MySQL grouping with detail - mysql

Question

I have a table that looks like this...
user_id, match_id, points_won
1 14 10
1 8 12
1 12 80
2 8 10
3 14 20
3 2 25
I want to write a MYSQL script that pulls back the most points a user has won in a single match and includes the match_id in the results - in other words...
user_id, match_id, max_points_won
1 12 80
2 8 10
3 2 25
Of course if I didn't need the match_id I could just do...
select user_id, max(points_won)
from table
group by user_id
But as soon as I add match_id to the "select" and "group by" I have a row for every match, and if I only add the match_id to the "select" (and not the "group by") then it won't correctly relate to the points_won.
Ideally I don't want to do the following either because it doesn't feel particularly safe (e.g. if the user has won the same amount of points on multiple matches)...
SELECT t.user_id, max(t.points_won) max_points_won
, (select t2.match_id
from table t2
where t2.user_id = t.user_id
and t2.points_won = max_points_won) as 'match_of_points_maximum'
FROM table t
GROUP BY t.user_id
Are there any more elegant options for this problem?

Answer 1

This is harder than it needs to be in MySQL. One method is a bit of a hack but it works in most circumstances. That is the group_concat()/substring_index() trick:
select user_id, max(points_won),
substring_index(group_concat(match_id order by points_won desc), ',', 1)
from table
group by user_id;
The group_concat() concatenates together all the match_ids, ordered by the points descending. The substring_index() then takes the first one.
Two important caveats:
The resulting expression has a type of string, regardless of the internal type.
The group_concat() uses an internal buffer, whose length -- by default -- is 1,024 characters. This default length can be changed.

Answer 2

You can use the query:
select user_id, max(points_won)
from table
group by user_id
as a derived table. Joining this to the original table gets you what you want:
select t1.user_id, t1.match_id, t2.max_points_won
from table as t1
join (
select user_id, max(points_won) as max_points_won
from table
group by user_id
) as t2 on t1.user_id = t2.user_id and t1.points_won = t2.max_points_won

Answer 3

I think you can optimize your query by add limit 1 in the inner query.
SELECT t.user_id, max(t.points_won) max_points_won
, (select t2.match_id
from table t2
where t2.user_id = t.user_id
and t2.points_won = max_points_won limit 1) as 'match_of_points_maximum'
FROM table t
GROUP BY t.user_id

Answer 4

EDIT : only for postgresql, sql-server, oracle
You could use row_number :
SELECT USER_ID, MATCH_ID, POINTS_WON
FROM
(
SELECT user_id, match_id, points_won, row_number() over (partition by user_id order by points_won desc) rn
from table
) q
where q.rn = 1
For a similar function, have a look at Gordon Linoff's answer or at this article.
In your example, you partition your set of result per user then you order by points_won desc to obtain highest winning point first.