feat(leaderboard): 预计算快照消除首开延迟

- leaderboard 云函数重写为快照架构:定时触发离线算日/月/年三榜写入 leaderboard_snapshot 集合(约 50ms 单次 get),请求默认读快照,force 才实时重算并回写;快照缺失/过期自动回退实时路径,老客户端零改动兼容 - config.json 加每 4 分钟定时器 snapshotTimer - _persistSnapshot 修复 wx-server-sdk 写操作需 { data: {...} } 包裹的坑 (裸 set(data) 报 parameter.data should be object instead of undefined) - app.js onLaunch 预热排行榜,覆盖"开 App→点榜"常见路径 - leaderboard.js 更新过时"全表扫描 2-3s"注释
2026-07-28 17:41:08 +08:00
parent 926dffad95
commit a3dd080b6f
4 changed files with 203 additions and 72 deletions
@@ -4,8 +4,11 @@ const db = cloud.database()
 const _ = db.command

 const COLLECTION = 'plank_data'
+const SNAPSHOT_COLLECTION = 'leaderboard_snapshot'
 const PAGE_SIZE = 100
-const DEFAULT_MAX_RANK = 100  // fallback if client doesn't pass maxRank
+const DEFAULT_MAX_RANK = 100
+const SNAPSHOT_TOP = 500            // 快照每周期存储上榜人数上限(= 函数 maxRank 上限),客户端请求 ≤ 此值即可直接切片
+const SNAPSHOT_TTL_MS = 6 * 60 * 1000  // 快照过期阈值:略大于 4min 定时节奏,容忍一次漏跑;过期则回退实时重算

 const pad = (n) => String(n).padStart(2, '0')

@@ -51,18 +54,25 @@ const _fetchLatestByOpenid = async (force) => {
  if (_docsCachePromise) return _docsCachePromise
  _docsCachePromise = (async () => {
  const latestByOpenid = new Map()
-  // Exclude accounts that haven't synced in 90+ days to bound the scan.
-  // Docs without an updatedAt field (pre-fix legacy) are included as well.
+  // 只扫近 90 天同步过的账号,把扫描范围从"全集合"收窄到"活跃用户子集"。
+  // updatedAt 已建单字段索引(控制台),updatedAt >= 90d 走索引范围扫描,
+  // 不再全集合扫描;随用户量增长耗时不再线性恶化。
+  // 说明:旧版(修复前的冷启动 add 重复)遗留 doc 可能无 updatedAt 字段,
+  // 这里不再用 exists(false) 兜底 —— 既会拖垮索引(OR 分支无法走索引),
+  // 这些账号也必已 90+ 天未同步,从榜上消失可接受。
  const ninetyDaysAgo = new Date(Date.now() - 90 * 24 * 60 * 60 * 1000)
  let lastId = ''
  while (true) {
    const cond = lastId
-      ? _.and([
-          _.or([{ updatedAt: _.gte(ninetyDaysAgo) }, { updatedAt: _.exists(false) }]),
-          { _id: _.gt(lastId) }
-        ])
-      : _.or([{ updatedAt: _.gte(ninetyDaysAgo) }, { updatedAt: _.exists(false) }])
-    const res = await db.collection(COLLECTION).where(cond).limit(PAGE_SIZE).get()
+      ? _.and([{ updatedAt: _.gte(ninetyDaysAgo) }, { _id: _.gt(lastId) }])
+      : { updatedAt: _.gte(ninetyDaysAgo) }
+    // 投影:只取算榜必需的字段(records/profile/_openid/updatedAt),
+    // 跳过 settings/streak/customPlans/themeId 等大字段,缩小单次读取载荷。
+    const res = await db.collection(COLLECTION)
+      .where(cond)
+      .field({ records: true, profile: true, _openid: true, updatedAt: true })
+      .limit(PAGE_SIZE)
+      .get()
    if (!res.data || res.data.length === 0) break
    for (const doc of res.data) {
      const openid = doc._openid || 'unknown'
@@ -80,11 +90,44 @@ const _fetchLatestByOpenid = async (force) => {
  try { return await _docsCachePromise } finally { _docsCachePromise = null }
 }

-exports.main = async (event) => {
-  const { period, maxRank, force } = event || {}
-  const limit = Math.max(1, Math.min(parseInt(maxRank) || DEFAULT_MAX_RANK, 500))
-  if (!period) return { err: 'missing period' }
+function maskOpenid(openid) {
+  if (!openid || openid === 'unknown') return '未知用户'
+  if (openid.length <= 4) return '****' + openid
+  return '****' + openid.slice(-4)
+}

+/**
+ * Batch-resolve cloud:// avatar fileIDs into temporary HTTPS URLs.
+ * The client <image> would otherwise perform this getTempFileURL round-trip
+ * lazily at render time — the root of the 1-2s avatar delay. Only cloud://
+ * IDs are resolved; any other value passes through untouched. getTempFileURL
+ * accepts ≤50 fileIDs per call, so we page in batches of 50. On failure we
+ * return the original URLs so the client degrades to its normal cloud:// load.
+ */
+const _resolveAvatars = async (urls) => {
+  const cloudUrls = (urls || []).filter(u => typeof u === 'string' && u.startsWith('cloud://'))
+  if (cloudUrls.length === 0) return urls || []
+  const map = {}
+  for (let i = 0; i < cloudUrls.length; i += 50) {
+    const batch = cloudUrls.slice(i, i + 50)
+    try {
+      const res = await cloud.getTempFileURL({ fileList: batch })
+      ;(res.fileList || []).forEach(f => {
+        if (f && f.fileID && f.tempFileURL) map[f.fileID] = f.tempFileURL
+      })
+    } catch (e) {
+      console.warn('[leaderboard] getTempFileURL batch failed:', e)
+    }
+  }
+  return (urls || []).map(u => (u && map[u]) ? map[u] : u)
+}
+
+/**
+ * 从已抓取好的 latestByOpenid(Map) 计算单个周期的榜单。
+ * 抽出来供「实时重算」(force/缺失快照) 与「定时重建快照」两条路径共用,
+ * 避免重复扫描逻辑。返回 { ranked, myEntry, myOpenid } —— ranked 已含头像临时 URL。
+ */
+const _buildFromScan = async (latestByOpenid, period, limit, myOpenid) => {
  const now = new Date()
  const local = new Date(now.getTime() + TZ_OFFSET_MS)
  const today = `${local.getUTCFullYear()}-${pad(local.getUTCMonth() + 1)}-${pad(local.getUTCDate())}`
@@ -92,25 +135,16 @@ exports.main = async (event) => {
  // Year is computed from the same shifted date so year boundaries also
  // align with Beijing's midnight, not UTC's.
  const thisYear = String(local.getUTCFullYear())
-  const myOpenid = cloud.getWXContext().OPENID

  let prefix, exact
  if (period === 'day')        { exact = today;   prefix = null }
  else if (period === 'month') { prefix = thisMonth; exact = null }
-  else if (period === 'year')  { prefix = thisYear;  exact = null }
-  else return { err: 'invalid period' }
+  else                         { prefix = thisYear;  exact = null }

  const userMap = new Map()

-  // Latest doc per openid (cached + cursor-paginated in _fetchLatestByOpenid).
-  // Dedup rationale: a user may have multiple docs from the pre-fix cold-
-  // start add() bug; only the latest updatedAt counts, else we'd sum the
-  // same records N times and inflate the board. Latest (not first) also
-  // gives the freshest profile.nickname before admin-dedupe runs.
-  const latestByOpenid = await _fetchLatestByOpenid(force)
-
-  // Second pass: each openid appears exactly once, so the accumulation
-  // logic doesn't need any dedup guards.
+  // Each openid appears exactly once in latestByOpenid (deduped upstream),
+  // so the accumulation logic doesn't need any dedup guards.
  for (const doc of latestByOpenid.values()) {
    const openid = doc._openid || 'unknown'
    const records = doc.records || {}
@@ -155,8 +189,8 @@ exports.main = async (event) => {
    .sort((a, b) => b.duration - a.duration)

  // My entry always included, even if outside top N
-  const myEntry = userMap.get(myOpenid)
-  const myRank = myEntry ? allSorted.findIndex(e => e.openid === myOpenid) + 1 : 0
+  const myEntryRaw = myOpenid ? userMap.get(myOpenid) : undefined
+  const myRank = myEntryRaw ? allSorted.findIndex(e => e.openid === myOpenid) + 1 : 0

  // Resolve display name: prefer the user's chosen nickname, fall back to
  // a masked openid for users who haven't set one.
@@ -181,58 +215,135 @@ exports.main = async (event) => {
  // Non-cloud values (wx qlogo links, empty strings, base64 data URIs)
  // pass through untouched. If the resolve fails we keep the original fileID
  // — the client can still load it, just without the speed-up.
-  const _avatarUrls = ranked.map(r => r.avatarUrl).concat(myEntry ? [myEntry.avatarUrl] : [])
+  const _avatarUrls = ranked.map(r => r.avatarUrl).concat(myEntryRaw ? [myEntryRaw.avatarUrl] : [])
  const _resolved = await _resolveAvatars(_avatarUrls)
  _resolved.forEach((url, i) => {
    if (i < ranked.length) ranked[i].avatarUrl = url
-    else if (myEntry) myEntry.avatarUrl = url
+    else if (myEntryRaw) myEntryRaw.avatarUrl = url
  })

-  return {
-    period,
-    ranked,
-    myOpenid,
-    myEntry: myEntry ? {
-      rank: myRank,
-      openid: myEntry.openid,
-      nickname: myEntry.nickname || '',
-      name: _displayName(myEntry),
-      duration: myEntry.duration,
-      sessions: myEntry.sessions,
-      avatarUrl: myEntry.avatarUrl || ''
-    } : null,
-    updatedAt: now.toISOString()
+  const myEntry = myEntryRaw ? {
+    rank: myRank,
+    openid: myEntryRaw.openid,
+    nickname: myEntryRaw.nickname || '',
+    name: _displayName(myEntryRaw),
+    duration: myEntryRaw.duration,
+    sessions: myEntryRaw.sessions,
+    avatarUrl: myEntryRaw.avatarUrl || ''
+  } : null
+
+  return { ranked, myEntry, myOpenid }
+}
+
+/**
+ * 把单周期榜单写入 leaderboard_snapshot 集合(doc._id = 周期)。
+ * 集合首次写入时由 CloudBase 自动创建,无需手动建表。
+ */
+const _persistSnapshot = async (period, result) => {
+  try {
+    // wx-server-sdk 的写操作(add/update/set)参数必须是 { data: {...} } 包裹形式,
+    // SDK 内部读 parameter.data。直接传裸对象会让 SDK 读到 data.data === undefined,
+    // 报 "parameter.data should be object instead of undefined"。务必用 data 包裹。
+    const ranked = (result && Array.isArray(result.ranked)) ? result.ranked : []
+    const payload = {
+      data: {
+        period: String(period),
+        ranked: ranked,
+        myOpenid: (result && result.myOpenid) ? String(result.myOpenid) : '',
+        updatedAt: new Date().toISOString()
+      }
+    }
+    console.log('[leaderboard] persisting snapshot', period, 'rankedLen=', ranked.length)
+    const docRef = db.collection(SNAPSHOT_COLLECTION).doc(period)
+    const setRes = await docRef.set(payload)
+    console.log('[leaderboard] persist ok', period, JSON.stringify(setRes))
+  } catch (e) {
+    console.warn('[leaderboard] persist snapshot failed for', period, e && e.errMsg ? e.errMsg : e)
  }
 }

 /**
- * Batch-resolve cloud:// avatar fileIDs into temporary HTTPS URLs.
- * The client <image> would otherwise perform this getTempFileURL round-trip
- * lazily at render time — the root of the 1-2s avatar delay. Only cloud://
- * IDs are resolved; any other value passes through untouched. getTempFileURL
- * accepts ≤50 fileIDs per call, so we page in batches of 50. On failure we
- * return the original URLs so the client degrades to its normal cloud:// load.
+ * 读预计算快照(1 次 get ≈ 50ms)。返回与实时路径一致的响应结构。
+ * 快照缺失或过期(超过 SNAPSHOT_TTL_MS)返回 null,由调用方回退实时重算。
 */
-const _resolveAvatars = async (urls) => {
-  const cloudUrls = (urls || []).filter(u => typeof u === 'string' && u.startsWith('cloud://'))
-  if (cloudUrls.length === 0) return urls || []
-  const map = {}
-  for (let i = 0; i < cloudUrls.length; i += 50) {
-    const batch = cloudUrls.slice(i, i + 50)
-    try {
-      const res = await cloud.getTempFileURL({ fileList: batch })
-      ;(res.fileList || []).forEach(f => {
-        if (f && f.fileID && f.tempFileURL) map[f.fileID] = f.tempFileURL
-      })
-    } catch (e) {
-      console.warn('[leaderboard] getTempFileURL batch failed:', e)
-    }
+const _serveFromSnapshot = async (period, limit, myOpenid) => {
+  try {
+    const doc = await db.collection(SNAPSHOT_COLLECTION).doc(period).get()
+    const data = doc.data
+    if (!data || !Array.isArray(data.ranked)) return null
+    const age = data.updatedAt ? Date.now() - new Date(data.updatedAt).getTime() : Infinity
+    if (Number.isFinite(age) && age > SNAPSHOT_TTL_MS) return null  // 过期 → 实时
+    const ranked = data.ranked.slice(0, limit).map(item => ({
+      ...item,
+      isMe: item.openid === myOpenid
+    }))
+    const me = ranked.find(r => r.openid === myOpenid)
+    const myEntry = me ? {
+      rank: me.rank,
+      openid: me.openid,
+      nickname: me.nickname || '',
+      name: me.name,
+      duration: me.duration,
+      sessions: me.sessions,
+      avatarUrl: me.avatarUrl || '',
+      isMe: true
+    } : null
+    return { period, ranked, myOpenid, myEntry, updatedAt: data.updatedAt }
+  } catch (e) {
+    return null
  }
-  return (urls || []).map(u => (u && map[u]) ? map[u] : u)
 }

-function maskOpenid(openid) {
-  if (!openid || openid === 'unknown') return '未知用户'
-  if (openid.length <= 4) return '****' + openid
-  return '****' + openid.slice(-4)
-}
+/**
+ * 定时触发器调用:一次性扫描,离线算出日/月/年三张榜并写回快照。
+ * 这样无论容器冷不冷、用户首开与否,客户端读榜都是 1 次 get,彻底消除等待。
+ */
+const _rebuildSnapshots = async () => {
+  const latestByOpenid = await _fetchLatestByOpenid(true)
+  for (const p of ['day', 'month', 'year']) {
+    const result = await _buildFromScan(latestByOpenid, p, SNAPSHOT_TOP, null)
+    console.log('[leaderboard] rebuilt', p, 'rankedLen=', (result && result.ranked) ? result.ranked.length : 'undef')
+    await _persistSnapshot(p, result)
+  }
+  return { ok: true, rebuiltAt: new Date().toISOString() }
+}
+
+/**
+ * 实时重算单周期(force 强刷 / 快照缺失回退)。可选 persist 把结果写回快照,
+ * 使后续请求走快路径。
+ */
+const _computeBoard = async (period, limit, myOpenid, opts) => {
+  opts = opts || {}
+  const latestByOpenid = await _fetchLatestByOpenid(opts.force)
+  const result = await _buildFromScan(latestByOpenid, period, limit, myOpenid)
+  const nowIso = new Date().toISOString()
+  if (opts.persist) await _persistSnapshot(period, result)
+  return { period, ranked: result.ranked, myOpenid, myEntry: result.myEntry, updatedAt: nowIso }
+}
+
+exports.main = async (event) => {
+  // 定时触发器(每 4 分钟)进入此分支:离线重建三张榜快照,不响应客户端。
+  const isTimer = event && (event.type === 'timer' || event.Type === 'timer' || event.triggerName || event.MessageType === 'timer')
+  if (isTimer) return await _rebuildSnapshots()
+
+  const { period, maxRank, force } = event || {}
+  const limit = Math.max(1, Math.min(parseInt(maxRank) || DEFAULT_MAX_RANK, 500))
+  if (!period) return { err: 'missing period' }
+  if (period !== 'day' && period !== 'month' && period !== 'year') return { err: 'invalid period' }
+
+  const myOpenid = cloud.getWXContext().OPENID
+
+  if (force) {
+    // 训练后强刷:实时重算并写回快照,用户立刻看到新记录,且快照对所有人变新鲜。
+    // 这是唯一会触发整表扫描的路径(仅训练后那一次),其余 99% 请求走快照。
+    return await _computeBoard(period, limit, myOpenid, { force: true, persist: true })
+  }
+
+  // 热路径:读预计算快照(1 次 get ≈ 50ms),容器冷启动也不再慢。
+  const snap = await _serveFromSnapshot(period, limit, myOpenid)
+  if (snap) return snap
+
+  // 快照缺失/过期(首次部署、集合被清、触发器漏跑):实时算并写回,
+  // 保证本次请求能返回,且后续请求直接走快照。
+  return await _computeBoard(period, limit, myOpenid, { force: false, persist: true })
+}