ActiveRecord find_each combined with limit and order

I'm trying to run a query of about 50,000 records using ActiveRecord's find_each method, but it seems to be ignoring my other parameters like so:

Thing.active.order("created_at DESC").limit(50000).find_each {|t| puts t.id }

Instead of stopping at 50,000 I'd like and sorting by created_at, here's the resulting query that gets executed over the entire dataset:

Thing Load (198.8ms)  SELECT "things".* FROM "things" WHERE "things"."active" = 't' AND ("things"."id" > 373343) ORDER BY "things"."id" ASC LIMIT 1000

Is there a way to get similar behavior to find_each but with a total max limit and respecting my sort criteria?

The documentation says that find_each and find_in_batches don't retain sort order and limit because:

• Sorting ASC on the PK is used to make the batch ordering work.
• Limit is used to control the batch sizes.

You could write your own version of this function like @rorra did. But you can get into trouble when mutating the objects. If for example you sort by created_at and save the object it might come up again in one of the next batches. Similarly you might skip objects because the order of results has changed when executing the query to get the next batch. Only use that solution with read only objects.

Now my primary concern was that I didn't want to load 30000+ objects into memory at once. My concern was not the execution time of the query itself. Therefore I used a solution that executes the original query but only caches the ID's. It then divides the array of ID's into chunks and queries/creates the objects per chunk. This way you can safely mutate the objects because the sort order is kept in memory.

Here is a minimal example similar to what I did:

batch_size = 512
ids = Thing.order('created_at DESC').pluck(:id) # Replace .order(:created_at) with your own scope
ids.each_slice(batch_size) do |chunk|
    Thing.find(chunk, :order => "field(id, #{chunk.join(',')})").each do |thing|
      # Do things with thing
    end
end

The trade-offs to this solution are:

• The complete query is executed to get the ID's
• An array of all the ID's is kept in memory
• Uses the MySQL specific FIELD() function

Hope this helps!

find_each uses find_in_batches under the hood.

Its not possible to select the order of the records, as described in find_in_batches, is automatically set to ascending on the primary key (“id ASC”) to make the batch ordering work.

However, the criteria is applied, what you can do is:

Thing.active.find_each(batch_size: 50000) { |t| puts t.id }

Regarding the limit, it wasn't implemented yet: https://github.com/rails/rails/pull/5696

---

Answering to your second question, you can create the logic yourself:

total_records = 50000
batch = 1000
(0..(total_records - batch)).step(batch) do |i|
  puts Thing.active.order("created_at DESC").offset(i).limit(batch).to_sql
end

Retrieving the ids first and processing the in_groups_of

ordered_photo_ids = Photo.order(likes_count: :desc).pluck(:id)

ordered_photo_ids.in_groups_of(1000, false).each do |photo_ids|
  photos = Photo.order(likes_count: :desc).where(id: photo_ids)

  # ...
end

It's important to also add the ORDER BY query to the inner call.

Rails 6.1 adds support for descending order in find_each, find_in_batches and in_batches.

One option is to put an implementation tailored for your particular model into the model itself (speaking of which, id is usually a better choice for ordering records, created_at may have duplicates):

class Thing < ActiveRecord::Base
  def self.find_each_desc limit
    batch_size = 1000
    i = 1
    records = self.order(created_at: :desc).limit(batch_size)
    while records.any?
      records.each do |task|
        yield task, i
        i += 1
        return if i > limit
      end
      records = self.order(created_at: :desc).where('id < ?', records.last.id).limit(batch_size)
    end
  end
end

Or else you can generalize things a bit, and make it work for all the models:

lib/active_record_extensions.rb:

ActiveRecord::Batches.module_eval do
  def find_each_desc limit
    batch_size = 1000
    i = 1
    records = self.order(id: :desc).limit(batch_size)
    while records.any?
      records.each do |task|
        yield task, i
        i += 1
        return if i > limit
      end
      records = self.order(id: :desc).where('id < ?', records.last.id).limit(batch_size)
    end
  end
end

ActiveRecord::Querying.module_eval do
  delegate :find_each_desc, :to => :all
end

config/initializers/extensions.rb:

require "active_record_extensions"

P.S. I'm putting the code in files according to this answer.

You can iterate backwards by standard ruby iterators:

Thing.last.id.step(0,-1000) do |i|
  Thing.where(id: (i-1000+1)..i).order('id DESC').each do |thing|
    #...
  end
end

Note: +1 is because BETWEEN which will be in query includes both bounds but we need include only one.

Sure, with this approach there could be fetched less than 1000 records in batch because some of them are deleted already but this is ok in my case.

As remarked by @Kirk in one of the comments, find_each supports limit as of version 5.1.0.

Example from the changelog:

Post.limit(10_000).find_each do |post|
  # ...
end

The documentation says:

> Limits are honored, and if present there is no requirement for the batch size: it can be less than, equal to, or greater than the limit.

(setting a custom order is still not supported though)

I was looking for the same behaviour and thought up of this solution. This DOES NOT order by created_at but I thought I would post anyways.

max_records_to_retrieve = 50000
last_index = Thing.count
start_index = [(last_index - max_records_to_retrieve), 0].max
Thing.active.find_each(:start => start_index) do |u|
    # do stuff
end

Drawbacks of this approach:

• You need 2 queries (first one should be fast)
• This guarantees a max of 50K records but if ids are skipped you will get less.

You can try ar-as-batches Gem.

From their documentation you can do something like this

Users.where(country_id: 44).order(:joined_at).offset(200).as_batches do |user|
  user.party_all_night!
end

Using Kaminari or something other it will be easy.

Create batch loader class.

module BatchLoader
  extend ActiveSupport::Concern

  def batch_by_page(options = {})
    options = init_batch_options!(options)

    next_page = 1

    loop do
      next_page = yield(next_page, options[:batch_size])

      break next_page if next_page.nil?
    end
  end

  private

  def default_batch_options
    {
      batch_size: 50
    }
  end

  def init_batch_options!(options)
    options ||= {}
    default_batch_options.merge!(options)
  end
end

Create Repository

class ThingRepository
  include BatchLoader

  # @param [Integer] per_page
  # @param [Proc] block
  def batch_changes(per_page=100, &block)
    relation = Thing.active.order("created_at DESC")

    batch_by_page do |next_page|
      query = relation.page(next_page).per(per_page)
      yield query if block_given?
      query.next_page
    end
  end
end

Use the repository

repo = ThingRepository.new
repo.batch_changes(5000).each do |g|
  g.each do |t|
    #...
  end
end

Adding find_in_batches_with_order did solve my usecase, where I was having ids already but need batching and ordering. It was inspired by @dirk-geurs solution

# Create file config/initializers/find_in_batches_with_order.rb with follwing code.
ActiveRecord::Batches.class_eval do
  ## Only flat order structure is supported now
  ## example: [:forename, :surname] is supported but [:forename, {surname: :asc}] is not supported
  def find_in_batches_with_order(ids: nil, order: [], batch_size: 1000)
    relation = self
    arrangement = order.dup
    index = order.find_index(:id)

    unless index
      arrangement.push(:id)
      index = arrangement.length - 1
    end

    ids ||= relation.order(*arrangement).pluck(*arrangement).map{ |tupple| tupple[index] }
    ids.each_slice(batch_size) do |chunk_ids|
      chunk_relation = relation.where(id: chunk_ids).order(*order)
      yield(chunk_relation)
    end
  end
end

Leaving Gist here https://gist.github.com/the-spectator/28b1176f98cc2f66e870755bb2334545

I had the same problem with a query with DISTINCT ON where you need an ORDER BY with that field, so this is my approach with Postgres:

def filtered_model_ids
  Model.joins(:father_model)
       .select('DISTINCT ON (model.field) model.id')
       .order(:field)
       .map(&:id)
end

def processor
  filtered_model_ids.each_slice(BATCH_SIZE).lazy.each do |batch|
    Model.find(batch).each do |record|
      # Code
    end
  end
end

My code

batch_size = 100
total_count = klass.count
offset = 0
processed_count = 0
while processed_count < total_count
  relation = klass.order({ active_at: :asc, created_at: :desc }).offset(offset).limit(batch_size)
  relation.each do |record|
    record.process
  end
  processed_count += batch_size
end

Do it in one query and avoid iterating:

will product a query like this