Isolator

Detect non-atomic interactions within DB transactions.

Examples:

# HTTP calls within transaction
User.transaction do
  user = User.new(user_params)
  user.save!
  # HTTP API call
  PaymentsService.charge!(user)
end

#=> raises Isolator::HTTPError

# background job
User.transaction do
  user.update!(confirmed_at: Time.now)
  UserMailer.successful_confirmation(user).deliver_later
end

#=> raises Isolator::BackgroundJobError

Of course, Isolator can detect implicit transactions too. Consider this pretty common bad practice–enqueueing background job from after_create callback:

class Comment < ApplicationRecord
  # the good way is to use after_create_commit
  # (or not use callbacks at all)
  after_create :notify_author

  private

  def notify_author
    CommentMailer.comment_created(self).deliver_later
  end
end

Comment.create(text: "Mars is watching you!")
#=> raises Isolator::BackgroundJobError

Isolator is supposed to be used in tests and on staging.

Installation

Add this line to your application's Gemfile:

# We suppose that Isolator is used in development and test
# environments.
group :development, :test do
  gem "isolator"
end

# Or you can add it to Gemfile with `require: false`
# and require it manually in your code.
#
# This approach is useful when you want to use it in staging env too.
gem "isolator", require: false

Usage

Isolator is a plug-n-play tool, so, it begins to work right after required.

However, there are some potential caveats:

1. Isolator tries to detect the environment automatically and includes only necessary adapters. Thus the order of loading gems matters: make sure that isolator is required in the end (NOTE: in Rails, all adapters loaded after application initialization).

2. Isolator does not distinguish framework-level adapters. For example, :active_job spy doesn't take into account which AJ adapter you use; if you are using a safe one (e.g. Que) just disable the :active_job adapter to avoid false negatives. You can do this by adding an initializer:

   require "active_job/base"
   Isolator.adapters.active_job.disable!
   

3. Isolator tries to detect the test environment and slightly change its behavior: first, it respect transactional tests; secondly, error raising is turned on by default (see below).

4. Experimental multiple databases has been added in v0.7.0. Please, let us know if you encounter any issues.

Configuration

Isolator.configure do |config|
  # Specify a custom logger to log offenses
  config.logger = nil

  # Raise exception on offense
  config.raise_exceptions = false # true in test env

  # Send notifications to uniform_notifier
  config.send_notifications = false

  # Customize backtrace filtering (provide a callable)
  # By default, just takes the top-5 lines
  config.backtrace_filter = ->(backtrace) { backtrace.take(5) }

  # Define a custom ignorer class (must implement .prepare)
  # uses a row number based list from the .isolator_todo.yml file
  config.ignorer = Isolator::Ignorer

  # Turn on/off raising exceptions for simultaneous transactions to different databases
  config.disallow_per_thread_concurrent_transactions = false

  # Limit the amount of allowed nested transactions
  config.max_subtransactions_depth = 5
end

Isolator relies on [uniform_notifier][] to send custom notifications.

NOTE: uniform_notifier should be installed separately (i.e., added to Gemfile).

Callbacks

Isolator different callbacks so you can inject your own logic or build custom extensions.

# This callback is called when Isolator enters the "danger zone"—a within-transaction context
Isolator.before_isolate do
  puts "Entering a database transaction. Be careful!"
end

# This callback is called when Isolator leaves the "danger zone"
Isolator.after_isolate do
  puts "Leaving a database transaction. Everything is fine. Feel free to call slow HTTP APIs"
end

# This callback is called every time a new transaction is open (root or nested)
Isolator.on_transaction_begin do |event|
  puts "New transaction from #{event[:connection_id]}. " \
       "Current depth: #{event[:depth]}"
end

# This callback is called every time a transaction is completed
Isolator.on_transaction_end do |event|
  puts "Transaction completed from #{event[:connection_id]}. " \
       "Current depth: #{event[:depth]}"
end

Transactional tests support

• Rails' baked-in use_transactional_tests
• database_cleaner gem. Make sure that you require isolator after database_cleaner.

Supported ORMs

• ActiveRecord >= 6.0 (see older versions of Isolator for previous versions)
• ROM::SQL (only if Active Support instrumentation extension is loaded)

Adapters

Isolator has a bunch of built-in adapters:

• :http – built on top of [Sniffer][]
• :active_job
• :sidekiq
• :resque
• :resque_scheduler
• :sucker_punch
• :mailer
• :webmock – track mocked HTTP requests (unseen by Sniffer) in tests
• :action_cable

You can dynamically enable/disable adapters, e.g.:

# Disable HTTP adapter == do not spy on HTTP requests
Isolator.adapters.http.disable!

# Enable back

Isolator.adapters.http.enable!

For active_job, be sure to first require "active_job/base".

Fix Offenses

For the actions that should be executed only after successful transaction commit (which is mostly always so), you can try to use the after_commit callback from [after_commit_everywhere] gem (or use native AR callback in models if it's applicable).

Ignore Offenses

Since Isolator adapter is just a wrapper over original code, it may lead to false positives when there is another library patching the same behaviour. In that case you might want to ignore some offenses.

Consider an example: we use Sidekiq along with sidekiq-postpone–gem that patches Sidekiq::Client#raw_push and allows you to postpone jobs enqueueing (e.g. to enqueue everything when a transaction is commited–we don't want to raise exceptions in such situation).

To ignore offenses when sidekiq-postpone is active, you can add an ignore proc:

Isolator.adapters.sidekiq.ignore_if { Thread.current[:sidekiq_postpone] }

You can add as many ignores as you want, the offense is registered iff all of them return false.

Using with sidekiq/testing

If you require sidekiq/testing in your tests after isolator is required then it will blow away isolator's hooks, so you need to require isolator after requiring sidekiq/testing.

If you're using Rails and want to use isolator in development and staging, then here is a way to do this.

# Gemfile
gem "isolator", require: false # so it delays loading till after sidekiq/testing

# config/initializers/isolator.rb
require "sidekiq/testing" if Rails.env.test?

unless Rails.env.production? # so we get it in staging too
  require "isolator"
  Isolator.configure do |config|
    config.send_notifications = true # ...
  end
end

Using with legacy Rails codebases

If you already have a huge Rails project it can be tricky to turn Isolator on because you'll immediately get a lot of failed specs. If you want to fix detected issues one by one, you can list all of them in the special files .isolator_todo.yml and .isolator_ignore.yml in the following way:

sidekiq:
  - app/models/user.rb:20
  - app/models/sales/**/*.rb

You can ignore the same files in multiple adapters using YML aliases in the following way:

http_common: &http_common
  - app/models/user.rb:20

http: *http_common
webmock: *http_common

All the exceptions raised in the listed lines will be ignored.

The .isolator_todo.yml file is intended to point to the code that should be fixed later, and .isolator_ignore.yml points to the code that for some reasons is not expected to be fixed. (See https://github.com/palkan/isolator/issues/40)

Using with legacy Ruby codebases

If you are not using Rails, you'll have to load ignores from file manually, using Isolator::Ignorer.prepare(path:), for instance Isolator::Ignorer.prepare(path: "./config/.isolator_todo.yml")

Custom Adapters

An adapter is just a combination of a method wrapper and lifecycle hooks.

Suppose that you have a class Danger with a method #explode, which is not safe to be run within a DB transaction. Then you can isolate it (i.e., register with Isolator):

# The first argument is a unique adapter id,
# you can use it later to enable/disable the adapter
#
# The second argument is the method owner and
# the third one is a method name.
Isolator.isolate :danger, Danger, :explode, options

# NOTE: if you want to isolate a class method, use singleton_class instead
Isolator.isolate :danger, Danger.singleton_class, :explode, options

Possible options are:

• exception_class – an exception class to raise in case of offense
• exception_message – custom exception message (could be specified without a class)
• details_message – a block to generate additional exception message information:

Isolator.isolate :active_job,
  target: ActiveJob::Base,
  method_name: :enqueue,
  exception_class: Isolator::BackgroundJobError,
  details_message: ->(obj) {
    "#{obj.class.name}(#{obj.arguments})"
  }

Isolator.isolate :promoter,
  target: UserPromoter,
  method_name: :call,
  details_message: ->