ruby - Changing table name at query run time in a Rails application

Question

I have a fat multi-tenant Rails app on Apache + mod_passenger that outputs product prices from a PostgreSQL table as follows:

Table "public.products"
Column | Type
id     | bigint
name   | character varying(100)
price  | numeric(8,2)

Then inside products.rb I have...

class Product < PostgresDatabase
     self.table_name = "products"

     # ... yadda yadda

end

What I want is to partition the "products" table in a very specific way so that I end up with something like products_TENANT-ID for each tenant (basically views of the main products table but that's another story) and be able to query like this:

Products.for_tenant(TENANT-ID).where(:name => "My product")......

I figure I can just create a method:

class Product < PostgresDatabase
     self.table_name = "products"

     # ... yadda yadda
     def for_tenant(tid)
          self.table_name = "products_" + tid.to_s
          self
     end
end

But what kind of impact could this have on the application considering there's lots of traffic (thousands of requests per second)? Is there something I am missing? Should I try a different strategy?

Thank you very much for any feedback/thoughts!

Answer 1

The method

def self.for_tenant(tid)
  self.table_name = "products_" + tid.to_s
  self
end

makes sense, however, it has a side effect: it changes table name for Product class. When this class is used later in the same request, for example, in this way:

Product.where(name: "My other product") ...

the table name won't be products as you may expect; it will stay the same as changed by for_tenant method previously.

To avoid this ambiguity and keep code clean you can use another strategy:

1) Define a module which holds all logic of work with tenant partitions:

# app/models/concerns/partitionable.rb

module Partitionable
  def self.included(base)
    base.class_eval do
      def self.tenant_model(tid)
        partition_suffix = "_#{tid}"

        table = "#{table_name}#{partition_suffix}"

        exists = connection.select_one("SELECT EXISTS (SELECT 1 FROM pg_tables WHERE schemaname = 'public' AND tablename = '#{table}')")
        unless exists['exists'] == 't' || exists['exists'] == true  # different versions of pg gem give different answers
          return self # returning original model class
        end

        class_name = "#{name}#{partition_suffix}"

        model_class = Class.new(self)

        model_class.define_singleton_method(:table_name) do
          table
        end

        model_class.define_singleton_method(:name) do
          class_name
        end

        model_class
      end
    end
  end
end

2) Include this module in your model class:

class Product < PostgresDatabase
  include Partitionable

  ...
end

3) Use it the same way as you intended:

Product.tenant_model(TENANT_ID).where(name: "My product")...

What's happened there:

Method tenant_model(TENANT_ID) creates another model class for the tenant with ID TENANT_ID. This class has name Product_<TENANT_ID>, works with table products_<TENANT_ID> and inherits all methods of Product class. So it could be used like a regular model. And class Product itself remains untouched: its table_name is still products.