我有两张桌子：
表registrations：
event_time | name | --------------------------- 2014-07-16 11:40 Bob1 2014-07-16 10:00 Bob2 2014-07-16 09:20 Bob3 2014-07-15 11:20 Bob4 2014-07-15 10:20 Bob5 2014-07-15 09:00 Bob6
表session_log：
event_time | name | games_played | level_at_end | --------------------------------------------------------- 2014-07-16 11:40 Bob1 12 2 2014-07-16 10:00 Bob2 0 0 2014-07-16 09:20 Bob3 146 9 2014-07-15 11:20 Bob4 11 2 2014-07-15 10:20 Bob5 0 0 2014-07-15 09:00 Bob6 1 0
每次用户login..play..logout我都会将记录写入session_log。因此，每天可能有一个用户的许多条目。
用户可能会注册但不能登录。
我的系统中的级别从0开始。
我需要建立一个这样的报告：
day | registrations | logged_in | played_users | lvl1 | lvl2 | lvl3 | lvl4 | lvl5 | lvl10 ------------------------------------------------------------------------------------------------ 2014-07-29 23 21 14 14 10 9 4 2 0 2014-07-28 18 17 15 14 11 9 3 1 1
其中：
day - 报告聚合日registrations - 指定日期的注册数量logged_in - 注册后24小时内登录的用户数played_users - 注册后24小时内播放的用户数lvl1 - 注册后24小时内达到1级的用户数lvl2 - 注册后24小时内达到2级的用户数lvl3 - 注册后24小时内达到3级的用户数lvl4 - 注册后24小时内达到4级的用户数lvl5 - 注册后24小时内达到5级的用户数lvl10 - 注册后24小时内达到10级的用户数所以，我写了一个这样的查询：
SELECT
date(r.event_time) AS day,
count(DISTINCT r.name) AS registrations,
count(DISTINCT s1.name) AS logged_in,
count(DISTINCT s2.name) AS played_users,
count(DISTINCT sl_1.name) AS lvl1,
count(DISTINCT sl_2.name) AS lvl2,
count(DISTINCT sl_3.name) AS lvl3,
count(DISTINCT sl_4.name) AS lvl4,
count(DISTINCT sl_5.name) AS lvl5,
count(DISTINCT sl_10.name) AS lvl10
FROM registrations AS r
LEFT JOIN session_log AS s1
ON r.name = s1.name
AND s1.event_time >= r.event_time
AND s1.event_time < date_trunc('day', r.event_time + INTERVAL '1 days')
LEFT JOIN session_log as s2
ON r.name = s2.name
AND s2.event_time >= r.event_time
AND s2.event_time < date_trunc('day', r.event_time + INTERVAL '1 days')
AND s2.games_played > 0
LEFT JOIN session_log as sl_1
ON r.name = sl_1.name
AND sl_1.event_time >= r.event_time
AND sl_1.event_time < date_trunc('day', r.event_time + INTERVAL '1 days')
AND sl_1.level_at_end > 0
LEFT JOIN session_log as sl_2
ON r.name = sl_2.name
AND sl_2.event_time >= r.event_time
AND sl_2.event_time < date_trunc('day', r.event_time + INTERVAL '1 days')
AND sl_2.level_at_end > 1
LEFT JOIN session_log as sl_3
ON r.name = sl_3.name
AND sl_3.event_time >= r.event_time
AND sl_3.event_time < date_trunc('day', r.event_time + INTERVAL '1 days')
AND sl_3.level_at_end > 2
LEFT JOIN session_log as sl_4
ON r.name = sl_4.name
AND sl_4.event_time >= r.event_time
AND sl_4.event_time < date_trunc('day', r.event_time + INTERVAL '1 days')
AND sl_4.level_at_end > 3
LEFT JOIN session_log as sl_5
ON r.name = sl_5.name
AND sl_5.event_time >= r.event_time
AND sl_5.event_time < date_trunc('day', r.event_time + INTERVAL '1 days')
AND sl_5.level_at_end > 4
LEFT JOIN session_log as sl_10
ON r.name = sl_10.name
AND sl_10.event_time >= r.event_time
AND sl_10.event_time < date_trunc('day', r.event_time + INTERVAL '1 days')
AND sl_10.level_at_end > 9
WHERE r.event_time >= '2014-07-01'
AND r.event_time < '2014-07-30'
GROUP BY day
ORDER BY day DESC;
它有效，但速度很慢。有没有办法加快这个查询？
答案 0 :(得分：0)
因此，我得到的查询如下：
SELECT
date(reg_time) AS day,
count(glid) AS registrations,
sum(CASE WHEN games_played IS NOT NULL THEN 1 ELSE 0 END) AS logged_in,
sum(CASE WHEN games_played > 0 THEN 1 ELSE 0 END) AS played_users,
sum(CASE WHEN level_at_end > 0 THEN 1 ELSE 0 END) AS lvl1,
sum(CASE WHEN level_at_end > 1 THEN 1 ELSE 0 END) AS lvl2,
sum(CASE WHEN level_at_end > 2 THEN 1 ELSE 0 END) AS lvl3,
sum(CASE WHEN level_at_end > 3 THEN 1 ELSE 0 END) AS lvl4,
sum(CASE WHEN level_at_end > 4 THEN 1 ELSE 0 END) AS lvl5,
sum(CASE WHEN level_at_end > 9 THEN 1 ELSE 0 END) AS lvl10
FROM (SELECT
max(r.event_time) AS reg_time,
r.glid,
max(s.level_at_end) AS level_at_end,
sum(s.games_played) AS games_played
FROM registrations AS r
LEFT JOIN session_log AS s
ON r.glid = s.glid
AND s.event_time >= r.event_time
AND s.event_time < date_trunc('day', r.event_time + INTERVAL '1 day')
GROUP BY r.glid
ORDER BY reg_time DESC) AS foo
GROUP BY day
ORDER BY day DESC;
密钥是嵌套的select，并且left join简单地预先计算了所有需要的值。
答案 1 :(得分：-1)
如何使用CASE？ F.e：
SELECT
date(r.event_time) AS day,
count(DISTINCT r.name) AS registrations,
count(distinct CASE WHEN s.event_time < date_trunc('day', r.event_time + INTERVAL '1 days') THEN s.name ELSE null END) AS logged_in,
count(distinct CASE WHEN s.event_time < date_trunc('day', r.event_time + INTERVAL '1 days') AND s.games_played > 0) THEN s.name ELSE null END) AS played_users
count(distinct CASE WHEN s.event_time < date_trunc('day', r.event_time + INTERVAL '1 days') AND s.level_at_end > 0) THEN s.name ELSE null END) AS lvl1
count(distinct CASE WHEN s.event_time < date_trunc('day', r.event_time + INTERVAL '1 days') AND s.level_at_end > 1) THEN s.name ELSE null END) AS lvl2
count(distinct CASE WHEN s.event_time < date_trunc('day', r.event_time + INTERVAL '1 days') AND s.level_at_end > 2) THEN s.name ELSE null END) AS lvl3
count(distinct CASE WHEN s.event_time < date_trunc('day', r.event_time + INTERVAL '1 days') AND s.level_at_end > 3) THEN s.name ELSE null END) AS lvl4
count(distinct CASE WHEN s.event_time < date_trunc('day', r.event_time + INTERVAL '1 days') AND s.level_at_end > 4) THEN s.name ELSE null END) AS lvl5
count(distinct CASE WHEN s.event_time < date_trunc('day', r.event_time + INTERVAL '1 days') AND s.level_at_end > 5) THEN s.name ELSE null END) AS lvl10
FROM registrations AS r
LEFT JOIN session_log AS s
ON r.name = s.name
AND s.event_time >= r.event_time
WHERE r.event_time >= '2014-07-01'
AND r.event_time < '2014-07-30'
GROUP BY day
ORDER BY day DESC;