PostgreSQL provides an hstore data type for storing arbitrarily complex structures in a column. ActiveRecord 4.0 supports Hstore but casts all valus in the store to a string. Further, ActiveRecord does not provide discrete fields to access values directly in the hstore column. The HstoreAccessor gem solves both of these issues.
Add this line to your application's Gemfile:
gem 'hstore_accessor'
And then execute:
$ bundle
Or install it yourself as:
$ gem install hstore_accessor
The hstore_accessor
method accepts the name of the hstore column you'd
like to use and a hash with keys representing fields and values
indicating the type to be stored in that field. The available types
are: string
, integer
, float
, time
, boolean
, array
, and hash
.
class Product < ActiveRecord::Base
hstore_accessor :options,
color: :string,
weight: :integer,
price: :float,
built_at: :time,
tags: :array,
ratings: :hash
end
Now you can interact with the fields stored in the hstore directly.
p = Product.new
p.color = "green"
p.weight = 34
p.price = 99.95
p.built_at = Time.now - 10.days
p.popular = true
p.tags = ["housewares", "kitchen"]
p.ratings = { user_a: 3, user_b: 4 }
Reading these fields works as well.
p.color #=> "green"
p.tags #=> ["housewares", "kitchen"]
In order to reduce the storage overhead of hstore keys (especially when indexed) you can specify an alternate key.
hstore_accessor :options,
color: { data_type: :string, store_key: "c" },
weight: { data_type: :integer, store_key: "w" }
In the above example you can continue to interact with the fields using
their full name but when saved to the database the field will be set
using the store_key
.
The hstore_accessor
macro also creates scopes for string
, integer
,
float
, time
, boolean
, and array
fields.
For string
types, a with_<key>
scope is created which checks for
equality.
Product.with_color("green")
For integer
and float
types five scopes are created:
Product.price_lt(240.00) # price less than
Product.price_lte(240.00) # price less than or equal to
Product.price_eq(240.00) # price equal to
Product.price_gte(240.00) # price greater than or equal to
Product.price_gt(240.00) # price greater than
For time
fileds, three scopes are provided:
Product.built_at_before(Time.now) # built before the given time
Product.built_at_eq(Time.now - 10.days) # built at an exact time
Product.built_at_after(Time.now - 4.days) # built after the given time
For array
types, two scopes are created:
Product.tags_eq(["housewares", "kitchen"]) # tags equaling
Product.tags_contains("kitchen") # tags containing a single value
Product.tags_contains(["housewares", "kitchen"]) # tags containing a number of values
Two scopes are created for boolean
fields:
Product.is_popular # => when populer is set to true
Product.not_popular # => when populer is set to false
Predicate methods are also available on instances:
product = Product.new(popular: true)
product.popular? #=> true
One of the big issues with ActiveRecord
single-table inheritance (STI)
is sparse columns. Essentially, as sub-types of the original table
diverge further from their parent more columns are left empty in a given
table. Postgres' hstore
type provides part of the solution in that
the values in an hstore
column does not impose a structure - different
rows can have different values.
We set up our table with an hstore field:
# db/migration/<timestamp>_create_players_table.rb
class CreateVehiclesTable < ActiveRecord::Migration
def change
create_table :vehicles do |t|
t.string :make
t.string :model
t.integer :model_year
t.string :type
t.hstore :data
end
end
end
And for our models:
# app/models/vehicle.rb
class Vehicle < ActiveRecord::Base
end
# app/models/vehicles/automobile.rb
class Automobile < Vehicle
hstore_accessor :data,
axle_count: :integer,
weight: :float,
engine_details: :hash
end
# app/models/vehicles/airplane.rb
class Airplane < Vehicle
hstore_accessor :data,
engine_type: :string,
safety_rating: :integer,
features: :hash
end
From here any attributes specific to any sub-class can be stored in the
hstore
column avoiding sparse data. Indices can also be created on
individual fields in an hstore
column.
This approach was originally concieved by Joe Hirn in this blog post.
- Fork it
- Create your feature branch (
git checkout -b my-new-feature
) - Commit your changes (
git commit -am 'Add some feature'
) - Push to the branch (
git push origin my-new-feature
) - Create new Pull Request