wx_pbzc/cloudfunctions/leaderboard/index.js

const cloud = require('wx-server-sdk')
cloud.init({ env: cloud.DYNAMIC_CURRENT_ENV })
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
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')

// Fixed UTC+8 offset — the user base is China-only and `new Date()` in
// the cloud function runs in UTC, so without this shift "today" rolls
// over at 08:00 Beijing instead of 00:00. If you ever support users in
// other timezones, switch back to a client-provided date.
const TZ_OFFSET_MS = 8 * 60 * 60 * 1000

// In-instance cache of the latest doc per openid. The full-collection
// scan + dedup is the expensive part of building the board; cache it for
// a short TTL so repeated board opens (and day/month/year switches within
// a session) reuse one scan. Cold starts just re-scan. Not persistent
// across instances, which is fine - the board needn't be real-time.
// 5min: the board is non-real-time, and the expensive part is the serial
// full-collection scan (100/page). A longer TTL means far fewer re-scans
// while the instance stays warm; the trade-off is a freshly-trained user
// may not see themselves for up to 5min, acceptable for a motivational
// board. The client also keeps its own display cache, so re-entry paints
// instantly regardless of this TTL.
const CACHE_TTL = 5 * 60 * 1000
let _docsCache = null  // { latestByOpenid: Map, ts }
let _docsCachePromise = null  // in-flight 扫描 promise,防缓存击穿

const tsOf = (d) => {
  if (!d.updatedAt) return 0
  // CloudBase returns Date objects in the SDK, but be defensive in case
  // a stringified ISO timestamp sneaks through (older SDKs / migrations).
  if (d.updatedAt instanceof Date) return d.updatedAt.getTime()
  const t = new Date(d.updatedAt).getTime()
  return Number.isFinite(t) ? t : 0
}

// Page through the collection keeping only the most-recently-updated doc
// per openid. Uses cursor pagination (_id > lastId) instead of skip(N):
// skip is O(N) and degrades as the collection grows; a cursor is constant
// cost per page, and default get() order (_id ascending) won't drop/dup.
const _fetchLatestByOpenid = async (force) => {
  if (!force && _docsCache && (Date.now() - _docsCache.ts) < CACHE_TTL) {
    return _docsCache.latestByOpenid
  }
  // 缓存过期或强刷(force):复用 in-flight promise,避免并发请求各自全表扫描(缓存击穿)
  if (_docsCachePromise) return _docsCachePromise
  _docsCachePromise = (async () => {
  const latestByOpenid = new Map()
  // 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([{ 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'
      const current = latestByOpenid.get(openid)
      if (!current || tsOf(doc) > tsOf(current)) {
        latestByOpenid.set(openid, doc)
      }
    }
    if (res.data.length < PAGE_SIZE) break
    lastId = res.data[res.data.length - 1]._id
  }
  _docsCache = { latestByOpenid, ts: Date.now() }
  return latestByOpenid
  })()
  try { return await _docsCachePromise } finally { _docsCachePromise = null }
}

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())}`
  const thisMonth = today.substring(0, 7)
  // 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())

  let prefix, exact
  if (period === 'day')        { exact = today;   prefix = null }
  else if (period === 'month') { prefix = thisMonth; exact = null }
  else                         { prefix = thisYear;  exact = null }

  const userMap = new Map()

  // 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 || {}
    let duration = 0
    let sessions = 0

    for (const monthKey of Object.keys(records)) {
      if (prefix && !monthKey.startsWith(prefix)) continue
      const arr = records[monthKey]
      if (!Array.isArray(arr)) continue
      for (const r of arr) {
        // 防御脏数据:r.date 缺失/非字符串、duration 非数字时跳过,
        // 避免单条坏记录抛 TypeError 导致整个排行榜对所有人失败
        if (!r || typeof r.date !== 'string') continue
        const dur = Number(r.duration) || 0
        if (period === 'day') {
          // Use startsWith rather than === to tolerate date strings with
          // trailing time/zone info (e.g. "2026-06-11T09:11:24.587Z" if
          // some legacy path stored an ISO date) and to be timezone-
          // agnostic when client local date differs from server UTC date.
          if (r.date.startsWith(exact)) { duration += dur; sessions++ }
        } else {
          if (r.date.startsWith(prefix)) { duration += dur; sessions++ }
        }
      }
    }

    if (duration > 0) {
      const profile = doc.profile || {}
      userMap.set(openid, {
        openid,
        duration,
        sessions,
        nickname: profile.nickname || '',
        avatarUrl: profile.avatarUrl || ''
      })
    }
  }

  // Full sorted list (for ranking)
  const allSorted = Array.from(userMap.values())
    .sort((a, b) => b.duration - a.duration)

  // My entry always included, even if outside top N
  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.
  const _displayName = (entry) =>
    (entry && entry.nickname && entry.nickname.trim()) ||
    maskOpenid(entry && entry.openid)

  // Top N for the board
  const ranked = allSorted.slice(0, limit).map((item, i) => ({
    rank: i + 1,
    isMe: item.openid === myOpenid,
    nickname: item.nickname || '',
    name: _displayName(item),
    duration: item.duration,
    sessions: item.sessions,
    avatarUrl: item.avatarUrl || ''
  }))

  // Pre-resolve cloud:// avatar fileIDs into temporary HTTPS URLs so the
  // client <image> doesn't pay a getTempFileURL round-trip at render time
  // (that lazy round-trip is what made avatars lag 1-2s behind the text).
  // 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(myEntryRaw ? [myEntryRaw.avatarUrl] : [])
  const _resolved = await _resolveAvatars(_avatarUrls)
  _resolved.forEach((url, i) => {
    if (i < ranked.length) ranked[i].avatarUrl = url
    else if (myEntryRaw) myEntryRaw.avatarUrl = url
  })

  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)
  }
}

/**
 * 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 _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
  }
}

/**
 * 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 })
}