Poor performance on a 'leaderboard' query

Question

Consider the following 2 tables:

players:  | rooms: 
id        | id
name      | type
room_id   | embed_code_id
points    |

My query needs to fetch the 100 players with the most points from a given room type and embed_code_id.

here's what I have currently:

    select `players`.`id`,
       `players`.`points`
from `players`
       inner join `rooms` on `players`.`room_id` = `rooms`.`id`
where `rooms`.`embed_code_id` = 678
  and `rooms`.`type` = 'single'
order by `players`.`points` desc, `players`.`id` asc
limit 100;

This query results in poor performance (1.5 - 10 seconds, depends on the parameters passed)

Notes: * Both tables have about 1M records * when i remove the ORDER BY clause, the time drops below 100ms * I tried setting an index on players.points & players.id, which didn't help * I tried setting an index on embed_code_id & type which also didn't help

The EXPLAIN output suggests Using index; Using temporary; Using filesort which i don't know how to fix:

1   SIMPLE  rooms   NULL    ref PRIMARY,embed_code_id,type  type    6   const,const 478797  100.00  Using index; Using temporary; Using filesort
1   SIMPLE  players NULL    ref room_id room_id 4   playgo.rooms.id 1   100.00  NULL

Can this query be more efficient or i'm missing some index?

Thanks

Answer 1

Converting the query to cartesian form:

select `players`.`id`,
       `players`.`points`
from `players`, `rooms` 
where `players`.`room_id` = `rooms`.`id`
  and `rooms`.`embed_code_id` = 678
  and `rooms`.`type` = 'single'
order by `players`.`points` desc, `players`.`id` asc
limit 100;

allows to understand that you need in indices:

rooms (embed_code_id, type, id) or rooms (type, embed_code_id, id)

players (room_id, id, points) or players (room_id, points, id)

The index on rooms is effective. The index on players is covering.

Answer 2

You need a combined index on rooms(embed_code_id,type).

Only one index can be used at a time. type is a poor index.