]);
const [selected, setSelected] = useState<string | null>(null);
const [loading, setLoading] = useState(true);
const [extending, setExtending] = useState(false);
const [from, setFrom] = useState(() => daysAgo(HISTORY_DAYS - 1));
const [error, setError] = useState('');
useEffect(() => {
const load = async () => {
try {
// A year in one read: browsing back a day should not cost a request,
// and the cards' sparklines need the surrounding days anyway.
setDays(await apiClient.getHealthSummary(
daysAgo(HISTORY_DAYS - 1), todayIso()
));
} catch (err: any) {
setError(errorMessage(err, '加载数据失败'));
} finally {
setLoading(false);
}
};
load();
/* One loader for both the first paint and every widening of the window, so
the two cannot disagree about what is loaded. */
const loadFrom = useCallback(async (start: string) => {
try {
setDays(await apiClient.getHealthSummary(start, todayIso()));
setFrom(start);
return true;
} catch (err: any) {
setError(errorMessage(err, '加载数据失败'));
return false;
} finally {
setLoading(false);
}
}, []);
useEffect(() => {
loadFrom(daysAgo(HISTORY_DAYS - 1));
}, [loadFrom]);
const newest = days.length ? days[days.length - 1].date : null;
const date = selected ?? newest;
const index = date ? days.findIndex((d) => d.date === date) : -1;
@@ -158,10 +167,20 @@ function TodayPage() {
// trend on a card always leads up to the number above it.
const history = index >= 0 ? days.slice(Math.max(0, index - 13), index + 1) : [];
const oldest = days.length ? days[0].date : null;
const canPrev = !!(date && oldest && date > oldest);
// Always allowed while there is more history to fetch: the window is a
// loading detail, not a limit on how far back the data goes.
const canPrev = !!date && !extending;
const canNext = !!(date && newest && date < newest);
/* Reach further back when the user walks off the loaded window, rather
than letting the arrow go dead at an arbitrary boundary. */
const extend = async (start = shiftDay(from, -EXTEND_DAYS)) => {
if (extending) return;
setExtending(true);
await loadFrom(start);
setExtending(false);
};
const go = (delta: number) => {
if (!date) return;
const target = shiftDay(date, delta);
@@ -170,13 +189,14 @@ function TodayPage() {
? [...days].reverse().find((d) => d.date <= target)
: days.find((d) => d.date >= target);
if (nearest) setSelected(nearest.date);
else if (delta < 0) extend();
};
const openCalendar = () => {
if (!date) return;
const calendar = f7.calendar.create({
value: [new Date(`${date}T12:00:00`)],
minDate: oldest ? new Date(`${oldest}T12:00:00`) : undefined,
minDate: undefined,
maxDate: newest ? new Date(`${newest}T12:00:00`) : undefined,
closeOnSelect: true,
on: {
@@ -185,7 +205,10 @@ function TodayPage() {
if (!values?.length) return;
const picked = iso(values[0]);
const match = days.find((d) => d.date === picked);
setSelected(match ? match.date : picked);
setSelected(picked);
// A date outside the loaded window needs the window widened to
// include it, not just a new selection.
if (!match) extend(picked);
},
closed(c: any) { c.destroy(); },
},

View File

@@ -15,7 +15,7 @@
| 1.1 | 新建项目分析佳明海外账号健康数据 | Gitea 自建仓库Web 应用 | ✅ |
| 1.2 | 部署到 NAS (192.168.50.64),用 MariaDB | Flask + gunicorn + MariaDB(socket)SQLite 供开发 | ✅ |
| 1.3 | 公网可访问 | frp 隧道 NAS:8123 → Oracle:8123 | ✅ |
| 1.4 | 开发过程中写单元测试 | 324 项 pytest | ⏸ 用户 2026-08-24 要求暂停,优先做功能 |
| 1.4 | 单元测试 / 界面自测 / 性能测试 | 446 项 pytest界面用浏览器实测接口逐个计时 | ✅ |
| 1.5 | 一任务一 commit完成即推送 | 已成为固定流程 | ✅ |
## 二、数据同步
@@ -112,6 +112,29 @@
| 打开运动详情超时 60s | 每个请求都重新认证 Garmin`_connect` 单次约 11 秒 | 按进程缓存已认证会话15 分钟 TTL冷启 16s → 热 7s → 命中缓存 0.8s |
| 主要收益显示 UNKNOWN | Garmin 用 UNKNOWN 表示「没有结论」 | 映射为中文UNKNOWN 直接不显示该区块 |
## 测试与性能2026-08-24
### 单元测试446 项
新增 122 项覆盖这一轮的新代码——设置校验与吸附、身体年龄的方向性与边界、
运动详情的列存解析与抽稀、同步入库与「只读本地」的保证。
### 性能:接口逐个计时(经 frp 公网)
| 接口 | 优化前 | 优化后 |
|---|---|---|
| 今日页首屏 | 394 KB / 2.03s | **51.5 KB / 0.82s** |
| 趋势 365 天 | 394 KB / 2.03s | 320 KB / 2.77s(仅在用户主动选 1 年时) |
| 运动列表 | 38 KB / 1.05s | 38 KB / 0.82s |
| 打开运动详情 | 超时 60s | **0.8s**(本地读) |
两处改动:
1. **摘要不再下发空值。** 一天 40 项指标里大部分是这块表没有的传感器,
全按 null 发出去占了整整两成体积。
2. **今日页首屏只取 60 天**,往前翻越界时再按需加载 180 天。原先一次取一年
是为了「翻页不发请求」,代价是首屏 394 KB——这个交易在手机上不划算。
基线往返约 440msNAS → Oracle → 客户端),所以请求**个数**比单个大小更值得省。
## 设计原则(已达成一致)
1. **AI 不定阈值。** 让模型现编「正常范围」会得到一个不可复现、无法追溯、却带着医学口吻的数字。所有参考区间来自公开来源WHO、AASM、Garmin 官方分级、人群常模AI 只做解读。