Added EachBatch for iterating tables in batches

This module provides a class method called `each_batch` that can be used
to iterate tables in batches in a more efficient way compared to Rails'
`in_batches` method. This commit also includes a RuboCop cop to
blacklist the use of `in_batches` in favour of this new method.

app/models/concerns/each_batch.rb

+module EachBatch
+  extend ActiveSupport::Concern
+
+  module ClassMethods
+    # Iterates over the rows in a relation in batches, similar to Rails'
+    # `in_batches` but in a more efficient way.
+    #
+    # Unlike `in_batches` provided by Rails this method does not support a
+    # custom start/end range, nor does it provide support for the `load:`
+    # keyword argument.
+    #
+    # This method will yield an ActiveRecord::Relation to the supplied block, or
+    # return an Enumerator if no block is given.
+    #
+    # Example:
+    #
+    #     User.each_batch do |relation|
+    #       relation.update_all(updated_at: Time.now)
+    #     end
+    #
+    # The supplied block is also passed an optional batch index:
+    #
+    #     User.each_batch do |relation, index|
+    #       puts index # => 1, 2, 3, ...
+    #     end
+    #
+    # You can also specify an alternative column to use for ordering the rows:
+    #
+    #     User.each_batch(column: :created_at) do |relation|
+    #       ...
+    #     end
+    #
+    # This will produce SQL queries along the lines of:
+    #
+    #     User Load (0.7ms)  SELECT  "users"."id" FROM "users" WHERE ("users"."id" >= 41654)  ORDER BY "users"."id" ASC LIMIT 1 OFFSET 1000
+    #       (0.7ms)  SELECT COUNT(*) FROM "users" WHERE ("users"."id" >= 41654) AND ("users"."id" < 42687)
+    #
+    # of - The number of rows to retrieve per batch.
+    # column - The column to use for ordering the batches.
+    def each_batch(of: 1000, column: primary_key)
+      unless column
+        raise ArgumentError,
+          'the column: argument must be set to a column name to use for ordering rows'
+      end
+
+      start = except(:select)
+        .select(column)
+        .reorder(column => :asc)
+        .take
+
+      return unless start
+
+      start_id = start[column]
+      arel_table = self.arel_table
+
+      1.step do |index|
+        stop = except(:select)
+          .select(column)
+          .where(arel_table[column].gteq(start_id))
+          .reorder(column => :asc)
+          .offset(of)
+          .limit(1)
+          .take
+
+        relation = where(arel_table[column].gteq(start_id))
+
+        if stop
+          stop_id = stop[column]
+          start_id = stop_id
+          relation = relation.where(arel_table[column].lt(stop_id))
+        end
+
+        # Any ORDER BYs are useless for this relation and can lead to less
+        # efficient UPDATE queries, hence we get rid of it.
+        yield relation.except(:order), index
+
+        break unless stop
+      end
+    end
+  end
+end

doc/development/README.md

@@ -56,6 +56,7 @@
 - [Single Table Inheritance](single_table_inheritance.md)
 - [Background Migrations](background_migrations.md)
 - [Storing SHA1 Hashes As Binary](sha1_as_binary.md)
+- [Iterating Tables In Batches](iterating_tables_in_batches.md)
 
 ## i18n
 

doc/development/iterating_tables_in_batches.md

+# Iterating Tables In Batches
+
+Rails provides a method called `in_batches` that can be used to iterate over
+rows in batches. For example:
+
+```ruby
+User.in_batches(of: 10) do |relation|
+  relation.update_all(updated_at: Time.now)
+end
+```
+
+Unfortunately this method is implemented in a way that is not very efficient,
+both query and memory usage wise.
+
+To work around this you can include the `EachBatch` module into your models,
+then use the `each_batch` class method. For example:
+
+```ruby
+class User < ActiveRecord::Base
+  include EachBatch
+end
+
+User.each_batch(of: 10) do |relation|
+  relation.update_all(updated_at: Time.now)
+end
+```
+
+This will end up producing queries such as:
+
+```
+User Load (0.7ms)  SELECT  "users"."id" FROM "users" WHERE ("users"."id" >= 41654)  ORDER BY "users"."id" ASC LIMIT 1 OFFSET 1000
+  (0.7ms)  SELECT COUNT(*) FROM "users" WHERE ("users"."id" >= 41654) AND ("users"."id" < 42687)
+```
+
+The API of this method is similar to `in_batches`, though it doesn't support
+all of the arguments that `in_batches` supports. You should always use
+`each_batch` _unless_ you have a specific need for `in_batches`.

lib/gitlab/otp_key_rotator.rb

@@ -34,7 +34,7 @@ module Gitlab
 
       write_csv do |csv|
         ActiveRecord::Base.transaction do
-          User.with_two_factor.in_batches do |relation|
+          User.with_two_factor.in_batches do |relation| # rubocop: disable Cop/InBatches
             rows = relation.pluck(:id, :encrypted_otp_secret, :encrypted_otp_secret_iv, :encrypted_otp_secret_salt)
             rows.each do |row|
               user = %i[id ciphertext iv salt].zip(row).to_h

rubocop/cop/in_batches.rb

+require_relative '../model_helpers'
+
+module RuboCop
+  module Cop
+    # Cop that prevents the use of `in_batches`
+    class InBatches < RuboCop::Cop::Cop
+      MSG = 'Do not use `in_batches`, use `each_batch` from the EachBatch module instead'.freeze
+
+      def on_send(node)
+        return unless node.children[1] == :in_batches
+
+        add_offense(node, :selector)
+      end
+    end
+  end
+end

rubocop/rubocop.rb

@@ -5,6 +5,7 @@ require_relative 'cop/redirect_with_status'
 require_relative 'cop/polymorphic_associations'
 require_relative 'cop/project_path_helper'
 require_relative 'cop/active_record_dependent'
+require_relative 'cop/in_batches'
 require_relative 'cop/migration/add_column'
 require_relative 'cop/migration/add_column_with_default_to_large_table'
 require_relative 'cop/migration/add_concurrent_foreign_key'

spec/models/concerns/each_batch_spec.rb

+require 'spec_helper'
+
+describe EachBatch do
+  describe '.each_batch' do
+    let(:model) do
+      Class.new(ActiveRecord::Base) do
+        include EachBatch
+
+        self.table_name = 'users'
+      end
+    end
+
+    before do
+      5.times { create(:user, updated_at: 1.day.ago) }
+    end
+
+    it 'yields an ActiveRecord::Relation when a block is given' do
+      model.each_batch do |relation|
+        expect(relation).to be_a_kind_of(ActiveRecord::Relation)
+      end
+    end
+
+    it 'yields a batch index as the second argument' do
+      model.each_batch do |_, index|
+        expect(index).to eq(1)
+      end
+    end
+
+    it 'accepts a custom batch size' do
+      amount = 0
+
+      model.each_batch(of: 1) { amount += 1 }
+
+      expect(amount).to eq(5)
+    end
+
+    it 'does not include ORDER BYs in the yielded relations' do
+      model.each_batch do |relation|
+        expect(relation.to_sql).not_to include('ORDER BY')
+      end
+    end
+
+    it 'allows updating of the yielded relations' do
+      time = Time.now
+
+      model.each_batch do |relation|
+        relation.update_all(updated_at: time)
+      end
+
+      expect(model.where(updated_at: time).count).to eq(5)
+    end
+  end
+end

spec/rubocop/cop/in_batches_spec.rb

+require 'spec_helper'
+require 'rubocop'
+require 'rubocop/rspec/support'
+require_relative '../../../rubocop/cop/in_batches'
+
+describe RuboCop::Cop::InBatches do
+  include CopHelper
+
+  subject(:cop) { described_class.new }
+
+  it 'registers an offense when in_batches is used' do
+    inspect_source(cop, 'foo.in_batches do; end')
+
+    aggregate_failures do
+      expect(cop.offenses.size).to eq(1)
+      expect(cop.offenses.map(&:line)).to eq([1])
+    end
+  end
+end