[inferiorhuman]

I cut my teeth on Ruby and Rails, I've screwed around with Phoenix, and I'm currently infatuated with Rust. Earlier on I thought that, yeah, Rust is a great replacement for C and not something want to write a web app in. So, sure, I've got a bunch of little doodads written in Rust (mostly CLI tools). Now? The fluff I've written in Elixir is getting ported to Rust.

The Rust tooling and community are just that good. I'd rather have to work on some extra verbosity and maybe even maintain my own helper library in Rust than deal with with Elixir/Phoenix at this point. The number one reason is that Elixir deployments are nothing short of a nightmare. I've found the community not particularly helpful or knowledgeable about how things work (and given how many moving pieces are in an Elixir deployment, and how brittle Elixir deployments are, this is a terrible thing). And you're definitely not going to write CLI tools in Elixir (it straight doesn't work with escript in the first place).

Compiling down into a single "static" binary like Rust (and Go) do by default is a huge win for things like this. The quick startup time of a Rust or Go app compared to Ruby or anything JVM based is great for rapid iteration and testing. The tooling around Rust is, in general, fantastic. Being that much of a pleasure to use really makes up for how inappropriate it is to use Rust for most high level web app things.

Phoenix had a lot of promise, but these days I'd much rather go back to Rails or Django (with an eye towards how to migrate to a compiled language) if I needed to churn something out quickly.

[AlchemistCamp]

This is quite a coincidence as deployment was exactly what I covered in my last YT video. I hear the complaint about deployment a lot, but it really doesn't make sense to me.

If you run mix in production, then it's just like running Rails or Node. You don't get certain features like hot upgrades in production while keeping what's in memory on the server, but you don't get that from Rails/Node/PHP/Python either.

If you want those BEAM releases, then you have to use the Elixir (or Erlang) tools for them. I also encountered a great deal of frustration when working from a different OS, but no more than I did when building Rails apps several years ago.

Now I generally, build with Gitlab CI/CD and deploy from that Gitlab container to my server with Distillery and Edeliver. It lets me use the standard CI/CD tools and have stateful upgrades. I am running Ubuntu on the server, though:

https://www.youtube.com/watch?v=-mm44ADU3kc

[Fire-Dragon-DoL]

Sorry, I don't get your point for deployment. There are complaints you can make fot everything, but deploying an elixir application boils down to compile, upload, start.

Sure, 2 days of work the first time, which is a negligible amount of time for any real world project (otherwise you are on the wrong language). This if you want a good pipeline. Otherwise it's a couple of hours, but even with rails i always invested a couple of days to have a good pipeline.

Doesn't mean rust does not do it better, but it's really not the point for choosing a language over another. In a big project you have a dedicated person. In a medium project, the investment is very small. In a small project, you are better served by prototyping tools (rails)

[uxcolumbo]

I'm currently looking into learning Elixir, but your comment sprinkled some doubts on that.

If deployment was easier in the Elixir world, would you have stuck with it?

[robotmay]

Honestly Elixir has been one of the most enjoyable learning experiences I've had in years. Deployment is fine, but there are a few different things to consider (because you can do some really cool stuff using BEAM). But you can just chuck it on Heroku if you want something easy.

I also like Rust a lot, but I find Elixir pretty much frictionless to work in. Rust is basically you vs the compiler for quite a while. I've been using Rust for IoT work but I vastly prefer Elixir for web. And since I've been enjoying it so much I'm gonna start looking at Nerves (https://github.com/nerves-project/nerves) for IoT work as well.

[inferiorhuman]

If deployment was easier in the Elixir world, would you have stuck with it?

Perhaps, but Elixir was a very niche product for me. I'd use it for web apps and that's about it. Even then the ecosystem was fairly immature, and I'd only really want to use it in a setting that played to the strengths of Erlang. One of the big problems with the deployments being a gigantic beast is that there are maybe one or two people who understand the tooling around it. Great when it works, really lousy when it doesn't.

With the other languages being discussed (e.g. Rust, Go, Python, Ruby, Clojure, Kotlin) I've found far broader uses. I'm not going to write TUI/GUI apps nor would I write shared libraries with Elixir. Unless escript support was implemented you simply can't write things that depend on a shebang.

There's no harm in learning it, but the key is in being able to go beyond the hype and evaluate when it's worth using the knowledge you've just acquired.

[Gazler]

Not sure when you were using Elixir, but releases are a good option for deployment now. This is supported with the distillery library (https://hexdocs.pm/distillery/home.html).

The master version of Elixir also has support for releases https://hexdocs.pm/mix/master/Mix.Tasks.Release.html

[inferiorhuman]

Last I checked distillery was Linux only (e.g. does not work on BSD), which is a deal breaker for me. But I wouldn't consider using distillery (and thus Elixir/Phoenix) on any platform given the haphazard nature of the support for distillery. There's exactly one person who knows anything about it and he supports it when he doesn't have anything better to do. Sure, in a smaller shop you could dig into distillery and try to fix things up. But when shit hits the fan, do you want to discover the hidden dependencies (e.g. bash vs POSIX sh)? I wouldn't want to bet on a single point of failure like that.

And the big problem for Elixir is that you don't see that level of mess or siloed knowledge with other languages. Say what you will about something like Capistrano, but it's an order of magnitude easier to dig into and support because it's not mashing together a variety of languages (erlang, bash, elixir). Some languages like anything JVM based (that compile down to JARs), Rust, and Go compile down into single artifacts quite easily. Others like Ruby and Python can be moved around with standard system tools easily enough.

[josevalim]

If those are your concerns, then the fact we are adding releases to Elixir core in the upcoming version should solve most of them.

The single point of failure is gone. The code is now maintained by the Elixir Core Team and it should garner more general attention from the community. The amount and complexity of scripts have been reduced drastically and they are statically verified to run on `sh` (no bash). As with everything else in Elixir, we do our best to keep the Erlang <-> Elixir interface pretty clean. You should expect the same level of quality and polish as the remaining of the Elixir tooling.

I am not sure about this point though:

> Others like Ruby and Python can be moved around with standard system tools easily enough.

If you are not using releases, and simply Mix, is Elixir any harder to move around then Ruby and Python? From my understanding, using Mix to run Elixir in prod is a similar experience that both Ruby and Python would provide.

[inferiorhuman]

If those are your concerns, then the fact we are adding releases to Elixir core in the upcoming version should solve most of them.

I've articulated my experience in my previous reply to you, but I'll pose this response: that it's taken so long for Elixir to come up with a manageable deployment story makes it seem like deployments are an afterthought with Elixir. If things are better now that's great, but as an ops guy by trade I think that deployments are one of the most important user stories out there.

I suspect one of the reasons people are gravitating towards Rust for things it's not particularly well suited for (like web apps with a ton of business logic) is that the Rust team focuses on important user stories like:

- tooling (human readable error messages, dependency management, portability)

- deployments

- IDE user experience

These are things that Elixir and Phoenix aren't focused on. For example I still get the occasional email from github about the poor souls trying to use distillery on FreeBSD so I don't buy that things have improved that much.

[pmarreck]

The Elixir tooling is fantastic, not sure where you pulled that one out from. IDE's are unnecessary cruft to the point that I'd call it a language-design smell if you really felt compelled to use one; VSCode is great. Deployments are trivial if you use Gigalixir (which has a free tier) and will be easier in the general case going forward.

[josevalim]

I am sorry to hear that you had bad experiences while deploying Elixir. To avoid having others running into the same bad experiences, I would like to point out to some resources and recent improvements in the area for those interested in deploying Elixir.

There are mostly two ways to deploy Elixir. One is the provisioned approach: you install Erlang/Elixir (or you assume it is installed in the machine), then you fetch your project source code, fetch deps, compile them and run your project (in the same way you would run it locally).

The other approach is via releases. A release is a way to package the Erlang VM, Elixir and all of your application code into something you can dropped into production. The downside is that you need to match the target OS and architecture but there are many documented approaches to do so. Today the main tool for performing releases is Distillery and it has great documentation covering AWS, Docker, etc: https://hexdocs.pm/distillery/home.html

In Elixir v1.9, coming out this July, we are incorporating releases as part of the language. You can read more about it in our documentation for master: https://hexdocs.pm/mix/master/Mix.Tasks.Release.html

If you are using a PaaS such as Heroku, it is likely there is already tooling that will take care of it for you, such as the Heroku Elixir buildpack: https://github.com/HashNuke/heroku-buildpack-elixir. There is also Gigalixir, a PaaS specific to Erlang/Elixir: https://gigalixir.com/

Finally, if you want a more structured resource, there is also the Adopting Elixir book (disclaimer! I am one of the authors): https://pragprog.com/book/tvmelixir/adopting-elixir - One third of the book is about production concerns, covering deployment, performance, metrics, monitoring, etc.

If none of this works, you can always reach out to the community in elixirforum.com. I assure you there are more than 2 developers that know about deployment :) and if you still feel like none of the replies about deployment there have been satisfactory, feel free to add me to the conversation there.

> And you're definitely not going to write CLI tools in Elixir (it straight doesn't work with escript in the first place).

I do not understand this comment as Elixir works with escripts just fine. You can build an escript for your project by simply calling `mix escript.build`. This has been available since v1.0 (https://github.com/elixir-lang/elixir/blob/v1.0.0/lib/mix/li...). You can even install escripts directly from Hex (our package manager) since Elixir v1.4 (2.5 years old).

[inferiorhuman]

FWIW the last time I tried to use Elixir with escript was with 1.3. The parts that make Elixir interesting for web apps aren't necessarily things you'd be leveraging in scripts and so, unfortunately, instead of waiting around for Elixir to do the things I was interested in I've moved on. There just aren't any killer features in Elixir that would revisit it. If I need lightweight concurrency primitives there's go. If I want immutability by default there's rust. If I want to hack up some scripts that leverage my web app there's Ruby and Python.

I'm deploying to BSD, which distillery does not support. Perhaps there are other people who can grok deployments on Elixir, but none of them have seen fit to (or know how to) resolve these issues. I've seen scattered reports of success in the issues, bitwalker's attitude and general willful lack of engagement was enough to get me to move on from Elixir and not look back. By the time I gave up on Elixir, deployments on BSD had been known broken for months.

I'm used to hitting speed bumps by not deploying to Linux, but Elixir was by far the bumpiest road for the least benefit. For comparison the rust community has been extremely responsive to BSD users of all stripes (incl. dfbsd).

[josevalim]

> FWIW the last time I tried to use Elixir with escript was with 1.3.

Interesting, escripts should have worked by then. I assume this was quite some time ago, so you don't have it around anymore, but if you do recall what went wrong I would love to take a look at it. But just to put things in perspective, v1.3 is almost three years old and things have generally evolved since.

> There just aren't any killer features in Elixir that would revisit it.

Well, sometimes the killer feature is having all of that in the same package: immutability, concurrent primitives, scripts, etc.

There are also killer features (although I guess "killer" will depend on the person) that we get from running on the Erlang VM: focus on fault-tolerance and supervisors, concurrency and distribution out-of-the-box, and generally other features coming from processes and the actor model. For example, Phoenix makes an excellent use of all those features to build performant web applications with a focus on real time (Phoenix LiveView being a recent great example). Nerves uses the fault tolerance bits for building embedded software. Etc.

[inferiorhuman]

But just to put things in perspective, v1.3 is almost three years old and things have generally evolved since.

But who wants to sit idly and wait for three years for a language to evolve (and regress)? At what point does saying "deployments have been broken with no fix or root cause in sight, let's just sit tight and hope for the best" sound ridiculous?

Were the alternatives to Elixir mediocre I'd consider revisiting it. Were I to see an interesting opening at an Elixir shop, I'd consider it. Would I ever stake my reputation on suggesting Elixir for a new project in a professional environment? Absolutely, 100% no.

Well, sometimes the killer feature is having all of that in the same package: immutability, concurrent primitives, scripts, etc.

I get all of that with rust and I'm not tied to Linux or a petulant maintainer. vOv

I'd be careful about touting performance versus a compiled language like Rust though. Going from Phoenix to Rocket I saw a pretty dramatic speedup in the time it took to serve requests even on my simplest web app (an image gallery). Phoenix wasn't slow, but Rocket was consistently faster.

Edit: OK async on rust is still a shit show, but hey nothing's perfect, right?

[josevalim]

> But who wants to sit idly and wait for three years for a language to evolve?

That's not what I meant at all.

Look, I am not trying to invalidate your experience. Nor I am implying you should wait 3 years for things to get fixed. Nor I am saying that Elixir is better than Rust.

I just want to point out that, for someone starting with or using Elixir today, their experience may be different because 3 years is a lot for a language and ecosystem where v1.0 was only 5.5 years ago. Deployment involves many concerns and even if some particular scenarios have not evolved accordingly, you can be sure others have. The same applies to Rust and other languages, they are improving all the time.

[inferiorhuman]

I just want to point out that, for someone starting with or using Elixir today, their experience may be different because 3 years is a lot for a language and ecosystem where v1.0 was only 5.5 years ago.

Distillery was (is?) broken less than a year ago. Ending up in a situation where such a key component (deployments) is developed and understood by only one person for such a long period of time speaks very poorly to the oversight and development process of Elixir. Beyond that it creates a distrust that Elixir will continue to work or prioritized appropriately — deployments worked on BSD for some period of time.

The last message I got from Github from BSD user commenting about distillery not working was in February 2019. After that it devolved into "here's how you get distillery working in Docker." That's the kind of low quality engagement I walked away from, and that's the reason I continue to recommend against Elixir or Phoenix for new projects.