Fly.io: the Reclaimer of Heroku's Magic

Heroku was catalytic to my career. It's been hard to watch the fall from grace. Don't get me wrong, Heroku still works, but it's obviously been in maintenance mode for years. When I worked there, there was a goal that just kind of grew in scope over and over without reaching an end state: the Dogwood stack.

In Heroku each "stack" is the substrate the dynos run on. It encompasses the AWS runtime, the HTTP router, the logging pipeline and a bunch of the other infrastructure like the slug builder and the deployment infrastructure. The three stacks Heroku has used are named after trees: Aspen, Bamboo and Cedar. Every Heroku app today runs on the Cedar stack, and compared to Bamboo it was a generational leap in capability. Cedar was what introduced buildpacks and support for any language under the sun. Prior stacks railroaded you into Ruby on Rails (Heroku used to be a web IDE for making Rails apps). However there were always plans to improve with another generational leap. This ended up being called the "Dogwood stack", but Dogwood never totally materialized because it was too ambitious for Heroku to handle post-acquisition. Parts of Dogwood's roadmap ended up being used in the implementation of Private Spaces, but as a whole I don't expect Dogwood to materialize in Heroku in the way we all had hoped.

However, I can confidently say that fly.io seems like a viable inheritor of the mantle of responsibility that Heroku has left into the hands of the cloud. fly.io is a Platform-as-a-Service that hosts your applications on top of physical dedicated servers run all over the world instead of being a reseller of AWS. This allows them to get your app running in multiple regions for a lot less than it would cost to run it on Heroku. They also use anycasting to allow your app to use the same IP address globally. The internet itself will load balance users to the nearest instance using BGP as the load balancing substrate.

<Cadey> People have been asking me what I would suggest using instead of Heroku. I have been unable to give a good option until now. If you are dissatisfied with the neglect of Heroku in the wake of the Salesforce acquisition, take a look at fly.io. Its free tier is super generous. I worked at Heroku and I am beyond satisfied with it. I'm considering using it for hosting some personal services that don't need something like NixOS.

Applications can be built either using cloud native buildpacks, Dockerfiles or arbitrary docker images that you generated with something like Nix's pkgs.dockerTools.buildLayeredImage. This gives you freedom to do whatever you want like the Cedar stack, but at a fraction of the cost. Its default instance size is likely good enough to run the blog you are reading right now and would be able to do that for $2 a month plus bandwidth costs (I'd probably estimate that to be about $3-5, depending on how many times I get on the front page of Hacker News).

You can have persistent storage in the form of volumes, poke the internal DNS server fly.io uses for service discovery, run apps that use arbitrary TCP/UDP ports (even a DNS server!), connect to your internal network over WireGuard, ssh into your containers, and import Heroku apps into fly.io without having to rebuild them. This is what the Dogwood stack should have been. This represents a generational leap in the capabilities of what a Platform as a Service can do.

The stream VOD in the footer of this post contains my first impressions using fly.io to try and deploy an app written with Deno to the cloud. I ended up creating a terrible CRUD app on stream using SQLite that worked perfectly beyond expectations. I was able to restart the app and my SQLite database didn't get blown away. I could easily imagine myself combining something like litestream into my docker images to automate offsite backups of SQLite databases like this. It was magical.

<Mara> If you've never really used Heroku, for context each dyno has a mutable filesystem. However that filesystem gets blown away every time a dyno reboots. Having something that is mutable and persistent is mind-blowing.

Everything else you expect out of Heroku works like you'd expect in fly.io. The only things I can see missing are automated Redis hosting by the platform (however this seems intentional as fly.io is generic enough to just run redis directly for you) and the marketplace. The marketplace being absent is super reasonable, seeing as Heroku's marketplace only really started existing as a result of them being the main game in town with all the mindshare. fly.io is a voice among a chorus, so it's understandable that it wouldn't have the same treatment.

Overall, I would rate fly.io as a worthy inheritor of Heroku's mantle as the platform as a service that is just magic. It Just Works™️. There was no fighting it at a platform level, it just worked. Give it a try.

<Cadey> Don't worry @tqbf, fly.io put in a good showing. I still wanna meet you at some conference.