Add finder to find unsynced projects using FDW queries

ee/app/finders/geo/project_unsynced_finder.rb

+# frozen_string_literal: true
+
+# Finder for retrieving unsynced projects that belong to a specific
+# shard using FDW queries.
+#
+# Basic usage:
+#
+#     Geo::ProjectUnsyncedFinder
+#       .new(current_node: Gitlab::Geo.current_node, shard_name: 'default', batch_size: 1000)
+#       .execute.
+module Geo
+  class ProjectUnsyncedFinder
+    def initialize(current_node:, shard_name:, batch_size:)
+      @current_node = Geo::Fdw::GeoNode.find(current_node.id)
+      @shard_name = shard_name
+      @batch_size = batch_size
+    end
+
+    # rubocop:disable CodeReuse/ActiveRecord
+    def execute
+      return Geo::Fdw::Project.none unless valid_shard?
+
+      current_node
+        .projects
+        .missing_project_registry
+        .within_shards(shard_name)
+        .limit(batch_size)
+    end
+    # rubocop:enable CodeReuse/ActiveRecord
+
+    private
+
+    attr_reader :current_node, :shard_name, :batch_size
+
+    def valid_shard?
+      return true unless current_node.selective_sync_by_shards?
+
+      current_node.selective_sync_shards.include?(shard_name)
+    end
+  end
+end

ee/app/models/geo/fdw/project.rb

@@ -8,6 +8,11 @@ module Geo
       self.table_name = Gitlab::Geo::Fdw.foreign_table_name('projects')
 
       class << self
+        def missing_project_registry
+          left_outer_join_project_registry
+            .where(Geo::ProjectRegistry.arel_table[:project_id].eq(nil))
+        end
+
         # Searches for a list of projects based on the query given in `query`.
         #
         # On PostgreSQL this method uses "ILIKE" to perform a case-insensitive
@@ -25,6 +30,17 @@ module Geo
         def within_shards(shard_names)
           where(repository_storage: Array(shard_names))
         end
+
+        private
+
+        def left_outer_join_project_registry
+          join_statement =
+            arel_table
+              .join(Geo::ProjectRegistry.arel_table, Arel::Nodes::OuterJoin)
+              .on(arel_table[:id].eq(Geo::ProjectRegistry.arel_table[:project_id]))
+
+          joins(join_statement.join_sources)
+        end
       end
     end
   end

ee/spec/finders/geo/project_unsynced_finder_spec.rb

+# frozen_string_literal: true
+
+require 'spec_helper'
+
+describe Geo::ProjectUnsyncedFinder, :geo do
+  # Disable transactions via :delete method because a foreign table
+  # can't see changes inside a transaction of a different connection.
+  describe '#execute', :delete do
+    let(:node) { create(:geo_node) }
+    let(:group_1) { create(:group) }
+    let(:group_2) { create(:group) }
+    let(:nested_group_1) { create(:group, parent: group_1) }
+    let!(:project_1) { create(:project, group: group_1) }
+    let!(:project_2) { create(:project, group: nested_group_1) }
+    let!(:project_3) { create(:project, group: group_2) }
+    let!(:project_4) { create(:project, group: group_1) }
+
+    before do
+      skip('FDW is not configured') unless Gitlab::Geo::Fdw.enabled?
+
+      project_4.update_column(:repository_storage, 'foo')
+    end
+
+    subject { described_class.new(current_node: node, shard_name: 'default', batch_size: 100) }
+
+    context 'without selective sync' do
+      it 'returns projects without an entry on the tracking database' do
+        create(:geo_project_registry, :synced, project: project_2)
+
+        expect(subject.execute).to match_ids(project_1, project_3)
+      end
+    end
+
+    context 'with selective sync by namespace' do
+      it 'returns projects that belong to the namespaces without an entry on the tracking database' do
+        create(:geo_project_registry, :synced, project: project_4)
+
+        node.update!(selective_sync_type: 'namespaces', namespaces: [group_1, nested_group_1])
+
+        expect(subject.execute).to match_ids(project_1, project_2)
+      end
+    end
+
+    context 'with selective sync by shard' do
+      before do
+        node.update!(selective_sync_type: 'shards', selective_sync_shards: ['foo'])
+      end
+
+      it 'does not return registries when selected shards to sync does not include the shard_name' do
+        subject = described_class.new(current_node: node, shard_name: 'default', batch_size: 100)
+
+        expect(subject.execute).to be_empty
+      end
+
+      it 'returns projects that belong to the shards without an entry on the tracking database' do
+        project_5 = create(:project, group: group_1)
+        project_5.update_column(:repository_storage, 'foo')
+        create(:geo_project_registry, :synced, project: project_4)
+
+        subject = described_class.new(current_node: node, shard_name: 'foo', batch_size: 100)
+
+        expect(subject.execute).to match_ids(project_5)
+      end
+    end
+  end
+end