How to organise methods in a self join model - ruby-on-rails

In designing a data model, you will sometimes find a model that should have a relation to itself. Ruby On Rails Guide provides a neat example. I am using it as a template for my example
For example, you may want to store all users in a single database model, but be able to trace relationships such as between affiliate and users. This situation can be modeled with self-joining associations:
class User < ApplicationRecord
has_many :users, :through => :referrals
has_one :affiliate, :through => :referral
end
This allows me to keep both users and affiliate in a same database table which is correct because fundamentally they are all individual users.
Problem arises when the model grows. Affiliate has its own set of methods - earnings, expected_earnings etc. These methods are very specific to Affiliate and I have my qualm keeping them with other user methods.
Loading object in correctly named variable helps:
affiliate = User.find 1
affiliate.earnings # used in context of affiliate
user = User.find 1
user.subscriptions # mostly in context to user
But when I read the User model, Affiliate related methods feels out-of-place.
Is there a way to namespace these methods correctly? What is the standard way of organizing self join model methods?

One way to solve this is with Single Table Inheritance. Before accepting this approach, I would recommend searching the web for "single table inheritance rails" and reading up on the pros and cons of it. A lot of digital ink has been spent on this subject.
With the caveat out of the way, Single Table Inheritance (STI) allows you to let multiple Rails models share one database table. You do this by adding a string field called type to your database table. Rails will interpret this as the subclass of your model. You would then create several models that inherit from User.
In your specific case, the type field would either contain user or affiliate. You would also create an Affliliate class which inherits from User. All of your Affiliate specific methods would be put in the Affiliate class. Rails is smart enough to use the type field in the database to identify records from the appropriate class.
Here is the migration you would run:
class AddTypeToUsers < ActiveRecord::Migration[5.1]
def change
add_column :users, :type, :string
add_index :users, :type
end
end
Next you would add an Affiliate class:
# app/models/affliliate.rb
class Affiliate < User
# Affiliate specific methods here.
end
You may also want to create a class for non-affiliate users. Call it customers:
# app/models/customer.rb
class Customer < User
# Customer specific methods here.
end
Use the appropriate class name when creating new records and rails will automatically populate the type field in the database.
You would then moving your associations to the appropriate model:
# app/models/affiliate.rb
class Affiliate < User
has_many :customers, through: :referrals, foreign_key: :user_id
end
# app/models/customer.rb
class Customer < User
has_one :affiliate, through: :referral, foreign_key: :user_id
end
I have not tested this, but it should work.

Related

Rails associations logical vs code

I am currently building my first application with Ruby on Rails (version 5.2.4.2 with ruby 2.6.3) and am having an issue with one of the associations. In my model there are applications in which you supply a single company that you are applying to. Logically I would like to be able to get an application and say application.company = Company.find_by(...).
To make this work, I have
# app/models/Application.rb
class Application < ApplicationRecord
has_one: company
end
# app/models/Company.rb
class Company < ApplicationRecord
end
# database migration
Class AddCompanyToApplication
def change
add_reference :companies, :application, foreign_key: true
end
end
Doing this allows me to use the desired syntax but has one issue. When I create a second application for the same company, the row in the companies table is changed that removes the company from the first application.
The list of companies is predetermined and I just want to refer to a company from within the application. Is there a way to be able to use the assignment syntax while having the foreign key placed in application rather than companies? From a database perspective I feel like application should hold the foreign key, but this breaks the assignment syntax. It seems like Ruby wants me to make Company have has_many :applications, but this breaks the semantics of what I am trying to accomplish. Is there a way to change the foreign key placement or should I forgo the assignment syntax and stick to SQL and assigning directly to id's?
Right now, you have a 1 to 1 relationship. An application has_one :company and a company belongs_to :application. That's why when you try to create a second application for the same company, the application_id in the companies table gets overridden. If you change your company association to has_many :applications, you should be able to add more than one application to a company. Then you can call application.company to return the application's current company or application.company = ... to assign a company to the current application.
class Application < ApplicationRecord
has_one :company
end
class Company < ApplicationRecord
has_many :applications
end
To make this work, you have to create a new migration:
Class AddCompanyIdToApplication
def change
add_reference :application, :company, foreign_key: true
end
end
Before you do that, you have to rollback your existing AddCompanyToApplication migration (which is named wrong, by the way, because you're not adding the company to the application, you're adding the application to the company) and delete the AddCompanyToApplication migration.
Read more about associations
Read more about add_reference

Creating a many-to-many record on Rails

I have a simple task list app that has users and lists on it, the users are managed by Devise, and can create task lists, as well as favorite lists created by other users, or by themself. The relation of ownership between users and lists were easy to establish, but I am having trouble setting up the relation of a user favoriting a list. I envision it being a many-to-many relation after all, a user can favorite many lists and a list can be favorited by many users, this relationship happening on top of another already existing one-to-many relationship of list ownership by a user gave me some pause as to whether this is good practice to do, but I proceeded with my attempt regardless.
Currently I have two models, one for the user, and one for the list, and I tried to create a migration for the favorites by running rails g migration CreateJoinTableFavorites users lists, which resulted in the following migration
class CreateJoinTableFavorites < ActiveRecord::Migration[5.2]
def change
create_join_table :users, :lists do |t|
t.index [:user_id, :list_id] <-- I uncommented this line
# t.index [:list_id, :user_id]
t.timestamps <-- I added this line
end
end
end
I thought this would create a table named "Favorites" that would automatically link users and lists, but instead it created a table called "lists_users". Now I am stuck as to what to do next. I have read that I need to create a model for this join table, but I don't know how to go about doing that. What command do I run? rails g model Favorites? rails g model ListsUsers? do I also inform the fields I want to add such as rails g model Favorites user_id:integer list_id:integer, or is there another better way to do it such as perhaps rails g model Favorites user:references list:references? What's the best practice here
Beyond that, I have added a button inside my list#show view for the user to click to add that list to their favorites, and had some trouble routing it. What I did was create a button like this:
<%= button_to 'Add to favorites', add_favorites_path({list_id: #list.id}), method: :post %>
as well as a new route:
post 'add_favorites', to: 'lists#add_favorites'
Though this I managed to have access to the list id and user id in that action, now I don't know how to proceed to create the "favorite" database entry in my lists_users table. To illustrate, I'll paste here my "add_favorite" action
def add_favorites
user_id = current_user.id
list_id = params[:list_id]
#TODO: create the relation in lists_items table
end
I'm aware that I can't get this to work without the model for the join table, but even if I had that model, I haven't had much luck searching for what to do within the controller to create that relation. Anyway, my models are as follows:
class List < ApplicationRecord
belongs_to :user
has_many :users, through: :lists_users
end
class User < ApplicationRecord
devise :database_authenticatable, :registerable,
:recoverable, :rememberable, :validatable
has_many :lists
has_many :lists, through: :lists_users
end
So to summarize, I am aware that I am missing a model for the join table, and would like a step-by-step as to how to create it, what name to give it, etc, as well as how to proceed within my action in my controller to create a new favorite entry
There are two ways to create a many-to-many relation in Rails. What you're doing seems to conflate the two, which I suspect is the source of your problem.
Briefly, the two methods are:
1) has_many :other_models, through: :relation or
2) has_and_belongs_to_many :other_models
The main difference being that the "has_many through" method expects the join table to be a separate model which can be handled independently of this relationship if need be, while the "has_and_belongs_to_many" method does not require the join table to have a corresponding model. In the latter case, you will not be able to deal with the join table independently. (This makes timestamps on the join table useless, by the way.)
Which method you should go with depends on your use case. The docs summarize the criteria nicely:
The simplest rule of thumb is that you should set up a has_many :through relationship if you need to work with the relationship model as an independent entity. If you don't need to do anything with the relationship model, it may be simpler to set up a has_and_belongs_to_many relationship (though you'll need to remember to create the joining table in the database). (emphasis added)
Now for your question: When you use create_join_table, you're treating it as though you're setting things up for a has_and_belongs_to_many relation. create_join_table will create a table named "#{table1}_#{table2}" with ids pointing to those tables. It alphabetizes them too, which is why you got "lists_users" instead of "users_lists". This is in fact the standard naming convention for rails join tables if you are planning on using has_and_belongs_to_many, and generally shouldn't be renamed.
If you really want to use has_and_belongs_to_many, keep the migration with the create_join_table and just do the following in your models:
# user.rb
class User
has_and_belongs_to_many :lists
end
# list.rb
class List
has_and_belongs_to_many :users
end
And voila. No Favorite model is needed, and rails is smart enough to handle the relationships through the table on its own. Although a bit easier, the downside is, as stated above, that you won't be able to deal with the join table as an independent model. (Again, timestamps on the join table are useless in this case, as Rails won't set them.)
Edit: Since you can't directly touch lists_users, you'd create relationships by setting the lists relation on a user, or by setting the users relation on lists, like so:
def add_favorites
list = List.find(params[:list_id])
current_user.lists << list # creates the corresponding entry in lists_users
# Don't forget to test how this works when the current_user has already favorited a list!
# If you want to prevent that from happening, try
# current_user.lists << list unless current_user.lists.include?(list)
# Alternatively you can do the assignment in reverse:
# list.users << current_user
# Again, because the join table is not an independent model, Rails won't be able to do much to other columns on lists_users out of the box.
# This includes timestamps
end
On the other hand, if you want to use "has_many through", don't use create_join_table. If you're using has_many through, the join table should be thought of almost as an entirely separate model, that just happens to have two foreign keys and tie two other models together in a many-to-many relationship. In this case, you'd do something like:
# migration
class CreateFavorites < ActiveRecord::Migration[5.2]
def change
create_table :favorites do |t|
t.references :list
t.references :user
t.timestamps
end
end
end
# user.rb
class User
has_many :favorites
has_many :lists, through: :favorites
end
# list.rb
class List
has_many :favorites
has_many :users, through: :favorites
end
# favorite.rb
class Favorite
belongs_to :list
belongs_to :user
end
# controller
def add_favorites
# You actually have a Favorite model in this case, while you don't in the other. The Favorite model can be more or less independent of the List and User, and can be given other attributes like timestamps.
# It's the rails methods like `save`, `create`, and `update` that set timestamps, so this will track those for you as any other model.
Favorite.create(list_id: params[:list_id], user: current_user)
end
You might want to reflect on which method to use. Again, this really depends on your use case, and on the criteria above. Personally, when I'm not sure, I prefer the "has_many through" method as it gives you more tools to work with and is generally more flexible.
You may try following :
class User
has_and_belongs_to_many :lists
end
class List
has_and_belongs_to_many :users
end
class CreateUsersAndLists
def change
create_table :users do |t|
# Code
end
create_table :lists do |t|
# Code
end
create_table :users_lists id: false do |t|
t.belongs_to :user, index: true
t.belongs_to :list, index: true
t.boolean :is_favourite
end
end
end

setting up a belongs_to relation when the foreign key is stored in metadata

In a Rails 4 application, I have an STI model that stores metadata in a jsonb column.
Base Class:
class Post < ActiveRecord::Base
...
end
Subclass:
class JobPost < Post
# has a jsonb column for metadata
end
One of the data attributes in the metadata column of a JobPost is a foreign_key reference to another table (company_id). I'd like to add a belongs_to :company reference in the JobPost model. It seems like this should be possible by doing something like
class JobPost < Post
belongs_to :company do
Company.find_by_id self.metadata['company_id']
end
end
but that doesn't appear to work. Help?
Note: I am not necessarily intent on using belongs_to rather than writing needed methods like def company by hand, but I do need a way to eager load companies when listing job posts. If there's a way to do that eager loading without a belongs_to relation I'm all ears.
Update1
I have also tried the following, which doesn't appear to work either:
class JobPost < Post
belongs_to :company, foreign_key: "(posts.metadata->>'company_id')::integer".to_sym
end
Update2
To be more clear about my intentions and need:
1) A JobPost belongs_to a Company, but a Post (and other subclasses of Post) does not. I'd prefer not to jankily add the company_id column to the posts table when it won't be used by the other subclasses.
2) A JobPost could justify having it's own table (perhaps the relationship with a Company is enough to justify it). There are reasons why this wouldn't be ideal, but if that's the only answer I'm open to it. I'd, however, like a more definitive "what you're trying to do can't be done" response before going down this road, though.
The primary question is whether you can customize belongs_to so that it uses the metadata column rather than expecting the foreign key to be a column in the table.
The secondary question is whether you can eager load companies alongside job posts without having that belongs_to relation set up.
EDIT
UPD 2
You need to add "company_id" column to the base class of your STI table. If JobPost inherits from Post, and it should have "company_id" then add the "company_id" column to Post (base table).
Remember STI stands for "Single Table Inheritance" so there is only one table on database schema level. Imagine a column of a Post table, where few data records are the foreign key entries for Companies with company_id and what about the other records of this column with non JobPost subclass types, are they null/empty? Hence the relationship is defined with parent STI table and subclass inherits these relations. Additional type column in STI defines the subclass type.
Check here
You may need to dig further on Polymorphic classes instead of STI if both JobPost and Post have relationship with Company, else create two separate model, as they tend do have some unique relationships and column fields.
UPD
Based on updated ask
app/model/company.rb
class Company < ActiveRecord::Base
has_many :posts
delegate :jobposts, to: :posts
end
app/model/post.rb
class Post < ActiveRecord::Base
belongs_to :company
self.inheritance_column = :ptype
scope :job_posts, -> { where(ptype: 'JobPost') }
def self.ptype
%w(JobPost)
end
end
app/models/jobpost.rb
class JobPost < Post; end
Create a company
company = Company.create!(company_params)
Create some posts and add them to the company
company.posts << JobPost.new(jobpost_params)
To fetch jobpost by company relationship
company.job_posts
In case you are storing company_id in jsonb in any which column, just format your jobpost_params hash input accordingly and it should do the deed for you
OLD ASK
To find by primary key
Company.find(id)
In your case, id is self.metadata['company_id']
To find by other keys
Company.find_by(key: value)
Company.find_by_id is no more recommended
Please remove do and end after belongs_to in your model, instead in your controller you can write:
Jobpost.all.each do |x|
# your do
end
regarding foreign key, as rails is convention over configuration, it by default includes company_id reference to Jobpost which you can change in your Company.rb model

Include a record by default in any association

I have a Client Model as below:
class Client < ActiveRecord::Base
has_many :custodians,:dependent => :destroy
I have a Custodian Model as below:
class Custodian < ActiveRecord::Base
belongs_to :client
In my custodians table I have record with id = 0 , name = 'N/A' that I want to include in all my collection_selects irrespective of the client_id.
e.g for client_id = 10 I want the following in collection_select
Custodian.where('client_id = 10 or client_id = 0')
I know I can do it in my views but I have too many views so it is not practical. Plus I want a more DRY method on either Custodian model or associations. I tried default_scope on Custodian model but could not get it to work.
Basically I am looking for way to always include custodian with id=0 in each association and collection_select.
You can't do what you want using a has_many and belongs_to approach. To implement a belongs_to relationship, the Custodian record has to have a single client_id field. Your logic requires that the custodian_id=0 record belong to many Client records, so it would have to have many client_id fields, but it can only have one. See the Rails Guides-Active Record Associations-The belongs_to Association (http://guides.rubyonrails.org/association_basics.html)
You can accomplish what you want by using a has_and_belongs_to_many relationship. By making both the Custodian and Client models has_and_belongs_to_many to each other, you will be able to have the custodian_id=0 record belong to many Client records and all the other Custodian records will only belong to one client (even though they could belong to many, your program logic must only allow them to belong to one.) See the has_and_belongs_to_many section of the above Rails Guide. To be clear, here is how your models would look:
class Client < ActiveRecord::Base
has_many_and_belongs_to_many :custodians
end
class Custodian < ActiveRecord::Base
has_many_and_belongs_to_many :client
end
Also, because of your special case on custodian_id=0, you will need to establish the look-up table record for the custodian_id=0 record relationship using an active_record callback (probably before_validation or before_create) when you create a new Client record.
Similarly, you will need to implement your own :dependent => :destroy functionality using the before_destroy callback to preserve the custodian_id=0 record and delete all the other associated Custodian records. You'll also have to destroy the corresponding look-up table entries.
This sounds like a lot of work, but if you absolutely must have the custodian_id=0 record associated with every Client, this is the only way I can see it being done. You may want to evaluate it this is really necessary. There may be other program logic that could allow you to get to similar results without going through this process.
You could use an instance or class method:
#app/models/client.rb
Class Client < ActiveRecord::Base
has_many :custodians,:dependent => :destroy
def inc_zero(id)
where("client_id = ? OR client_id = 0", id)
end
def self.inc_zero_custodians(id)
joins(:custodians).where("client_id = ? OR client_id = 0", id)
end
end
#-> Client.custodians.inc_zero(10)
#-> Client.inc_zero_custodians(10)

Rails Active Record - How to model a user/profile scenario

I have a rails application that has three different types of users and I need them all to share the same common profile information. However, each different user also has unique attributes themselves. I'm not sure how to separate out the different fields.
Admin (site wide admin)
Owner (of a store/etc)
Member (such as a member of a co-op)
I'm using devise for authentication and cancan for authorization. Therefore I have a User model with a set of roles that can be applied to the user. This class looks this:
class User < ActiveRecord::Base
# ... devise stuff omitted for brevity ...
# Roles association
has_many :assignments
has_many :roles, :through => :assignments
# For cancan: https://github.com/ryanb/cancan/wiki/Separate-Role-Model
def has_role?(role_sym)
roles.any? { |r| r.name.underscore.to_sym == role_sym }
end
end
Each user has a profile that includes:
First & Last Name
Address Info (city/st/zip/etc)
Phone
I do not want to pollute the User model with this info so I'm throwing it into a Profile model. This part is fairly simple. This turns the User model into something like this:
class User < ActiveRecord::Base
# ... devise stuff omitted for brevity ...
# ... cancan stuff omitted for brevity ...
has_one :profile
end
The additional fields is where I have some uneasy feelings about how to model are ...
If a user is an admin, they'll have unique fields such as:
admin_field_a:string
admin_field_b:string
etc
If a user is a Owner they'll have unique fields ...
stripe_api_key:string
stripe_test_api_key:string
stripe_account_number:string
has_one :store # AR Refence to another model that Admin and Member do not have.
If a user is a member they'll have a few additional fields as such:
stripe_account_number:string
belongs_to :store # the store that they are a member of
has_many :note
...
and a Store model will contain a has_many on the members so we work the the members of the store.
The issue is around the additional fields. Do I set these up as different classes? Put them into a different
I've currently tried a few different ways to set this up:
One way is to set up the User Model as aggregate root
class User < ActiveRecord::Base
# ...
# Roles association
has_many :assignments
has_many :roles, :through => :assignments
# Profile and other object types
has_one :profile
has_one :admin
has_one :owner
has_one :member
# ...
end
The benefit of this approach is the User model is the root and can access everything. The downfall is that if the user is a "owner" then the "admin" and "member" references will be nil (and the cartesian of the other possibilities - admin but not owner or member, etc).
The other option I was thinking of was to have each type of user inherit from the User model as such:
class User < ActiveRecord::Base
# ... other code removed for brevity
has_one :profile
end
class Admin < User
# admin fields
end
class Owner < User
# owner fields
end
class Member < User
# member fields
end
Problem with this is that I'm polluting the User object with all kinds of nil's in the table where one type doesn't need the values from another type/etc. Just seems kind of messy, but I'm not sure.
The other option was to create each account type as the root, but have the user as a child object as shown below.
class Admin
has_one :user
# admin fields go here.
end
class Owner
has_one :user
# owner fields go here.
end
class Member
has_one :user
# member fields go here.
end
The problem with the above is I'm not sure how to load up the proper class once the user logs in. I'll have their user_id and I'll be able to tell which role they are (because of the role association on the user model), but I'm not sure how to go from user UP to a root object. Methods? other?
Conclusion
I have a few different ways to do this, but I'm not sure what the correct "rails" approach is. What is the correct way to model this in rails AR? (MySQL backend). If there is not a "right" approach, whats the best of the above (I'm also open to other ideas).
Thanks!
My answer assumes that a given user can only be one type of user - e.g. ONLY an Admin or ONLY a Member. If so, this seems like a perfect job for ActiveRecord's Polymorphic association.
class User < ActiveRecord::Base
# ...
belongs_to :privilege, :polymorphic => true
end
This association gives User an accessor called 'privilege' (for lack of a better term and to avoid naming confusion which will become apparent later). Because it is polymorphic, it can return a variety of classes. The polymorphic relationship requires two columns on the corresponding table - one (accessor)_type and (accessor)_id. In my example, the User table would gain two fields: privilege_type and privilege_id which ActiveRecord combines to find the associated entry during lookups.
Your Admin, Owner and Member classes look like this:
class Admin
has_one :user, :as => :privilege
# admin fields go here.
end
class Owner
has_one :user, :as => :privilege
# owner fields go here.
end
class Member
has_one :user, :as => :privilege
# member fields go here.
end
Now you can do things like this:
u = User.new(:attribute1 => user_val1, ...)
u.privilege = Admin.new(:admin_att1 => :admin_val1, ...)
u.save!
# Saves a new user (#3, for example) and a new
# admin entry (#2 in my pretend world).
u.privilege_type # Returns 'Admin'
u.privilege_id # Returns 2
u.privilege # returns the Admin#2 instance.
# ActiveRecord's SQL behind the scenes:
# SELECT * FROM admin WHERE id=2
u.privilege.is_a? Admin # returns true
u.privilege.is_a? Member # returns false
Admin.find(2).user # returns User#3
# ActiveRecord's SQL behind the scenes:
# SELECT * FROM user WHERE privilege_type='Admin'
# AND privilege_id=2
I would recommend you make the (accessor)_type field on the database an ENUM if you expect it to be a known set of values. An ENUM, IMHO, is a better choice than a VARCHAR255 which Rails would normally default to, is easier/faster/smaller to index but makes changes down the road more difficult/timeconsuming when you've got millions of users. Also, index the association properly:
add_column :privilege_type, "ENUM('Admin','Owner','Member')", :null => false
add_column :privilege_id, :integer, :null => false
add_index :user, [:privilege_type, :privilege_id], :unique => true
add_index :user, :privilege_type
The first index allows ActiveRecord to rapidly find the reverse association (e.g. find the user that has a privilege of Admin#2) and the second index allows you to find all Admins or all Members.
This RailsCast is a bit dated but a good tutorial on polymorphic relationships nonetheless.
One last note - in your question, you indicated Admin, Owner or Member was the user's type which is appropriate enough, but as you probably see, I'd have to explain that your user table would then have a user_type_type field.
I'm probably not going to give you a rails approved suggestion, but ..
Separating your profile is a good call. Consider using the Decorator pattern for the a role. You can have an AdminUserDecorator, OwnerUserDecorator, or MemberOwnerDecorator. You could also dynamically add the additional fields directly on the instance (it is Ruby after all), but I think that would get ugly and complicated. (If you really want to do bad things, use a visitor object to give you an instance of your decorator from a method on the user class.)
Also, why put the stripe or payment config on the Owner instead of being part of the store information? Unless perhaps an owner can have multiple stores and use the same payment info for each store?
UPDATE: I should also suggest using TDD to flush out what works.
You've already accepted an answer, but for what it's worth, I have a similar situation and chose to go with your "User Model as aggregate root" approach. My User model contains all the "profile" info, and the User, to use a fictitious example, has_one :buyer and has_one :seller. I use a simple tinyint field as bitflags for which roles the user holds, since users could be both buyers and sellers (or TBD other roles I need in the future). If a bit is set, you can assume the corresponding association is not nil (which hasn't been an issue for me since I always check the bitflags before using the association reference). I don't actually have too many unique fields in my real subordinate models, but it's very useful to keep things decluttered when each subordinate model has additional associations, like if seller has_one :merchant_account" and a buyer has_one :purchase_history, etc. I haven't gone live yet, but when I do, I'll follow this post up with any issues I encounter.

Resources