Celluloid VS Async Ruby

Speaking as someone who has used Erlang longer than most, who created a pre-Elixir-like language for Erlang's BEAM VM, and who routinely listens to Carl Hewitt's rants about why Erlang actors are bad, and who tried to make a Ruby actor library after using innumerable other library-level actor solutions...

Celluloid

Recently I've started Codecrafters' Build your own Redis (which I highly recommend!). One of the first steps is to make a TCP server that can accept more than one connection. I noticed that all the provided solutions use threads or IO#Select, and none of them use a fiber-based solution with Async. In this really short article, I'll show how one could achieve the result using Async.

Since the threading problem is specific to the Puma web server, let's look at another option: Falcon. This is a new, highly concurrent Rack-compliant web server built on the async gem. It uses Ruby Fibers instead of Threads, which are cheaper to create and have much lower overhead.

Async is a composable asynchronous I/O framework for Ruby. It allows you to do things concurrently using Fibers. Since 3.0, Ruby has a fiber scheduler and Ruby core supports it. This means you can have non-blocking I/O without much effort, for example, when using Net::HTTP. If you perform a blocking operation, such as an HTTP call, inside a fiber, it will immediately yield so that another fiber can become active and do some useful work instead of blocking and waiting for the HTTP call to complete.

The Async gem is the natural successor, It's actively maintained, and allows you write synchronous code is if it wasn't non-blocking, and most libraries don't need any special support for Async (exceptions are gems with C extensions that do I/O and DB libraries with connection pooling that would otherwise be thread-based).

https://github.com/socketry/async uses coroutines and I think in general it’s been a great model with very few downsides in practice.

There's async I/O. Here's a library that leans on Ruby 3's fiber scheduler.

Hey folks, wanted to show this off and get feedback. Still early/experimental but there are quite a few concepts I'm excited about here. This project came about while writing a program in Go and loving its approach to concurrency. Being a long-time Rubyist I immediately started to think about what similar concepts might look like in Ruby.

I set out with two main design constraints:

1. Lightweight: I didn't want routines to be backed by fibers or threads. Having been involved some in the async project (https://github.com/socketry/async), I had some experience using fibers for concurrency but was curious if they could be avoided.

2. Explicitness: Routine behavior must be written to describe exactly how it is to behave. I always felt like concurrent code was hard to fully understand because of the indirection involved. On the spectrum between tedium and magical I wanted to err more on the side of tedium with Goru.

Goru routines are just blocks that are called once for every tick of the reactor. It is up to the developer to implement behavior in terms of a state machine, where on each tick the routine takes some action and then updates the state of the routine for the next tick. This fulfills both design constraints:

1. Because routines are just blocks, they weigh in at about ~345 bytes of memory overhead.

2. Routine behavior is explicit because it is written as a state machine inside the block.

Couple more features worth noting:

* Goru includes channels for buffered reading/writing (similar to channels in Go).

* Goru ships with primitives for non-blocking IO to easily build things like http servers.

Curious your thoughts!

For those who are interested here is the question.