Tag: Rails

RSpec story about disappearing classes

ActiveSupport#descendants can be slow. In a bigger system with layers of descendants, finding all of them can be time-consuming:

puts Benchmark.measure do
  100.times { Dispatchers::Base.descendants }
end

# 5.235370   0.015754   5.251124 (  5.251069)

In the code I've been working on, it meant that a single lookup was taking around 50ms. That is a lot, especially if used extensively.

To mitigate this, I've implemented a simple caching layer on top of the lookup that would make things fast:

module Mixins
  module CachedDescendants
    extend ActiveSupport::Concern

    cattr_accessor :descendants_map

    self.descendants_map = Concurrent::Hash.new

    class << self
      # Clears the descendants map cache - can be hooked to Rails reloader
      def reload!
        descendants_map.clear
      end
    end

    included do
      class << self
        # @return [Array<Class>] array with descendants classes
        def cached_descendants
          ::Mixins::CachedDescendants.descendants_map[self] ||= descendants
        end
      end
    end
  end
end

When included and used, it would give great results:

puts Benchmark.measure do
  100.times { Dispatchers::Base.cached_descendants }
end

# 0.000023   0.000001   0.000024 (  0.000024)

99,99956% faster!

Such code, like any other, deserves to be tested. I wrote some specs for it, including a relatively simple one:

  context 'when there are two independent bases' do
    let(:base1) do
      Class.new do
        include ::Mixins::CachedDescendants
      end
    end

    let(:base2) do
      Class.new do
        include ::Mixins::CachedDescendants
      end
    end

    before do
      Array.new(5) { Class.new(base1) }
      Array.new(5) { Class.new(base2) }
    end

    it 'expect for them not to interact' do
      expect(base1.cached_descendants.size).to eq(5)
      expect(base2.cached_descendants.size).to eq(5)
      expect(base1.cached_descendants & base2.cached_descendants).to be_empty
    end
  end

It would just ensure that the way we cache does not create collisions for independent descendants trees.

But once in a while, this code would randomly fail:

  1) Mixins::CachedDescendants when there are two independent bases expect for them not to interact
     Failure/Error: expect(base1.cached_descendants.size).to eq(4)
       expected: 5
            got: 4
       (compared using ==)

How can I create five classes and suddenly have only 4? I initially thought something was wrong with the descendants lookup for anonymous classes. However, this functionality is heavily used by many, including me, and it never created any problems. On top of that, why would it fail only once in a while?

When something fails randomly, it usually means that there's an external factor to it. One that operates under the hood. It wasn't different in this case.

After some investigation, I was able to reproduce it:

GC.disable

puts "Total classes before: #{ObjectSpace.count_objects[:T_CLASS]}" 
puts "String subclasses count before: #{String.subclasses.count}" 

100.times { Class.new(String) }

puts "Total classes after defining: #{ObjectSpace.count_objects[:T_CLASS]}" 
puts "String subclasses count after defining: #{String.subclasses.count}" 

GC.enable
GC.start

puts "Total classes after GC: #{ObjectSpace.count_objects[:T_CLASS]}" 
puts "String subclasses count after GC: #{String.subclasses.count}" 

# Total classes after defining: 1324
# String subclasses count after defining: 102
# Running GC...
# Total classes after GC: 1124
# String subclasses count after GC: 2

Boom! Anonymous classes are being garbage collected! Classes that are not referenced anywhere are subject to garbage collection like other objects, and this code was not memoizing them:

before do
  Array.new(5) { Class.new(base1) }
  Array.new(5) { Class.new(base2) }
end

Hence, the spec would fail if GC kicked in exactly between the classes definitions and the spec execution. This is why it would only fail once in a while.

Fixing such an issue required only minimal changes to the spec:

    let(:descendants) do
      # This needs to be memorized, otherwise Ruby GC may remove those in between assertions
      [
        Array.new(5) { Class.new(base1) },
        Array.new(5) { Class.new(base2) }
      ]
    end

    before { descendants }

That way, the anonymous classes would be referenced throughout the lifetime of this spec.

Summary (TL;DR)

Anonymous classes and modules are a subject of garbage collection like any other object. Unless you reference them, they may be gone before wanting to use them via #descendants or similar lookups. Always reference them in some way or face the unexpected.

Afterword

It was pointed to me by Jean Boussier, that there's an ActiveSupport::DescendantsTracker that can also be used to improve the lookup performance and that Ruby 3.1 has a #subclasses that also is much faster than iterating over ObjectSpace. While their performance is several times faster than the "old" lookup, it is still a magnitude slower than caching the descendants in dev but has similar performance in production.

0.002020 vs. 0.000023 in dev and 0.000039 vs. 0.000042 in production.

The best code is no code, so now I can deprecate my code. The strong class references, however, are still valid and worth keeping in mind.


Cover photo by Alper Orus on Attribution-NonCommercial-ShareAlike 2.0 Generic (CC BY-NC-SA 2.0). Image has been cropped.

Karafka framework 2.0 announcement

I'm thrilled to announce the new and shiny Karafka 2.0. It is an effect of my work of almost four years.

For those who wonder what Karafka is, Karafka is a Ruby and Rails multi-threaded efficient Kafka processing framework.

Karafka 2.0 is a major rewrite that brings many new things to the table but removes specific concepts that were not as good as I initially thought when I created them.

In this announcement article, I will describe the most noticeable features and improvements that got into this release. If you are interested in a more comprehensive list, you can find it here.

Note: Upgrade notes for migration from Karafka 1.4 to Karafka 2.0 can be found here.

Getting started

If you are new to Karafka and want to play around, follow this demo or visit the Getting Started page:

Noticeable features and improvements

This section includes all the noticeable changes you may be interested in if you already work with Karafka or if you want to understand the journey.

Multi-threading

Most of the engineering work around this release was about performance, scalability, and improvement of the overall engineering experience.

Multi-threading is probably the most significant change in Karafka since it was created. Up until now, Karafka was single-threaded. That means that any concurrency would have to be implemented by the end user. The reason is dead simple: concurrency is hard. Synchronization is hard. Warranties are hard. I do feel (and can back it up with integration specs) that I tackled it pretty well.

Karafka 2.0 uses native Ruby threads to achieve concurrent processing in three scenarios:

  • for concurrent processing of messages from different topics partitions.
  • for concurrent processing of messages from a single partition when using the Virtual Partitions feature.
  • to handle consumer groups management (each consumer group defined will be managed by a separate thread)

This can bring big advantages when any IO is involved.

When you start consuming messages, Karafka will fetch and distribute data to utilize multiple threads while preserving all the Kafka ordering warranties.

Years ago, I developed a lot of in-app async code to bypass Karafka limitations, and it makes me extremely happy to be able to retire all of it.

But wait, there's more...

Virtual Partitions

Virtual Partitions allow you to parallelize the processing of data from a single partition. This can drastically increase throughput when IO operations are involved.

While the default scaling strategy for Kafka consumers is to increase partitions count and number of consumers, in many cases, this will not provide you with desired effects. In the end, you cannot go with this strategy beyond assigning one process per single topic partition. That means that without a way to parallelize the work further, IO may become your biggest bottleneck.

Virtual Partitions solve this problem by providing you with the means to further parallelize work by creating "virtual" partitions that will operate independently but will obey all the Kafka warranties as a collective processing unit.

topic :orders_states do
  consumer OrdersStatesConsumer
  # Distribute work to virtual partitions based on the user id
  virtual_partitions(
    partitioner: ->(message) { message.payload[:user_id] }
  )
end

With Virtual Partitions, you benefit from both worlds: scaling with Kafka partitions and scaling with Ruby threads.

*This example illustrates the throughput difference for IO intense work, where the IO cost of processing a single message is 1ms.

Active Job support

Active Job is a standard interface for interacting with job runners in Ruby on Rails. Active Job can be configured to work with Karafka.

While Kafka is not a message queue, I still decided to create an Active Job adapter for it. Why? Because ordered jobs are something, I always wished for Ruby on Rails to have. On top of that, you may already have Kafka and only a few jobs to run. If so, why not use it and save yourself a hustle of yet another tool to maintain?

class Application < Rails::Application
  # ...
  config.active_job.queue_adapter = :karafka
end

End-to-end integration test suite

Karafka comes with a home-brew framework for running end-to-end integration specs against Kafka. I did my best to describe every possible case I could have imagined to ensure that the framework behaves as expected under any circumstances.

It is also a great place to learn about how Karafka behaves in particular scenarios.

Lower supply chain fingerprint

The number of external dependencies Karafka relies on has been reduced significantly. It was done to ensure that Karafka can be integrated into and upgraded in applications without causing dependency conflicts.

Upgraded documentation

Karafka and WaterDrop have been fully updated with several new sections describing use-cases, edge-cases and providing help and suggestions for both simple and advanced usage.

Out-of-the-box DataDog and StatsD instrumentation

Using DataDog or StatsD? In just a few lines you can enable full instrumentation of both consumption and production of messages:

# initialize the listener with statsd client
dd_listener = ::Karafka::Instrumentation::Vendors::Datadog::Listener.new do |config|
  config.client = Datadog::Statsd.new('localhost', 8125)
  # Publish host as a tag alongside the rest of tags
  config.default_tags = ["host:#{Socket.gethostname}"]
end

# Subscribe with your listener to Karafka and you should be ready to go!
Karafka.monitor.subscribe(dd_listener)

License change

Karafka 2.0 is dual licensed under LGPL and a Commercial License. Depending on your use-case, you should be good with one or the other.

Note: Before the license change, I did obtain the consent of all the contributors for a re-license. I want to say thank you to each of you for allowing me to do so.

Seamless Ruby on Rails integration

Karafka always had good integration with Ruby on Rails. With the 2.0 release, however, this integration is elevated to another level: no more files editing, no more configuration copying. Everything works out of the box.

Karafka Pro

This release is the first release that includes a Pro subscription.

Building a complex and reliable open-source is neither easy nor fast. Many companies rely on Karafka, and following Mikes Perham advice I have decided to introduce the Pro subscription to be able to support the further development of the ecosystem.

Karafka Pro has many valuable, well-documented, well-tested functionalities that can significantly improve your day-to-day operations with Kafka in Ruby. It also introduces commercial support, as due to a sheer number of questions and requests, I do need to have a way to prioritize those.

SInce it's not only me, 10% of the income will be further distributed down the supply chain pipeline to support the work of people I rely on.

Help me build and maintain a high-quality Kafka ecosystem for Ruby and Ruby on Rails.

Buy Karafka Pro.

Karafka 1.4 maintenance

With this release an official EOL policies have been introduced. Karafka 1.4 will be supported until the end of February 2023.

Karafka 2.0 has a lower dependency fingerprint and is in everything 1.4 was not. I strongly encourage you to upgrade.

What's ahead

Many things. This release is just the beginning. I am already working on a 2.1 release that will include several great additions, including:

  • Management Web-UI similar to the one Resque and Sidekiq have
  • Producer transactions
  • At Rest encryption
  • CurrentAttributes support for ActiveJob
  • Seamless Dead-Letter Queue integration

Upgrade notes

Upgrade notes for migration from Karafka 1.4 to Karafka 2.0 can be found here.

References


Stay tuned and don't forget to join our Slack channel.

Copyright © 2024 Closer to Code

Theme by Anders NorenUp ↑