debian-mirror-gitlab/app/models/namespaces/traversal/linear.rb

252 lines
9 KiB
Ruby
Raw Normal View History

2021-04-29 21:17:54 +05:30
# frozen_string_literal: true
#
# Query a recursively defined namespace hierarchy using linear methods through
# the traversal_ids attribute.
#
# Namespace is a nested hierarchy of one parent to many children. A search
# using only the parent-child relationships is a slow operation. This process
# was previously optimized using Postgresql recursive common table expressions
# (CTE) with acceptable performance. However, it lead to slower than possible
# performance, and resulted in complicated queries that were difficult to make
# performant.
#
# Instead of searching the hierarchy recursively, we store a `traversal_ids`
# attribute on each node. The `traversal_ids` is an ordered array of Namespace
# IDs that define the traversal path from the root Namespace to the current
# Namespace.
#
# For example, suppose we have the following Namespaces:
#
# GitLab (id: 1) > Engineering (id: 2) > Manage (id: 3) > Access (id: 4)
#
# Then `traversal_ids` for group "Access" is [1, 2, 3, 4]
#
# And we can match against other Namespace `traversal_ids` such that:
#
# - Ancestors are [1], [1, 2], [1, 2, 3]
# - Descendants are [1, 2, 3, 4, *]
# - Root is [1]
# - Hierarchy is [1, *]
#
# Note that this search method works so long as the IDs are unique and the
# traversal path is ordered from root to leaf nodes.
#
# We implement this in the database using Postgresql arrays, indexed by a
# generalized inverted index (gin).
module Namespaces
module Traversal
module Linear
extend ActiveSupport::Concern
2021-10-27 15:23:28 +05:30
include LinearScopes
2021-04-29 21:17:54 +05:30
UnboundedSearch = Class.new(StandardError)
included do
2021-06-08 01:23:25 +05:30
before_update :lock_both_roots, if: -> { sync_traversal_ids? && parent_id_changed? }
2021-04-29 21:17:54 +05:30
after_update :sync_traversal_ids, if: -> { sync_traversal_ids? && saved_change_to_parent_id? }
2022-04-04 11:22:00 +05:30
# sync traversal_ids on namespace create, which can happen quite early within a transaction, thus keeping the lock on root namespace record
# for a relatively long time, e.g. creating the project namespace when a project is being created.
after_create :sync_traversal_ids, if: -> { sync_traversal_ids? && !sync_traversal_ids_before_commit? }
# This uses rails internal before_commit API to sync traversal_ids on namespace create, right before transaction is committed.
# This helps reduce the time during which the root namespace record is locked to ensure updated traversal_ids are valid
before_commit :sync_traversal_ids, on: [:create], if: -> { sync_traversal_ids? && sync_traversal_ids_before_commit? }
2021-04-29 21:17:54 +05:30
end
def sync_traversal_ids?
Feature.enabled?(:sync_traversal_ids, root_ancestor, default_enabled: :yaml)
end
2022-04-04 11:22:00 +05:30
def sync_traversal_ids_before_commit?
Feature.enabled?(:sync_traversal_ids_before_commit, root_ancestor, default_enabled: :yaml)
end
2021-04-29 21:17:54 +05:30
def use_traversal_ids?
2021-10-27 15:23:28 +05:30
return false unless Feature.enabled?(:use_traversal_ids, default_enabled: :yaml)
2021-06-08 01:23:25 +05:30
traversal_ids.present?
2021-04-29 21:17:54 +05:30
end
2022-03-02 08:16:31 +05:30
def use_traversal_ids_for_self_and_hierarchy?
return false unless use_traversal_ids?
return false unless Feature.enabled?(:use_traversal_ids_for_self_and_hierarchy, root_ancestor, default_enabled: :yaml)
traversal_ids.present?
end
2021-09-30 23:02:18 +05:30
def use_traversal_ids_for_ancestors?
return false unless use_traversal_ids?
return false unless Feature.enabled?(:use_traversal_ids_for_ancestors, root_ancestor, default_enabled: :yaml)
traversal_ids.present?
end
2022-01-26 12:08:38 +05:30
def use_traversal_ids_for_ancestors_upto?
return false unless use_traversal_ids?
return false unless Feature.enabled?(:use_traversal_ids_for_ancestors_upto, root_ancestor, default_enabled: :yaml)
traversal_ids.present?
end
2021-09-30 23:02:18 +05:30
def use_traversal_ids_for_root_ancestor?
return false unless Feature.enabled?(:use_traversal_ids_for_root_ancestor, default_enabled: :yaml)
2021-09-04 01:27:46 +05:30
2021-09-30 23:02:18 +05:30
traversal_ids.present?
end
def root_ancestor
return super unless use_traversal_ids_for_root_ancestor?
2021-09-04 01:27:46 +05:30
strong_memoize(:root_ancestor) do
2021-11-11 11:23:49 +05:30
if parent_id.nil?
2021-09-30 23:02:18 +05:30
self
else
Namespace.find_by(id: traversal_ids.first)
end
2021-09-04 01:27:46 +05:30
end
end
2021-04-29 21:17:54 +05:30
def self_and_descendants
2021-06-08 01:23:25 +05:30
return super unless use_traversal_ids?
lineage(top: self)
end
2021-09-04 01:27:46 +05:30
def self_and_descendant_ids
return super unless use_traversal_ids?
self_and_descendants.as_ids
end
2021-06-08 01:23:25 +05:30
def descendants
return super unless use_traversal_ids?
self_and_descendants.where.not(id: id)
end
2022-03-02 08:16:31 +05:30
def self_and_hierarchy
return super unless use_traversal_ids_for_self_and_hierarchy?
self_and_descendants.or(ancestors)
end
2021-06-08 01:23:25 +05:30
def ancestors(hierarchy_order: nil)
2021-09-30 23:02:18 +05:30
return super unless use_traversal_ids_for_ancestors?
2021-06-08 01:23:25 +05:30
return self.class.none if parent_id.blank?
lineage(bottom: parent, hierarchy_order: hierarchy_order)
2021-04-29 21:17:54 +05:30
end
2021-09-30 23:02:18 +05:30
def ancestor_ids(hierarchy_order: nil)
return super unless use_traversal_ids_for_ancestors?
hierarchy_order == :desc ? traversal_ids[0..-2] : traversal_ids[0..-2].reverse
end
2022-01-26 12:08:38 +05:30
# Returns all ancestors upto but excluding the top.
# When no top is given, all ancestors are returned.
# When top is not found, returns all ancestors.
#
# This copies the behavior of the recursive method. We will deprecate
# this behavior soon.
def ancestors_upto(top = nil, hierarchy_order: nil)
return super unless use_traversal_ids_for_ancestors_upto?
# We can't use a default value in the method definition above because
# we need to preserve those specific parameters for super.
hierarchy_order ||= :desc
# Get all ancestor IDs inclusively between top and our parent.
top_index = top ? traversal_ids.find_index(top.id) : 0
ids = traversal_ids[top_index...-1]
ids_string = ids.map { |id| Integer(id) }.join(',')
# WITH ORDINALITY lets us order the result to match traversal_ids order.
from_sql = <<~SQL
unnest(ARRAY[#{ids_string}]::bigint[]) WITH ORDINALITY AS ancestors(id, ord)
INNER JOIN namespaces ON namespaces.id = ancestors.id
SQL
self.class
.from(Arel.sql(from_sql))
.order('ancestors.ord': hierarchy_order)
end
2021-09-30 23:02:18 +05:30
def self_and_ancestors(hierarchy_order: nil)
return super unless use_traversal_ids_for_ancestors?
return self.class.where(id: id) if parent_id.blank?
lineage(bottom: self, hierarchy_order: hierarchy_order)
end
def self_and_ancestor_ids(hierarchy_order: nil)
return super unless use_traversal_ids_for_ancestors?
hierarchy_order == :desc ? traversal_ids : traversal_ids.reverse
end
2021-04-29 21:17:54 +05:30
private
# Update the traversal_ids for the full hierarchy.
#
# NOTE: self.traversal_ids will be stale. Reload for a fresh record.
def sync_traversal_ids
# Clear any previously memoized root_ancestor as our ancestors have changed.
clear_memoization(:root_ancestor)
2021-09-30 23:02:18 +05:30
Namespace::TraversalHierarchy.for_namespace(self).sync_traversal_ids!
2021-04-29 21:17:54 +05:30
end
2021-06-08 01:23:25 +05:30
# Lock the root of the hierarchy we just left, and lock the root of the hierarchy
# we just joined. In most cases the two hierarchies will be the same.
def lock_both_roots
parent_ids = [
parent_id_was || self.id,
parent_id || self.id
].compact
roots = Gitlab::ObjectHierarchy
.new(Namespace.where(id: parent_ids))
.base_and_ancestors
.reorder(nil)
.where(parent_id: nil)
Namespace.lock.select(:id).where(id: roots).order(id: :asc).load
end
2021-04-29 21:17:54 +05:30
# Search this namespace's lineage. Bound inclusively by top node.
2021-06-08 01:23:25 +05:30
def lineage(top: nil, bottom: nil, hierarchy_order: nil)
raise UnboundedSearch, 'Must bound search by either top or bottom' unless top || bottom
2021-04-29 21:17:54 +05:30
2021-12-11 22:18:48 +05:30
skope = self.class
2021-04-29 21:17:54 +05:30
2021-06-08 01:23:25 +05:30
if top
2021-10-27 15:23:28 +05:30
skope = skope.where("traversal_ids @> ('{?}')", top.id)
2021-06-08 01:23:25 +05:30
end
if bottom
2022-01-26 12:08:38 +05:30
skope = skope.where(id: bottom.traversal_ids)
2021-06-08 01:23:25 +05:30
end
# The original `with_depth` attribute in ObjectHierarchy increments as you
# walk away from the "base" namespace. This direction changes depending on
# if you are walking up the ancestors or down the descendants.
if hierarchy_order
depth_sql = "ABS(#{traversal_ids.count} - array_length(traversal_ids, 1))"
2021-11-11 11:23:49 +05:30
skope = skope.select(skope.default_select_columns, "#{depth_sql} as depth")
2021-10-27 15:23:28 +05:30
# The SELECT includes an extra depth attribute. We wrap the SQL in a
# standard SELECT to avoid mismatched attribute errors when trying to
# chain future ActiveRelation commands, and retain the ordering.
skope = self.class
.from(skope, self.class.table_name)
2021-11-11 11:23:49 +05:30
.select(skope.arel_table[Arel.star])
2021-10-27 15:23:28 +05:30
.order(depth: hierarchy_order)
2021-06-08 01:23:25 +05:30
end
skope
2021-04-29 21:17:54 +05:30
end
end
end
end