sql problem,challenge

I want to get

id   a    b     c
------------开发者_Python百科--------
1    1    100   90
6    2    50    100

...from:

id   a    b     c
--------------------
1    1    100   90
2    1    300   50
3    1    200   20
4    2    200   30
5    2    300   70
6    2    50    100

It's the row with the smallest b group by a.

How to do it with sql?

EDIT

I thought it can be achieved by

select * from table group by a having min(b);

which I found later it's wrong.

But is it possible to do it with having statement?

I'm using MySQL

SELECT t1.*
FROM mytable t1
LEFT OUTER JOIN mytable t2
  ON (t1.a=t2.a AND t1.b>t2.b)
WHERE t2.a IS NULL;

This works because there should be no matching row t2 with the same a and a lesser b.

update: This solution has the same issue with ties that other folks have identified. However, we can break ties:

SELECT t1.*
FROM mytable t1
LEFT OUTER JOIN mytable t2
  ON (t1.a=t2.a AND (t1.b>t2.b OR t1.b=t2.b AND t1.id>t2.id))
WHERE t2.a IS NULL;

Assuming for instance that in the case of a tie, the row with the lower id should be the row we choose.

This doesn't do the trick:

select * from table group by a having min(b);

Because HAVING MIN(b) only tests that the least value in the group is not false (which in MySQL means not zero). The condition in a HAVING clause is for excluding groups from the result, not for choosing the row within the group to return.

In MySQL:

select t1.* from test as t1
inner join
(select t2.a, min(t2.b) as min_b from test as t2 group by t2.a) as subq
on subq.a=t1.a and subq.min_b=t1.b;

Here is the proof:

mysql> create table test (id int unsigned primary key auto_increment, a int unsigned not null, b int unsigned not null, c int unsigned not null) engine=innodb;
Query OK, 0 rows affected (0.55 sec)

mysql> insert into test (a,b,c) values (1,100,90), (1,300,50), (1,200,20), (2,200,30), (2,300,70), (2,50,100);
Query OK, 6 rows affected (0.39 sec)
Records: 6  Duplicates: 0  Warnings: 0

mysql> select * from test;
+----+---+-----+-----+
| id | a | b   | c   |
+----+---+-----+-----+
|  1 | 1 | 100 |  90 |
|  2 | 1 | 300 |  50 |
|  3 | 1 | 200 |  20 |
|  4 | 2 | 200 |  30 |
|  5 | 2 | 300 |  70 |
|  6 | 2 |  50 | 100 |
+----+---+-----+-----+
6 rows in set (0.00 sec)

mysql> select t1.* from test as t1 inner join (select t2.a, min(t2.b) as min_b from test as t2 group by t2.a) as subq on subq.a=t1.a and subq.min_b=t1.b;
+----+---+-----+-----+
| id | a | b   | c   |
+----+---+-----+-----+
|  1 | 1 | 100 |  90 |
|  6 | 2 |  50 | 100 |
+----+---+-----+-----+
2 rows in set (0.00 sec)

Use:

SELECT DISTINCT
       x.*
  FROM TABLE x
  JOIN (SELECT t.a,
               MIN(t.b) 'min_b'
          FROM TABLE T
      GROUP BY t.a) y ON y.a = x.a
                     AND y.min_b = x.b

You're right. select min(b), a from table group by a. If you want the entire row, then you've use analytics function. That depends on database s/w.

It depends on the implementation, but this is usually faster than the self-join method:

SELECT id, a, b, c
FROM
    (
        SELECT id, a, b, c
        , ROW_NUMBER() OVER(PARTITION BY a ORDER BY b ASC) AS [b IN a]
    ) As SubqueryA
WHERE [b IN a] = 1

Of course it does require that you SQL implementation be fairly up-to-date with the standard.