Populate user_highest_roles table

The user_highest_roles table is populated with a background migration
which schedules ~530 jobs with a 5 minutes gap to create an entry for
each active User.

changelogs/unreleased/populate_user_highest_roles_table.yml

+---
+title: Populate user_highest_roles table
+merge_request: 27127
+author:
+type: added

db/migrate/20200317110602_add_migrating_user_highest_roles_table_index_to_users.rb

+# frozen_string_literal: true
+
+# See https://docs.gitlab.com/ee/development/migration_style_guide.html
+# for more information on how to write migrations for GitLab.
+
+class AddMigratingUserHighestRolesTableIndexToUsers < ActiveRecord::Migration[6.0]
+  include Gitlab::Database::MigrationHelpers
+
+  DOWNTIME = false
+  INDEX_NAME = 'index_for_migrating_user_highest_roles_table'
+
+  disable_ddl_transaction!
+
+  def up
+    add_concurrent_index :users,
+                         :id,
+                         where: "state = 'active' AND user_type IS NULL AND bot_type IS NULL AND ghost IS NOT TRUE",
+                         name: INDEX_NAME
+  end
+
+  def down
+    remove_concurrent_index :users, :id, name: INDEX_NAME
+  end
+end

db/post_migrate/20200311130802_schedule_populate_user_highest_roles_table.rb

+# frozen_string_literal: true
+
+class SchedulePopulateUserHighestRolesTable < ActiveRecord::Migration[6.0]
+  include Gitlab::Database::MigrationHelpers
+
+  BATCH_SIZE = 10_000
+  DELAY = 5.minutes.to_i
+  DOWNTIME = false
+  MIGRATION = 'PopulateUserHighestRolesTable'
+
+  disable_ddl_transaction!
+
+  class User < ActiveRecord::Base
+    include EachBatch
+
+    scope :active, -> {
+      where(state: 'active', user_type: nil, bot_type: nil)
+        .where('ghost IS NOT TRUE')
+    }
+  end
+
+  def up
+    # We currently have ~5_300_000 users with the state active on GitLab.com.
+    # This means it'll schedule ~530 jobs (10k Users each) with a 5 minutes gap,
+    # so this should take ~44 hours for all background migrations to complete.
+    User.active.each_batch(of: BATCH_SIZE) do |batch, index|
+      range = batch.pluck(Arel.sql('MIN(id)'), Arel.sql('MAX(id)')).first
+      delay = index * DELAY
+
+      migrate_in(delay.seconds, MIGRATION, [*range])
+    end
+  end
+
+  def down
+    # nothing
+  end
+end

db/structure.sql

@@ -9030,6 +9030,8 @@ CREATE UNIQUE INDEX index_feature_gates_on_feature_key_and_key_and_value ON publ
 
 CREATE UNIQUE INDEX index_features_on_key ON public.features USING btree (key);
 
+CREATE INDEX index_for_migrating_user_highest_roles_table ON public.users USING btree (id) WHERE (((state)::text = 'active'::text) AND (user_type IS NULL) AND (bot_type IS NULL) AND (ghost IS NOT TRUE));
+
 CREATE INDEX index_for_resource_group ON public.ci_builds USING btree (resource_group_id, id) WHERE (resource_group_id IS NOT NULL);
 
 CREATE INDEX index_for_status_per_branch_per_project ON public.merge_trains USING btree (target_project_id, target_branch, status);
@@ -12811,6 +12813,7 @@ COPY "schema_migrations" (version) FROM STDIN;
 20200311084025
 20200311093210
 20200311094020
+20200311130802
 20200311141053
 20200311141943
 20200311154110
@@ -12825,6 +12828,7 @@ COPY "schema_migrations" (version) FROM STDIN;
 20200316111759
 20200316162648
 20200316173312
+20200317110602
 20200317142110
 20200318140400
 20200318152134

lib/gitlab/background_migration/populate_user_highest_roles_table.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module BackgroundMigration
+    # This background migration creates records on user_highest_roles according to
+    # the given user IDs range. IDs will load users with a left outer joins to
+    # have a record for users without a Group or Project. One INSERT per ID is
+    # issued.
+    class PopulateUserHighestRolesTable
+      BATCH_SIZE = 100
+
+      # rubocop:disable Style/Documentation
+      class User < ActiveRecord::Base
+        self.table_name = 'users'
+
+        scope :active, -> {
+          where(state: 'active', user_type: nil, bot_type: nil)
+            .where('ghost IS NOT TRUE')
+        }
+      end
+
+      def perform(from_id, to_id)
+        (from_id..to_id).each_slice(BATCH_SIZE) do |ids|
+          execute(
+            <<-EOF
+              INSERT INTO user_highest_roles (updated_at, user_id, highest_access_level)
+              #{select_sql(from_id, to_id)}
+              ON CONFLICT (user_id) DO
+              UPDATE SET highest_access_level = EXCLUDED.highest_access_level
+            EOF
+          )
+        end
+      end
+
+      private
+
+      def select_sql(from_id, to_id)
+        User
+          .select('NOW() as updated_at, users.id, MAX(access_level) AS highest_access_level')
+          .joins('LEFT OUTER JOIN members ON members.user_id = users.id AND members.requested_at IS NULL')
+          .where(users: { id: active_user_ids(from_id, to_id) })
+          .group('users.id')
+          .to_sql
+      end
+
+      def active_user_ids(from_id, to_id)
+        User.active.where(users: { id: from_id..to_id }).pluck(:id)
+      end
+
+      def execute(sql)
+        @connection ||= ActiveRecord::Base.connection
+        @connection.execute(sql)
+      end
+    end
+  end
+end

spec/lib/gitlab/background_migration/populate_user_highest_roles_table_spec.rb

+# frozen_string_literal: true
+
+require 'spec_helper'
+
+describe Gitlab::BackgroundMigration::PopulateUserHighestRolesTable, schema: 20200311130802 do
+  let(:members) { table(:members) }
+  let(:users) { table(:users) }
+  let(:user_highest_roles) { table(:user_highest_roles) }
+
+  def create_user(id, params = {})
+    user_params = {
+      id: id,
+      state: 'active',
+      user_type: nil,
+      bot_type: nil,
+      ghost: nil,
+      email: "user#{id}@example.com",
+      projects_limit: 0
+    }.merge(params)
+
+    users.create(user_params)
+  end
+
+  def create_member(id, access_level, params = {})
+    params = {
+      user_id: id,
+      access_level: access_level,
+      source_id: 1,
+      source_type: 'Group',
+      notification_level: 0
+    }.merge(params)
+
+    members.create(params)
+  end
+
+  before do
+    create_user(1)
+    create_user(2, state: 'blocked')
+    create_user(3, user_type: 2)
+    create_user(4)
+    create_user(5, bot_type: 1)
+    create_user(6, ghost: true)
+    create_user(7, ghost: false)
+    create_user(8)
+
+    create_member(1, 40)
+    create_member(7, 30)
+    create_member(8, 20, requested_at: Time.current)
+
+    user_highest_roles.create(user_id: 1, highest_access_level: 50)
+  end
+
+  describe '#perform' do
+    it 'creates user_highest_roles rows according to users', :aggregate_failures do
+      expect { subject.perform(1, 8) }.to change(UserHighestRole, :count).from(1).to(4)
+
+      created_or_updated_rows = [
+        { 'user_id' => 1, 'highest_access_level' => 40 },
+        { 'user_id' => 4, 'highest_access_level' => nil },
+        { 'user_id' => 7, 'highest_access_level' => 30 },
+        { 'user_id' => 8, 'highest_access_level' => nil }
+      ]
+
+      rows = user_highest_roles.order(:user_id).map do |row|
+        row.attributes.slice('user_id', 'highest_access_level')
+      end
+
+      expect(rows).to match_array(created_or_updated_rows)
+    end
+  end
+end

spec/migrations/schedule_populate_user_highest_roles_table_spec.rb

+# frozen_string_literal: true
+
+require 'spec_helper'
+require Rails.root.join('db', 'post_migrate', '20200311130802_schedule_populate_user_highest_roles_table.rb')
+
+describe SchedulePopulateUserHighestRolesTable do
+  let(:users) { table(:users) }
+
+  def create_user(id, params = {})
+    user_params = {
+      id: id,
+      state: 'active',
+      user_type: nil,
+      bot_type: nil,
+      ghost: nil,
+      email: "user#{id}@example.com",
+      projects_limit: 0
+    }.merge(params)
+
+    users.create!(user_params)
+  end
+
+  it 'correctly schedules background migrations' do
+    create_user(1)
+    create_user(2, state: 'blocked')
+    create_user(3, user_type: 2)
+    create_user(4)
+    create_user(5, bot_type: 1)
+    create_user(6, ghost: true)
+    create_user(7, ghost: false)
+
+    stub_const("#{described_class.name}::BATCH_SIZE", 2)
+
+    Sidekiq::Testing.fake! do
+      Timecop.freeze do
+        migrate!
+
+        expect(described_class::MIGRATION).to be_scheduled_delayed_migration(5.minutes, 1, 4)
+
+        expect(described_class::MIGRATION).to be_scheduled_delayed_migration(10.minutes, 7, 7)
+
+        expect(BackgroundMigrationWorker.jobs.size).to eq(2)
+      end
+    end
+  end
+end