Merge branch 'refactor-modules-for-import-export' into 'master'

Refactor import_export

Closes #207846

See merge request gitlab-org/gitlab!25623

app/models/ci/pipeline.rb

@@ -599,7 +599,7 @@ module Ci
     # Manually set the notes for a Ci::Pipeline
     # There is no ActiveRecord relation between Ci::Pipeline and notes
     # as they are related to a commit sha. This method helps importing
-    # them using the +Gitlab::ImportExport::ProjectRelationFactory+ class.
+    # them using the +Gitlab::ImportExport::Project::RelationFactory+ class.
     def notes=(notes)
       notes.each do |note|
         note[:id] = nil

app/services/groups/import_export/export_service.rb

@@ -37,7 +37,7 @@ module Groups
       end
 
       def tree_exporter
-        Gitlab::ImportExport::GroupTreeSaver.new(group: @group, current_user: @current_user, shared: @shared, params: @params)
+        Gitlab::ImportExport::Group::TreeSaver.new(group: @group, current_user: @current_user, shared: @shared, params: @params)
       end
 
       def file_saver

app/services/groups/import_export/import_service.rb

@@ -34,7 +34,7 @@ module Groups
       end
 
       def restorer
-        @restorer ||= Gitlab::ImportExport::GroupTreeRestorer.new(user: @current_user,
+        @restorer ||= Gitlab::ImportExport::Group::TreeRestorer.new(user: @current_user,
                                                                   shared: @shared,
                                                                   group: @group,
                                                                   group_hash: nil)

app/services/projects/import_export/export_service.rb

@@ -54,7 +54,7 @@ module Projects
       end
 
       def project_tree_saver
-        Gitlab::ImportExport::ProjectTreeSaver.new(project: project, current_user: current_user, shared: shared, params: params)
+        Gitlab::ImportExport::Project::TreeSaver.new(project: project, current_user: current_user, shared: shared, params: params)
       end
 
       def uploads_saver

doc/development/import_project.md

@@ -81,7 +81,7 @@ The last option is to import a project using a Rails console:
    sudo -u git -H bundle exec rails console RAILS_ENV=production
    ```
 
-1. Create a project and run `ProjectTreeRestorer`:
+1. Create a project and run `Project::TreeRestorer`:
 
    ```ruby
    shared_class = Struct.new(:export_path) do
@@ -98,7 +98,7 @@ The last option is to import a project using a Rails console:
    begin
      #Enable Request store
      RequestStore.begin!
-     Gitlab::ImportExport::ProjectTreeRestorer.new(user: user, shared: shared, project: project).restore
+     Gitlab::ImportExport::Project::TreeRestorer.new(user: user, shared: shared, project: project).restore
    ensure
      RequestStore.end!
      RequestStore.clear!
@@ -128,11 +128,11 @@ The last option is to import a project using a Rails console:
 For Performance testing, we should:
 
 - Import a quite large project, [`gitlabhq`](https://gitlab.com/gitlab-org/quality/performance-data#gitlab-performance-test-framework-data) should be a good example.
-- Measure the execution time of `ProjectTreeRestorer`.
+- Measure the execution time of `Project::TreeRestorer`.
 - Count the number of executed SQL queries during the restore.
 - Observe the number of GC cycles happening.
 
-You can use this [snippet](https://gitlab.com/gitlab-org/gitlab/snippets/1924954), which will restore the project, and measure the execution time of `ProjectTreeRestorer`, number of SQL queries and number of GC cycles happening.
+You can use this [snippet](https://gitlab.com/gitlab-org/gitlab/snippets/1924954), which will restore the project, and measure the execution time of `Project::TreeRestorer`, number of SQL queries and number of GC cycles happening.
 
 You can execute the script from the `gdk/gitlab` directory like this:
 

doc/user/project/settings/import_export.md

@@ -89,7 +89,7 @@ The following items will NOT be exported:
 
 NOTE: **Note:**
 For more details on the specific data persisted in a project export, see the
-[`import_export.yml`](https://gitlab.com/gitlab-org/gitlab/blob/master/lib/gitlab/import_export/import_export.yml) file.
+[`import_export.yml`](https://gitlab.com/gitlab-org/gitlab/blob/master/lib/gitlab/import_export/project/import_export.yml) file.
 
 ## Exporting a project and its data
 

ee/lib/ee/gitlab/import_export/group_project_object_builder.rb

-# frozen_string_literal: true
-
-module EE
-  module Gitlab
-    module ImportExport
-      module GroupProjectObjectBuilder
-        extend ::Gitlab::Utils::Override
-        extend ActiveSupport::Concern
-
-        private
-
-        override :where_clause_for_klass
-        def where_clause_for_klass
-          return attrs_to_arel(attributes.slice('filename')).and(table[:issue_id].eq(nil)) if design?
-
-          super
-        end
-
-        def design?
-          klass == DesignManagement::Design
-        end
-      end
-    end
-  end
-end

ee/lib/ee/gitlab/import_export/project/object_builder.rb

+# frozen_string_literal: true
+
+module EE
+  module Gitlab
+    module ImportExport
+      module Project
+        module ObjectBuilder
+          extend ::Gitlab::Utils::Override
+          extend ActiveSupport::Concern
+
+          private
+
+          override :where_clause_for_klass
+          def where_clause_for_klass
+            return attrs_to_arel(attributes.slice('filename')).and(table[:issue_id].eq(nil)) if design?
+
+            super
+          end
+
+          def design?
+            klass == DesignManagement::Design
+          end
+        end
+      end
+    end
+  end
+end

ee/lib/ee/gitlab/import_export/project/relation_factory.rb

+# frozen_string_literal: true
+
+module EE
+  module Gitlab
+    module ImportExport
+      module Project
+        module RelationFactory
+          extend ActiveSupport::Concern
+
+          EE_OVERRIDES = {
+            design: 'DesignManagement::Design',
+            designs: 'DesignManagement::Design',
+            design_versions: 'DesignManagement::Version',
+            actions: 'DesignManagement::Action',
+            deploy_access_levels: 'ProtectedEnvironment::DeployAccessLevel',
+            unprotect_access_levels: 'ProtectedBranch::UnprotectAccessLevel'
+          }.freeze
+
+          EE_EXISTING_OBJECT_RELATIONS = %i[DesignManagement::Design].freeze
+
+          class_methods do
+            extend ::Gitlab::Utils::Override
+
+            override :overrides
+            def overrides
+              super.merge(EE_OVERRIDES)
+            end
+
+            override :existing_object_relations
+            def existing_object_relations
+              super + EE_EXISTING_OBJECT_RELATIONS
+            end
+          end
+        end
+      end
+    end
+  end
+end

ee/lib/ee/gitlab/import_export/project_relation_factory.rb

-# frozen_string_literal: true
-
-module EE
-  module Gitlab
-    module ImportExport
-      module ProjectRelationFactory
-        extend ActiveSupport::Concern
-
-        EE_OVERRIDES = {
-          design: 'DesignManagement::Design',
-          designs: 'DesignManagement::Design',
-          design_versions: 'DesignManagement::Version',
-          actions: 'DesignManagement::Action',
-          deploy_access_levels: 'ProtectedEnvironment::DeployAccessLevel',
-          unprotect_access_levels: 'ProtectedBranch::UnprotectAccessLevel'
-        }.freeze
-
-        EE_EXISTING_OBJECT_RELATIONS = %i[DesignManagement::Design].freeze
-
-        class_methods do
-          extend ::Gitlab::Utils::Override
-
-          override :overrides
-          def overrides
-            super.merge(EE_OVERRIDES)
-          end
-
-          override :existing_object_relations
-          def existing_object_relations
-            super + EE_EXISTING_OBJECT_RELATIONS
-          end
-        end
-      end
-    end
-  end
-end

ee/spec/lib/ee/gitlab/import_export/group_tree_restorer_spec.rb → ee/spec/lib/ee/gitlab/import_export/group/tree_restorer_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::GroupTreeRestorer do
+describe Gitlab::ImportExport::Group::TreeRestorer do
   include ImportExport::CommonUtil
 
   let(:user) { create(:user) }

ee/spec/lib/ee/gitlab/import_export/group_tree_saver_spec.rb → ee/spec/lib/ee/gitlab/import_export/group/tree_saver_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::GroupTreeSaver do
+describe Gitlab::ImportExport::Group::TreeSaver do
   describe 'saves the group tree into a json object' do
     let_it_be(:user) { create(:user) }
     let_it_be(:group) { create(:group) }

ee/spec/lib/ee/gitlab/import_export/project_tree_restorer_spec.rb → ee/spec/lib/ee/gitlab/import_export/project/tree_restorer_spec.rb

 # frozen_string_literal: true
 require 'spec_helper'
 
-describe Gitlab::ImportExport::ProjectTreeRestorer do
+describe Gitlab::ImportExport::Project::TreeRestorer do
   include ImportExport::CommonUtil
   using RSpec::Parameterized::TableSyntax
 

ee/spec/lib/ee/gitlab/import_export/project_tree_saver_spec.rb → ee/spec/lib/ee/gitlab/import_export/project/tree_saver_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::ProjectTreeSaver do
+describe Gitlab::ImportExport::Project::TreeSaver do
   describe 'saves the project tree into a json object' do
     let_it_be(:user) { create(:user) }
     let_it_be(:group) { create(:group) }

ee/spec/lib/gitlab/import_export/group_relation_factory_spec.rb → ee/spec/lib/gitlab/import_export/group/relation_factory_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::GroupRelationFactory do
+describe Gitlab::ImportExport::Group::RelationFactory do
   let(:group) { create(:group) }
   let(:members_mapper) { double('members_mapper').as_null_object }
   let(:user) { create(:admin) }
@@ -11,7 +11,7 @@ describe Gitlab::ImportExport::GroupRelationFactory do
     described_class.create(relation_sym:   relation_sym,
                            relation_hash:  relation_hash,
                            members_mapper: members_mapper,
-                           object_builder: Gitlab::ImportExport::GroupObjectBuilder,
+                           object_builder: Gitlab::ImportExport::Group::ObjectBuilder,
                            user:           user,
                            importable:     group,
                            excluded_keys:  excluded_keys)

ee/spec/lib/gitlab/import_export/group_project_object_builder_spec.rb → ee/spec/lib/gitlab/import_export/project/object_builder_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::GroupProjectObjectBuilder do
+describe Gitlab::ImportExport::Project::ObjectBuilder do
   let!(:group) { create(:group, :private) }
   let!(:subgroup) { create(:group, :private, parent: group) }
   let!(:project) do

lib/gitlab/import_export.rb

@@ -43,7 +43,7 @@ module Gitlab
     end
 
     def config_file
-      Rails.root.join('lib/gitlab/import_export/import_export.yml')
+      Rails.root.join('lib/gitlab/import_export/project/import_export.yml')
     end
 
     def version_filename
@@ -77,7 +77,7 @@ module Gitlab
     end
 
     def group_config_file
-      Rails.root.join('lib/gitlab/import_export/group_import_export.yml')
+      Rails.root.join('lib/gitlab/import_export/group/import_export.yml')
     end
   end
 end

lib/gitlab/import_export/attribute_cleaner.rb

@@ -4,8 +4,8 @@ module Gitlab
   module ImportExport
     class AttributeCleaner
       ALLOWED_REFERENCES = [
-        *ProjectRelationFactory::PROJECT_REFERENCES,
-        *ProjectRelationFactory::USER_REFERENCES,
+        *Gitlab::ImportExport::Project::RelationFactory::PROJECT_REFERENCES,
+        *Gitlab::ImportExport::Project::RelationFactory::USER_REFERENCES,
         'group_id',
         'commit_id',
         'discussion_id',

lib/gitlab/import_export/base/object_builder.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Base
+      # Base class for Group & Project Object Builders.
+      # This class is not intended to be used on its own but
+      # rather inherited from.
+      #
+      # Cache keeps 1000 entries at most, 1000 is chosen based on:
+      #    - one cache entry uses around 0.5K memory, 1000 items uses around 500K.
+      #      (leave some buffer it should be less than 1M). It is afforable cost for project import.
+      #    - for projects in Gitlab.com, it seems 1000 entries for labels/milestones is enough.
+      #      For example, gitlab has ~970 labels and 26 milestones.
+      LRU_CACHE_SIZE = 1000
+
+      class ObjectBuilder
+        def self.build(*args)
+          new(*args).find
+        end
+
+        def initialize(klass, attributes)
+          @klass = klass.ancestors.include?(Label) ? Label : klass
+          @attributes = attributes
+
+          if Gitlab::SafeRequestStore.active?
+            @lru_cache = cache_from_request_store
+            @cache_key = [klass, attributes]
+          end
+        end
+
+        def find
+          find_with_cache do
+            find_object || klass.create(prepare_attributes)
+          end
+        end
+
+        protected
+
+        def where_clauses
+          raise NotImplementedError
+        end
+
+        # attributes wrapped in a method to be
+        # adjusted in sub-class if needed
+        def prepare_attributes
+          attributes
+        end
+
+        private
+
+        attr_reader :klass, :attributes, :lru_cache, :cache_key
+
+        def find_with_cache
+          return yield unless lru_cache && cache_key
+
+          lru_cache[cache_key] ||= yield
+        end
+
+        def cache_from_request_store
+          Gitlab::SafeRequestStore[:lru_cache] ||= LruRedux::Cache.new(LRU_CACHE_SIZE)
+        end
+
+        def find_object
+          klass.where(where_clause).first
+        end
+
+        def where_clause
+          where_clauses.reduce(:and)
+        end
+
+        def table
+          @table ||= klass.arel_table
+        end
+
+        # Returns Arel clause:
+        # `"{table_name}"."{attrs.keys[0]}" = '{attrs.values[0]} AND {table_name}"."{attrs.keys[1]}" = '{attrs.values[1]}"`
+        # from the given Hash of attributes.
+        def attrs_to_arel(attrs)
+          attrs.map do |key, value|
+            table[key].eq(value)
+          end.reduce(:and)
+        end
+
+        # Returns Arel clause `"{table_name}"."title" = '{attributes['title']}'`
+        # if attributes has 'title key, otherwise `nil`.
+        def where_clause_for_title
+          attrs_to_arel(attributes.slice('title'))
+        end
+
+        # Returns Arel clause `"{table_name}"."description" = '{attributes['description']}'`
+        # if attributes has 'description key, otherwise `nil`.
+        def where_clause_for_description
+          attrs_to_arel(attributes.slice('description'))
+        end
+
+        # Returns Arel clause `"{table_name}"."created_at" = '{attributes['created_at']}'`
+        # if attributes has 'created_at key, otherwise `nil`.
+        def where_clause_for_created_at
+          attrs_to_arel(attributes.slice('created_at'))
+        end
+      end
+    end
+  end
+end

lib/gitlab/import_export/base/relation_factory.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Base
+      class RelationFactory
+        include Gitlab::Utils::StrongMemoize
+
+        IMPORTED_OBJECT_MAX_RETRIES = 5.freeze
+
+        OVERRIDES = {}.freeze
+        EXISTING_OBJECT_RELATIONS = %i[].freeze
+
+        # This represents all relations that have unique key on `project_id` or `group_id`
+        UNIQUE_RELATIONS = %i[].freeze
+
+        USER_REFERENCES = %w[
+           author_id
+           assignee_id
+           updated_by_id
+           merged_by_id
+           latest_closed_by_id
+           user_id
+           created_by_id
+           last_edited_by_id
+           merge_user_id
+           resolved_by_id
+           closed_by_id
+           owner_id
+         ].freeze
+
+        TOKEN_RESET_MODELS = %i[Project Namespace Group Ci::Trigger Ci::Build Ci::Runner ProjectHook].freeze
+
+        def self.create(*args)
+          new(*args).create
+        end
+
+        def self.relation_class(relation_name)
+          # There are scenarios where the model is pluralized (e.g.
+          # MergeRequest::Metrics), and we don't want to force it to singular
+          # with #classify.
+          relation_name.to_s.classify.constantize
+        rescue NameError
+          relation_name.to_s.constantize
+        end
+
+        def initialize(relation_sym:, relation_hash:, members_mapper:, object_builder:, user:, importable:, excluded_keys: [])
+          @relation_name = self.class.overrides[relation_sym]&.to_sym || relation_sym
+          @relation_hash = relation_hash.except('noteable_id')
+          @members_mapper = members_mapper
+          @object_builder = object_builder
+          @user = user
+          @importable = importable
+          @imported_object_retries = 0
+          @relation_hash[importable_column_name] = @importable.id
+
+          # Remove excluded keys from relation_hash
+          # We don't do this in the parsed_relation_hash because of the 'transformed attributes'
+          # For example, MergeRequestDiffFiles exports its diff attribute as utf8_diff. Then,
+          # in the create method that attribute is renamed to diff. And because diff is an excluded key,
+          # if we clean the excluded keys in the parsed_relation_hash, it will be removed
+          # from the object attributes and the export will fail.
+          @relation_hash.except!(*excluded_keys)
+        end
+
+        # Creates an object from an actual model with name "relation_sym" with params from
+        # the relation_hash, updating references with new object IDs, mapping users using
+        # the "members_mapper" object, also updating notes if required.
+        def create
+          return if invalid_relation?
+
+          setup_base_models
+          setup_models
+
+          generate_imported_object
+        end
+
+        def self.overrides
+          self::OVERRIDES
+        end
+
+        def self.existing_object_relations
+          self::EXISTING_OBJECT_RELATIONS
+        end
+
+        private
+
+        def invalid_relation?
+          false
+        end
+
+        def setup_models
+          raise NotImplementedError
+        end
+
+        def unique_relations
+          # define in sub-class if any
+          self.class::UNIQUE_RELATIONS
+        end
+
+        def setup_base_models
+          update_user_references
+          remove_duplicate_assignees
+          reset_tokens!
+          remove_encrypted_attributes!
+        end
+
+        def update_user_references
+          self.class::USER_REFERENCES.each do |reference|
+            if @relation_hash[reference]
+              @relation_hash[reference] = @members_mapper.map[@relation_hash[reference]]
+            end
+          end
+        end
+
+        def remove_duplicate_assignees
+          return unless @relation_hash['issue_assignees']
+
+          # When an assignee did not exist in the members mapper, the importer is
+          # assigned. We only need to assign each user once.
+          @relation_hash['issue_assignees'].uniq!(&:user_id)
+        end
+
+        def generate_imported_object
+          imported_object
+        end
+
+        def reset_tokens!
+          return unless Gitlab::ImportExport.reset_tokens? && self.class::TOKEN_RESET_MODELS.include?(@relation_name)
+
+          # If we import/export to the same instance, tokens will have to be reset.
+          # We also have to reset them to avoid issues when the gitlab secrets file cannot be copied across.
+          relation_class.attribute_names.select { |name| name.include?('token') }.each do |token|
+            @relation_hash[token] = nil
+          end
+        end
+
+        def remove_encrypted_attributes!
+          return unless relation_class.respond_to?(:encrypted_attributes) && relation_class.encrypted_attributes.any?
+
+          relation_class.encrypted_attributes.each_key do |key|
+            @relation_hash[key.to_s] = nil
+          end
+        end
+
+        def relation_class
+          @relation_class ||= self.class.relation_class(@relation_name)
+        end
+
+        def importable_column_name
+          importable_class_name.concat('_id')
+        end
+
+        def importable_class_name
+          @importable.class.to_s.downcase
+        end
+
+        def imported_object
+          if existing_or_new_object.respond_to?(:importing)
+            existing_or_new_object.importing = true
+          end
+
+          existing_or_new_object
+        rescue ActiveRecord::RecordNotUnique
+          # as the operation is not atomic, retry in the unlikely scenario an INSERT is
+          # performed on the same object between the SELECT and the INSERT
+          @imported_object_retries += 1
+          retry if @imported_object_retries < IMPORTED_OBJECT_MAX_RETRIES
+        end
+
+        def parsed_relation_hash
+          @parsed_relation_hash ||= Gitlab::ImportExport::AttributeCleaner.clean(relation_hash: @relation_hash,
+                                                                                 relation_class: relation_class)
+        end
+
+        def existing_or_new_object
+          # Only find existing records to avoid mapping tables such as milestones
+          # Otherwise always create the record, skipping the extra SELECT clause.
+          @existing_or_new_object ||= begin
+            if existing_object?
+              attribute_hash = attribute_hash_for(['events'])
+
+              existing_object.assign_attributes(attribute_hash) if attribute_hash.any?
+
+              existing_object
+            else
+              # Because of single-type inheritance, we need to be careful to use the `type` field
+              # See https://gitlab.com/gitlab-org/gitlab/issues/34860#note_235321497
+              inheritance_column = relation_class.try(:inheritance_column)
+              inheritance_attributes = parsed_relation_hash.slice(inheritance_column)
+              object = relation_class.new(inheritance_attributes)
+              object.assign_attributes(parsed_relation_hash)
+              object
+            end
+          end
+        end
+
+        def attribute_hash_for(attributes)
+          attributes.each_with_object({}) do |hash, value|
+            hash[value] = parsed_relation_hash.delete(value) if parsed_relation_hash[value]
+            hash
+          end
+        end
+
+        def existing_object
+          @existing_object ||= find_or_create_object!
+        end
+
+        def unique_relation_object
+          unique_relation_object = relation_class.find_or_create_by(importable_column_name => @importable.id)
+          unique_relation_object.assign_attributes(parsed_relation_hash)
+          unique_relation_object
+        end
+
+        def find_or_create_object!
+          return unique_relation_object if unique_relation?
+
+          # Can't use IDs as validation exists calling `group` or `project` attributes
+          finder_hash = parsed_relation_hash.tap do |hash|
+            if relation_class.attribute_method?('group_id') && @importable.is_a?(::Project)
+              hash['group'] = @importable.group
+            end
+
+            hash[importable_class_name] = @importable if relation_class.reflect_on_association(importable_class_name.to_sym)
+            hash.delete(importable_column_name)
+          end
+
+          @object_builder.build(relation_class, finder_hash)
+        end
+
+        def setup_note
+          set_note_author
+          # attachment is deprecated and note uploads are handled by Markdown uploader
+          @relation_hash['attachment'] = nil
+        end
+
+        # Sets the author for a note. If the user importing the project
+        # has admin access, an actual mapping with new project members
+        # will be used. Otherwise, a note stating the original author name
+        # is left.
+        def set_note_author
+          old_author_id = @relation_hash['author_id']
+          author = @relation_hash.delete('author')
+
+          update_note_for_missing_author(author['name']) unless has_author?(old_author_id)
+        end
+
+        def has_author?(old_author_id)
+          admin_user? && @members_mapper.include?(old_author_id)
+        end
+
+        def missing_author_note(updated_at, author_name)
+          timestamp = updated_at.split('.').first
+          "\n\n *By #{author_name} on #{timestamp} (imported from GitLab project)*"
+        end
+
+        def update_note_for_missing_author(author_name)
+          @relation_hash['note'] = '*Blank note*' if @relation_hash['note'].blank?
+          @relation_hash['note'] = "#{@relation_hash['note']}#{missing_author_note(@relation_hash['updated_at'], author_name)}"
+        end
+
+        def admin_user?
+          @user.admin?
+        end
+
+        def existing_object?
+          strong_memoize(:_existing_object) do
+            self.class.existing_object_relations.include?(@relation_name) || unique_relation?
+          end
+        end
+
+        def unique_relation?
+          strong_memoize(:unique_relation) do
+            importable_foreign_key.present? &&
+              (has_unique_index_on_importable_fk? || uses_importable_fk_as_primary_key?)
+          end
+        end
+
+        def has_unique_index_on_importable_fk?
+          cache = cached_has_unique_index_on_importable_fk
+          table_name = relation_class.table_name
+          return cache[table_name] if cache.has_key?(table_name)
+
+          index_exists =
+            ActiveRecord::Base.connection.index_exists?(
+              relation_class.table_name,
+              importable_foreign_key,
+              unique: true)
+
+          cache[table_name] = index_exists
+        end
+
+        # Avoid unnecessary DB requests
+        def cached_has_unique_index_on_importable_fk
+          Thread.current[:cached_has_unique_index_on_importable_fk] ||= {}
+        end
+
+        def uses_importable_fk_as_primary_key?
+          relation_class.primary_key == importable_foreign_key
+        end
+
+        def importable_foreign_key
+          relation_class.reflect_on_association(importable_class_name.to_sym)&.foreign_key
+        end
+      end
+    end
+  end
+end

lib/gitlab/import_export/base_object_builder.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    # Base class for Group & Project Object Builders.
-    # This class is not intended to be used on its own but
-    # rather inherited from.
-    #
-    # Cache keeps 1000 entries at most, 1000 is chosen based on:
-    #    - one cache entry uses around 0.5K memory, 1000 items uses around 500K.
-    #      (leave some buffer it should be less than 1M). It is afforable cost for project import.
-    #    - for projects in Gitlab.com, it seems 1000 entries for labels/milestones is enough.
-    #      For example, gitlab has ~970 labels and 26 milestones.
-    LRU_CACHE_SIZE = 1000
-
-    class BaseObjectBuilder
-      def self.build(*args)
-        new(*args).find
-      end
-
-      def initialize(klass, attributes)
-        @klass = klass.ancestors.include?(Label) ? Label : klass
-        @attributes = attributes
-
-        if Gitlab::SafeRequestStore.active?
-          @lru_cache = cache_from_request_store
-          @cache_key = [klass, attributes]
-        end
-      end
-
-      def find
-        find_with_cache do
-          find_object || klass.create(prepare_attributes)
-        end
-      end
-
-      protected
-
-      def where_clauses
-        raise NotImplementedError
-      end
-
-      # attributes wrapped in a method to be
-      # adjusted in sub-class if needed
-      def prepare_attributes
-        attributes
-      end
-
-      private
-
-      attr_reader :klass, :attributes, :lru_cache, :cache_key
-
-      def find_with_cache
-        return yield unless lru_cache && cache_key
-
-        lru_cache[cache_key] ||= yield
-      end
-
-      def cache_from_request_store
-        Gitlab::SafeRequestStore[:lru_cache] ||= LruRedux::Cache.new(LRU_CACHE_SIZE)
-      end
-
-      def find_object
-        klass.where(where_clause).first
-      end
-
-      def where_clause
-        where_clauses.reduce(:and)
-      end
-
-      def table
-        @table ||= klass.arel_table
-      end
-
-      # Returns Arel clause:
-      # `"{table_name}"."{attrs.keys[0]}" = '{attrs.values[0]} AND {table_name}"."{attrs.keys[1]}" = '{attrs.values[1]}"`
-      # from the given Hash of attributes.
-      def attrs_to_arel(attrs)
-        attrs.map do |key, value|
-          table[key].eq(value)
-        end.reduce(:and)
-      end
-
-      # Returns Arel clause `"{table_name}"."title" = '{attributes['title']}'`
-      # if attributes has 'title key, otherwise `nil`.
-      def where_clause_for_title
-        attrs_to_arel(attributes.slice('title'))
-      end
-
-      # Returns Arel clause `"{table_name}"."description" = '{attributes['description']}'`
-      # if attributes has 'description key, otherwise `nil`.
-      def where_clause_for_description
-        attrs_to_arel(attributes.slice('description'))
-      end
-
-      # Returns Arel clause `"{table_name}"."created_at" = '{attributes['created_at']}'`
-      # if attributes has 'created_at key, otherwise `nil`.
-      def where_clause_for_created_at
-        attrs_to_arel(attributes.slice('created_at'))
-      end
-    end
-  end
-end

lib/gitlab/import_export/base_relation_factory.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    class BaseRelationFactory
-      include Gitlab::Utils::StrongMemoize
-
-      IMPORTED_OBJECT_MAX_RETRIES = 5.freeze
-
-      OVERRIDES = {}.freeze
-      EXISTING_OBJECT_RELATIONS = %i[].freeze
-
-      # This represents all relations that have unique key on `project_id` or `group_id`
-      UNIQUE_RELATIONS = %i[].freeze
-
-      USER_REFERENCES = %w[
-         author_id
-         assignee_id
-         updated_by_id
-         merged_by_id
-         latest_closed_by_id
-         user_id
-         created_by_id
-         last_edited_by_id
-         merge_user_id
-         resolved_by_id
-         closed_by_id
-         owner_id
-       ].freeze
-
-      TOKEN_RESET_MODELS = %i[Project Namespace Group Ci::Trigger Ci::Build Ci::Runner ProjectHook].freeze
-
-      def self.create(*args)
-        new(*args).create
-      end
-
-      def self.relation_class(relation_name)
-        # There are scenarios where the model is pluralized (e.g.
-        # MergeRequest::Metrics), and we don't want to force it to singular
-        # with #classify.
-        relation_name.to_s.classify.constantize
-      rescue NameError
-        relation_name.to_s.constantize
-      end
-
-      def initialize(relation_sym:, relation_hash:, members_mapper:, object_builder:, user:, importable:, excluded_keys: [])
-        @relation_name = self.class.overrides[relation_sym]&.to_sym || relation_sym
-        @relation_hash = relation_hash.except('noteable_id')
-        @members_mapper = members_mapper
-        @object_builder = object_builder
-        @user = user
-        @importable = importable
-        @imported_object_retries = 0
-        @relation_hash[importable_column_name] = @importable.id
-
-        # Remove excluded keys from relation_hash
-        # We don't do this in the parsed_relation_hash because of the 'transformed attributes'
-        # For example, MergeRequestDiffFiles exports its diff attribute as utf8_diff. Then,
-        # in the create method that attribute is renamed to diff. And because diff is an excluded key,
-        # if we clean the excluded keys in the parsed_relation_hash, it will be removed
-        # from the object attributes and the export will fail.
-        @relation_hash.except!(*excluded_keys)
-      end
-
-      # Creates an object from an actual model with name "relation_sym" with params from
-      # the relation_hash, updating references with new object IDs, mapping users using
-      # the "members_mapper" object, also updating notes if required.
-      def create
-        return if invalid_relation?
-
-        setup_base_models
-        setup_models
-
-        generate_imported_object
-      end
-
-      def self.overrides
-        self::OVERRIDES
-      end
-
-      def self.existing_object_relations
-        self::EXISTING_OBJECT_RELATIONS
-      end
-
-      private
-
-      def invalid_relation?
-        false
-      end
-
-      def setup_models
-        raise NotImplementedError
-      end
-
-      def unique_relations
-        # define in sub-class if any
-        self.class::UNIQUE_RELATIONS
-      end
-
-      def setup_base_models
-        update_user_references
-        remove_duplicate_assignees
-        reset_tokens!
-        remove_encrypted_attributes!
-      end
-
-      def update_user_references
-        self.class::USER_REFERENCES.each do |reference|
-          if @relation_hash[reference]
-            @relation_hash[reference] = @members_mapper.map[@relation_hash[reference]]
-          end
-        end
-      end
-
-      def remove_duplicate_assignees
-        return unless @relation_hash['issue_assignees']
-
-        # When an assignee did not exist in the members mapper, the importer is
-        # assigned. We only need to assign each user once.
-        @relation_hash['issue_assignees'].uniq!(&:user_id)
-      end
-
-      def generate_imported_object
-        imported_object
-      end
-
-      def reset_tokens!
-        return unless Gitlab::ImportExport.reset_tokens? && self.class::TOKEN_RESET_MODELS.include?(@relation_name)
-
-        # If we import/export to the same instance, tokens will have to be reset.
-        # We also have to reset them to avoid issues when the gitlab secrets file cannot be copied across.
-        relation_class.attribute_names.select { |name| name.include?('token') }.each do |token|
-          @relation_hash[token] = nil
-        end
-      end
-
-      def remove_encrypted_attributes!
-        return unless relation_class.respond_to?(:encrypted_attributes) && relation_class.encrypted_attributes.any?
-
-        relation_class.encrypted_attributes.each_key do |key|
-          @relation_hash[key.to_s] = nil
-        end
-      end
-
-      def relation_class
-        @relation_class ||= self.class.relation_class(@relation_name)
-      end
-
-      def importable_column_name
-        importable_class_name.concat('_id')
-      end
-
-      def importable_class_name
-        @importable.class.to_s.downcase
-      end
-
-      def imported_object
-        if existing_or_new_object.respond_to?(:importing)
-          existing_or_new_object.importing = true
-        end
-
-        existing_or_new_object
-      rescue ActiveRecord::RecordNotUnique
-        # as the operation is not atomic, retry in the unlikely scenario an INSERT is
-        # performed on the same object between the SELECT and the INSERT
-        @imported_object_retries += 1
-        retry if @imported_object_retries < IMPORTED_OBJECT_MAX_RETRIES
-      end
-
-      def parsed_relation_hash
-        @parsed_relation_hash ||= Gitlab::ImportExport::AttributeCleaner.clean(relation_hash: @relation_hash,
-                                                                               relation_class: relation_class)
-      end
-
-      def existing_or_new_object
-        # Only find existing records to avoid mapping tables such as milestones
-        # Otherwise always create the record, skipping the extra SELECT clause.
-        @existing_or_new_object ||= begin
-          if existing_object?
-            attribute_hash = attribute_hash_for(['events'])
-
-            existing_object.assign_attributes(attribute_hash) if attribute_hash.any?
-
-            existing_object
-          else
-            # Because of single-type inheritance, we need to be careful to use the `type` field
-            # See https://gitlab.com/gitlab-org/gitlab/issues/34860#note_235321497
-            inheritance_column = relation_class.try(:inheritance_column)
-            inheritance_attributes = parsed_relation_hash.slice(inheritance_column)
-            object = relation_class.new(inheritance_attributes)
-            object.assign_attributes(parsed_relation_hash)
-            object
-          end
-        end
-      end
-
-      def attribute_hash_for(attributes)
-        attributes.each_with_object({}) do |hash, value|
-          hash[value] = parsed_relation_hash.delete(value) if parsed_relation_hash[value]
-          hash
-        end
-      end
-
-      def existing_object
-        @existing_object ||= find_or_create_object!
-      end
-
-      def unique_relation_object
-        unique_relation_object = relation_class.find_or_create_by(importable_column_name => @importable.id)
-        unique_relation_object.assign_attributes(parsed_relation_hash)
-        unique_relation_object
-      end
-
-      def find_or_create_object!
-        return unique_relation_object if unique_relation?
-
-        # Can't use IDs as validation exists calling `group` or `project` attributes
-        finder_hash = parsed_relation_hash.tap do |hash|
-          if relation_class.attribute_method?('group_id') && @importable.is_a?(Project)
-            hash['group'] = @importable.group
-          end
-
-          hash[importable_class_name] = @importable if relation_class.reflect_on_association(importable_class_name.to_sym)
-          hash.delete(importable_column_name)
-        end
-
-        @object_builder.build(relation_class, finder_hash)
-      end
-
-      def setup_note
-        set_note_author
-        # attachment is deprecated and note uploads are handled by Markdown uploader
-        @relation_hash['attachment'] = nil
-      end
-
-      # Sets the author for a note. If the user importing the project
-      # has admin access, an actual mapping with new project members
-      # will be used. Otherwise, a note stating the original author name
-      # is left.
-      def set_note_author
-        old_author_id = @relation_hash['author_id']
-        author = @relation_hash.delete('author')
-
-        update_note_for_missing_author(author['name']) unless has_author?(old_author_id)
-      end
-
-      def has_author?(old_author_id)
-        admin_user? && @members_mapper.include?(old_author_id)
-      end
-
-      def missing_author_note(updated_at, author_name)
-        timestamp = updated_at.split('.').first
-        "\n\n *By #{author_name} on #{timestamp} (imported from GitLab project)*"
-      end
-
-      def update_note_for_missing_author(author_name)
-        @relation_hash['note'] = '*Blank note*' if @relation_hash['note'].blank?
-        @relation_hash['note'] = "#{@relation_hash['note']}#{missing_author_note(@relation_hash['updated_at'], author_name)}"
-      end
-
-      def admin_user?
-        @user.admin?
-      end
-
-      def existing_object?
-        strong_memoize(:_existing_object) do
-          self.class.existing_object_relations.include?(@relation_name) || unique_relation?
-        end
-      end
-
-      def unique_relation?
-        strong_memoize(:unique_relation) do
-          importable_foreign_key.present? &&
-            (has_unique_index_on_importable_fk? || uses_importable_fk_as_primary_key?)
-        end
-      end
-
-      def has_unique_index_on_importable_fk?
-        cache = cached_has_unique_index_on_importable_fk
-        table_name = relation_class.table_name
-        return cache[table_name] if cache.has_key?(table_name)
-
-        index_exists =
-          ActiveRecord::Base.connection.index_exists?(
-            relation_class.table_name,
-            importable_foreign_key,
-            unique: true)
-
-        cache[table_name] = index_exists
-      end
-
-      # Avoid unnecessary DB requests
-      def cached_has_unique_index_on_importable_fk
-        Thread.current[:cached_has_unique_index_on_importable_fk] ||= {}
-      end
-
-      def uses_importable_fk_as_primary_key?
-        relation_class.primary_key == importable_foreign_key
-      end
-
-      def importable_foreign_key
-        relation_class.reflect_on_association(importable_class_name.to_sym)&.foreign_key
-      end
-    end
-  end
-end

lib/gitlab/import_export/group/object_builder.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Group
+      # Given a class, it finds or creates a new object at group level.
+      #
+      # Example:
+      #   `Group::ObjectBuilder.build(Label, label_attributes)`
+      #    finds or initializes a label with the given attributes.
+      class ObjectBuilder < Base::ObjectBuilder
+        def self.build(*args)
+          ::Group.transaction do
+            super
+          end
+        end
+
+        def initialize(klass, attributes)
+          super
+
+          @group = @attributes['group']
+
+          update_description
+        end
+
+        private
+
+        attr_reader :group
+
+        # Convert description empty string to nil
+        # due to existing object being saved with description: nil
+        # Which makes object lookup to fail since nil != ''
+        def update_description
+          attributes['description'] = nil if attributes['description'] == ''
+        end
+
+        def where_clauses
+          [
+            where_clause_base,
+            where_clause_for_title,
+            where_clause_for_description,
+            where_clause_for_created_at
+          ].compact
+        end
+
+        # Returns Arel clause `"{table_name}"."group_id" = {group.id}`
+        def where_clause_base
+          table[:group_id].in(group_and_ancestor_ids)
+        end
+
+        def group_and_ancestor_ids
+          group.ancestors.map(&:id) << group.id
+        end
+      end
+    end
+  end
+end

lib/gitlab/import_export/group/relation_factory.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Group
+      class RelationFactory < Base::RelationFactory
+        OVERRIDES = {
+          labels:     :group_labels,
+          priorities: :label_priorities,
+          label:      :group_label,
+          parent:     :epic
+        }.freeze
+
+        EXISTING_OBJECT_RELATIONS = %i[
+          epic
+          epics
+          milestone
+          milestones
+          label
+          labels
+          group_label
+          group_labels
+        ].freeze
+
+        private
+
+        def setup_models
+          setup_note if @relation_name == :notes
+
+          update_group_references
+        end
+
+        def update_group_references
+          return unless self.class.existing_object_relations.include?(@relation_name)
+          return unless @relation_hash['group_id']
+
+          @relation_hash['group_id'] = @importable.id
+        end
+      end
+    end
+  end
+end

lib/gitlab/import_export/group/tree_restorer.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Group
+      class TreeRestorer
+        attr_reader :user
+        attr_reader :shared
+        attr_reader :group
+
+        def initialize(user:, shared:, group:, group_hash:)
+          @path = File.join(shared.export_path, 'group.json')
+          @user = user
+          @shared = shared
+          @group = group
+          @group_hash = group_hash
+        end
+
+        def restore
+          @tree_hash = @group_hash || read_tree_hash
+          @group_members = @tree_hash.delete('members')
+          @children = @tree_hash.delete('children')
+
+          if members_mapper.map && restorer.restore
+            @children&.each do |group_hash|
+              group = create_group(group_hash: group_hash, parent_group: @group)
+              shared = Gitlab::ImportExport::Shared.new(group)
+
+              self.class.new(
+                user: @user,
+                shared: shared,
+                group: group,
+                group_hash: group_hash
+              ).restore
+            end
+          end
+
+          return false if @shared.errors.any?
+
+          true
+        rescue => e
+          @shared.error(e)
+          false
+        end
+
+        private
+
+        def read_tree_hash
+          json = IO.read(@path)
+          ActiveSupport::JSON.decode(json)
+        rescue => e
+          @shared.logger.error(
+            group_id:   @group.id,
+            group_name: @group.name,
+            message:    "Import/Export error: #{e.message}"
+          )
+
+          raise Gitlab::ImportExport::Error.new('Incorrect JSON format')
+        end
+
+        def restorer
+          @relation_tree_restorer ||= RelationTreeRestorer.new(
+            user:             @user,
+            shared:           @shared,
+            importable:       @group,
+            tree_hash:        @tree_hash.except('name', 'path'),
+            members_mapper:   members_mapper,
+            object_builder:   object_builder,
+            relation_factory: relation_factory,
+            reader:           reader
+          )
+        end
+
+        def create_group(group_hash:, parent_group:)
+          group_params = {
+            name:      group_hash['name'],
+            path:      group_hash['path'],
+            parent_id: parent_group&.id,
+            visibility_level: sub_group_visibility_level(group_hash, parent_group)
+          }
+
+          ::Groups::CreateService.new(@user, group_params).execute
+        end
+
+        def sub_group_visibility_level(group_hash, parent_group)
+          original_visibility_level = group_hash['visibility_level'] || Gitlab::VisibilityLevel::PRIVATE
+
+          if parent_group && parent_group.visibility_level < original_visibility_level
+            Gitlab::VisibilityLevel.closest_allowed_level(parent_group.visibility_level)
+          else
+            original_visibility_level
+          end
+        end
+
+        def members_mapper
+          @members_mapper ||= Gitlab::ImportExport::MembersMapper.new(exported_members: @group_members, user: @user, importable: @group)
+        end
+
+        def relation_factory
+          Gitlab::ImportExport::Group::RelationFactory
+        end
+
+        def object_builder
+          Gitlab::ImportExport::Group::ObjectBuilder
+        end
+
+        def reader
+          @reader ||= Gitlab::ImportExport::Reader.new(
+            shared: @shared,
+            config: Gitlab::ImportExport::Config.new(
+              config: Gitlab::ImportExport.group_config_file
+            ).to_h
+          )
+        end
+      end
+    end
+  end
+end

lib/gitlab/import_export/group/tree_saver.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Group
+      class TreeSaver
+        attr_reader :full_path, :shared
+
+        def initialize(group:, current_user:, shared:, params: {})
+          @params       = params
+          @current_user = current_user
+          @shared       = shared
+          @group        = group
+          @full_path    = File.join(@shared.export_path, ImportExport.group_filename)
+        end
+
+        def save
+          group_tree = serialize(@group, reader.group_tree)
+          tree_saver.save(group_tree, @shared.export_path, ImportExport.group_filename)
+
+          true
+        rescue => e
+          @shared.error(e)
+          false
+        end
+
+        private
+
+        def serialize(group, relations_tree)
+          group_tree = tree_saver.serialize(group, relations_tree)
+
+          group.children.each do |child|
+            group_tree['children'] ||= []
+            group_tree['children'] << serialize(child, relations_tree)
+          end
+
+          group_tree
+        rescue => e
+          @shared.error(e)
+        end
+
+        def reader
+          @reader ||= Gitlab::ImportExport::Reader.new(
+            shared: @shared,
+            config: Gitlab::ImportExport::Config.new(
+              config: Gitlab::ImportExport.group_config_file
+            ).to_h
+          )
+        end
+
+        def tree_saver
+          @tree_saver ||= RelationTreeSaver.new
+        end
+      end
+    end
+  end
+end

lib/gitlab/import_export/group_object_builder.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    # Given a class, it finds or creates a new object at group level.
-    #
-    # Example:
-    #   `GroupObjectBuilder.build(Label, label_attributes)`
-    #    finds or initializes a label with the given attributes.
-    class GroupObjectBuilder < BaseObjectBuilder
-      def self.build(*args)
-        Group.transaction do
-          super
-        end
-      end
-
-      def initialize(klass, attributes)
-        super
-
-        @group = @attributes['group']
-
-        update_description
-      end
-
-      private
-
-      attr_reader :group
-
-      # Convert description empty string to nil
-      # due to existing object being saved with description: nil
-      # Which makes object lookup to fail since nil != ''
-      def update_description
-        attributes['description'] = nil if attributes['description'] == ''
-      end
-
-      def where_clauses
-        [
-          where_clause_base,
-          where_clause_for_title,
-          where_clause_for_description,
-          where_clause_for_created_at
-        ].compact
-      end
-
-      # Returns Arel clause `"{table_name}"."group_id" = {group.id}`
-      def where_clause_base
-        table[:group_id].in(group_and_ancestor_ids)
-      end
-
-      def group_and_ancestor_ids
-        group.ancestors.map(&:id) << group.id
-      end
-    end
-  end
-end

lib/gitlab/import_export/group_project_object_builder.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    # Given a class, it finds or creates a new object
-    # (initializes in the case of Label) at group or project level.
-    # If it does not exist in the group, it creates it at project level.
-    #
-    # Example:
-    #   `GroupProjectObjectBuilder.build(Label, label_attributes)`
-    #    finds or initializes a label with the given attributes.
-    #
-    # It also adds some logic around Group Labels/Milestones for edge cases.
-    class GroupProjectObjectBuilder < BaseObjectBuilder
-      def self.build(*args)
-        Project.transaction do
-          super
-        end
-      end
-
-      def initialize(klass, attributes)
-        super
-
-        @group = @attributes['group']
-        @project = @attributes['project']
-      end
-
-      def find
-        return if epic? && group.nil?
-
-        super
-      end
-
-      private
-
-      attr_reader :group, :project
-
-      def where_clauses
-        [
-          where_clause_base,
-          where_clause_for_title,
-          where_clause_for_klass
-        ].compact
-      end
-
-      # Returns Arel clause `"{table_name}"."project_id" = {project.id}` if project is present
-      # For example: merge_request has :target_project_id, and we are searching by :iid
-      # or, if group is present:
-      # `"{table_name}"."project_id" = {project.id} OR "{table_name}"."group_id" = {group.id}`
-      def where_clause_base
-        [].tap do |clauses|
-          clauses << table[:project_id].eq(project.id) if project
-          clauses << table[:group_id].in(group.self_and_ancestors_ids) if group
-        end.reduce(:or)
-      end
-
-      # Returns Arel clause for a particular model or `nil`.
-      def where_clause_for_klass
-        attrs_to_arel(attributes.slice('iid')) if merge_request?
-      end
-
-      def prepare_attributes
-        attributes.dup.tap do |atts|
-          atts.delete('group') unless epic?
-
-          if label?
-            atts['type'] = 'ProjectLabel' # Always create project labels
-          elsif milestone?
-            if atts['group_id'] # Transform new group milestones into project ones
-              atts['iid'] = nil
-              atts.delete('group_id')
-            else
-              claim_iid
-            end
-          end
-
-          atts['importing'] = true if klass.ancestors.include?(Importable)
-        end
-      end
-
-      def label?
-        klass == Label
-      end
-
-      def milestone?
-        klass == Milestone
-      end
-
-      def merge_request?
-        klass == MergeRequest
-      end
-
-      def epic?
-        klass == Epic
-      end
-
-      # If an existing group milestone used the IID
-      # claim the IID back and set the group milestone to use one available
-      # This is necessary to fix situations like the following:
-      #  - Importing into a user namespace project with exported group milestones
-      #    where the IID of the Group milestone could conflict with a project one.
-      def claim_iid
-        # The milestone has to be a group milestone, as it's the only case where
-        # we set the IID as the maximum. The rest of them are fixed.
-        milestone = project.milestones.find_by(iid: attributes['iid'])
-
-        return unless milestone
-
-        milestone.iid = nil
-        milestone.ensure_project_iid!
-        milestone.save!
-      end
-    end
-  end
-end
-
-Gitlab::ImportExport::GroupProjectObjectBuilder.prepend_if_ee('EE::Gitlab::ImportExport::GroupProjectObjectBuilder')

lib/gitlab/import_export/group_relation_factory.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    class GroupRelationFactory < BaseRelationFactory
-      OVERRIDES = {
-        labels:     :group_labels,
-        priorities: :label_priorities,
-        label:      :group_label,
-        parent:     :epic
-      }.freeze
-
-      EXISTING_OBJECT_RELATIONS = %i[
-        epic
-        epics
-        milestone
-        milestones
-        label
-        labels
-        group_label
-        group_labels
-      ].freeze
-
-      private
-
-      def setup_models
-        setup_note if @relation_name == :notes
-
-        update_group_references
-      end
-
-      def update_group_references
-        return unless self.class.existing_object_relations.include?(@relation_name)
-        return unless @relation_hash['group_id']
-
-        @relation_hash['group_id'] = @importable.id
-      end
-    end
-  end
-end

lib/gitlab/import_export/group_tree_restorer.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    class GroupTreeRestorer
-      attr_reader :user
-      attr_reader :shared
-      attr_reader :group
-
-      def initialize(user:, shared:, group:, group_hash:)
-        @path = File.join(shared.export_path, 'group.json')
-        @user = user
-        @shared = shared
-        @group = group
-        @group_hash = group_hash
-      end
-
-      def restore
-        @tree_hash = @group_hash || read_tree_hash
-        @group_members = @tree_hash.delete('members')
-        @children = @tree_hash.delete('children')
-
-        if members_mapper.map && restorer.restore
-          @children&.each do |group_hash|
-            group = create_group(group_hash: group_hash, parent_group: @group)
-            shared = Gitlab::ImportExport::Shared.new(group)
-
-            self.class.new(
-              user: @user,
-              shared: shared,
-              group: group,
-              group_hash: group_hash
-            ).restore
-          end
-        end
-
-        return false if @shared.errors.any?
-
-        true
-      rescue => e
-        @shared.error(e)
-        false
-      end
-
-      private
-
-      def read_tree_hash
-        json = IO.read(@path)
-        ActiveSupport::JSON.decode(json)
-      rescue => e
-        @shared.logger.error(
-          group_id:   @group.id,
-          group_name: @group.name,
-          message:    "Import/Export error: #{e.message}"
-        )
-
-        raise Gitlab::ImportExport::Error.new('Incorrect JSON format')
-      end
-
-      def restorer
-        @relation_tree_restorer ||= RelationTreeRestorer.new(
-          user:             @user,
-          shared:           @shared,
-          importable:       @group,
-          tree_hash:        @tree_hash.except('name', 'path'),
-          members_mapper:   members_mapper,
-          object_builder:   object_builder,
-          relation_factory: relation_factory,
-          reader:           reader
-        )
-      end
-
-      def create_group(group_hash:, parent_group:)
-        group_params = {
-          name:      group_hash['name'],
-          path:      group_hash['path'],
-          parent_id: parent_group&.id,
-          visibility_level: sub_group_visibility_level(group_hash, parent_group)
-        }
-
-        ::Groups::CreateService.new(@user, group_params).execute
-      end
-
-      def sub_group_visibility_level(group_hash, parent_group)
-        original_visibility_level = group_hash['visibility_level'] || Gitlab::VisibilityLevel::PRIVATE
-
-        if parent_group && parent_group.visibility_level < original_visibility_level
-          Gitlab::VisibilityLevel.closest_allowed_level(parent_group.visibility_level)
-        else
-          original_visibility_level
-        end
-      end
-
-      def members_mapper
-        @members_mapper ||= Gitlab::ImportExport::MembersMapper.new(exported_members: @group_members, user: @user, importable: @group)
-      end
-
-      def relation_factory
-        Gitlab::ImportExport::GroupRelationFactory
-      end
-
-      def object_builder
-        Gitlab::ImportExport::GroupObjectBuilder
-      end
-
-      def reader
-        @reader ||= Gitlab::ImportExport::Reader.new(
-          shared: @shared,
-          config: Gitlab::ImportExport::Config.new(
-            config: Gitlab::ImportExport.group_config_file
-          ).to_h
-        )
-      end
-    end
-  end
-end

lib/gitlab/import_export/group_tree_saver.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    class GroupTreeSaver
-      attr_reader :full_path, :shared
-
-      def initialize(group:, current_user:, shared:, params: {})
-        @params       = params
-        @current_user = current_user
-        @shared       = shared
-        @group        = group
-        @full_path    = File.join(@shared.export_path, ImportExport.group_filename)
-      end
-
-      def save
-        group_tree = serialize(@group, reader.group_tree)
-        tree_saver.save(group_tree, @shared.export_path, ImportExport.group_filename)
-
-        true
-      rescue => e
-        @shared.error(e)
-        false
-      end
-
-      private
-
-      def serialize(group, relations_tree)
-        group_tree = tree_saver.serialize(group, relations_tree)
-
-        group.children.each do |child|
-          group_tree['children'] ||= []
-          group_tree['children'] << serialize(child, relations_tree)
-        end
-
-        group_tree
-      rescue => e
-        @shared.error(e)
-      end
-
-      def reader
-        @reader ||= Gitlab::ImportExport::Reader.new(
-          shared: @shared,
-          config: Gitlab::ImportExport::Config.new(
-            config: Gitlab::ImportExport.group_config_file
-          ).to_h
-        )
-      end
-
-      def tree_saver
-        @tree_saver ||= RelationTreeSaver.new
-      end
-    end
-  end
-end

lib/gitlab/import_export/importer.rb

@@ -49,7 +49,7 @@ module Gitlab
       end
 
       def project_tree
-        @project_tree ||= Gitlab::ImportExport::ProjectTreeRestorer.new(user: current_user,
+        @project_tree ||= Gitlab::ImportExport::Project::TreeRestorer.new(user: current_user,
                                                                         shared: shared,
                                                                         project: project)
       end
@@ -125,7 +125,7 @@ module Gitlab
 
       def project_to_overwrite
         strong_memoize(:project_to_overwrite) do
-          Project.find_by_full_path("#{project.namespace.full_path}/#{original_path}")
+          ::Project.find_by_full_path("#{project.namespace.full_path}/#{original_path}")
         end
       end
     end

lib/gitlab/import_export/members_mapper.rb

@@ -91,9 +91,9 @@ module Gitlab
 
       def relation_class
         case @importable
-        when Project
+        when ::Project
           ProjectMember
-        when Group
+        when ::Group
           GroupMember
         end
       end

lib/gitlab/import_export/project/object_builder.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Project
+      # Given a class, it finds or creates a new object
+      # (initializes in the case of Label) at group or project level.
+      # If it does not exist in the group, it creates it at project level.
+      #
+      # Example:
+      #   `ObjectBuilder.build(Label, label_attributes)`
+      #    finds or initializes a label with the given attributes.
+      #
+      # It also adds some logic around Group Labels/Milestones for edge cases.
+      class ObjectBuilder < Base::ObjectBuilder
+        def self.build(*args)
+          ::Project.transaction do
+            super
+          end
+        end
+
+        def initialize(klass, attributes)
+          super
+
+          @group = @attributes['group']
+          @project = @attributes['project']
+        end
+
+        def find
+          return if epic? && group.nil?
+
+          super
+        end
+
+        private
+
+        attr_reader :group, :project
+
+        def where_clauses
+          [
+            where_clause_base,
+            where_clause_for_title,
+            where_clause_for_klass
+          ].compact
+        end
+
+        # Returns Arel clause `"{table_name}"."project_id" = {project.id}` if project is present
+        # For example: merge_request has :target_project_id, and we are searching by :iid
+        # or, if group is present:
+        # `"{table_name}"."project_id" = {project.id} OR "{table_name}"."group_id" = {group.id}`
+        def where_clause_base
+          [].tap do |clauses|
+            clauses << table[:project_id].eq(project.id) if project
+            clauses << table[:group_id].in(group.self_and_ancestors_ids) if group
+          end.reduce(:or)
+        end
+
+        # Returns Arel clause for a particular model or `nil`.
+        def where_clause_for_klass
+          attrs_to_arel(attributes.slice('iid')) if merge_request?
+        end
+
+        def prepare_attributes
+          attributes.dup.tap do |atts|
+            atts.delete('group') unless epic?
+
+            if label?
+              atts['type'] = 'ProjectLabel' # Always create project labels
+            elsif milestone?
+              if atts['group_id'] # Transform new group milestones into project ones
+                atts['iid'] = nil
+                atts.delete('group_id')
+              else
+                claim_iid
+              end
+            end
+
+            atts['importing'] = true if klass.ancestors.include?(Importable)
+          end
+        end
+
+        def label?
+          klass == Label
+        end
+
+        def milestone?
+          klass == Milestone
+        end
+
+        def merge_request?
+          klass == MergeRequest
+        end
+
+        def epic?
+          klass == Epic
+        end
+
+        # If an existing group milestone used the IID
+        # claim the IID back and set the group milestone to use one available
+        # This is necessary to fix situations like the following:
+        #  - Importing into a user namespace project with exported group milestones
+        #    where the IID of the Group milestone could conflict with a project one.
+        def claim_iid
+          # The milestone has to be a group milestone, as it's the only case where
+          # we set the IID as the maximum. The rest of them are fixed.
+          milestone = project.milestones.find_by(iid: attributes['iid'])
+
+          return unless milestone
+
+          milestone.iid = nil
+          milestone.ensure_project_iid!
+          milestone.save!
+        end
+      end
+    end
+  end
+end
+
+Gitlab::ImportExport::Project::ObjectBuilder.prepend_if_ee('EE::Gitlab::ImportExport::Project::ObjectBuilder')

lib/gitlab/import_export/project/relation_factory.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Project
+      class RelationFactory < Base::RelationFactory
+        prepend_if_ee('::EE::Gitlab::ImportExport::Project::RelationFactory') # rubocop: disable Cop/InjectEnterpriseEditionModule
+
+        OVERRIDES = { snippets: :project_snippets,
+                      ci_pipelines: 'Ci::Pipeline',
+                      pipelines: 'Ci::Pipeline',
+                      stages: 'Ci::Stage',
+                      statuses: 'commit_status',
+                      triggers: 'Ci::Trigger',
+                      pipeline_schedules: 'Ci::PipelineSchedule',
+                      builds: 'Ci::Build',
+                      runners: 'Ci::Runner',
+                      hooks: 'ProjectHook',
+                      merge_access_levels: 'ProtectedBranch::MergeAccessLevel',
+                      push_access_levels: 'ProtectedBranch::PushAccessLevel',
+                      create_access_levels: 'ProtectedTag::CreateAccessLevel',
+                      labels: :project_labels,
+                      priorities: :label_priorities,
+                      auto_devops: :project_auto_devops,
+                      label: :project_label,
+                      custom_attributes: 'ProjectCustomAttribute',
+                      project_badges: 'Badge',
+                      metrics: 'MergeRequest::Metrics',
+                      ci_cd_settings: 'ProjectCiCdSetting',
+                      error_tracking_setting: 'ErrorTracking::ProjectErrorTrackingSetting',
+                      links: 'Releases::Link',
+                      metrics_setting: 'ProjectMetricsSetting' }.freeze
+
+        BUILD_MODELS = %i[Ci::Build commit_status].freeze
+
+        GROUP_REFERENCES = %w[group_id].freeze
+
+        PROJECT_REFERENCES = %w[project_id source_project_id target_project_id].freeze
+
+        EXISTING_OBJECT_RELATIONS = %i[
+          milestone
+          milestones
+          label
+          labels
+          project_label
+          project_labels
+          group_label
+          group_labels
+          project_feature
+          merge_request
+          epic
+          ProjectCiCdSetting
+          container_expiration_policy
+        ].freeze
+
+        def create
+          @object = super
+
+          # We preload the project, user, and group to re-use objects
+          @object = preload_keys(@object, PROJECT_REFERENCES, @importable)
+          @object = preload_keys(@object, GROUP_REFERENCES, @importable.group)
+          @object = preload_keys(@object, USER_REFERENCES, @user)
+        end
+
+        private
+
+        def invalid_relation?
+          # Do not create relation if it is:
+          #   - An unknown service
+          #   - A legacy trigger
+          unknown_service? ||
+            (!Feature.enabled?(:use_legacy_pipeline_triggers, @importable) && legacy_trigger?)
+        end
+
+        def setup_models
+          case @relation_name
+          when :merge_request_diff_files then setup_diff
+          when :notes then setup_note
+          when :'Ci::Pipeline' then setup_pipeline
+          when *BUILD_MODELS then setup_build
+          end
+
+          update_project_references
+          update_group_references
+        end
+
+        def generate_imported_object
+          if @relation_name == :merge_requests
+            MergeRequestParser.new(@importable, @relation_hash.delete('diff_head_sha'), super, @relation_hash).parse!
+          else
+            super
+          end
+        end
+
+        def update_project_references
+          # If source and target are the same, populate them with the new project ID.
+          if @relation_hash['source_project_id']
+            @relation_hash['source_project_id'] = same_source_and_target? ? @relation_hash['project_id'] : MergeRequestParser::FORKED_PROJECT_ID
+          end
+
+          @relation_hash['target_project_id'] = @relation_hash['project_id'] if @relation_hash['target_project_id']
+        end
+
+        def same_source_and_target?
+          @relation_hash['target_project_id'] && @relation_hash['target_project_id'] == @relation_hash['source_project_id']
+        end
+
+        def update_group_references
+          return unless existing_object?
+          return unless @relation_hash['group_id']
+
+          @relation_hash['group_id'] = @importable.namespace_id
+        end
+
+        def setup_build
+          @relation_hash.delete('trace') # old export files have trace
+          @relation_hash.delete('token')
+          @relation_hash.delete('commands')
+          @relation_hash.delete('artifacts_file_store')
+          @relation_hash.delete('artifacts_metadata_store')
+          @relation_hash.delete('artifacts_size')
+        end
+
+        def setup_diff
+          @relation_hash['diff'] = @relation_hash.delete('utf8_diff')
+        end
+
+        def setup_pipeline
+          @relation_hash.fetch('stages', []).each do |stage|
+            stage.statuses.each do |status|
+              status.pipeline = imported_object
+            end
+          end
+        end
+
+        def unknown_service?
+          @relation_name == :services && parsed_relation_hash['type'] &&
+            !Object.const_defined?(parsed_relation_hash['type'])
+        end
+
+        def legacy_trigger?
+          @relation_name == :'Ci::Trigger' && @relation_hash['owner_id'].nil?
+        end
+
+        def preload_keys(object, references, value)
+          return object unless value
+
+          references.each do |key|
+            attribute = "#{key.delete_suffix('_id')}=".to_sym
+            next unless object.respond_to?(key) && object.respond_to?(attribute)
+
+            if object.read_attribute(key) == value&.id
+              object.public_send(attribute, value) # rubocop:disable GitlabSecurity/PublicSend
+            end
+          end
+
+          object
+        end
+      end
+    end
+  end
+end

lib/gitlab/import_export/project/tree_loader.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Project
+      class TreeLoader
+        def load(path, dedup_entries: false)
+          tree_hash = ActiveSupport::JSON.decode(IO.read(path))
+
+          if dedup_entries
+            dedup_tree(tree_hash)
+          else
+            tree_hash
+          end
+        end
+
+        private
+
+        # This function removes duplicate entries from the given tree recursively
+        # by caching nodes it encounters repeatedly. We only consider nodes for
+        # which there can actually be multiple equivalent instances (e.g. strings,
+        # hashes and arrays, but not `nil`s, numbers or booleans.)
+        #
+        # The algorithm uses a recursive depth-first descent with 3 cases, starting
+        # with a root node (the tree/hash itself):
+        # - a node has already been cached; in this case we return it from the cache
+        # - a node has not been cached yet but should be; descend into its children
+        # - a node is neither cached nor qualifies for caching; this is a no-op
+        def dedup_tree(node, nodes_seen = {})
+          if nodes_seen.key?(node) && distinguishable?(node)
+            yield nodes_seen[node]
+          elsif should_dedup?(node)
+            nodes_seen[node] = node
+
+            case node
+            when Array
+              node.each_index do |idx|
+                dedup_tree(node[idx], nodes_seen) do |cached_node|
+                  node[idx] = cached_node
+                end
+              end
+            when Hash
+              node.each do |k, v|
+                dedup_tree(v, nodes_seen) do |cached_node|
+                  node[k] = cached_node
+                end
+              end
+            end
+          else
+            node
+          end
+        end
+
+        # We do not need to consider nodes for which there cannot be multiple instances
+        def should_dedup?(node)
+          node && !(node.is_a?(Numeric) || node.is_a?(TrueClass) || node.is_a?(FalseClass))
+        end
+
+        # We can only safely de-dup values that are distinguishable. True value objects
+        # are always distinguishable by nature. Hashes however can represent entities,
+        # which are identified by ID, not value. We therefore disallow de-duping hashes
+        # that do not have an `id` field, since we might risk dropping entities that
+        # have equal attributes yet different identities.
+        def distinguishable?(node)
+          if node.is_a?(Hash)
+            node.key?('id')
+          else
+            true
+          end
+        end
+      end
+    end
+  end
+end

lib/gitlab/import_export/project/tree_restorer.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Project
+      class TreeRestorer
+        LARGE_PROJECT_FILE_SIZE_BYTES = 500.megabyte
+
+        attr_reader :user
+        attr_reader :shared
+        attr_reader :project
+
+        def initialize(user:, shared:, project:)
+          @user = user
+          @shared = shared
+          @project = project
+          @tree_loader = TreeLoader.new
+        end
+
+        def restore
+          @tree_hash = read_tree_hash
+          @project_members = @tree_hash.delete('project_members')
+
+          RelationRenameService.rename(@tree_hash)
+
+          if relation_tree_restorer.restore
+            import_failure_service.with_retry(action: 'set_latest_merge_request_diff_ids!') do
+              @project.merge_requests.set_latest_merge_request_diff_ids!
+            end
+
+            true
+          else
+            false
+          end
+        rescue => e
+          @shared.error(e)
+          false
+        end
+
+        private
+
+        def large_project?(path)
+          File.size(path) >= LARGE_PROJECT_FILE_SIZE_BYTES
+        end
+
+        def read_tree_hash
+          path = File.join(@shared.export_path, 'project.json')
+          dedup_entries = large_project?(path) &&
+            Feature.enabled?(:dedup_project_import_metadata, project.group)
+
+          @tree_loader.load(path, dedup_entries: dedup_entries)
+        rescue => e
+          Rails.logger.error("Import/Export error: #{e.message}") # rubocop:disable Gitlab/RailsLogger
+          raise Gitlab::ImportExport::Error.new('Incorrect JSON format')
+        end
+
+        def relation_tree_restorer
+          @relation_tree_restorer ||= RelationTreeRestorer.new(
+            user: @user,
+            shared: @shared,
+            importable: @project,
+            tree_hash: @tree_hash,
+            object_builder: object_builder,
+            members_mapper: members_mapper,
+            relation_factory: relation_factory,
+            reader: reader
+          )
+        end
+
+        def members_mapper
+          @members_mapper ||= Gitlab::ImportExport::MembersMapper.new(exported_members: @project_members,
+                                                                      user: @user,
+                                                                      importable: @project)
+        end
+
+        def object_builder
+          Project::ObjectBuilder
+        end
+
+        def relation_factory
+          Project::RelationFactory
+        end
+
+        def reader
+          @reader ||= Gitlab::ImportExport::Reader.new(shared: @shared)
+        end
+
+        def import_failure_service
+          @import_failure_service ||= ImportFailureService.new(@project)
+        end
+      end
+    end
+  end
+end

lib/gitlab/import_export/project/tree_saver.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module ImportExport
+    module Project
+      class TreeSaver
+        attr_reader :full_path
+
+        def initialize(project:, current_user:, shared:, params: {})
+          @params       = params
+          @project      = project
+          @current_user = current_user
+          @shared       = shared
+          @full_path    = File.join(@shared.export_path, ImportExport.project_filename)
+        end
+
+        def save
+          project_tree = tree_saver.serialize(@project, reader.project_tree)
+          fix_project_tree(project_tree)
+          tree_saver.save(project_tree, @shared.export_path, ImportExport.project_filename)
+
+          true
+        rescue => e
+          @shared.error(e)
+          false
+        end
+
+        private
+
+        # Aware that the resulting hash needs to be pure-hash and
+        # does not include any AR objects anymore, only objects that run `.to_json`
+        def fix_project_tree(project_tree)
+          if @params[:description].present?
+            project_tree['description'] = @params[:description]
+          end
+
+          project_tree['project_members'] += group_members_array
+
+          RelationRenameService.add_new_associations(project_tree)
+        end
+
+        def reader
+          @reader ||= Gitlab::ImportExport::Reader.new(shared: @shared)
+        end
+
+        def group_members_array
+          group_members.as_json(reader.group_members_tree).each do |group_member|
+            group_member['source_type'] = 'Project' # Make group members project members of the future import
+          end
+        end
+
+        def group_members
+          return [] unless @current_user.can?(:admin_group, @project.group)
+
+          # We need `.where.not(user_id: nil)` here otherwise when a group has an
+          # invitee, it would make the following query return 0 rows since a NULL
+          # user_id would be present in the subquery
+          # See http://stackoverflow.com/questions/129077/not-in-clause-and-null-values
+          non_null_user_ids = @project.project_members.where.not(user_id: nil).select(:user_id)
+
+          GroupMembersFinder.new(@project.group).execute.where.not(user_id: non_null_user_ids)
+        end
+
+        def tree_saver
+          @tree_saver ||= RelationTreeSaver.new
+        end
+      end
+    end
+  end
+end

lib/gitlab/import_export/project_relation_factory.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    class ProjectRelationFactory < BaseRelationFactory
-      prepend_if_ee('::EE::Gitlab::ImportExport::ProjectRelationFactory') # rubocop: disable Cop/InjectEnterpriseEditionModule
-
-      OVERRIDES = { snippets: :project_snippets,
-                    ci_pipelines: 'Ci::Pipeline',
-                    pipelines: 'Ci::Pipeline',
-                    stages: 'Ci::Stage',
-                    statuses: 'commit_status',
-                    triggers: 'Ci::Trigger',
-                    pipeline_schedules: 'Ci::PipelineSchedule',
-                    builds: 'Ci::Build',
-                    runners: 'Ci::Runner',
-                    hooks: 'ProjectHook',
-                    merge_access_levels: 'ProtectedBranch::MergeAccessLevel',
-                    push_access_levels: 'ProtectedBranch::PushAccessLevel',
-                    create_access_levels: 'ProtectedTag::CreateAccessLevel',
-                    labels: :project_labels,
-                    priorities: :label_priorities,
-                    auto_devops: :project_auto_devops,
-                    label: :project_label,
-                    custom_attributes: 'ProjectCustomAttribute',
-                    project_badges: 'Badge',
-                    metrics: 'MergeRequest::Metrics',
-                    ci_cd_settings: 'ProjectCiCdSetting',
-                    error_tracking_setting: 'ErrorTracking::ProjectErrorTrackingSetting',
-                    links: 'Releases::Link',
-                    metrics_setting: 'ProjectMetricsSetting' }.freeze
-
-      BUILD_MODELS = %i[Ci::Build commit_status].freeze
-
-      GROUP_REFERENCES = %w[group_id].freeze
-
-      PROJECT_REFERENCES = %w[project_id source_project_id target_project_id].freeze
-
-      EXISTING_OBJECT_RELATIONS = %i[
-        milestone
-        milestones
-        label
-        labels
-        project_label
-        project_labels
-        group_label
-        group_labels
-        project_feature
-        merge_request
-        epic
-        ProjectCiCdSetting
-        container_expiration_policy
-      ].freeze
-
-      def create
-        @object = super
-
-        # We preload the project, user, and group to re-use objects
-        @object = preload_keys(@object, PROJECT_REFERENCES, @importable)
-        @object = preload_keys(@object, GROUP_REFERENCES, @importable.group)
-        @object = preload_keys(@object, USER_REFERENCES, @user)
-      end
-
-      private
-
-      def invalid_relation?
-        # Do not create relation if it is:
-        #   - An unknown service
-        #   - A legacy trigger
-        unknown_service? ||
-          (!Feature.enabled?(:use_legacy_pipeline_triggers, @importable) && legacy_trigger?)
-      end
-
-      def setup_models
-        case @relation_name
-        when :merge_request_diff_files then setup_diff
-        when :notes then setup_note
-        when :'Ci::Pipeline' then setup_pipeline
-        when *BUILD_MODELS then setup_build
-        end
-
-        update_project_references
-        update_group_references
-      end
-
-      def generate_imported_object
-        if @relation_name == :merge_requests
-          MergeRequestParser.new(@importable, @relation_hash.delete('diff_head_sha'), super, @relation_hash).parse!
-        else
-          super
-        end
-      end
-
-      def update_project_references
-        # If source and target are the same, populate them with the new project ID.
-        if @relation_hash['source_project_id']
-          @relation_hash['source_project_id'] = same_source_and_target? ? @relation_hash['project_id'] : MergeRequestParser::FORKED_PROJECT_ID
-        end
-
-        @relation_hash['target_project_id'] = @relation_hash['project_id'] if @relation_hash['target_project_id']
-      end
-
-      def same_source_and_target?
-        @relation_hash['target_project_id'] && @relation_hash['target_project_id'] == @relation_hash['source_project_id']
-      end
-
-      def update_group_references
-        return unless existing_object?
-        return unless @relation_hash['group_id']
-
-        @relation_hash['group_id'] = @importable.namespace_id
-      end
-
-      def setup_build
-        @relation_hash.delete('trace') # old export files have trace
-        @relation_hash.delete('token')
-        @relation_hash.delete('commands')
-        @relation_hash.delete('artifacts_file_store')
-        @relation_hash.delete('artifacts_metadata_store')
-        @relation_hash.delete('artifacts_size')
-      end
-
-      def setup_diff
-        @relation_hash['diff'] = @relation_hash.delete('utf8_diff')
-      end
-
-      def setup_pipeline
-        @relation_hash.fetch('stages', []).each do |stage|
-          stage.statuses.each do |status|
-            status.pipeline = imported_object
-          end
-        end
-      end
-
-      def unknown_service?
-        @relation_name == :services && parsed_relation_hash['type'] &&
-          !Object.const_defined?(parsed_relation_hash['type'])
-      end
-
-      def legacy_trigger?
-        @relation_name == :'Ci::Trigger' && @relation_hash['owner_id'].nil?
-      end
-
-      def preload_keys(object, references, value)
-        return object unless value
-
-        references.each do |key|
-          attribute = "#{key.delete_suffix('_id')}=".to_sym
-          next unless object.respond_to?(key) && object.respond_to?(attribute)
-
-          if object.read_attribute(key) == value&.id
-            object.public_send(attribute, value) # rubocop:disable GitlabSecurity/PublicSend
-          end
-        end
-
-        object
-      end
-    end
-  end
-end

lib/gitlab/import_export/project_tree_loader.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    class ProjectTreeLoader
-      def load(path, dedup_entries: false)
-        tree_hash = ActiveSupport::JSON.decode(IO.read(path))
-
-        if dedup_entries
-          dedup_tree(tree_hash)
-        else
-          tree_hash
-        end
-      end
-
-      private
-
-      # This function removes duplicate entries from the given tree recursively
-      # by caching nodes it encounters repeatedly. We only consider nodes for
-      # which there can actually be multiple equivalent instances (e.g. strings,
-      # hashes and arrays, but not `nil`s, numbers or booleans.)
-      #
-      # The algorithm uses a recursive depth-first descent with 3 cases, starting
-      # with a root node (the tree/hash itself):
-      # - a node has already been cached; in this case we return it from the cache
-      # - a node has not been cached yet but should be; descend into its children
-      # - a node is neither cached nor qualifies for caching; this is a no-op
-      def dedup_tree(node, nodes_seen = {})
-        if nodes_seen.key?(node) && distinguishable?(node)
-          yield nodes_seen[node]
-        elsif should_dedup?(node)
-          nodes_seen[node] = node
-
-          case node
-          when Array
-            node.each_index do |idx|
-              dedup_tree(node[idx], nodes_seen) do |cached_node|
-                node[idx] = cached_node
-              end
-            end
-          when Hash
-            node.each do |k, v|
-              dedup_tree(v, nodes_seen) do |cached_node|
-                node[k] = cached_node
-              end
-            end
-          end
-        else
-          node
-        end
-      end
-
-      # We do not need to consider nodes for which there cannot be multiple instances
-      def should_dedup?(node)
-        node && !(node.is_a?(Numeric) || node.is_a?(TrueClass) || node.is_a?(FalseClass))
-      end
-
-      # We can only safely de-dup values that are distinguishable. True value objects
-      # are always distinguishable by nature. Hashes however can represent entities,
-      # which are identified by ID, not value. We therefore disallow de-duping hashes
-      # that do not have an `id` field, since we might risk dropping entities that
-      # have equal attributes yet different identities.
-      def distinguishable?(node)
-        if node.is_a?(Hash)
-          node.key?('id')
-        else
-          true
-        end
-      end
-    end
-  end
-end

lib/gitlab/import_export/project_tree_restorer.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    class ProjectTreeRestorer
-      LARGE_PROJECT_FILE_SIZE_BYTES = 500.megabyte
-
-      attr_reader :user
-      attr_reader :shared
-      attr_reader :project
-
-      def initialize(user:, shared:, project:)
-        @user = user
-        @shared = shared
-        @project = project
-        @tree_loader = ProjectTreeLoader.new
-      end
-
-      def restore
-        @tree_hash = read_tree_hash
-        @project_members = @tree_hash.delete('project_members')
-
-        RelationRenameService.rename(@tree_hash)
-
-        if relation_tree_restorer.restore
-          import_failure_service.with_retry(action: 'set_latest_merge_request_diff_ids!') do
-            @project.merge_requests.set_latest_merge_request_diff_ids!
-          end
-
-          true
-        else
-          false
-        end
-      rescue => e
-        @shared.error(e)
-        false
-      end
-
-      private
-
-      def large_project?(path)
-        File.size(path) >= LARGE_PROJECT_FILE_SIZE_BYTES
-      end
-
-      def read_tree_hash
-        path = File.join(@shared.export_path, 'project.json')
-        dedup_entries = large_project?(path) &&
-          Feature.enabled?(:dedup_project_import_metadata, project.group)
-
-        @tree_loader.load(path, dedup_entries: dedup_entries)
-      rescue => e
-        Rails.logger.error("Import/Export error: #{e.message}") # rubocop:disable Gitlab/RailsLogger
-        raise Gitlab::ImportExport::Error.new('Incorrect JSON format')
-      end
-
-      def relation_tree_restorer
-        @relation_tree_restorer ||= RelationTreeRestorer.new(
-          user: @user,
-          shared: @shared,
-          importable: @project,
-          tree_hash: @tree_hash,
-          object_builder: object_builder,
-          members_mapper: members_mapper,
-          relation_factory: relation_factory,
-          reader: reader
-        )
-      end
-
-      def members_mapper
-        @members_mapper ||= Gitlab::ImportExport::MembersMapper.new(exported_members: @project_members,
-                                                                    user: @user,
-                                                                    importable: @project)
-      end
-
-      def object_builder
-        Gitlab::ImportExport::GroupProjectObjectBuilder
-      end
-
-      def relation_factory
-        Gitlab::ImportExport::ProjectRelationFactory
-      end
-
-      def reader
-        @reader ||= Gitlab::ImportExport::Reader.new(shared: @shared)
-      end
-
-      def import_failure_service
-        @import_failure_service ||= ImportFailureService.new(@project)
-      end
-    end
-  end
-end

lib/gitlab/import_export/project_tree_saver.rb

-# frozen_string_literal: true
-
-module Gitlab
-  module ImportExport
-    class ProjectTreeSaver
-      attr_reader :full_path
-
-      def initialize(project:, current_user:, shared:, params: {})
-        @params       = params
-        @project      = project
-        @current_user = current_user
-        @shared       = shared
-        @full_path    = File.join(@shared.export_path, ImportExport.project_filename)
-      end
-
-      def save
-        project_tree = tree_saver.serialize(@project, reader.project_tree)
-        fix_project_tree(project_tree)
-        tree_saver.save(project_tree, @shared.export_path, ImportExport.project_filename)
-
-        true
-      rescue => e
-        @shared.error(e)
-        false
-      end
-
-      private
-
-      # Aware that the resulting hash needs to be pure-hash and
-      # does not include any AR objects anymore, only objects that run `.to_json`
-      def fix_project_tree(project_tree)
-        if @params[:description].present?
-          project_tree['description'] = @params[:description]
-        end
-
-        project_tree['project_members'] += group_members_array
-
-        RelationRenameService.add_new_associations(project_tree)
-      end
-
-      def reader
-        @reader ||= Gitlab::ImportExport::Reader.new(shared: @shared)
-      end
-
-      def group_members_array
-        group_members.as_json(reader.group_members_tree).each do |group_member|
-          group_member['source_type'] = 'Project' # Make group members project members of the future import
-        end
-      end
-
-      def group_members
-        return [] unless @current_user.can?(:admin_group, @project.group)
-
-        # We need `.where.not(user_id: nil)` here otherwise when a group has an
-        # invitee, it would make the following query return 0 rows since a NULL
-        # user_id would be present in the subquery
-        # See http://stackoverflow.com/questions/129077/not-in-clause-and-null-values
-        non_null_user_ids = @project.project_members.where.not(user_id: nil).select(:user_id)
-
-        GroupMembersFinder.new(@project.group).execute.where.not(user_id: non_null_user_ids)
-      end
-
-      def tree_saver
-        @tree_saver ||= RelationTreeSaver.new
-      end
-    end
-  end
-end

lib/gitlab/import_export/relation_tree_restorer.rb

@@ -69,7 +69,7 @@ module Gitlab
       def process_relation_item!(relation_key, relation_definition, relation_index, data_hash)
         relation_object = build_relation(relation_key, relation_definition, data_hash)
         return unless relation_object
-        return if importable_class == Project && group_model?(relation_object)
+        return if importable_class == ::Project && group_model?(relation_object)
 
         relation_object.assign_attributes(importable_class_sym => @importable)
 
@@ -110,7 +110,7 @@ module Gitlab
           excluded_keys:  excluded_keys_for_relation(importable_class_sym))
 
         @importable.assign_attributes(params)
-        @importable.drop_visibility_level! if importable_class == Project
+        @importable.drop_visibility_level! if importable_class == ::Project
 
         Gitlab::Timeless.timeless(@importable) do
           @importable.save!

spec/fixtures/trace/sample_trace

@@ -1442,7 +1442,7 @@ TodoService
     marks a single todo id as done
     caches the number of todos of a user
 
-Gitlab::ImportExport::ProjectTreeSaver
+Gitlab::ImportExport::Project::TreeSaver
   saves the project tree into a json object
     saves project successfully
     JSON

spec/lib/gitlab/import_export/base_object_builder_spec.rb → spec/lib/gitlab/import_export/base/object_builder_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::BaseObjectBuilder do
+describe Gitlab::ImportExport::Base::ObjectBuilder do
   let(:project) do
     create(:project, :repository,
            :builds_disabled,
@@ -11,7 +11,7 @@ describe Gitlab::ImportExport::BaseObjectBuilder do
            path: 'project')
   end
   let(:klass) { Milestone }
-  let(:attributes) { { 'title' => 'Test BaseObjectBuilder Milestone', 'project' => project } }
+  let(:attributes) { { 'title' => 'Test Base::ObjectBuilder Milestone', 'project' => project } }
 
   subject { described_class.build(klass, attributes) }
 

spec/lib/gitlab/import_export/base_relation_factory_spec.rb → spec/lib/gitlab/import_export/base/relation_factory_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::BaseRelationFactory do
+describe Gitlab::ImportExport::Base::RelationFactory do
   let(:user) { create(:admin) }
   let(:project) { create(:project) }
   let(:members_mapper) { double('members_mapper').as_null_object }
@@ -13,7 +13,7 @@ describe Gitlab::ImportExport::BaseRelationFactory do
   subject do
     described_class.create(relation_sym: relation_sym,
                            relation_hash: relation_hash,
-                           object_builder: Gitlab::ImportExport::GroupProjectObjectBuilder,
+                           object_builder: Gitlab::ImportExport::Project::ObjectBuilder,
                            members_mapper: members_mapper,
                            user: user,
                            importable: project,

spec/lib/gitlab/import_export/fork_spec.rb

@@ -24,11 +24,11 @@ describe 'forked project import' do
   end
 
   let(:saver) do
-    Gitlab::ImportExport::ProjectTreeSaver.new(project: project_with_repo, current_user: user, shared: shared)
+    Gitlab::ImportExport::Project::TreeSaver.new(project: project_with_repo, current_user: user, shared: shared)
   end
 
   let(:restorer) do
-    Gitlab::ImportExport::ProjectTreeRestorer.new(user: user, shared: shared, project: project)
+    Gitlab::ImportExport::Project::TreeRestorer.new(user: user, shared: shared, project: project)
   end
 
   before do

spec/lib/gitlab/import_export/group_object_builder_spec.rb → spec/lib/gitlab/import_export/group/object_builder_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::GroupObjectBuilder do
+describe Gitlab::ImportExport::Group::ObjectBuilder do
   let(:group) { create(:group) }
   let(:base_attributes) do
     {

spec/lib/gitlab/import_export/group_relation_factory_spec.rb → spec/lib/gitlab/import_export/group/relation_factory_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::GroupRelationFactory do
+describe Gitlab::ImportExport::Group::RelationFactory do
   let(:group) { create(:group) }
   let(:members_mapper) { double('members_mapper').as_null_object }
   let(:user) { create(:admin) }
@@ -11,7 +11,7 @@ describe Gitlab::ImportExport::GroupRelationFactory do
     described_class.create(relation_sym:   relation_sym,
                            relation_hash:  relation_hash,
                            members_mapper: members_mapper,
-                           object_builder: Gitlab::ImportExport::GroupObjectBuilder,
+                           object_builder: Gitlab::ImportExport::Group::ObjectBuilder,
                            user:           user,
                            importable:     group,
                            excluded_keys:  excluded_keys)

spec/lib/gitlab/import_export/group_tree_restorer_spec.rb → spec/lib/gitlab/import_export/group/tree_restorer_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::GroupTreeRestorer do
+describe Gitlab::ImportExport::Group::TreeRestorer do
   include ImportExport::CommonUtil
 
   let(:shared) { Gitlab::ImportExport::Shared.new(group) }

spec/lib/gitlab/import_export/group_tree_saver_spec.rb → spec/lib/gitlab/import_export/group/tree_saver_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::GroupTreeSaver do
+describe Gitlab::ImportExport::Group::TreeSaver do
   describe 'saves the group tree into a json object' do
     let(:shared) { Gitlab::ImportExport::Shared.new(group) }
     let(:group_tree_saver) { described_class.new(group: group, current_user: user, shared: shared) }
@@ -72,7 +72,7 @@ describe Gitlab::ImportExport::GroupTreeSaver do
     # except:
     # context 'with description override' do
     # context 'group members' do
-    # ^ These are specific for the groupTreeSaver
+    # ^ These are specific for the Group::TreeSaver
     context 'JSON' do
       let(:saved_group_json) do
         group_tree_saver.save

spec/lib/gitlab/import_export/importer_spec.rb

@@ -63,7 +63,7 @@ describe Gitlab::ImportExport::Importer do
       end
 
       it 'restores the ProjectTree' do
-        expect(Gitlab::ImportExport::ProjectTreeRestorer).to receive(:new).and_call_original
+        expect(Gitlab::ImportExport::Project::TreeRestorer).to receive(:new).and_call_original
 
         importer.execute
       end

spec/lib/gitlab/import_export/group_project_object_builder_spec.rb → spec/lib/gitlab/import_export/project/object_builder_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::GroupProjectObjectBuilder do
+describe Gitlab::ImportExport::Project::ObjectBuilder do
   let!(:group) { create(:group, :private) }
   let!(:subgroup) { create(:group, :private, parent: group) }
   let!(:project) do

spec/lib/gitlab/import_export/project_relation_factory_spec.rb → spec/lib/gitlab/import_export/project/relation_factory_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::ProjectRelationFactory do
+describe Gitlab::ImportExport::Project::RelationFactory do
   let(:group)   { create(:group) }
   let(:project) { create(:project, :repository, group: group) }
   let(:members_mapper) { double('members_mapper').as_null_object }
@@ -11,7 +11,7 @@ describe Gitlab::ImportExport::ProjectRelationFactory do
   let(:created_object) do
     described_class.create(relation_sym: relation_sym,
                            relation_hash: relation_hash,
-                           object_builder: Gitlab::ImportExport::GroupProjectObjectBuilder,
+                           object_builder: Gitlab::ImportExport::Project::ObjectBuilder,
                            members_mapper: members_mapper,
                            user: user,
                            importable: project,
@@ -243,11 +243,11 @@ describe Gitlab::ImportExport::ProjectRelationFactory do
   context 'Project references' do
     let(:relation_sym) { :project_foo_model }
     let(:relation_hash) do
-      Gitlab::ImportExport::ProjectRelationFactory::PROJECT_REFERENCES.map { |ref| { ref => 99 } }.inject(:merge)
+      Gitlab::ImportExport::Project::RelationFactory::PROJECT_REFERENCES.map { |ref| { ref => 99 } }.inject(:merge)
     end
 
     class ProjectFooModel < FooModel
-      attr_accessor(*Gitlab::ImportExport::ProjectRelationFactory::PROJECT_REFERENCES)
+      attr_accessor(*Gitlab::ImportExport::Project::RelationFactory::PROJECT_REFERENCES)
     end
 
     before do

spec/lib/gitlab/import_export/project_tree_loader_spec.rb → spec/lib/gitlab/import_export/project/tree_loader_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::ProjectTreeLoader do
+describe Gitlab::ImportExport::Project::TreeLoader do
   let(:fixture) { 'spec/fixtures/lib/gitlab/import_export/with_duplicates.json' }
   let(:project_tree) { JSON.parse(File.read(fixture)) }
 

spec/lib/gitlab/import_export/project_tree_restorer_spec.rb → spec/lib/gitlab/import_export/project/tree_restorer_spec.rb

 # frozen_string_literal: true
 
 require 'spec_helper'
-include ImportExport::CommonUtil
 
-describe Gitlab::ImportExport::ProjectTreeRestorer do
+describe Gitlab::ImportExport::Project::TreeRestorer do
   include ImportExport::CommonUtil
 
   let(:shared) { project.import_export_shared }

spec/lib/gitlab/import_export/project_tree_saver_spec.rb → spec/lib/gitlab/import_export/project/tree_saver_spec.rb

@@ -2,7 +2,7 @@
 
 require 'spec_helper'
 
-describe Gitlab::ImportExport::ProjectTreeSaver do
+describe Gitlab::ImportExport::Project::TreeSaver do
   describe 'saves the project tree into a json object' do
     let(:shared) { project.import_export_shared }
     let(:project_tree_saver) { described_class.new(project: project, current_user: user, shared: shared) }
@@ -75,7 +75,7 @@ describe Gitlab::ImportExport::ProjectTreeSaver do
     # except:
     # context 'with description override' do
     # context 'group members' do
-    # ^ These are specific for the ProjectTreeSaver
+    # ^ These are specific for the Project::TreeSaver
     context 'JSON' do
       let(:saved_project_json) do
         project_tree_saver.save

spec/lib/gitlab/import_export/relation_rename_service_spec.rb

@@ -22,7 +22,7 @@ describe Gitlab::ImportExport::RelationRenameService do
   end
 
   context 'when importing' do
-    let(:project_tree_restorer) { Gitlab::ImportExport::ProjectTreeRestorer.new(user: user, shared: shared, project: project) }
+    let(:project_tree_restorer) { Gitlab::ImportExport::Project::TreeRestorer.new(user: user, shared: shared, project: project) }
     let(:file_content) { IO.read(File.join(shared.export_path, 'project.json')) }
     let(:json_file) { ActiveSupport::JSON.decode(file_content) }
 
@@ -99,7 +99,7 @@ describe Gitlab::ImportExport::RelationRenameService do
     let(:relation_tree_saver) { Gitlab::ImportExport::RelationTreeSaver.new }
 
     let(:project_tree_saver) do
-      Gitlab::ImportExport::ProjectTreeSaver.new(
+      Gitlab::ImportExport::Project::TreeSaver.new(
         project: project, current_user: user, shared: shared)
     end
 

spec/lib/gitlab/import_export/relation_tree_restorer_spec.rb

@@ -39,8 +39,8 @@ describe Gitlab::ImportExport::RelationTreeRestorer do
   context 'when restoring a project' do
     let(:path) { 'spec/fixtures/lib/gitlab/import_export/complex/project.json' }
     let(:importable) { create(:project, :builds_enabled, :issues_disabled, name: 'project', path: 'project') }
-    let(:object_builder) { Gitlab::ImportExport::GroupProjectObjectBuilder }
-    let(:relation_factory) { Gitlab::ImportExport::ProjectRelationFactory }
+    let(:object_builder) { Gitlab::ImportExport::Project::ObjectBuilder }
+    let(:relation_factory) { Gitlab::ImportExport::Project::RelationFactory }
     let(:reader) { Gitlab::ImportExport::Reader.new(shared: shared) }
     let(:tree_hash) { importable_hash }
 

spec/services/groups/import_export/export_service_spec.rb

@@ -19,7 +19,7 @@ describe Groups::ImportExport::ExportService do
     end
 
     it 'saves the models' do
-      expect(Gitlab::ImportExport::GroupTreeSaver).to receive(:new).and_call_original
+      expect(Gitlab::ImportExport::Group::TreeSaver).to receive(:new).and_call_original
 
       service.execute
     end

spec/services/projects/import_export/export_service_spec.rb

@@ -27,7 +27,7 @@ describe Projects::ImportExport::ExportService do
     end
 
     it 'saves the models' do
-      expect(Gitlab::ImportExport::ProjectTreeSaver).to receive(:new).and_call_original
+      expect(Gitlab::ImportExport::Project::TreeSaver).to receive(:new).and_call_original
 
       service.execute
     end

spec/support/import_export/common_util.rb

@@ -34,13 +34,13 @@ module ImportExport
     end
 
     def get_project_restorer(project, import_path)
-      Gitlab::ImportExport::ProjectTreeRestorer.new(
+      Gitlab::ImportExport::Project::TreeRestorer.new(
         user: project.creator, shared: get_shared_env(path: import_path), project: project
       )
     end
 
     def get_project_saver(project, export_path)
-      Gitlab::ImportExport::ProjectTreeSaver.new(
+      Gitlab::ImportExport::Project::TreeSaver.new(
         project: project, current_user: project.creator, shared: get_shared_env(path: export_path)
       )
     end

spec/support/import_export/configuration_helper.rb

@@ -36,8 +36,8 @@ module ConfigurationHelper
   end
 
   def relation_class_for_name(relation_name)
-    relation_name = Gitlab::ImportExport::ProjectRelationFactory.overrides[relation_name.to_sym] || relation_name
-    Gitlab::ImportExport::ProjectRelationFactory.relation_class(relation_name)
+    relation_name = Gitlab::ImportExport::Project::RelationFactory.overrides[relation_name.to_sym] || relation_name
+    Gitlab::ImportExport::Project::RelationFactory.relation_class(relation_name)
   end
 
   def parsed_attributes(relation_name, attributes, config: Gitlab::ImportExport.config_file)

spec/support/shared_examples/lib/gitlab/import_export/project_tree_restorer_shared_examples.rb

 # frozen_string_literal: true
 
-# Shared examples for ProjectTreeRestorer (shared to allow the testing
+# Shared examples for Project::TreeRestorer (shared to allow the testing
 # of EE-specific features)
 RSpec.shared_examples 'restores project successfully' do |**results|
   it 'restores the project' do