I like how Phoenix had the stones to replace webpack after all the additional support requests it was bringing in. I've been using esbuild with 1.6.0-dev and it is a lot more ergonomic for the end user dev - https://sergiotapia.com/phoenix-160-liveview-esbuild-tailwin...
What has me really excited was that I was looking at Fly.io to create a global mesh of servers for a liveview app so people can connect to whatever is closest, and guess who just joined Fly.io's team? Chris! I think this is going to be a very productive stack that will get you far far ahead of the competition once you learn it's patterns. Complicated features become trivial to implement with great UX to your customers.
I've gotta agree. We replaced webpack with esbuild in our Phoenix 1.5 app. The migration took a bit of time to get right, and I was questioning whether it was worth it, but having seen the results -- faster builds, fewer dependencies, generally not having to worry about webpack at all -- I'm thrilled this is becoming the default.
Chris: how do you feel about what's needed to make Phoenix as "core" of a technology like other similar projects like Django or Rails? The latter are used by huge companies, which have vested interest in keeping these technologies around and evolving. Is it corporate sponsorship from companies using Phoenix? Note, I've used Phoenix/Elixir as the core technology in the last of the 3 companies I've been in.
We're already there. We have huge companies using elixir and phoenix, with phoenix powering the web platform for the majority of them. Several great case studies exist on elixir-lang.org if you're interested, ie:
And of course Bleacher Report, which has been on Phoenix since 2014, well before 1.0.
On the sponsorship side, DockYard invested in my work for the last six years where I spent the majority of the year working on oss, and I've just joined https://fly.io where I'll spend 85% of my time working on Phoenix and related libraries. So we are well covered on the corporate sponsorship side. We also have Michael Crumm on the Phoenix team who works at Cargo Sense, where they support a portion of his time working on LiveView. Of course it would be great for other large companies leveraging value out of our work to give back, but we're well established on the sponsorship side :)
It's worked out really well for us at PepsiCo, it's basically our go to for in house application development on the eCommerce side. We'll be hiring some elixir devs again soon.
We're not hiring any FTEs outside of the US yet, but we probably will in 2022. Keep an eye out, there should be some interesting stuff coming up in the future.
Ruby companies like AirBNB, Stripe ended up edging out their competitors using the relatively glacial Java or C++ frameworks. We will need to see a new generation of Phoenix winners before Elixir is taken seriously.
It could be that the marginal performance gain from using Elixir is not worth the hassle of training and hiring in a whole new language. It remains to be seen.
Just a small nit-pick here. If we're talking REST/graphql stateless APIs, sure the scalability gains may be a harder sell vs throwing hardware at the problem, but the value-add depends on what's being built. I built Phoenix first and foremost for real-time, stateful web applications. So anything that requires websockets, frequent updates, PubSub, presence, etc, you will be extremely hard to match what our stack provides out of the box, performance wise, and distribution wise. Not everyone needs these things but increasingly companies do, ie chats, social feeds, dashboards, and now LiveView-like interactive applications all change the equation from "marginal" perf gains.
I am an Elixir evangelist, to be clear, but it's one thing to have believers and another to change the world. I think you are right that Phoenix will eventually win the day in regards to real-time, stateful Web applications. "Traditional" REST APIs may be a harder sell, and perhaps Elixir doesn't need to eat the entire Web to considered successful. There are still many Java developers at work after all.
I am really confused by this comment. "Companies leveraging a quick to develop in, batteries included framework, with really poor performance, beat out companies using highly performant but less quick to develop in languages/frameworks, is reason why I'm suspect about a quick to develop in, batteries included framework that also prioritizes real time communication, with quite solid performance, and really predictable latencies".
I mean...the argument about "the hassle of training and hiring in a whole new language" applied just as much when Rails took off. While Ruby had been around since the mid-90s, it was definitely an esoteric language for the decade or so prior to Rails.
Will Phoenix penetrate as broadly as Rails? Probably not, since Rails already exists, and now benefits from having a lot of people who know it. But being able to reduce hardware costs by 90% while giving up little of the flexibility, and bundling in a lot of real time features can be quite compelling; we'll just have to see.
> Will Phoenix penetrate as broadly as Rails? Probably not, since Rails already exists
I hope it does one day, but not to replace Rails.
It's a really hard sell because I think eventually we're all going to converge to the same real-time, low server cost setups. At this point it becomes how fast is it to build my app and what is it like to maintain it.
Rails 7 is coming out with a neat feature that will build in encrypted fields at the DB level for Active Record with a way to situationally decrypt them in case you need to debug something specific in production. Then, any time a developer accesses those decrypted fields an audit log will be generated. It'll also audit log any Rails console access[0].
These are really big things to not have to deal with. Will Ecto ever get this sort of thing? Maybe, maybe not. But I think the bar is raising overall where if things like this aren't added in the future you have a less chance of being adopted by the masses because not everyone has the resources to spend months developing stuff like this in house.
Did we exist without it in the past? Absolutely, but I think we need to be comfortable with adjusting our expectations that higher standards are coming as there's more framework competition and languages coming out. LiveView is a great example of this, only a few months after it was announced a bunch of popular frameworks released their take on it that play to the strengths of that framework of choice. Now there's kind of an expectation that it shouldn't take an SPA to develop a mostly document driven application with a nice user experience.
Personally I would want features that help me build apps faster and more securely while having to pay $40 / month to host my app instead of a framework that doesn't have these features but it costs me $20 / month to host it instead. Sure if you take this to completion and say but if you have 10,000 servers running it's double the cost but for most applications you can host a pretty popular and profitable app for less than $100 / month with any modern tech stack without making performance compromises (ie. <= 100ms page responses for the p95).
Stripe, Shopify and GitHub are still largely relying on Ruby as far as I know (and while there are reimplementations of some of their services in some languages, I don't think C++ or Java have been publicly mentioned often).
That said, I think one important asset of Elixir (in my view) is not the raw performance, but the overall "total cost of ownership" (including maintenance work, and software is a lot of maintenance).
Being able to simplify architectures is something that is nicely done with Elixir, actually.
If a part of an app needs to play a "proxy" role (like here https://github.com/etalab/transport-site/tree/master/apps/un...), then I just add a component to the app, and I can keep the incoming connections under hand, all while issuing external queries etc.
If there is a need to do some rich interactive dashboards, I can use LiveView.
If I need to demonstrate some simple stuff, I can use "Mix.install" to create one-off scripts.
If I need to do more data-science, I can tap into LiveBook, VegaLite etc.
All this with a small team or as a solo developer, is quite great, much more than raw performance, in my eyes.
Technically "marginal" refers to the gains made by an incremental improvement, in this case the adoption of Phoenix over Rails, and not the use of Phoenix over nothing at all. So we still refer to the gains from Phoenix as marginal, even though they are substantial in many cases.
Can talk about what (if) any plans exist to make deploying Phoenix apps at fly easier.
Said another way, I’d love to click a button to deploy my app to be (a) globally distributed, (b) managed global database, (c) managed operating systems, (d) managed dns.
Yes! What you're describing is important, and roughly matches what we want for all apps. Managing apps + databases across multiple regions should be as simple as `fly region add syd`. This takes like 5 steps right now, it should take one.
We're also improving the Phoenix deploy process. I don't like it when people have to see Dockerfiles. Modern app deployment needs frameworks and infrastructure to work together. Our Elixir deployment doc (https://fly.io/docs/getting-started/elixir/) will someday be replaced with, hopefully, "mix phx.deploy".
a. Will pre-prod environments be included in those plans you outlined above (eg Dev, Test, Stage)
b. Any plans to make it easier on Developers to have local/remote environments for development? (Eg GitHub Codespaces)
c. Do customers typically also use 3rd party security offerings like Cloudflare in conjunction with fly.io - or do you see more customer just using fly.io without any additional protection?
d. One thing that Cloudflare Workers does that’s interesting is that it deploys your webasm code to every pop/region they have & them orchestrate the communication between the regions to keep data consistent. Is similar functionality on roadmap?
e. Any rough order of magnitude on timing for what you described?
I've dabbled with Phoenix 1.6-dev (after seeing it in action and being mindblown in the livebook code), and think the new HEEX components are great.
Building a LiveView app out of components, and sending messages back to the server and re-rendering, feels a lot like building a react app! This is a real intuitive and easy to use flow. (I remember when react came on the scene and its "one way dataflow" was so groundbreaking to me.) That said, a react app where every state update is a round-trip to the server is probably not ideal. I know that there's the PETAL stack for this sort of thing. I guess my question is: with so much functionality available in pure Elixir/LiveView, what's your decision process when embarking on a new app about when to try to get by with just that, vs bringing in react or a full SPA frontend and simple API backend?
Separate question: what's the best dev environment and tooling for Elixir/Phoenix/LiveView these days? My VSCode setup is pretty great for React, particularly with format-on-save nesting and indenting my JSX properly. I miss that when working in (h)eex templates.
I rewrote a large Rails+React+GraphQL with LiveView. LiveView replaced most of React. I still use React for rich text editors and complex browser based UI. LiveView plays well with this. I am happy not using GraphQL at all. It also consolidated all MVC code in one place. Phoenix made a few other things possible: in place chat, live updates (via websockets. No more polling).
That's really the goal, and for a large class of applications we are already there. Higher-fidelity UIs is pretty vague and there is absolutely a limit to where you necessarily need a bunch of code running on the client. Google maps & docs? Millions of lines of JS is inescapable, but dashboards, feeds, chats? LiveView excels here. We don't yet know where the limit is, but if you look at Livebook, we have a collaborative code notebook with presence, realtime updates across users and operational-transform based content updates powered by LiveView, with only ~3k lines of basic JavaScript for the client-side bits like the Monaco editor.
Have you done any comprehensive profiling of Livebook? It would be really interesting to see how it runs on low end or average cloud hosting providers under load.
I'd love to see how many thousands of concurrent users could be interactively using and collaborating on a Livebook that's running on something like a $15 / month server. You could show off the memory, CPU usage and the latency of responses with varied amounts of load.
IMO that would be a very useful thing to look at and could push folks into using Phoenix more so than they already do because having a bunch of tiny websocket connections open (the old 2 million connection benchmark) isn't really relatable to a typical web application that you mentioned people are building like dashboards, feeds and other things. Livebook is a concrete application with features that might appear in many other web apps.
Something like that would be worthy of a demo video at a talk or even your YouTube channel if you were ever fishing around for what people in the community might want to see.
The total cost to run this is on Fly.io is $38/mo (~$2/mo per region). We've seen as many as 10k concurrent users pushing 30k "clicks" per second. People are fond of writing JavaScript in their browser console to simulate very fast clicks, so it turned into a funny stress test.
My browser slowed down, the Elixir processes never did.
Thanks, I remember seeing that one a few months ago.
I hope this doesn't come off too harshly but that demo feels like it's meant to showcase high requests per second, not be a representation of how a real application would perform. It's broadcasting a counter over a websocket connection. It feels very similar to the old 2 million websocket connection demo. In other words I can't use your demo to make a meaningful conclusion on how LV would perform in my app. Also according to dev tools your demo page takes 300ms for me to load it from NY but I get 10-15ms response times on many other sites in the north east region of the US. A 300ms page response is usually what I would get hitting a server in Australia. The initial page load feels very delayed to load, although the counter increment was lightning fast.
I like the idea of using Livebook as a test case because it's an actual application, it's something relatable to what you might build instead of something that feels like it was created in a vacuum to promote high numbers. Sort of like how you could benchmark nginx with a hello world response. It's cool to see the high numbers but it doesn't help folks understand how things would work in a real app.
Having a breakdown of the average and p95/p99 response times at 100, 500, 1000, 2000, 10000, 25000 and 100000 connections all sending messages at varying rates along with seeing CPU and memory usage on a certain tier of VPS would be amazingly useful. You could take that data and factor in how your app might deal with similar conditions.
It's not too harsh at all! The demo is really meant to show liveview interaction latency, we didn't really ever think it would be useful for load.
That 300ms load time is actually because it queries Australia when you first load the page. Each region keeps its own counter and it grabs the current value from all of them at load time. It's probably worth optimizing that. :)
When learning elixir and phoenix, it is hard to find updated tutorials. Whenever I try an youtube tutorial, there is always an error. Can you share some updated links/books for a beginner?
Not op, but for me the official elixir site's guide help a lot. I actually read that side by side with Sasa Juric's Elixir in Action and felt like I got a pretty good understanding of elixir.
For Phoenix, I've just been using the guide in phoenix's hexdocs and reading some random blogs about anything specific.
These are great suggestions. The getting started guides on Elixir and Phoenix get you moving with the basics. Beyond that, Programming Elixir, Elixir in Action, and Programming Phoenix are great book options.
Yesterday we launched a big feature built using Elixir/Phoenix/LiveView/Surface and it has been really great.
We are a SAAS for managing sports teams and clubs https://www.sportmember.com and our current stack is Ruby on Rails, GraphQL and React Native for our app.
When deciding to develop this new feature we narrowed it down to these two options:
1) Rails/GraphQL/React
2) Elixir and LiveView
What's great about Elixir and LiveView is
1) The whole GraphQL API layer was not necessary to build. The best API is no API ;-)
2) No need to duplicate code like for example for validation on the frontend and the backend
3) Developing the whole feature using just Elixir instead of having to switch between TypeScript and Ruby
This means that we have been able to develop this feature in about 2/3 of the time which is pretty nice!
"Phoenix 1.6 ships with a new phx.gen.auth command for a complete authentication solution bootstrapped into your application" - hope it's a decent implementation. Setting up a JWT authentication took me days while learning phoenix/elixir and it is the sole reason why I never went with Phoenix for any other project. So many components involved.
In one of the HEEx examples in this post it mentions components and uses a weather component as an example with: <MyApp.Weather.city name="Kraków"/>
I'm curious, what would the syntax look like if you wanted to create something like a card component that has an optional header, body and footer where you would want your arguments to be blocks of arbitrary HTML that you pass in and not a single string variable?
Basically a way to do something like this (a more simplified version since I don't know what the API would be):
<Foo name="bar">
<p>Custom stuff!</p>
</Foo>
And then the component itself would have a named "yield" that places in your arbitrary blocks of code in the correct spot.
I think this might be called slots in other frameworks, but they seem essential to be able to create re-usable components unless HEEx has an undocumented way to do this? It looks like Surface has this at https://surface-ui.org/slots.
What are the benefits of using HEEx over Surface? I know they're working together and in the end will share similar functionality and maybe Surface will use HEEx under the hood but as an end user building web applications with Phoenix is there a single case where I would use HEEx instead of Surface? If not, why isn't Surface part of core Phoenix or not heavily pushed as what to use for a templating solution in the docs?
I chose to stick to base LiveView in my apps because it works fine (slots are nice but not essential), and I want to avoid too much API churn and learning curve for other devs on the same team. The best Surface features are being gradually refined and merged into LV. HEEx is a great example of that.
Thanks, I think that would very likely handle the basic case of having 1 block.
But in a re-usable card component you might decide to have a title, body and footer block where the footer is optional and perhaps falls back with a default value if it's undefined. Slots with Surface handle this use case based on their docs. I think there's a lot of component use cases where you'd want the idea of having multiple blocks.
The docs you linked don't cover if HEEx supports anything like that, or being able to use multiple named blocks in 1 component. I tried to read the source but render_block is a macro and I couldn't proceed to trace it.
"Phoenix also now includes swoosh by default for mailer support and a new phx.gen.notifier task for generating a notifier for sending email along with a dev mailbox for local development."
Crap. I literally just built this out for a personal Phoenix project two weeks ago (using Bamboo instead of Swoosh because I could get it working for SES more easily). Guess I'll see about adopting the official and probably far better solution. Since there are also published upgrade steps. CURSE YOU FOR BEING SO HELPFUL, PHOENIX DEVS!
I'm loving the direction that Phoenix and Elixir is taking. I am well invested in my current company but one day if I get a chance to work with it professionally I will be very happy. The tooling is evolving at a quick pace and is super powerful already out of the box.
Follow-up question: having built a mobile app backend on Phoenix (with Channels) and having played a bit with Absinthe and LiveView -- is there a future of Phoenix that's focused not only on web but also mobile?
What has me really excited was that I was looking at Fly.io to create a global mesh of servers for a liveview app so people can connect to whatever is closest, and guess who just joined Fly.io's team? Chris! I think this is going to be a very productive stack that will get you far far ahead of the competition once you learn it's patterns. Complicated features become trivial to implement with great UX to your customers.