– Working on it during nights and weekends as a part time project

– Motivation for using Phoenix and Elixir

– Ian is confident that if it came down to it, he could hire Elixir developers later

– Generally speaking it’s a monolithic app but it has a few tiny services broken out

– At the moment there’s not much benefit in breaking it out into umbrella apps

– The main app is about 8,000 lines of Elixir and it’s using Phoenix contexts

– Phoenix has some opinions but you still need to make a lot of decisions

– A couple of context names that are used in the main app

– Parsing PDF files and doing some work in the background

– Phoenix channels are used to ferry back the parser’s status to the user

– Live View isn’t currently being used

– It’s a split app style with a Phoenix API and a React front-end

– Live View currently doesn’t have enough features to replace a React app

– The PDF parsing rabbit hole goes quite deep

– AWS ECS, AWS Lambda and a service called Spotinst helps run and scale the app

– Docker isn’t being used in development but tests are run in Docker

– Gitlab CI handles testing, building images and pushing them to ECR

– Why exactly isn’t Docker being used in development?

– Testing various AWS services locally using Local Stack

– Local Stack is open source and free but they also have a paid tier if you want it

– PostgreSQL is the primary database and S3 is used for storing flat files

– The ex_aws library is used to connect to AWS using Elixir

– Caching is being done directly in Elixir with a custom GenServer approach

– What exactly is being cached? Mainly the PDF documents

– nginx isn’t being used because AWS’ API Gateway and load balancers fill that role

– Static files are being hosted with AWS CloudFront (CDN)

– Going all-in with AWS and being very happy with it for productivity

– 3 to 6 EC2 instances are used in the ECS cluster depending on the load

– Scaling up is automated and takes about 30 seconds

– Spotinst helps with that by having idle machines that are ready to go

– The highest load comes from uploading many fairly large PDF files in parallel

– Between AWS’ and Spotinst’ logs and alarms, it’s easy to keep an eye on the load

– Stripe is being used to handle payments and the payment strategy is interesting

– Moving to Stripe’s hosted checkout eventually would remove a lot of stress

– Currently PaymentIntents and SCA isn’t supported by choice

– Handling payments was the last feature that was added to the app before shipping

– PayPal isn’t supported yet because there’s only so many hours in the day

– The only emails being sent out are for user actions which is handled by Cognito

– Walking us through a deployment from development to production

– ECS has gotten a little nicer to work with in regards to updating services

– Having issues with AWS App Mesh and Envoy due to issues with websockets

– Secrets are managed with env variables hard coded into the task definition files

– The AWS web console is starting to become quite good

– Rolling restarts are done over ECS to deploy without downtime

– How do you deal with draining worker connections to avoid losing partial uploads?

– Rihanna is an Elixir job processing library backed by PostgreSQL

– You usually end up needing a job processing library even with Elixir

– As for backups, all of the data and flat files are backed up and could be recovered

– IP bans at the firewall level helps with denial of service attacks

– Everything together on AWS costs about $400 / a month

– Gust Launch gave him $15,000 in AWS credits for starting a business with them

– Trying to crash things on purpose by throwing massive traffic at it

– Thinking that all users are out to get you and designing your app to be robust

– Plug.Upload goes straight to S3 to handle file uploads

– Best tips? Profile and load test your app before you launch your app publicly

– If you have a lot of task based jobs, look into using Lambda early on

– Check out Scrollkeeper, also Ian is on Twitter and GitHub

– Ian also wrote a blog post series on writing your own web crawler in Elixir