Backfill LfsObjectsProject records of forks

To fix the behavior of existing forks, `LfsObjectsProject` records
need to be backfilled. This is so we can remove the neeed to
depend on the source when looking for LFS objects from forks.

changelogs/unreleased/55487-backfill-lfs-objects-projects-sql.yml

+---
+title: Backfill LfsObjectsProject records of forks
+merge_request: 24767
+author:
+type: other

db/post_migrate/20200210062432_schedule_link_lfs_objects.rb

+# frozen_string_literal: true
+
+class ScheduleLinkLfsObjects < ActiveRecord::Migration[6.0]
+  include Gitlab::Database::MigrationHelpers
+
+  DOWNTIME = false
+  MIGRATION = 'LinkLfsObjects'
+  BATCH_SIZE = 1_000
+
+  disable_ddl_transaction!
+
+  class Project < ActiveRecord::Base
+    include EachBatch
+
+    self.table_name = 'projects'
+  end
+
+  def up
+    fork_network_members =
+      Gitlab::BackgroundMigration::LinkLfsObjects::ForkNetworkMember
+        .select(1)
+        .with_non_existing_lfs_objects
+        .where('fork_network_members.project_id = projects.id')
+
+    forks = Project.where('EXISTS (?)', fork_network_members)
+
+    queue_background_migration_jobs_by_range_at_intervals(
+      forks,
+      MIGRATION,
+      BackgroundMigrationWorker.minimum_interval,
+      batch_size: BATCH_SIZE
+    )
+  end
+
+  def down
+    # no-op
+  end
+end

lib/gitlab/background_migration/link_lfs_objects.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module BackgroundMigration
+    # Create missing LfsObjectsProject records for forks
+    class LinkLfsObjects
+      # Model definition used for migration
+      class ForkNetworkMember < ActiveRecord::Base
+        self.table_name = 'fork_network_members'
+
+        def self.with_non_existing_lfs_objects
+          joins('JOIN lfs_objects_projects lop ON fork_network_members.forked_from_project_id = lop.project_id')
+            .where(
+              <<~SQL
+                NOT EXISTS (
+                  SELECT 1
+                  FROM lfs_objects_projects
+                  WHERE lfs_objects_projects.project_id = fork_network_members.project_id
+                  AND lfs_objects_projects.lfs_object_id = lop.lfs_object_id
+                )
+              SQL
+            )
+        end
+      end
+
+      def perform(start_id, end_id)
+        select_query =
+          ForkNetworkMember
+            .select('lop.lfs_object_id, fork_network_members.project_id')
+            .with_non_existing_lfs_objects
+            .where(project_id: start_id..end_id)
+
+        return if select_query.empty?
+
+        execute <<-SQL
+          INSERT INTO lfs_objects_projects (lfs_object_id, project_id)
+          #{select_query.to_sql}
+        SQL
+      end
+
+      private
+
+      def execute(sql)
+        ::ActiveRecord::Base.connection.execute(sql)
+      end
+    end
+  end
+end

spec/lib/gitlab/background_migration/link_lfs_objects_spec.rb

+# frozen_string_literal: true
+
+require 'spec_helper'
+
+describe Gitlab::BackgroundMigration::LinkLfsObjects, :migration, schema: 2020_02_10_062432 do
+  let(:namespaces) { table(:namespaces) }
+  let(:projects) { table(:projects) }
+  let(:fork_networks) { table(:fork_networks) }
+  let(:fork_network_members) { table(:fork_network_members) }
+  let(:lfs_objects) { table(:lfs_objects) }
+  let(:lfs_objects_projects) { table(:lfs_objects_projects) }
+  let(:namespace) { namespaces.create(name: 'GitLab', path: 'gitlab') }
+  let(:source_project) { projects.create(namespace_id: namespace.id) }
+  let(:another_source_project) { projects.create(namespace_id: namespace.id) }
+  let(:project) { projects.create(namespace_id: namespace.id) }
+  let(:another_project) { projects.create(namespace_id: namespace.id) }
+  let(:other_project) { projects.create(namespace_id: namespace.id) }
+  let(:linked_project) { projects.create(namespace_id: namespace.id) }
+  let(:fork_network) { fork_networks.create(root_project_id: source_project.id) }
+  let(:another_fork_network) { fork_networks.create(root_project_id: another_source_project.id) }
+  let(:lfs_object) { lfs_objects.create(oid: 'abc123', size: 100) }
+  let(:another_lfs_object) { lfs_objects.create(oid: 'def456', size: 200) }
+
+  before do
+    # Create links between projects
+    fork_network_members.create(fork_network_id: fork_network.id, project_id: source_project.id, forked_from_project_id: nil)
+
+    [project, another_project, linked_project].each do |p|
+      fork_network_members.create(
+        fork_network_id: fork_network.id,
+        project_id: p.id,
+        forked_from_project_id: fork_network.root_project_id
+      )
+    end
+
+    fork_network_members.create(fork_network_id: another_fork_network.id, project_id: another_source_project.id, forked_from_project_id: nil)
+    fork_network_members.create(fork_network_id: another_fork_network.id, project_id: other_project.id, forked_from_project_id: another_fork_network.root_project_id)
+
+    # Links LFS objects to some projects
+    [source_project, another_source_project, linked_project].each do |p|
+      lfs_objects_projects.create(lfs_object_id: lfs_object.id, project_id: p.id)
+      lfs_objects_projects.create(lfs_object_id: another_lfs_object.id, project_id: p.id)
+    end
+  end
+
+  it 'creates LfsObjectsProject records for forks within the specified range of project IDs' do
+    expect { subject.perform(project.id, other_project.id) }.to change { lfs_objects_projects.count }.by(6)
+
+    expect(lfs_object_ids_for(project)).to match_array(lfs_object_ids_for(source_project))
+    expect(lfs_object_ids_for(another_project)).to match_array(lfs_object_ids_for(source_project))
+    expect(lfs_object_ids_for(other_project)).to match_array(lfs_object_ids_for(another_source_project))
+
+    expect { subject.perform(project.id, other_project.id) }.not_to change { lfs_objects_projects.count }
+  end
+
+  context 'when it is not necessary to create LfsObjectProject records' do
+    it 'does not create LfsObjectProject records' do
+      expect { subject.perform(linked_project.id, linked_project.id) }
+        .not_to change { lfs_objects_projects.count }
+    end
+  end
+
+  def lfs_object_ids_for(project)
+    lfs_objects_projects.where(project_id: project.id).pluck(:lfs_object_id)
+  end
+end

spec/migrations/schedule_link_lfs_objects_spec.rb

+# frozen_string_literal: true
+
+require 'spec_helper'
+require Rails.root.join('db', 'post_migrate', '20200210062432_schedule_link_lfs_objects.rb')
+
+describe ScheduleLinkLfsObjects, :migration, :sidekiq do
+  let(:namespaces) { table(:namespaces) }
+  let(:projects) { table(:projects) }
+  let(:fork_networks) { table(:fork_networks) }
+  let(:fork_network_members) { table(:fork_network_members) }
+  let(:lfs_objects) { table(:lfs_objects) }
+  let(:lfs_objects_projects) { table(:lfs_objects_projects) }
+  let(:namespace) { namespaces.create(name: 'GitLab', path: 'gitlab') }
+  let(:fork_network) { fork_networks.create(root_project_id: source_project.id) }
+  let(:another_fork_network) { fork_networks.create(root_project_id: another_source_project.id) }
+  let(:source_project) { projects.create(namespace_id: namespace.id) }
+  let(:another_source_project) { projects.create(namespace_id: namespace.id) }
+  let(:project) { projects.create(namespace_id: namespace.id) }
+  let(:another_project) { projects.create(namespace_id: namespace.id) }
+  let(:other_project) { projects.create(namespace_id: namespace.id) }
+  let(:linked_project) { projects.create(namespace_id: namespace.id) }
+  let(:lfs_object) { lfs_objects.create(oid: 'abc123', size: 100) }
+  let(:another_lfs_object) { lfs_objects.create(oid: 'def456', size: 200) }
+
+  before do
+    # Create links between projects
+    fork_network_members.create(fork_network_id: fork_network.id, project_id: source_project.id, forked_from_project_id: nil)
+
+    [project, another_project, linked_project].each do |p|
+      fork_network_members.create(
+        fork_network_id: fork_network.id,
+        project_id: p.id,
+        forked_from_project_id: fork_network.root_project_id
+      )
+    end
+
+    fork_network_members.create(fork_network_id: another_fork_network.id, project_id: another_source_project.id, forked_from_project_id: nil)
+    fork_network_members.create(fork_network_id: another_fork_network.id, project_id: other_project.id, forked_from_project_id: another_fork_network.root_project_id)
+  end
+
+  context 'when there are forks to be backfilled' do
+    before do
+      stub_const("#{described_class.name}::BATCH_SIZE", 2)
+
+      # Links LFS objects to some projects
+      [source_project, another_source_project, linked_project].each do |p|
+        lfs_objects_projects.create(lfs_object_id: lfs_object.id, project_id: p.id)
+        lfs_objects_projects.create(lfs_object_id: another_lfs_object.id, project_id: p.id)
+      end
+    end
+
+    it 'schedules background migration to link LFS objects' do
+      Sidekiq::Testing.fake! do
+        migrate!
+
+        expect(BackgroundMigrationWorker.jobs.size).to eq(2)
+        expect(described_class::MIGRATION)
+          .to be_scheduled_delayed_migration(2.minutes, project.id, another_project.id)
+        expect(described_class::MIGRATION)
+          .to be_scheduled_delayed_migration(4.minutes, other_project.id, other_project.id)
+      end
+    end
+  end
+
+  context 'when there are no forks to be backfilled' do
+    before do
+      # Links LFS objects to all projects
+      projects.all.each do |p|
+        lfs_objects_projects.create(lfs_object_id: lfs_object.id, project_id: p.id)
+        lfs_objects_projects.create(lfs_object_id: another_lfs_object.id, project_id: p.id)
+      end
+    end
+
+    it 'does not schedule any job' do
+      Sidekiq::Testing.fake! do
+        migrate!
+
+        expect(BackgroundMigrationWorker.jobs.size).to eq(0)
+      end
+    end
+  end
+end