LightRecord
ActiveRecord extension to kick the speed of allocating ActiveRecord object, and it allows fetching DB rows sequentially to reduce memory usage
Supports: Rails 5, 6 and 7
Note: it overrides some internal methods of active record, and in some rare cases it not work correctly, please test it well in your project before using it in production
How it works
It provides functionality to load ActiveRecord records with patched attribute related methods. This make AR objects as read-only but it makes up to 5 times less object allocations.
Each time when you retrieve objects via .light_records
it will create anonymous class to work with given set of attributes.
LightRecord Extension Class
↓
Your AR Model
↓
ActiveRecord::Base
Installation
gem 'light_record', github: 'paxa/light_record'
scope.light_records
records = User.limit(1_000_000).light_records
records # => array of records. Very fast and very memory efficient
Idea is to skip all magic related to attributes and object initialization. This creates new class inherited from your model. That allows us to create only one extra object when we initialize new record.
Simply it become something like this:
class User_light_record < User
def initialize(attributes)
@attributes = attributes # hash of data "as is" from database library
end
def email
@attributes[:email]
end
end
scope.light_records_each
Other method: .light_records_each
, it will utilize stream: true
feature from mysql2 client. So it will initialize objects one by one for every interaction:
User.limit(1_000_000).light_records_each do |user|
user.do_something
end
This allow you to interate big amount of data without using find_each
or find_in_batches
because with light_records_each
it will use very low memory. Or allow you to use find_in_batches
with bigger batch size
* Please note that time will be as a ruby Time object, instead of TimeWithZone. To make it in correct time zone you can call it as:
record.created_at.in_time_zone(Time.zone)
Benchmarks
Still on a way, but I try to use in some project and it gives 3-5 times improvement, and 2-3 times less memory usage
Sometimes this can break functionality because it will override attribute methods and disable some of features in activerecord.
There is mechanism to override attribute methods created by LightRecord:
class User < ActiveRecord::Base
# this module will be included in extending class when we use light_records and light_records_each
module LightRecord
def sometihng
end
def success
attributes[:success] == 1
end
def success_time
return attributes[:success_time] unless attributes[:success_time]
@success_time ||= attributes[:success_time].in_time_zone(Time.zone)
end
def success_time=(val)
@success_time = nil
super(val)
end
end
end
Note: when you use LightRecord instances it will break type casting
This gem supports MySQL and PostgreSQL