GROUP_CONCAT with limit

问题:

I have table with player-s in many-to-many relation with skill-s

The goal is to list the players and their "top 3 skills" with a single query.

fiddle

create table player(
  id int primary key
);

create table skill(
  id int primary key,
  title varchar(100)
);

create table player_skills (
  id int primary key,
  player_id int,
  skill_id int,
  value int
);

Query:

SELECT 
p.id,  
group_concat(s.title  SEPARATOR ', ') as skills

FROM player p
LEFT JOIN player_skills ps ON ps.player_id = p.id
LEFT JOIN skill s ON s.id = ps.skill_id

WHERE ps.value > 2
-- skills limit 3 some how ...
group by p.id 
order by s.id


-- expected result
-- player_ID, skills
-- 1 , 'one'
-- 2 , 'one'
-- 3 , 'two, three, four'

As you can see in the fiddle the result of the query is missing only the limit of 3 skills.
I tried several variation of sub queries.. joins and so but with no effect.

回答1:

One somewhat hacky way to do it is to post-process the result of GROUP_CONCAT:

substring_index(group_concat(s.title SEPARATOR ','), ',', 3) as skills

Of course this assumes that your skill names don't contain commas and that their amount is reasonably small.

fiddle

A feature request for GROUP_CONCAT to support an explicit LIMIT clause is unfortunately still not resolved.

UPDATE: As user Strawberry points out, the table player_skills should have the tuple (player_id, skill_id) as its primary key, otherwise the schema allows for the same skill to be assigned to a player multiple times, in which case group_concat would not work as expected.

回答2:

Increase GROUP_CONCAT function length using GLOBAL group_concat_max_len GROUP_CONCAT() maximum length is 1024 characters.
What you can do is to set the GLOBAL group_concat_max_len in mysql

SET GLOBAL group_concat_max_len = 1000000;

Try this and it will work for sure.

回答3:

There is a much cleaner solution. Wrap it inside another SELECT statement.

SELECT GROUP_CONCAT(id) FROM (
    SELECT DISTINCT id FROM people LIMIT 4
) AS ids;

/* Result 134756,134754,134751,134750 */

回答4:

Here's another solution. It includes an arbitrary mechanism for resolving ties, and employes a schema slightly differing from yours...

SELECT a.player_id
     , GROUP_CONCAT(s.title ORDER BY rank) skills
  FROM
     ( SELECT x.*, COUNT(*) rank
         FROM player_skills x
         JOIN player_skills y 
           ON y.player_id = x.player_id
          AND (y.value > x.value
           OR (y.value = x.value AND y.skill_id <= x.skill_id))
        GROUP 
           BY player_id, value, skill_id
       HAVING COUNT(*) <= 3
     ) a
  JOIN skill s
    ON s.skill_id = a.skill_id
 GROUP 
    BY player_id;

http://sqlfiddle.com/#!2/34497/18

Incidentally, if you have a presentation layer/application-level code, then consider doing all the GROUP_CONCAT stuff there. It's more flexible.

回答5:

You can simulate the partitioned row_number using user variables and then limit rows and apply group_concat:

select p.id,
    group_concat(s.title separator ', ') as skills
from player p
left join (
    select distinct ps.player_id,
        ps.skill_id,
        @rn := if(@player_id = player_id, @rn+1, if(@player_id := player_id, 1, 1)) as seqnum
    from player_skills ps
    cross join (select @rn := 0, @player_id := null) x
    where ps.value > 2
    order by player_id, value desc
    ) ps on p.id = ps.player_id and ps.seqnum <= 3
left join skill s on ps.skill_id = s.id
group by p.id;

Demo

This method doesn't require any table to read more than once.