Ruby and Elixir: Polyglottin' FTW

Ruby and Elixir: Polyglottin' FTW(neo.com)

55 points by linc01n 11 years ago | 23 comments

rubiquity 11 years ago |

I've done a setup similar to this in the past and really enjoy it. If you ever find yourself writing logic in your Resque/Sidekiq jobs that has to do with manually managing a data structure in Redis with regard to your jobs, your app is screaming for a language with better tools for modeling complex workflows.

A lot of people are hitting on the performance improvements (25 Sidekiq threads/workers versus 10k-100k+ Elixir/Erlang processes) but when I find myself doing this it's actually because my background process flows are really complex. The drastic increase in concurrency is just the cherry on top.

Background work is asynchronous by nature and often times you have dependencies between jobs, groups/batches jobs and also need robust error handling/recovery. This is where Elixir/Erlang really shines and using Actors & OTP to model these complex flows is the huge win in my experience. Sidekiq Pro offers support for batches which covers some of this, but ultimately Elixir/Erlang processes are just a huge improvement in modeling such systems.

A concrete example is sending off 10 jobs. When all 10 are done, move on to the next step. Each of these 10 jobs can also send data to the waiting step which it can then use as arguments. If any job should fail N times, cancel and rollback all of the jobs and try again. After the next step, split those 10 jobs into two groups of 5 and do those 5 sequentially. So on and so forth. Try doing this in Ruby. You'll likely have to manage your own data structures in Redis. In Elixir you just use GenServers and sometimes ets tables. It's all handled in the language itself, rather than external data stores.

Sidekiq is probably suitable for 90% of Ruby apps doing basic fire and forget and batching with Sidekiq Pro. But if your domain has seriously complicated background processing, use a tool more suited for that.

themartorana 11 years ago |

Best thing I ever learned - probably 15 years ago at this point - is just use the right tool for the job.

We're a mobile game company, and we use C, C#, Objective C, Java, Python, Go, Erlang, JavaScript, Lua, and more where they make sense.

I love language talk and debates (and hate the language hate) here on HN. But if you're only using a single language in production, you're likely not that big, or more likely have a lot to gain by reworking certain bits with other technologies where they make sense.

pjmlp 11 years ago | |

I never got those "I am language X developer".

It is like "I only do hammers and nails. For drilling ask the guy over there."

dagw 11 years ago | | |

On the other hand if I want some intricate decorative wood carving, I'm going to seek out a guy who specializes in intricate decorative wood carving, not a guy who says "well I've never done any carving, but I've sawed, drilled and hammered lot of wood in my day and carving is basically the same thing."

ankurpatel 11 years ago |

Why are you adding an extra level of abstraction in between Rails and Sidekiq? Why not just have Rails push to redis queue and elixir process finishes the job of sending email or doing what it has to do. Why have Sidekiq run at all? It seems adding multiple layers as such will make it harder to understand, debug and maintain in future.

EDIT: Also your example of polling on Redis queue using elixir is very inefficient and makes your article not so credible in my eyes or eyes of other good software architects. Redis is meant for pub sub and not polling.

wcummings 11 years ago | |

Redis has blocking list primitives which you can use to build an efficient queue (push and pop in O(1)), if you look at the code you'll see he's using brpop

ankurpatel 11 years ago | | |

I agree but there is no need to poll. Polling is a poor mans solution in a world where you can have persistent connection and get push notification of when a new item is added to the queue and there is no need to throttle for 10 seconds before polling again.

sanderjd 11 years ago | |

I think it makes some sense to use the Sidekiq client, because it does a lot of the tedious work of getting data into redis in a well-defined and battle-tested fashion. But certainly it doesn't make sense to run the Sidekiq daemon in this configuration.

mmcclure 11 years ago |

We've done similar things at work (using a Redis queue and a Go co-process), but I actually really like the idea of using a fairly common and well-liked delayed job interface in Rails and swapping out the worker underneath. Cool idea.

One minor suggestion for the author (post doesn't have a comment section). `:timer.sleep(10)` is explained as "sleep for 10 seconds", but sleep takes millis as an argument (http://www.erlang.org/doc/man/timer.html#sleep-1).

Dirlewanger 11 years ago |

Excuse my poor understanding of Ruby's GIL, but is the need for an Elixir worker because Ruby is by default single-threaded, and therefore actual default concurrent Sidekiq workers are not possible? Or this a problem scaled up to the degree that we're talking ~thousands of jobs per second that Ruby's limitations start to show?

phamilton 11 years ago | |

Since Ruby's GIL means you can't do true parallel processing, it's common to kick off multiple worker processes and have them each take one job at a time.

If the job involves an expensive API call it will tie up a worker for the duration of the call.

So it's not so much ~thousands of jobs per second as it is a limitation of only 10 ruby processes per 1GB of memory. (A ruby process with the Rails env loaded is about 100MB.)

Elixir gives you the ability to make a few orders of magnitude more calls in a tiny amount of memory. (100,000 Elixir "workers" can be had in less than 1GB).

mperham 11 years ago | |

Sidekiq runs 25 worker threads by default. MRI will context switch upon I/O so typical server-side applications will get plenty of concurrency, even with the GIL, because they spend lots of time talking to the DB, memcached, redis, 3rd party APIs, etc.

tl;dr - Sidekiq can do a LOT of work, even on MRI with the GIL.

lshemesh 11 years ago |

I recently did a Sidekiq background job in ruby passing messages to a python script using zeromq. It was a blast.

spot 11 years ago |

Polyglot web IDE that allows you to mix Python, Javascript, R, Ruby, Scala, Java: http://BeakerNotebook.com work with them all in the same page. As a corollary, you can work with Python2 and Python3 in the same notebook, and share data.

I totally agree that each language has its strength and each facet of your problem may best be solved in a different language.

dozzie 11 years ago |

Yeah, be polyglot by speaking American English and British English! I mean, Ruby and Elixir!

http://satwcomic.com/multi-language/moby-is-a-dick

pselbert 11 years ago | |

That's like claiming any two C syntax inspired languages are the same because they share curly braces, semicolons and naming conventions.

dagw 11 years ago | |

While Ruby and Elixir have quite similar syntax, their underlying run time models are fundamentally very different.

pmontra 11 years ago | | |

Yes very different model, and you realize that the syntax is not so similar once you really start working with Elixir: you start using constructs that don't belong to procedural languages at all.

Similarities don't go much further than def do end (but there are too many do and too many different def in Elixir) and the sensible naming of some Library.functions() to match classes and methods in Ruby's standard library. But hey, making a good first impression is extremely important and Elixir has been engineered well in that regard. It's not one of those I'll-never-ever-be-able-to-read-it languages.