random_batches.rb

    module BatchesRandomStart
      # `find_in_batches`, but starting from a random position in the set of
      # records, then continuing from the beginning of the set.  Give up early
      # if `time_limit` elapses.
      def find_in_batches_random_start(batch_size: 10, time_limit: nil)
        relation = scoped

        c = relation.count
        return if c == 0

        # start a random amount through the set of records
        start_id = offset(rand(c)).first.try(:id)
        return unless start_id

        start_time = Time.now
        timesup = ->{ time_limit && (start_time + time_limit < Time.now) }

        relation.find_in_batches(batch_size: batch_size, start: start_id) do |batch|
          yield batch unless batch.empty?
          return if timesup.()
        end

        # start again from 0, processing only records with ids less than where
        # we started
        relation = relation.where(relation.table[relation.primary_key].lt(start_id))
        relation.find_in_batches(batch_size: batch_size, start: 0) do |batch|
          yield batch unless batch.empty?
          return if timesup.()
        end
      end

      def find_each_random_start(batch_size: 10, time_limit: nil, &block)
        find_in_batches_random_start(batch_size: batch_size, time_limit: time_limit) do |batch|
          batch.each(&block)
        end
      end
    end

    ActiveRecord::Base.extend(BatchesRandomStart)