Sometimes you might need to use SQL UNION in your application, in this article we explan you how to abstract SQL union creation within ActiveRecord.

The main idea is to create an abstaction which receives in input an array of ActiveRecord relations and then transform the given array into SQL, eventually we’ll use ActiveRecord.from method in order to load the given SQL as an ActiveRecord relation.

Please note that I extracted this examples from gitlab codebase.

module Gitlab
  module SQL
    # Class for building SQL UNION statements.
    # ORDER BYs are dropped from the relations as the final sort order is not
    # guaranteed any way.
    # Example usage:
    #     union =[user.personal_projects, user.projects])
    #     sql   = union.to_sql
    #     Project.from("(#{union}) projects")
    class Union
      def initialize(relations, remove_duplicates: true)
        @relations = relations
        @remove_duplicates = remove_duplicates

      def to_sql
        # Some relations may include placeholders for prepared statements, these
        # aren't incremented properly when joining relations together this way.
        # By using "unprepared_statements" we remove the usage of placeholders
        # (thus fixing this problem), at a slight performance cost.
        fragments = ActiveRecord::Base.connection.unprepared_statement do
 { |rel| rel.reorder(nil).to_sql }.reject(&:blank?)

        if fragments.any?
          "(" + fragments.join(")\n#{union_keyword}\n(") + ")"

      def union_keyword
        @remove_duplicates ? 'UNION' : 'UNION ALL'

With the above class you can create very quiclky an UNION given an array of AR relations, for example as explained in the comment you can do:

union =[user.personal_projects, user.projects])
sql   = union.to_sql

projects = Project.from("(#{union}) projects")

Worth noting that we are dealing always with AR relations which imples that the query is run only if needed; this also implies that the AR relation can be used as parameter to other queries.

Also note that this code will run without loading data in memory (until is needed) thus the code is performant!