Question

The question may be long but the problem is actually simple. I have 3 models: List, Appearance and Movie. A List has many Movies through an Appearance, and they should be sorted by the attribute rank of the joins model. So what indexes should I be using? This is how my models currently look:

# Models
class Movie < ActiveRecord::Base  
  has_many :appearances, :dependent => :destroy
  has_many :lists, :through => :appearances
end

class Appearance < ActiveRecord::Base
  belongs_to :list
  belongs_to :movie
end

class List < ActiveRecord::Base
  has_many :appearances, :dependent => :destroy
  has_many :movies, :through => :appearances

  def self.find_complete(id)
    List.includes({:appearances => :movie}).where("appearances.rank IS NOT NULL").order("appearances.rank ASC").find(id)
  end
end

And here are the indexes I already have. Do I need a composite index? Or just an index by rank?

# Tables
class CreateAppearances < ActiveRecord::Migration
  def change
    create_table :appearances do |t|
      t.integer :list_id, :null => false
      t.integer :movie_id, :null => false
      t.integer :rank
    end
    add_index :appearances, :list_id
    add_index :appearances, :movie_id
  end
end

Finally, is there any way to refactor that find_complete List method? Like using a default_scope? Don’t forget that a rank may be null.

Answer 1

ActiveRecord is doing two queries.

SELECT DISTINCT `lists`.id
FROM `lists` LEFT OUTER JOIN `appearances` ON `appearances`.`list_id` = `lists`.`id`
             LEFT OUTER JOIN `movies` ON `movies`.`id` = `appearances`.`movie_id`
WHERE `lists`.`id` = 1 AND (appearances.rank IS NOT NULL)
ORDER BY appearances.rank ASC LIMIT 1

and

SELECT `lists`.`id` AS t0_r0, `lists`.`name` AS t0_r1, `lists`.`created_at` AS t0_r2, `lists`.`updated_at` AS t0_r3, `appearances`.`id` AS t1_r0, `appearances`.`list_id` AS t1_r1, `appearances`.`movie_id` AS t1_r2, `appearances`.`rank` AS t1_r3, `appearances`.`created_at` AS t1_r4, `appearances`.`updated_at` AS t1_r5, `movies`.`id` AS t2_r0, `movies`.`name` AS t2_r1, `movies`.`created_at` AS t2_r2, `movies`.`updated_at` AS t2_r3
FROM `lists` LEFT OUTER JOIN `appearances` ON `appearances`.`list_id` = `lists`.`id`
             LEFT OUTER JOIN `movies` ON `movies`.`id` = `appearances`.`movie_id`
WHERE `lists`.`id` = 1 AND `lists`.`id` IN (1) AND (appearances.rank IS NOT NULL)
ORDER BY appearances.rank ASC

In both cases what you need is an index with both lists.id and appearances.rank. Some databases will create indexes on multiple tables. I also don’t know if your database will be smart enough to use an index on appearances.list_id and appearances.rank, but that is worth a try.

add_index :appearances, [:list_id, :rank]

If that doesn’t help, at least index :rank, so the database might be able to avoid the sort.