Skip to content

Latest commit

 

History

History
146 lines (104 loc) · 5.17 KB

README.md

File metadata and controls

146 lines (104 loc) · 5.17 KB
Raw

Examples for Ruby 3 Concurrency

In this repo you have some example code to test the new concurrency features in Ruby 3: The Fiber Scheduler and Ractors.

There's code to cover 3 characteristic scenarios

  • A combination of IO and CPU tasks
  • A bunch of CPU intensive tasks
  • A sequence of Download-Process-Upload steps, akin to accessing a paginated API

You can find a live run through these examples in this presentation:

Code organization

The examples themselves run the tasks and write timing information to STDOUT.

The actual Ruby logic for the tasks is the same for all examples, and lives in the logic.rb file, that's required by all examples.

The code that tracks the timings lives in trace.rb and, again, it's required by the scripts.

In order to make results easier to visualize, there's also a script that uses tk_component to diplay graphical bars with the timings for each task (tk_component is a Ruby Gem that provides a component approach to build Ruby/TK applications).

Running the code

In order to run any of the tests, use the run_test.rb script.

For instance, to run the process_plain.rb example, you would run:

bundle exec ./run_test.rb process_async.rb

That should produce results like these:

Example output

In case the script needs some arguments, enclose both the script name and the arguments in quotes, like here:

bundle exec ./run_test.rb "paging_threads.rb 6"

Scenario 1: Download, process and upload data from/to several sources

The script will download a big text (War and Peace) from Project Guttemberg and some words from an HTTP API.

Once it has both, it will do a global replace of the strings across the whole text.

When done, it will save the results in a local file and upload part of it to an HTTP API.

Takeaway for Scenario 1

Both Threads and Fiber Scheduler provide concurrency.

The Threads example provides more concurrency because:

  • Local filesystem IO operations are still blocking under the Fiber Scheduler.
  • DNS lookups are blocking, too.

The Async example could be made more concurrent by inverting the order of the the items in the last step: uploading and then saving instead of saving and then uploading, as the uploading will not block while saving will.

Scenario 2: String substitution in parallel

The script will download again "War and Peace" (but the time for that won't be included in the results) and then perform a string substitution on it for 5 times.

  • cpu_plain.rb No concurrency. Every step is done sequentially.
  • cpu_threads.rb. A Thread is created for each of the 5 tasks.
  • cpu_async.rb A Fiber Scheduler is used to run the 5 tasks asynchronously.
  • cpu_ractors.rb A Ractor (new in Ruby 3) is created for each of the tasks.

Takeaway for Scenario 2

As expected, Async doesn't give any concurrency here, given that there's no IO performed in the code.

Threads appear to be concurrent, but given they're running just Ruby code, the GIL (Global Interpreter Lock) in Ruby prevents more than one from being run at the same time, so threads just still the lock from one another and the total time is the same.

However, Ractors don't suffer from the GIL problem, so they can truly run concurrently and take advantage of multicore processors.

Scenario 3: Paging through an API

The script will read entries from a an HTTP API in batches (pages). For each page, it will do some string processing. When done, it will upload the results to another HTTP API endpoint.

If you pass an additional argument to the script, it will use it as the page size. If absent, page size defaults to 2 items per page.

  • paging_plain.rb No concurrency. Everything done sequentially.
  • paging_threads.rb The script uses 3 Threads and 2 Queues. The first Thread reads pages from the API and puts the results in one Queue. The second Thread gets data from the Queue, processes it and puts the results in the second Queue. The third Thread reads from the second Queue and uploads the data.
  • paging_async.rb Similar to the Threads script but using the Fiber Scheduler instead.

Takeaway for Scenario 3

The results are quite similar to Scenario 1. Again, Threads are a little more concurrent, but just sightly. Fiber Scheduler works fantastically well and provides very good concurrency without having to worry about thread safety of the code.

We couldn't run a Ractor based version because there is a problem with the URI class using class variables that makes it break when run under different Ractors.