Build Status Code Climate Coverage Status Gem Version

Create hierarchical queries using simple DSL, recursively traverse trees using single SQL query.

If a table contains hierarchical data, then you can select rows in hierarchical order using hierarchical query builder.


  • ActiveRecord >= 5.0, < 7.1
  • PostgreSQL >= 8.4
  • Postgres Gem >= 0.21, < 1.4

Note that though PostgresSQL 8.4 and up should work, this library is tested on PostgresSQL 10.5.

In a nutshell

Traverse trees

Let's say you've got an ActiveRecord model Category that related to itself:

class Category < ActiveRecord::Base
  belongs_to :parent, class_name: 'Category'
  has_many :children, foreign_key: :parent_id, class_name: 'Category'

# Table definition
# create_table :categories do |t|
#   t.integer :parent_id
#   t.string :name
# end

Traverse descendants

Category.join_recursive do |query|
  query.start_with(parent_id: nil)
       .connect_by(id: :parent_id)
end # returns ActiveRecord::Relation instance

Traverse ancestors

Category.join_recursive do |query|
  query.start_with(id: 42)
       .connect_by(parent_id: :id)

Show breadcrumbs using single SQL query

records = Category.join_recursive do |query|
    # assume that deepest node has depth=0
    .start_with(id: 42) { select('0 depth') }
    # for each ancestor decrease depth by 1, do not apply
    # following expression to first level of hierarchy
    .select(query.prior[:depth] - 1, start_with: false)
    .connect_by(parent_id: :id)
end.order('depth ASC')

# returns a regular ActiveRecord::Relation instance
# so methods like `pluck` all work as expected.

crumbs = records.pluck(:name).join(' / ')


Add this line to your application's Gemfile:

gem 'activerecord-hierarchical_query'

And then execute:

$ bundle

Or install it yourself as:

$ gem install activerecord-hierarchical_query

You'll then need to require the gem:

require 'active_record/hierarchical_query'

Alternatively, the require can be placed in the Gemfile:

gem 'activerecord-hierarchical_query', require: 'active_record/hierarchical_query'


Let's say you've got an ActiveRecord model Category with attributes id, parent_id and name. You can traverse nodes recursively starting from root rows connected by parent_id column ordered by name:

Category.join_recursive do
  start_with(parent_id: nil).
  connect_by(id: :parent_id).

Hierarchical queries consist of these important clauses:

  • START WITH clause

This clause specifies the root row(s) of the hierarchy.

  • CONNECT BY clause

This clause specifies relationship between parent rows and child rows of the hierarchy.


This clause specifies an order of rows in which they appear on each hierarchy level.

These terms are borrowed from Oracle hierarchical queries syntax.

Hierarchical queries are processed as follows:

  • First, root rows are selected -- those rows that satisfy START WITH condition in order specified by ORDER SIBLINGS clause. In example above it's specified by statements query.start_with(parent_id: nil) and query.order_siblings(:name).

  • Second, child rows for each root rows are selected. Each child row must satisfy condition specified by CONNECT BY clause with respect to one of the root rows (query.connect_by(id: :parent_id) in example above). Order of child rows is also specified by ORDER SIBLINGS clause.

  • Successive generations of child rows are selected with respect to CONNECT BY clause. First the children of each row selected in step 2 selected, then the children of those children and so on.


This clause is specified by start_with method:

Category.join_recursive { start_with(parent_id: nil) }
Category.join_recursive { start_with { where(parent_id: nil) } }
Category.join_recursive { start_with { |root_rows| root_rows.where(parent_id: nil) } }

All of these statements are equivalent.


This clause is necessary and specified by connect_by method:

# join parent table ID columns and child table PARENT_ID column
Category.join_recursive { connect_by(id: :parent_id) }

# you can use block to build complex JOIN conditions
Category.join_recursive do
  connect_by do |parent_table, child_table|
    parent_table[:id].eq child_table[:parent_id]


You can specify order in which rows on each hierarchy level should appear:

Category.join_recursive { order_siblings(:name) }

# you can reverse order
Category.join_recursive { order_siblings(name: :desc) }

# arbitrary strings and Arel nodes are allowed also
Category.join_recursive { order_siblings('name ASC') }
Category.join_recursive { |query| query.order_siblings(query.table[:name].asc) }

WHERE conditions

You can filter rows on each hierarchy level by applying WHERE conditions:

Category.join_recursive do
  connect_by(id: :parent_id).where('name LIKE ?', 'ruby %')

You can even refer to parent table, just don't forget to include columns in SELECT clause!

Category.join_recursive do |query|
  query.connect_by(id: :parent_id)
       .where(query.prior[:name].matches('ruby %'))

Or, if Arel semantics does not fit your needs:

Category.join_recursive do |query|
  query.connect_by(id: :parent_id)
       .where("#{}.name LIKE ?", 'ruby %')


Recursive query will loop if hierarchy contains cycles (your graph is not acyclic). NOCYCLE clause, which is turned off by default, could prevent it.

Loop example:

node_1 = Category.create
node_2 = Category.create(parent: node_1)

node_1.parent = node_2

node_1 and node_2 now link to each other, so the following query will not terminate:

Category.join_recursive do |query|
  query.connect_by(id: :parent_id)

#nocycle method will prevent endless loop:

Category.join_recursive do |query|
  query.connect_by(id: :parent_id)


By default, the union term in the Common Table Expression uses a UNION ALL. If you want to SELECT DISTINCT CTE values, add a query option for distinct:

Category.join_recursive do |query|
  query.connect_by(id: :parent_id)

If you want to join CTE terms by UNION DISTINCT, pass an option to join_recursive:

Category.join_recursive(union_type: :distinct) do |query|
  query.connect_by(id: :parent_id)

Generated SQL queries

Under the hood this extensions builds INNER JOIN to recursive subquery.

For example, this piece of code

Category.join_recursive do |query|
  query.start_with(parent_id: nil) { select('0 LEVEL') }
       .connect_by(id: :parent_id)
       .select(query.prior[:LEVEL] + 1, start_with: false)

Would generate following SQL:

SELECT "categories".*
FROM "categories" INNER JOIN (
    WITH RECURSIVE "categories__recursive" AS (
        SELECT depth,
               0 LEVEL,
               ARRAY["categories"."position"] AS __order_column,
               ARRAY["categories"."id"] AS __path
        FROM "categories"
        WHERE "categories"."parent_id" IS NULL

        UNION ALL

        SELECT "categories"."depth",
               "categories__recursive"."LEVEL" + 1,
               "categories__recursive"."__order_column" || "categories"."position",
               "categories__recursive"."__path" || "categories"."id"
        FROM "categories" INNER JOIN
             "categories__recursive" ON "categories__recursive"."id" = "categories"."parent_id"
        WHERE ("categories__recursive"."depth" <= 5) AND
              NOT ("categories"."id" = ANY("categories__recursive"."__path"))
    SELECT "categories__recursive".* FROM "categories__recursive"
) AS "categories__recursive" ON "categories"."id" = "categories__recursive"."id"
ORDER BY "categories__recursive"."__order_column" ASC

If you want to use a LEFT OUTER JOIN instead of an INNER JOIN, add a query option for outer_join_hierarchical. This option allows the query to return non-hierarchical entries:

  .join_recursive(outer_join_hierarchical: true)

If, when joining the recursive view to the main table, you want to change the foreign_key on the recursive view from the primary key of the main table to another column:

  .join_recursive(foreign_key: another_column)


Read through the short contributing guide.