Question

In my app articles have many children or parent articles through a self referential joining model article_relationships

class Article < ActiveRecord::Base

  has_many  :parent_child_relationships,
            :class_name   => "ArticleRelationship",
            :foreign_key  => :child_id,
            :dependent    => :destroy
  has_many  :parents,
            :through    => :parent_child_relationships,
            :source     => :parent

  has_many  :child_parent_relationships,
            :class_name   => "ArticleRelationship",
            :foreign_key  => :parent_id,
            :dependent    => :destroy
  has_many  :children,
            :through    => :child_parent_relationships,
            :source     => :child
end

class ArticleRelationship < ActiveRecord::Base
  belongs_to :parent, :class_name => "Article"
  belongs_to :child, :class_name => "Article"
end

I’ve got a fairly complex query on article_relationships that digs down into the article table

ArticleRelationship.joins(:parent, :child).where("((articles.type IN (:parent_article_type) AND parent_id IN (:ids)) OR (children_article_relationships.type IN (:child_article_type) AND child_id IN (:ids)) AND (article_relationships.created_at > :date OR article_relationships.updated_at > :date))", {:parent_article_type => "Emotion", :child_article_type => "Gateway", :ids => [1,2,3,4], :date => "2010-01-01"})

Is there any way I can index this effectively?

Answer 1

So, just for readability,

ArticleRelationship.joins(:parent, :child).
 where("((articles.type IN (:parent_article_type) AND parent_id IN (:ids)) OR
         (children_article_relationships.type IN (:child_article_type) AND child_id IN (:ids))
        AND (article_relationships.created_at > :date OR article_relationships.updated_at > :date))",
       { :parent_article_type => "Emotion",
         :child_article_type => "Gateway",
         :ids => [1,2,3,4], :date => "2010-01-01" })

The problem is the OR operators. Because of the OR, in both of your top level AND expressions, the database cannot use an index.

1. Since, when a record is created, it also sets the updated_at field to the same timestamp, you can remove the OR created_at, allowing you to index on update_at.
2. If you can filter the parent ids and child ids by article type, you can remove those ORs. However, make sure you can do so efficiently enough to not spend more time filtering the id lists as input, that it doesn’t take longer than the unindexed query. Try 1., first, and then see how long it takes with an updated_at index.

If you do both 1 and 2, be sure to create one index with all three fields, updated_at, parent_id, and child_id.