Use Phoenix Channels | Hacker News Reader

Use Phoenix Channels

info.codecast.io

146 pointsclessg47 comments

I'm so excited for Phoenix 1.6. I think Phoenix was almost onto gold with LiveView when introduced, but it was still a little rough around the edges. But Phoenix 1.6 should sand away all the roughness, and we'll be in an awesome place.

Basically, the issue before, IMO, was kind of a tension between "live controllers" and (they call them "dead" but I prefer "static") static controllers. You couldn't necessarily interleave them super well, and factored templates had issues when used in one place or another. (I'm not an expert here, this was just from some testing and dabbling.)

But now (or well, soon), there's going to be a unified "heex" format which works in both. You write components almost like React components, their structure is parsed to some extent to catch simple typos and bugs, and can pass in "props" as arguments. And said components can be re-used in both live and static controllers.

Basically, channels were Phoenix's killer feature, LiveView sort of abused them to show a promising new way of developing web apps, and now I'm hopeful we're just about at a point where Phoenix can settle down into a new normal with function components on hex.pm and easy and pervasive LiveView pages and widgets, added in easily to your existing, standard Phoenix apps.

And, combined with the recent announcement of the removal of node+npm+webpack by default, in favor of esbuild, I think Phoenix 1.6 is going to be a no-brainer choice for any new web apps going forward.

Wow, this is great news! I had not heard about that, but making live and static controllers work together has always been a pain point for me. Even with that pain point Phoenix has been my go to. But fixing that will be huge.

They're also separating out Phoenix.View so there will be less dependencies on using Phoenix view rendered templates for things like email or PDF generation outside of a web layer.

This is great news for me! Around 8 months ago I built a chess website using Phoenix (heavily using LiveView) and TailwindCSS. My two biggest complaints were exactly what you said:

1. LiveViews were kind of weird to fit in, as they were a separate format

2. Webpack was the bain of my existence, it was slow and hard to figure out as a first time webdev

Glad to hear this is being fixed! Looks like I'm going to have to go back and update the site

I haven't use LiveView since the introduction of components, it'll be a good occasion to try it again.

But to be honest about the "no-brainer" part, I'll probably stick to using Phoenix for apis and websockets because I'm personally firm on not using a framework that use the (IMHO) outdated layout->view model for displaying HTML.

> the (IMHO) outdated layout->view model for displaying HTML

What is the modern model that you do use? Phoenix is very flexible and I'm wondering if it could easily be added manually.

Template inheritance with blocks, as they do in Django or Symfony.

That way you don't have layouts anymore, but (usually) a base template that an action view overloads with its own content.

I know Ruby on Rails doesn't work like that so I guess that's why it didn't take in Phoenix (yet?).

perhaps you could have a glance at "surface" and see if it's what you want? I think the new .heex format also gives the basis to build what you want even if surface isn't the answer for you?

Agreed. Once you're used to template inheritance it's hard to go back.

I created a new app with Phoenix 1.6 using esbuild and mount LiveView components just like they were react components.

I'm done with client-side apps. I had no idea how SHIT the workflow was. I just got used to the stench. Good riddance to webpack too.

It is going to be good being able to pass props to these LiveView's. I remember that wasn't possible and I needed all sorts of workarounds.

Heex templates alone are going to be a game changer. I've been looking for 1.6 updates every day - can't wait for the release.

I'm hoping the development of community libraries will pick up again. Coming from Rails, gems like Pagy, Ransack & Pundit are hard to live without.

Do you know if HEEX will support slots like in Surface? It really makes it easier to rewrite usable components.

Not with LV 0.16, but slots is one of our top priorities after Phoenix 1.6 / LV 0.16 is out

Is there a slot-like alternative for the time being? Or maybe a gist'able module that can be added to get preliminary slot support into 1.6?

Yes, community libraries are a key part of the overall experience of using a framework and being productive.

For those coming into this that aren't of an Elixir background, here's some of the things that make BEAM (Erlang VM that Elixir is built on) interesting for soft real-time features:

  * BEAM provides out-of-the-box capability to network nodes together. This is extremely important in real-time flows because server A may originate a message that needs to get pushed to a user connected to server B + C (2 tabs open). The capability to both connect + send messages between nodes is built in.
  * The process model in the BEAM provides a really natural fit for WebSocket connections. Each connection is its own process, and other processes are spawned off of that for various needs. The benefit here is that a failure in one process (connection drop, code error, crash, etc) won't affect other users. It's not an issue of whether the code is written well to prevent it, it's just intrinsic to the runtime. State is also stored per-process, which is convenient when it comes to writing your application.

Phoenix itself has a really nice abstraction over WebSockets (Channels, which this post goes in depth on). I won't say too much since the article does, but it's just really enjoyable to work with. Also, Phoenix LiveView is built on channels, so knowing channels provides some benefit there.

If it's interesting to you, I wrote a book that basically walks you through "I don't know anything about Elixir" to building a demo "sneaker drop" store https://pragprog.com/titles/sbsockets/real-time-phoenix/

+1 for the RT Phoenix book!

> BEAM provides out-of-the-box capability to network nodes together.

It’s been half a decade since I last did Erlang, but wasn’t BEAM the Erlang VM and OTP the library that makes the networking possible?

Yes, that's a fair point. I should have written BEAM/OTP because they're separate things. BEAM enables the process model, but the implementation of things like networking is in OTP. I'm not sure how far deep the actual process implementation goes, (is it in BEAM or OTP?) I think it's BEAM but I could be wrong. I'm not really going that low in the stack.

Great book! +1

It's neat to see channels getting some love here! They were the main reason I started Phoenix seven or so years ago, and they were the first major feature to land in the framework. I'm happy to answer any questions here about channels or Phoenix in general.

Do you and José have any plans to release a new book or learning resource after Phoenix 1.6 and the new LiveView ships?

Check out Sophie and Bruce's new book on LiveView: https://pragprog.com/titles/liveview/programming-phoenix-liv...

Are there plans to implement a distributed stream messaging system (e.g. Kafka, RMQ streams) embeddable in a supervision tree like Phoenix PubSub?

Way easier said than done I'm sure, but I know personally there are a lot of use cases for the distributed stream guarantees.

Apparently I'm out of the loop, but it took a fair amount of reading to discover that Phoenix is a framework for Elixir. It's only mentioned in passing on the Phoenix project's homepage, very easy to miss.

Just strikes me as odd and makes it less approachable. Most other projects I've encountered present themselves as a part of the language's ecosystem where it seems like it's assumed here. E.g. React calls itself a JavaScript library.

This powers the realtime sync at my startup. its the closests thing Ive ever found to "just drop it in and it just works" for self hosted websocket systems.

Same here. It was a breeze to build a scalable, real-time auction system with Phoenix Channels at the previous startup I was working at. There are definitely "devil in the details" corner cases, but for the most part it works beautifully without much effort :)

> just drop it in

Have you looked at Pushpin [1]? (disclosure: I'm the lead dev). The goal is to be able to add it to anything, rather than requiring a certain language.

[1] https://github.com/fanout/pushpin

This seem pretty nice. How does it handle things like websocket reconnection?

Whenever a client connects (whether the first time, or on reconnect) or sends a message, Pushpin sends an HTTP request to a configured backend to retrieve WebSocket messages to be delivered to the client. This can be used to restore state on reconnect.

If you mean how is reconnection handled on the client side, that's up to the client. Pushpin is a server-side component with no client library. But the client could use a WebSocket library with reconnection ability, like reconnecting-websocket.

I haven't, thanks for the pointer! However, the main reason I chose Elixir/Phoenix for this project was not for the language itself, but for the runtime/VM (BEAM), which is battle-tested for such use cases.

Meteorjs was like this, even simpler I would say. But their Achilles heel was minimongo tie in and scalability issues.

aw how i miss you meteor

meteor is still around.

That said, phoenix/elixir's live view accomplishes the same thing in a database agnostic way. its sync system is built on channels and scales WAY better than meteor ever could.

yes meteor is still around but it's warts make it a nonstarter for me now that I know how it usually goes when in prod

try phoenix's liveview. Beam is built for scaling thousands to millions of stateful processes on a cluster. That at its core was the biggest bottleneck for meteor and for elixir, its more or less a solved problem unless you're talking about google or amazon levels of traffic

> "just drop it in and it just works" for self hosted websocket systems

I was also been underwhelmed by options in this area, so I've been working on a Rust library called Jamsocket[1]. The idea is that all you need to do is implement a trait, overload the functions for the events you want to handle (new connection, message, etc.) and deploy it.

I still need to work on production aspects of it, but it's in a state where you can play with it for local development, so I figured I'd share it here in case any Rust developers want to tinker with it.

[1] https://github.com/jamsocket/jamsocket

Not that familiar with current web technology. What do you mean with "self hosted" websocket systems?

it means the infrastructure for doing a thing is hosted on your machine as oppossed to being served by a third party vendor on their servers. Think pusher or pubnub for instance.

I normally don't like complaining about terminology but "real time" means something: deadlines. If you don't have a deadline, you don't have a real time system. The coopting of the word is making it more difficult to find and list resources if you don't already know how to prefix it with "soft" or "hard."

Wow, I'd never heard this terminology before but I just spent an unnecessary length of time reading the Wikipedia page on "Real-time computation" and it's interesting set of technical considerations I hadn't thought much about before. Thanks for making me aware of this.

I've been really interested in Phoenix since seeing some of those 1 million simultaneous connections demos, but I also know that it's possible to build unrealistic demos that produce great statistics.

So, I'm really curious if there are any people here who are able to get such significant performance benefits from the framework for "real" scenarios - high distributed traffic with occasional writes and lots of cached database reads, game lobbies, etc. - stuff that simply wouldn't be possible with another framework like Spring or Flask. Is the hype worth it?

This podcast episode[1] is about exactly this, scaling live chat to a few 100k users, on a small number of machines in a real word scenario where the users are all signing on in a short time span to live screen a film and chat about it.

[1] https://thinkingelixir.com/podcast-episodes/057-scaling-live...

I also went on the same podcast (Mark + co are great hosts!) and talked about some of the challenges of scaling in the real world: https://thinkingelixir.com/podcast-episodes/051-live-auction...

I'm not a seasoned Elixir dev but I wrote an article some months ago about how I used many Elixir & Phoenix features (including Channels & PubSub) for a new feature on my side project : https://conradfr.github.io/ProgRadio/

With also a demo https://livesongdemo.funkybits.fr/ that uses Channels to stream the Phoenix logs ;)

Tl;dr: Apparently Phoenix is an Erlang / Elixir thing, and something for those of us not on Erlang to pay no slight attention to.

It would have been better to put that in the first paragraph, or even the title, so we would not need to scroll past reams of breathless hype trying to discover any single thing that we might have some use for someday.

A million connections on one server is no great trick today. It's table stakes.

What protocol is this "channel" thing using behind? Because I went through the documentation and they don't mention anything of it. I can't understand why.

Reply on news.ycombinator.com