– The hex.pm site is a community effort with a few core contributors

– 5 reqs / second for the main site and 140 reqs / second for the repo side of things

– A majority of the traffic is pulling down packages with mix

– Eric started working on the Hex client in late 2013 before Elixir 1.0

– Hex launched with both the CLI and web front-end at the same time

– Motivation for using Elixir to build an Elixir package manager

– For the site, Plug was used initially because Phoenix wasn’t out yet at the time

– They try to keep the site updated to use the latest stable release of Phoenix & Ecto

– Why Eric decided to open source the platform on GitHub

– The Hex site’s source code helped Nick learn a bit more about Elixir and Phoenix

– Trying to use the latest best practices so the site is a good working example

– The site is mostly a monolith but there are a few dedicated services

– https://diff.hex.pm/ is one of the separate services along with billing and the docs

– Why did they decide to make the billing service closed source?

– There’s currently 10,000+ packages on Hex with 60k+ different versions combined

– Less than 10 GB of data makes up everything that’s stored on S3 for all packages

– The benefits of splitting up a few services outweigh the pain points

– Johanna Larsson (@joladev) initially created the Hex diff service

– The diff service’s search feature is powered by Phoenix Live View

– The main Hex website is mostly server rendered templates with a touch of JavaScript

– Breaking out smaller services lets you experiment with new technologies

– There is no admin dashboard web UI, it’s all powered by version controlled scripts

– PostgreSQL is the primary database

– Docker is used in production with the managed Kubernetes service on GCP

– There’s no need to cache anything because Phoenix is fast out of the box

– They rolled their own rate limiter with very little Elixir code

– Docker isn’t being used in development since their stack is so simple

– What made them choose Google Cloud (GCP)? Mainly Google’s managed Kubernetes

– (3) servers power everything, each having 2 CPU cores and 4 GB of memory

– When you install public packages from the CLI, you don’t hit the web API

– For private packages you do hit the web API but that’s going to change soon

– Fastly is their CDN and it lets you upload Varnish config files

– Fastly gives you unlimited and instant cache purges, most other CDNs limit you

– Amazon SES is used for transactional emails

– Rollback is used for error reporting and Stripe is being used for billing

– For logging and metrics, that’s all handled on Google Cloud and it’s decent

– Elixir Telemetry isn’t really being used now, but they might use it later

– Stripe SCA support is being worked on but it wasn’t fun to re-write everything for it

– What it’s like to push code from development to production

– Dealing with database migrations isn’t easy

– Putting your site into a read-only or maintenance mode during a migration

– Creating an Elixir library to toggle your app into read-only mode, make it so!

– DB backups are done through GCP, and they also use Tarsnap for S3 backups

– Health checks are in place with Google to get notified if a service goes down

– Best tips? Don’t start out with a fancy infrastructure, just get your app up and running

– Mistakes? It’s hard to test the Hex client (the CLI) and they are looking to improve that

– For testing, they mock out S3 with the local file system

– Hex.pm is open source on GitHub and you can find Eric on Twitter @emjii