transaction_isolation_continued

Set transaction isolation level in the ActiveRecord in a database agnostic way. Works with MySQL, PostgreSQL and SQLite as long as you are using new adapters mysql2, pg or sqlite3. Supports all ANSI SQL isolation levels: :serializable, :repeatable_read, :read_committed, :read_uncommitted.

See also transaction_retry gem for auto-retrying transactions on deadlocks and serialization errors.

Example

ActiveRecord::Base.isolation_level(:serializable) do
  # your code
end

Requirements

Rails: ActiveRecord 5.2+.
Database: MySQL(compatible with mysql 5 and 8), PostgreSQL, SQLite.
Ruby: MRI 2.5+.

Installation

Add this to your Gemfile:

gem 'transaction_isolation'

Then run:

bundle

It works out of the box with Ruby on Rails.

If you have a standalone ActiveRecord-based project you'll need to call:

TransactionIsolation.apply_activerecord_patch # after connecting to the database

after connecting to the database. This is because ActiveRecord loads adapters lazilly and only then they can be patched.

Features

Setting transaction isolation level: :serializable, :repeatable_read, :read_committed, :read_uncommitted
Auto-reverting to the original isolation level after the block
Database agnostic
MySQL, PostgreSQL and SQLite supported
Exception translation. All deadlocks and serialization errors are wrapped in a ActiveRecord::TransactionIsolationConflict exception
Use it in your Rails application or a standalone ActiveRecord-based project

Real world example

When implementing a table-based job queue you should ensure that only one worker process can pop a particular job from the queue.

Wrapping your code in a transaction is not enough because by default databases do not isolate transactions to the full extent, which leads to occasional phantom reads.

It is therefore necessary to manually raise the transaction isolation level.

The highest level of transaction isolation is called "serializable" and that's what we need here:

class QueuedJob < ActiveRecord::Base
  # Job status
  TODO = 1
  PROCESSING = 2
  DONE = 3

  # Returns first job from the queue or nil if the queue is empty
  def pop
    QueuedJob.isolation_level(:serializable) do
      QueuedJob.transaction do
        queued_job = find_by_status(TODO)
        if queud_job
          queued_job.update_attribute(:status, PROCESSING)
          return queued_job
        else
          return nil
        end
      end
    end
  rescue ActiveRecord::TransactionIsolationConflict => e
    logger.warn(e.message)
    retry
  end
end

Running tests

Run tests on the selected database (mysql2 by default):

# passing desired database, active record version and ruby version
docker compose run -e db=sqlite3 -e BUNDLE_GEMFILE=gemfiles/activerecord-7.0/Gemfile.sqlite3 ruby_2_7 bash -c ./docker/test-ruby.sh

# db options: mysql2, postgresql, sqlite3
# active record version options: 5.2, 6.0, 6.1, 7.0
# ruby version options: 2.5, 2.7, 3.0, 3.1

Run tests on all supported databases by ruby version:

docker compose up ruby_2_5
docker compose up ruby_2_7
docker compose up ruby_3_0
docker compose up ruby_3_1

Database configuration is hardcoded in test/db/db.rb; feel free to improve this and submit a pull request.

How intrusive is this gem?

You should be very suspicious about any gem that monkey patches your stock Ruby on Rails framework.

This gem is carefully written to not be more intrusive than it needs to be:

introduces several new methods to Mysql2Adapter, PostgreSQLAdapter, SQLite3Adapter; names are carefully taken to not collide with future changes
wraps #translate_exception method using alias_method_chain to add new translation
introduces new class ActiveRecord::TransactionIsolationConflict in the ActiveRecord module
introduces new convenience method ActiveRecord::Base.isolation_level akin to ActiveRecord::Base.transaction