Question

I’m trying to do a multi-table join that has a NOT IN component. Tables are

Post -> Term Relationship -> Term

Post
  has_many :term_relationships
  has_many :terms, :through => :term_relationships

TermRelationship
  belongs_to :post
  belongs_to :term

Term
  has_many :term_relationships
  has_many :posts, :through => :term_relationships

The goal is to get all posts except for those in “featured” let’s say. My current query would looks like:

WpPost.includes(:terms).where("terms.term NOT IN (?)", ["featured"])

This works great if the only term that it has attached is “featured”. If the post belongs to “featured” and “awesome” it will still show because of “awesome”.

Anyway to exclude a row entirely? Will it require a subquery? And if it does, how would I go about doing that in rails?

Thanks all!
Justin

Answer 1

You misuse the includes. It’s for eager loading, not for joining!

But you’re right about the approach. It can be used in your case. But Rails won’t issue nested request for NOT IN (?) even if it would be logical. You’ll get 2 queries instead (you’ll get NOT IN (id1, id2....,) instead of NOT IN (SELECT ....)).

So I would recommend you to use the squeel gem:

regular AR code (can also be prettified with squeel):

featured_posts = WpPost.joins(:terms).where(terms:{term: ['featured']}).uniq

and then use the sqeel’s power:

WpPost.where{id.not_in featured_posts}

(in and not_in are also aliased as >> and << but I didn’t want to scary anybody)

Note the using blocks and absence of symbols.

---

Some measurements based on Chinook Database under SQLite:

> Track.all
  Track Load (35.0ms)  SELECT "Track".* FROM "Track"

Relation with joins and like:

oldie = Track.joins{playlists}.where{playlists.name.like_any %w[%classic% %90%]}

Here’s NOT IN:

> Track.where{trackId.not_in oldie}.all
  Track Load (37.5ms)  SELECT "Track".* FROM "Track" WHERE "Track"."trackId" 
  NOT IN (SELECT "Track"."TrackId" FROM "Track" INNER JOIN "PlaylistTrack" ON
    "PlaylistTrack"."TrackId" = "Track"."TrackId" INNER JOIN "Playlist" ON 
    "Playlist"."PlaylistId" = "PlaylistTrack"."PlaylistId"
     WHERE (("Playlist"."name" LIKE '%classic%' OR "Playlist"."name" LIKE '%90%')))

FYI:

Track.where{trackId.not_in oldie}.count # => 1971
Track.count # => 3503
# join table:
PlaylistTrack.count # => 8715

Conclusion: I don’t see the overhead caused by NOT IN. 35.0 vs 37.5 isn’t noticeable difference. Few times 35.0 became 37.5 and vice verse.