Question

我有两个表：计算机和消息。这是一对多的关系;每台计算机有多条消息。

create table computers (
  `computer` varchar(45) not null,
  `status` varchar(25),
  primary key (`computer`)
  );

INSERT INTO computers (`computer`, `status`)
VALUES
('fred','completed'),
('barney','incomplete'),
('wilma','completed');

create table messages (
`id` int(11) NOT NULL AUTO_INCREMENT,
  `dateCreated` datetime NOT NULL,
  `computer` varchar(45) NOT NULL,
  `message` text,
  PRIMARY KEY (`id`)
);

INSERT INTO messages (datecreated, computer,message) 
VALUES
(now(),'fred','start something'),
(now(),'fred','doing something'),
(now(),'fred','end something'),
(now(),'barney','start something'),
(now(),'barney','doing something'),
(now(),'wilma','start something'),
(now(),'wilma','doing something'),
(now(),'wilma','end something');

我试图在每台计算机上获得一行，显示开始和结束消息的状态和日期。

fred  | complete   | 2018-01-29 08:00 | 2018-01-29 08:20
wilma | incomplete | 2018-01-29 08:10 | null

现在我正在使用子查询为每台计算机查询消息表两次。我知道这不是最好的方法。当有100台计算机时，它会固定数据库上的CPU。我试过工会并加入无济于事。

select C.computer, C.status,
  (select max(M.datecreated) from messages M where C.computer=M.computer and M.message like 'start%') as date_start,
  (select max(M.datecreated) from messages M where C.computer=M.computer and M.message like 'complete%') as date_complete
from computers C
order by computer;

这是一个相关的SQL小提琴：http://sqlfiddle.com/#!9/f5c929/1

Answer 1

您可以使用条件聚合来获取这些日期：

select computer,
   -- in case of multiple rows with start you might better use MIN for the start date
   max(case when message like 'start%' then datecreated end) as date_start,
   max(case when message like 'end%'   then datecreated end) as date_complete
from messages
group by computer

请参阅Fiddle

现在将其放在派生表中并将其加入computers

Answer 2

您可以使用GROUP BY和LEFT JOIN

select *
from computers C
left join
(
  select M.computer, max(M.datecreated) date_start
  from messages M
  where M.message like 'start%'
  group by M.computer
) s on C.computer = s.computer
left join
(
  select M.computer, max(M.datecreated) date_end
  from messages M
  where M.message like 'end%'
  group by M.computer
) e on C.computer = e.computer;

dbfiddle demo

从效率角度来看，您还需要索引messages(computer, message, datecreated)

Answer 3

左连接和最大

select C.computer, C.status,
  max(ms.datecreated) as date_start,
  max(me.datecreated) as date_complete
from computers C
left outer join messages ms on 
  ms.computer = c.computer
  and ms.message like 'start%'
left outer join messages me on
  me.computer = c.computer
  and me.message like 'end%'
group by c.computer, c.status 
order by c.computer;

Answer 4

条件聚合对此非常有用：

SELECT C.computer, C.status
     , MAX(CASE WHEN M.message LIKE 'start%' THEN M.datecreated ELSE NULL END) AS date_start
     , MAX(CASE WHEN M.message LIKE 'complete%' THEN M.datecreated ELSE NULL END) AS date_complete
FROM computers C
LEFT JOIN messages M ON C.computer = M.computer
GROUP BY C.computer, C.status
ORDER BY computer;

从技术上讲，ELSE NULL是多余的（省略的ELSE子句总是返回NULL）;但说明了意图。聚合函数忽略NULL值，因此表达式将过滤为适当的datecreated值。

Answer 5

您可以在gropued查询中使用几个内部联接来获取已启动和已完成的最大日期

select c.computer, c.status, t1.max_started, t2.max_finished
from computers c
inner join (

  select max(M.datecreated) max_started, computer
  from messages
  where message = 'started' 
  group by computer

) t1 on t1.computer = c.computer
inner join (

  select max(M.datecreated) max_finished, computer
  from messages
  where message = 'completed' 
  group by computer
) t2 on t2.computer = c.computer

一对多关系。必须是比使用子查询更好的方法

5 个答案: