feat(insights): 箱主数据面板——访问量、提问趋势、处理漏斗

新增 /user/insights，让箱主第一次能看到「有多少人来过我的箱、提问处理得怎么样」。数据来源： - 新表 box_views：一行 = 一访客×一天×一箱，views 累加，唯一索引 (user_id, viewed_on, visitor)。刻意不内嵌 dbutil.Model —— 软删除 + 唯一索引会让被删行永久占槽，且删过的槽位再也插不进去。 - 访客标识 = HMAC(日盐, IP) 前 32 hex：同日可去重（UV 准）、跨日不可关联（隐私）。代价是近 7/30 天 UV = 各日 UV 之和，页面上已写明口径。写入是异步的：buffered channel 512 + 2s 批量 flush，队列满即丢（统计尽力而为，不该因为记一笔访问就拖慢页面）。过滤箱主本人、空 UA、爬虫 UA、预取请求。密钥复用 server.xsrf_key；每日清理 400 天前的明细。页面（纯 CSS 画图，不引图表库）： - 今日/近 7 天/近 30 天 KPI + 上周对比 - 30 天 PV/UV 双柱趋势、24 小时时段分布 - 提问处理漏斗 + 回答率进度条 + 自动结论聚合层 internal/db/box_analytics.go 复用 analyticsDialectFor 的时区处理，与全站 stats 分开（那个不含单箱维度、按自然日缓存）。测试：box_views 4 项（含「同批同键先内存合并」——PG 的 ON CONFLICT 不允许同条 INSERT 二次命中同一行）、insights 的 pct/窗口切分/visitorID/爬虫过滤。
2026-10-29 18:25:53 +08:00
parent 5a05a61647
commit 4e574ac64a
19 changed files with 2255 additions and 2 deletions
@@ -0,0 +1,221 @@
+// Copyright 2026 tamakyi. All rights reserved.
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file.
+
+package db
+
+import (
+	"context"
+	"fmt"
+	"time"
+
+	"github.com/pkg/errors"
+	"gorm.io/gorm"
+)
+
+// BoxAnalyticsStore 是「箱主数据面板」里提问侧的聚合查询（访问侧见 BoxViewsStore）。
+//
+// 与 SiteAnalyticsStore 分开：那个是站方看的全站数据（不含任何单箱维度），
+// 这个是箱主看自己的。两边的 SQL 都按 user_id 走，混在一个接口里会让
+// 「全站聚合整天缓存」和「箱主实时看自己的」两种缓存策略纠缠在一起。
+type BoxAnalyticsStore interface {
+	// BoxTrend 返回 [from, to] 内每天的提问数与回答数。只返回有数据的天，缺日由上层补零。
+	BoxTrend(ctx context.Context, userID uint, from, to time.Time) ([]BoxDayCount, error)
+	// BoxHours 返回 since 之后该箱收到提问的 24 小时分布（下标 0..23）。
+	BoxHours(ctx context.Context, userID uint, since time.Time) ([24]int64, error)
+	// BoxFunnel 返回该箱的提问处理漏斗。
+	BoxFunnel(ctx context.Context, userID uint) (*BoxFunnel, error)
+}
+
+// BoxDayCount 某一天的提问 / 回答计数。
+type BoxDayCount struct {
+	Day      time.Time `gorm:"column:day"`
+	Questions int64    `gorm:"column:questions"`
+	Answers   int64    `gorm:"column:answers"`
+}
+
+// BoxFunnel 箱主的提问处理漏斗与响应速度。
+type BoxFunnel struct {
+	Total      int64 // 全部提问
+	Answered   int64 // 已回答
+	ReadOnly   int64 // 已读未答
+	Unread     int64 // 未读未答
+	Private    int64 // 私密提问
+	Anonymous  int64 // 匿名（未登录）提问
+	Pinned     int64 // 置顶
+	// 响应速度只统计已回答且 answered_at 有值的行（历史行退回 updated_at，
+	// 会有偏高，见 answered_at 的注释）。
+	RepliedIn24h int64
+	AvgReplyHour float64
+	// LastQuestionAt / LastAnswerAt 用于「你多久没看提问箱了」这类提示。
+	LastQuestionAt *time.Time
+	LastAnswerAt   *time.Time
+}
+
+var BoxAnalytics BoxAnalyticsStore
+
+var _ BoxAnalyticsStore = (*boxAnalytics)(nil)
+
+func NewBoxAnalyticsStore(db *gorm.DB) BoxAnalyticsStore {
+	return &boxAnalytics{db}
+}
+
+type boxAnalytics struct {
+	*gorm.DB
+}
+
+func (db *boxAnalytics) BoxTrend(ctx context.Context, userID uint, from, to time.Time) ([]BoxDayCount, error) {
+	d := analyticsDialectFor(db.DB, from)
+	from, to = dayOnly(from), dayOnly(to).AddDate(0, 0, 1)
+
+	// 提问与回答分两条 SQL 再在内存里合并：一条 SQL 里同时按 created_at 和
+	// answered_at 分组要写成两条子查询再 FULL JOIN，MySQL 没有 FULL JOIN，
+	// 而两张结果集都很小，合并成本可以忽略。
+	type row struct {
+		Key string
+		N   int64
+	}
+
+	var qs, as []row
+	if err := db.WithContext(ctx).Model(&Question{}).
+		Select(d.day+" AS key, COUNT(*) AS n").
+		Where("user_id = ? AND created_at >= ? AND created_at < ?", userID, from, to).
+		Group(d.day).
+		Scan(&qs).Error; err != nil {
+		return nil, errors.Wrap(err, "aggregate box questions by day")
+	}
+	if err := db.WithContext(ctx).Model(&Question{}).
+		Select(d.day+" AS key, COUNT(*) AS n").
+		Where("user_id = ? AND answer <> '' AND COALESCE(answered_at, updated_at) >= ? AND COALESCE(answered_at, updated_at) < ?",
+			userID, from, to).
+		Group(d.day).
+		Scan(&as).Error; err != nil {
+		return nil, errors.Wrap(err, "aggregate box answers by day")
+	}
+
+	// 用 map 合并，最后仍交给上层按日期补零（上层最清楚要展示哪些天）。
+	byDay := make(map[string]*BoxDayCount, len(qs))
+	order := make([]string, 0, len(qs)+len(as))
+	get := func(key string) *BoxDayCount {
+		if v, ok := byDay[key]; ok {
+			return v
+		}
+		t, err := time.ParseInLocation("2006-01-02", key, from.Location())
+		if err != nil {
+			t = from
+		}
+		v := &BoxDayCount{Day: t}
+		byDay[key] = v
+		order = append(order, key)
+		return v
+	}
+	for _, r := range qs {
+		get(r.Key).Questions = r.N
+	}
+	for _, r := range as {
+		get(r.Key).Answers = r.N
+	}
+
+	out := make([]BoxDayCount, 0, len(order))
+	for _, key := range order {
+		out = append(out, *byDay[key])
+	}
+	return out, nil
+}
+
+func (db *boxAnalytics) BoxHours(ctx context.Context, userID uint, since time.Time) ([24]int64, error) {
+	var out [24]int64
+	d := analyticsDialectFor(db.DB, since)
+
+	var rows []struct {
+		Hr int
+		N  int64
+	}
+	if err := db.WithContext(ctx).Model(&Question{}).
+		Select(d.hour+" AS hr, COUNT(*) AS n").
+		Where("user_id = ? AND created_at >= ?", userID, since).
+		Group(d.hour).
+		Scan(&rows).Error; err != nil {
+		return out, errors.Wrap(err, "aggregate box question hours")
+	}
+	for _, r := range rows {
+		if r.Hr >= 0 && r.Hr < 24 {
+			out[r.Hr] = r.N
+		}
+	}
+	return out, nil
+}
+
+func (db *boxAnalytics) BoxFunnel(ctx context.Context, userID uint) (*BoxFunnel, error) {
+	out := &BoxFunnel{}
+
+	var totals struct {
+		Total     int64      `gorm:"column:total"`
+		Answered  int64      `gorm:"column:answered"`
+		ReadOnly  int64      `gorm:"column:read_only"`
+		Unread    int64      `gorm:"column:unread"`
+		Private   int64      `gorm:"column:private"`
+		Anonymous int64      `gorm:"column:anonymous"`
+		Pinned    int64      `gorm:"column:pinned"`
+		LastQ     *time.Time `gorm:"column:last_q"`
+		LastA     *time.Time `gorm:"column:last_a"`
+	}
+	if err := db.WithContext(ctx).Model(&Question{}).
+		Select(`COUNT(*) AS total,
+			COALESCE(SUM(CASE WHEN answer <> '' THEN 1 ELSE 0 END), 0) AS answered,
+			COALESCE(SUM(CASE WHEN answer = '' AND is_read THEN 1 ELSE 0 END), 0) AS read_only,
+			COALESCE(SUM(CASE WHEN answer = '' AND NOT is_read THEN 1 ELSE 0 END), 0) AS unread,
+			COALESCE(SUM(CASE WHEN is_private THEN 1 ELSE 0 END), 0) AS private,
+			COALESCE(SUM(CASE WHEN asker_user_id = 0 THEN 1 ELSE 0 END), 0) AS anonymous,
+			COALESCE(SUM(CASE WHEN is_pinned THEN 1 ELSE 0 END), 0) AS pinned,
+			MAX(created_at) AS last_q,
+			MAX(CASE WHEN answer <> '' THEN COALESCE(answered_at, updated_at) END) AS last_a`).
+		Where("user_id = ?", userID).
+		Scan(&totals).Error; err != nil {
+		return nil, errors.Wrap(err, "aggregate box funnel")
+	}
+	out.Total = totals.Total
+	out.Answered = totals.Answered
+	out.ReadOnly = totals.ReadOnly
+	out.Unread = totals.Unread
+	out.Private = totals.Private
+	out.Anonymous = totals.Anonymous
+	out.Pinned = totals.Pinned
+	out.LastQuestionAt = totals.LastQ
+	out.LastAnswerAt = totals.LastA
+
+	// 响应速度：answered_at 为空的历史行退回 updated_at（口径偏大，只用于展示大致量级）。
+	replyMinutes := boxReplyMinutesExpr(db.DB, out.LastAnswerAt)
+	if replyMinutes == "" {
+		return out, nil
+	}
+	var speed struct {
+		In24h int64      `gorm:"column:in_24h"`
+		Avg   *float64   `gorm:"column:avg_min"`
+	}
+	if err := db.WithContext(ctx).Model(&Question{}).
+		Select(fmt.Sprintf(`COALESCE(SUM(CASE WHEN %[1]s <= 1440 THEN 1 ELSE 0 END), 0) AS in_24h,
+			AVG(%[1]s) AS avg_min`, replyMinutes)).
+		Where("user_id = ? AND answer <> ''", userID).
+		Scan(&speed).Error; err != nil {
+		return nil, errors.Wrap(err, "aggregate box reply speed")
+	}
+	out.RepliedIn24h = speed.In24h
+	if speed.Avg != nil && *speed.Avg > 0 {
+		out.AvgReplyHour = *speed.Avg / 60
+	}
+	return out, nil
+}
+
+// boxReplyMinutesExpr 返回「回答耗时（分钟）」的 SQL 表达式。
+//
+// 与 analyticsDialectFor 里的 replyMin 口径一致，但这里必须能固定到箱主自己那张表：
+// 表达式里出现的列名都来自 questions 表，不需要额外限定。
+func boxReplyMinutesExpr(db *gorm.DB, at *time.Time) string {
+	now := time.Now()
+	if at != nil {
+		now = *at
+	}
+	d := analyticsDialectFor(db, now)
+	return d.replyMin
+}