从开/关活动日志计算总活动时间
Posted
技术标签:
【中文标题】从开/关活动日志计算总活动时间【英文标题】：Calculating total active time from on/off activity log 【发布时间】：2016-07-08 14:50:34 【问题描述】：假设我有一个支持系统。当代理可以处理票证时，他们会自行开启。当它们不再可用时，它们会自行关闭。非常简单。每次有人打开或关闭自己时，该操作都存储在数据库表中。像这样：
USERID ACTION CREATED
1 ON 2016-01-10 12:00
2 ON 2016-01-10 13:00
2 OFF 2016-01-10 15:00
1 OFF 2016-01-10 17:00
1 ON 2016-01-11 10:00
1 OFF 2016-01-11 11:00
在上面的示例中，用户 1 总共活动了 6 小时。用户 2 总共活动了 2 小时。一个人如何编写一个查询来给我这些数据，像这样：
USERID TOTAL
1 6 hours
2 2 hours
查询还需要处理用户已打开自己但仍处于活动状态的情况（仅记录了一个 ON 操作，没有相应的 OFF）。
Tbh 我什至不知道从哪里开始。我在想我可能需要创建一系列时间戳，然后用它做一些事情。或者也许这只是通过编写代码更容易处理的那种情况。无论如何，任何帮助表示赞赏。
【问题讨论】：
检查LAG()和LEAD() analityc函数。
【参考方案1】：
设置：
create table a_table (userid int, action text, created timestamp);
insert into a_table values
(1, 'ON', '2016-01-10 12:00'),
(2, 'ON', '2016-01-10 13:00'),
(2, 'OFF', '2016-01-10 15:00'),
(1, 'OFF', '2016-01-10 17:00'),
(1, 'ON', '2016-01-11 10:00'),
(1, 'OFF', '2016-01-11 11:00'),
(1, 'ON', '2016-01-11 20:00');-- added
想法：
select distinct on (a1.userid, a1.created)
a1.userid, a1.action, a1.created,
a2.userid, a2.action, a2.created,
coalesce(a2.created, '2016-01-11 24:00:00')- a1.created as total -- '2016-01-11 24:00:00' = the end of reported period
from a_table a1
left join a_table a2
on a1.userid = a2.userid and a1.action = 'ON' and a2.action = 'OFF' and a1.created < a2.created
where a1.action = 'ON'
order by a1.userid, a1.created, a2.created;
userid | action | created | userid | action | created | total
--------+--------+---------------------+--------+--------+---------------------+----------
1 | ON | 2016-01-10 12:00:00 | 1 | OFF | 2016-01-10 17:00:00 | 05:00:00
1 | ON | 2016-01-11 10:00:00 | 1 | OFF | 2016-01-11 11:00:00 | 01:00:00
1 | ON | 2016-01-11 20:00:00 | | | | 04:00:00
2 | ON | 2016-01-10 13:00:00 | 2 | OFF | 2016-01-10 15:00:00 | 02:00:00
(4 rows)
查询：
select userid, sum(total)
from (
select distinct on (a1.userid, a1.created)
a1.userid,
coalesce(a2.created, '2016-01-11 24:00:00')- a1.created as total
from a_table a1
left join a_table a2
on a1.userid = a2.userid and a1.action = 'ON' and a2.action = 'OFF' and a1.created < a2.created
where a1.action = 'ON'
order by a1.userid, a1.created, a2.created
) s
group by 1
order by 1;
userid | sum
--------+----------
1 | 10:00:00
2 | 02:00:00
(2 rows)
【讨论】：
【参考方案2】：SQL Fiddle Demo
我为user_id = 3 添加了另一行 ON 但不是 OFF
nextAction 并不是真的需要它，但请帮忙看看LEAD 行是什么
基本查询：
SELECT "USERID", "ACTION", "CREATED",
LEAD("CREATED") OVER (PARTITION BY "USERID" ORDER BY "CREATED") nextDate,
LEAD("ACTION") OVER (PARTITION BY "USERID" ORDER BY "CREATED") nextAction
FROM activity
输出
| USERID | ACTION | CREATED | nextDate | nextAction |
|--------|--------|---------------------------|---------------------------|------------|
| 1 | ON | January, 10 2016 12:00:00 | January, 10 2016 17:00:00 | OFF |
| 1 | OFF | January, 10 2016 17:00:00 | January, 11 2016 10:00:00 | ON |
| 1 | ON | January, 11 2016 10:00:00 | January, 11 2016 11:00:00 | OFF |
| 1 | OFF | January, 11 2016 11:00:00 | (null) | (null) |
| 2 | ON | January, 10 2016 13:00:00 | January, 10 2016 15:00:00 | OFF |
| 2 | OFF | January, 10 2016 15:00:00 | (null) | (null) |
| 3 | ON | July, 08 2016 05:00:00 | (null) | (null) |
最终查询：
COALESCE(nextDate, NOW()) 用于不带OFF 的ON
SELECT "USERID",
SUM (COALESCE(nextDate, NOW()) - "CREATED") onTime
FROM (
SELECT "USERID", "ACTION", "CREATED",
LEAD("CREATED") OVER (PARTITION BY "USERID" ORDER BY "CREATED") nextDate,
LEAD("ACTION") OVER (PARTITION BY "USERID" ORDER BY "CREATED") nextaction
FROM activity
) T
WHERE "ACTION" = 'ON'
GROUP BY "USERID"
输出
| USERID | ontime |
|--------|-------------------------------------------------------|
| 1 | 0 years 0 mons 0 days 6 hours 0 mins 0.00 secs |
| 2 | 0 years 0 mons 0 days 2 hours 0 mins 0.00 secs |
| 3 | 0 years 0 mons 0 days 10 hours 47 mins 52.920889 secs |
【讨论】：
所有答案都可以，但我发现这是最优雅的解决方案。谢谢。【参考方案3】：您可以使用如下查询：
SELECT USERID,
CASE WHEN COUNT(*) = 1 THEN NOW() ELSE MAX(CREATED) END - MIN(CREATED) onTime
FROM (
SELECT USERID, ACTION, CREATED,
COUNT(CASE WHEN ACTION = 'ON' THEN 1 END) OVER
(PARTITION BY USERID ORDER BY CREATED) AS grp
FROM mytable) AS t
GROUP BY USERID, grp
Demo here
【讨论】：
【参考方案4】：测试表/数据：
drop table if exists a_table;
create table a_table (seq serial primary key, userid int, action text, created timestamp);
create index a_table_i on a_table (userid, action, created);
insert into a_table (userid, action, created) values
(1, 'ON' , '2016-01-10 12:00'),
(1, 'OFF' , '2016-01-10 17:00'),
(1, 'ON' , '2016-01-11 10:00'),
(1, 'OFF' , '2016-01-11 11:00'),
(1, 'ON' , '2016-01-11 20:00'),
(1, 'ON' , '2016-01-11 21:00'), -- an "ON" without an "OFF"
(1, 'OFF' , '2016-01-11 21:00'),
(1, 'OFF' , '2016-01-11 22:00'), -- an "OFF" without an "ON"
(1, 'ON' , '2016-01-11 21:10'),
(1, 'ON' , '2016-01-11 21:20'), -- an "ON" without an "OFF"
(1, 'OFF' , '2016-01-11 21:30'),
(2, 'ON' , '2016-01-10 13:00'),
(2, 'OFF' , '2016-01-10 15:00'),
(2, 'ON' , '2016-01-11 13:00'),
(2, 'OFF' , '2016-01-12 15:00');
查询表，记住可以存在许多“ON”用户操作而没有任何“OFF”（反之亦然）：
SELECT USERID, SUM(FINISHED - CREATED) AS ACTIVE_TIME
FROM (
SELECT *,
-- GET THE NEXT 'OFF' ACTION RIGTH AFTER IT
COALESCE((SELECT CREATED FROM a_table T1 WHERE T1.USERID = T.USERID AND T1.SEQ > T.SEQ ORDER BY T1.SEQ LIMIT 1),CURRENT_TIMESTAMP) AS FINISHED
FROM a_table T
-- LIST ONLY RECORDS WHERE ACTION IS 'ON' AND THE USER'S ACTION RIGHT BEFORE WAS 'OFF' OR 'NONE'
WHERE ACTION = 'ON'
AND COALESCE((SELECT ACTION FROM a_table T1 WHERE T1.USERID = T.USERID AND T1.SEQ < T.SEQ ORDER BY T1.SEQ DESC LIMIT 1),'OFF') = 'OFF'
) A
GROUP BY USERID
结果：
userid active_time
1 07:10:00
2 1 day 04:00:00
【讨论】：
以上是关于从开/关活动日志计算总活动时间的主要内容，如果未能解决你的问题，请参考以下文章
Android：根据列表视图中点击的项目计算价格。应用程序在总活动 Logcat 中崩溃：无效浮点数：“”