You could add an index without the proper foreign key constraint, or vice versa, but without looking at the data model my guess is that having both is probably best in this scenario.
You can't easily add a foreign key between reactions and articles because of the "polymorphic" nature of these relation. A reaction can be attached to a comment, to an article and other stuff I guess. So what Rails does it setup a pair of fields called reactable_id and reactable_type like this:
PracticalDeveloper_development=# \d reactions
Column | Type | Collation | Nullable | Default
id | integer | | not null | nextval('reactions_id_seq'::regclass)
category | character varying | | |
created_at | timestamp without time zone | | not null |
points | double precision | | | 1.0
reactable_id | integer | | |
reactable_type | character varying | | |
updated_at | timestamp without time zone | | not null |
user_id | integer | | |
"reactions_pkey" PRIMARY KEY, btree (id)
"index_reactions_on_category" btree (category)
"index_reactions_on_reactable_id" btree (reactable_id)
"index_reactions_on_reactable_type" btree (reactable_type)
"index_reactions_on_user_id" btree (user_id)
On top of those it mounts what it calles polymorphic associations. The gist is that every select triggered by something like article.reactions becomes a SELECT * FROM reactions WHERE reactable_type = 'Article' AND reactable_id = 1234, same for the other queries
SELECT * FROM reactions WHERE reactable_type = 'Article' AND reactable_id = 1234
Gross. The site runs on Postgres though iirc, so you could establish a foreign key relationship between reactions and reactables as a parent table extended by articles, posts, and so on. But I don't know if ActiveRecord would play nicely with that kind of specialized structure.
No you can't use Postgresql inheritance in Rails, not easily.
There's a concept of single table inheritance but it's all virtual. The various entities share the same table and are distinguished by a type column. You can find the explanation here medium.com/@dcordz/single-table-in... - I've used it once and it was used to map entities with differed only in name (sensors with a uniform API)
Another option is a complex junction table with reaction_id, article_id, post_id, and a CHECK constraint ensuring that only one reactable foreign key can have a value. But that does add enough complexity to at least give me pause.
The indexing in the current model could still be improved since reactable_id is unreliable on its own and reactable_type is low-cardinality. A single index on (reactable_type, reactable_id) would be much more useful (including for bulk delete!).
Yeah, I think the index on both keys is a good comprise
It might be nice longterm to merge reactables if they're similar enough. Articles are almost exactly posts without parents, after all. Ancillary information can go into a specialization table with an optional 1:1 relationship.
We're a place where coders share, stay up-to-date and grow their careers.
We strive for transparency and don't collect excess data.