Ancestry

Overview

Ancestry is a gem that allows rails ActiveRecord models to be organized as a tree structure (or hierarchy). It employs the materialized path pattern which allows operations to be performed efficiently.

Features

There are a few common ways of storing hierarchical data in a database: materialized path, closure tree table, adjacency lists, nested sets, and adjacency list with recursive queries.

Features from Materialized Path

• Store hierarchy in an easy to understand format. (e.g.: /1/2/3/)
• Store hierarchy in the original table with no additional tables.
• Single SQL queries for relations (ancestors, parent, root, children, siblings, descendants)
• Single query for creating records.
• Moving/deleting nodes only affect child nodes (rather than updating all nodes in the tree)

Features from Ancestry gem Implementation

• relations are implemented as scopes
• STI support
• Arrangement of subtrees into hashes
• Multiple strategies for querying materialized_path
• Multiple strategies for dealing with orphaned records
• depth caching
• depth constraints
• counter caches
• Multiple strategies for moving nodes
• Easy migration from parent_id based gems
• Integrity checking
• Integrity restoration
• Most queries use indexes on id or ancestry column. (e.g.: LIKE '#{ancestry}/%')

Since a Btree index has a limitaton of 2704 characters for the ancestry column, the maximum depth of an ancestry tree is 900 items at most. If ids are 4 digits long, then the max depth is 540 items.

When using STI all classes are returned from the scopes unless you specify otherwise using where(:type => "ChildClass").

Supported Rails versions

• Ancestry 2.x supports Rails 4.1 and earlier
• Ancestry 3.x supports Rails 4.2 and 5.0
• Ancestry 4.x supports Rails 5.2 through 7.0
• Ancestry 5.0 supports Rails 6.0 and higher Rails 5.2 with update_strategy=ruby is still being tested in 5.0.

Installation

Follow these steps to apply Ancestry to any ActiveRecord model:

Add to Gemfile

# Gemfile

gem 'ancestry'

$ bundle install

Add ancestry column to your table

$ rails g migration add_[ancestry]_to_[table] ancestry:string:index

class AddAncestryToTable < ActiveRecord::Migration[6.1]
  def change
    change_table(:table) do |t|
      # postgres
      t.string "ancestry", collation: 'C', null: false
      t.index "ancestry"
      # mysql
      t.string "ancestry", collation: 'utf8mb4_bin', null: false
      t.index "ancestry"
    end
  end
end

There are additional options for the columns in Ancestry Database Column and an explanation for opclass and collation.

$ rake db:migrate

Configure ancestry defaults

# config/initializers/ancestry.rb

# use the newer format
Ancestry.default_ancestry_format = :materialized_path2
# Ancestry.default_update_strategy = :sql

Add ancestry to your model

# app/models/[model.rb]

class [Model] < ActiveRecord::Base
   has_ancestry
end

Your model is now a tree!

Organising records into a tree

You can use parent_id and parent to add a node into a tree. They can be set as attributes or passed into methods like new, create, and update.

TreeNode.create! :name => 'Stinky', :parent => TreeNode.create!(:name => 'Squeeky')

Children can be created through the children relation on a node: node.children.create :name => 'Stinky'.

Tree Navigation

The node with the large border is the reference node (the node from which the navigation method is invoked.) The yellow nodes are those returned by the method.

parent	root1	ancestors
nil for a root node	self for a root node	root..parent
parent_id	root_id	ancestor_ids
has_parent?	is_root?	ancestors?
parent_of?	root_of?	ancestor_of?
children	descendants	indirects
child_ids	descendant_ids	indirect_ids
has_children?
child_of?	descendant_of?	indirect_of?
siblings	subtree	path
includes self	self..indirects	root..self
sibling_ids	subtree_ids	path_ids
has_siblings?
sibling_of?(node)

When using STI all classes are returned from the scopes unless you specify otherwise using where(:type => "ChildClass").

^{1. [other root records are considered siblings]↩}

has_ancestry options

The has_ancestry method supports the following options:

:ancestry_column       Column name to store ancestry
                       'ancestry' (default)
:ancestry_format       Format for ancestry column (see Ancestry Formats section):
                       :materialized_path   1/2/3, root nodes ancestry=nil (default)
                       :materialized_path2  /1/2/3/, root nodes ancestry=/ (preferred)
:orphan_strategy       How to handle children of a destroyed node:
                       :destroy   All children are destroyed as well (default)
                       :rootify   The children of the destroyed node become root nodes
                       :restrict  An AncestryException is raised if any children exist
                       :adopt     The orphan subtree is added to the parent of the deleted node
                                  If the deleted node is Root, then rootify the orphan subtree
                       :none      skip this logic. (add your own `before_destroy`)
:cache_depth           Cache the depth of each node: (See Depth Cache section)
                       false   Do not cache depth (default)
                       true    Cache depth in 'ancestry_depth'
                       String  Cache depth in the column referenced
:primary_key_format    Regular expression that matches the format of the primary key:
                       '[0-9]+'            integer ids (default)
                       '[-A-Fa-f0-9]{36}'  UUIDs
:touch                 Touch the ancestors of a node when it changes:
                       false  don't invalide nested key-based caches (default)
                       true   touch all ancestors of previous and new parents
:counter_cache         Create counter cache column accessor:
                       false  don't store a counter cache (default)
                       true   store counter cache in `children_count`.
                       String name of column to store counter cache.
:update_strategy       How to update descendants nodes:
                       :ruby  All descendants are updated using the ruby algorithm. (default)
                              This triggers update callbacks for each descendant node
                       :sql   All descendants are updated using a single SQL statement.
                              This strategy does not trigger update callbacks for the descendants.
                              This strategy is available only for PostgreSql implementations

Legacy configuration using acts_as_tree is still available. Ancestry defers to acts_as_tree if that gem is installed.

(Named) Scopes

The navigation methods return scopes instead of records, where possible. Additional ordering, conditions, limits, etc. can be applied and the results can be retrieved, counted, or checked for existence:

node.children.where(:name => 'Mary').exists?
node.subtree.order(:name).limit(10).each { ... }
node.descendants.count

A couple of class-level named scopes are included:

roots                   Root nodes
ancestors_of(node)      Ancestors of node, node can be either a record or an id
children_of(node)       Children of node, node can be either a record or an id
descendants_of(node)    Descendants of node, node can be either a record or an id
indirects_of(node)      Indirect children of node, node can be either a record or an id
subtree_of(node)        Subtree of node, node can be either a record or an id
siblings_of(node)       Siblings of node, node can be either a record or an id

It is possible thanks to some convenient rails magic to create nodes through the children and siblings scopes:

node.children.create
node.siblings.create!
TestNode.children_of(node_id).new
TestNode.siblings_of(node_id).create

Selecting nodes by depth

With depth caching enabled (see has_ancestry options), an additional five named scopes can be used to select nodes by depth:

before_depth(depth)     Return nodes that are less deep than depth (node.depth < depth)
to_depth(depth)         Return nodes up to a certain depth (node.depth <= depth)
at_depth(depth)         Return nodes that are at depth (node.depth == depth)
from_depth(depth)       Return nodes starting from a certain depth (node.depth >= depth)
after_depth(depth)      Return nodes that are deeper than depth (node.depth > depth)

Depth scopes are also available through calls to descendants, descendant_ids, subtree, subtree_ids, path and ancestors (with relative depth). Note that depth constraints cannot be passed to ancestor_ids or path_ids as both relations can be fetched directly from the ancestry column without needing a query. Use ancestors(depth_options).map(&:id) or ancestor_ids.slice(min_depth..max_depth) instead.

node.ancestors(:from_depth => -6, :to_depth => -4)
node.path.from_depth(3).to_depth(4)
node.descendants(:from_depth => 2, :to_depth => 4)
node.subtree.from_depth(10).to_depth(12)

Arrangement

arrange

A subtree can be arranged into nested hashes for easy navigation after database retrieval.

The resulting format is a hash of hashes

{
  #<TreeNode id: 100018, name: "Stinky", ancestry: nil> => {
    #<TreeNode id: 100019, name: "Crunchy", ancestry: "100018"> => {
      #<TreeNode id: 100020, name: "Squeeky", ancestry: "100018/100019"> => {}
    },
    #<TreeNode id: 100021, name: "Squishy", ancestry: "100018"> => {}
  }
}

There are many ways to call arrange:

TreeNode.find_by(:name => 'Crunchy').subtree.arrange
TreeNode.find_by(:name => 'Crunchy').subtree.arrange(:order => :name)

arrange_serializable

If a hash of arrays is preferred, arrange_serializable can be used. The results work well with to_json.

TreeNode.arrange_serializable(:order => :name)
# use an active model serializer
TreeNode.arrange_serializable { |parent, children| MySerializer.new(parent, children: children) }
TreeNode.arrange_serializable do |parent, children|
  {
     my_id: parent.id,
     my_children: children
  }
end

Sorting

The sort_by_ancestry class method: TreeNode.sort_by_ancestry(array_of_nodes) can be used to sort an array of nodes as if traversing in preorder. (Note that since materialized path trees do not support ordering within a rank, the order of siblings is dependant upon their original array order.)

Ancestry Database Column

Collation Indexes

Sorry, using collation or index operator classes makes this a little complicated. The root of the issue is that in order to use indexes, the ancestry column needs to compare strings using ascii rules.

It is well known that LIKE '/1/2/%' will use an index because the wildchard (i.e.: %) is on the right hand side of the LIKE. While that is true for ascii strings, it is not necessarily true for unicode. Since ancestry only uses ascii characters, telling the database this constraint will optimize the LIKE statemens.

Collation Sorting

As of 2018, standard unicode collation ignores punctuation for sorting. This ignores the ancestry delimiter (i.e.: /) and returns data in the wrong order. The exception being Postgres on a mac, which ignores proper unicode collation and instead uses ISO-8859-1 ordering (read: ascii sorting).

Using the proper column storage and indexes will ensure that data is returned from the database in the correct order. It will also ensure that developers on Mac or Windows will get the same results as linux production servers, if that is your setup.

Migrating Collation

If you are reading this and want to alter your table to add collation to an existing column, remember to drop existing indexes on the ancestry column and recreate them.

ancestry_format materialized_path and nulls

If you are using the legacy ancestry_format of :materialized_path, then you need to the collum to allow nulls. Change the column create accordingly: null: true.

Chances are, you can ignore this section as you most likely want to use :materialized_path2.

Postgres Storage Options

ascii field collation

The currently suggested way to create a postgres field is using 'C' collation:

t.string "ancestry", collation: 'C', null: false
t.index "ancestry"

ascii index

If you need to use a standard collation (e.g.: en_US), then use an ascii index:

t.string "ancestry", null: false
t.index  "ancestry", opclass: :varchar_pattern_ops

This option is mostly there for users who have an existing ancestry column and are more comfortable tweaking indexes rather than altering the ancestry column.

binary column

When the column is binary, the database doesn't convert strings using locales. Rails will convert the strings and send byte arrays to the database. At this time, this option is not suggested. The sql is not as readable, and currently this does not support the :sql update_strategy.

t.binary "ancestry", limit: 3000, null: false
t.index  "ancestry"

You may be able to alter the database to gain some readability:

ALTER DATABASE dbname SET bytea_output to 'escape';

Mysql Storage options

ascii field collation

The currently suggested way to create a postgres field is using 'C' collation:

t.string "ancestry", collation: 'utf8mb4_bin', null: false
t.index "ancestry"

binary collation

Collation of binary acts much the same way as the binary column:

t.string "ancestry", collate: 'binary', limit: 3000, null: false
t.index  "ancestry"

binary column

t.binary "ancestry", limit: 3000, null: false
t.index  "ancestry"

ascii character set

Mysql supports per column character sets. Using a character set of ascii will set this up.

ALTER TABLE table
  ADD COLUMN ancestry VARCHAR(2700) CHARACTER SET ascii;

Ancestry Formats

You can choose from 2 ancestry formats:

• :materialized_path - legacy format (currently the default for backwards compatibility reasons)
• :materialized_path2 - newer format. Use this if it is a new column

:materialized_path    1/2/3,  root nodes ancestry=nil
    descendants SQL: ancestry LIKE '1/2/3/%' OR ancestry = '1/2/3'
:materialized_path2  /1/2/3/, root nodes ancestry=/
    descendants SQL: ancestry LIKE '/1/2/3/%'

If you are unsure, choose :materialized_path2. It allows a not NULL column, faster descenant queries, has one less OR statement in the queries, and the path can be formed easily in a database query for added benefits.

There is more discussion in Internals or Migrating ancestry format For migrating from materialized_path to materialized_path2 see Ancestry Column

Migrating Ancestry Format

To migrate from materialized_path to materialized_path2:

klass = YourModel
# set all child nodes
klass.where.not(klass.arel_table[klass.ancestry_column].eq(nil)).update_all("#{klass.ancestry_column} = CONCAT('#{klass.ancestry_delimiter}', #{klass.ancestry_column}, '#{klass.ancestry_delimiter}')")
# set all root nodes
klass.where(klass.arel_table[klass.ancestry_column].eq(nil)).update_all("#{klass.ancestry_column} = '#{klass.ancestry_root}'")

change_column_null klass.table_name, klass.ancestry_column, false

Migrating from plugin that uses parent_id column

It should be relatively simple to migrating from a plugin that uses a parent_id column, (e.g.: awesome_nested_set, better_nested_set, acts_as_nested_set).

When running the installation steps, also remove the old gem from your Gemfile, and remove the old gem's macros from the model.

Then populate the ancestry column from rails console:

Model.build_ancestry_from_parent_ids!
# Model.rebuild_depth_cache!
Model.check_ancestry_integrity!

It is time to run your code. Most tree methods should work fine with ancestry and hopefully your tests only require a few minor tweaks to get up and runnnig.

Once you are happy with how your app is running, remove the old parent_id column:

$ rails g migration remove_parent_id_from_[table]

class RemoveParentIdFromToTable < ActiveRecord::Migration[6.1]
  def change
    remove_column "table", "parent_id", type: :integer
  end
end

$ rake db:migrate

Depth cache

Depth Cache Migration

To add depth_caching to an existing model:

Add column

class AddDepthCachToTable < ActiveRecord::Migration[6.1]
  def change
    change_table(:table) do |t|
      t.integer "ancestry_depth", default: 0
    end
  end
end

Add ancestry to your model

# app/models/[model.rb]

class [Model] < ActiveRecord::Base
   has_ancestry cache_depth: true
end

Update existing values

Add a custom script or run from rails console. Some use migrations, but that can make the migration suite fragile. The command of interest is:

Model.rebuild_depth_cache!

Running Tests

git clone [email protected]:stefankroes/ancestry.git
cd ancestry
cp test/database.example.yml test/database.yml
bundle
appraisal install
# all tests
appraisal rake test
# single test version (sqlite and rails 5.0)
appraisal sqlite3-ar-50 rake test