Elixir on AWS ECS definitely works very well.
I am somewhat puzzled by the way ECS and EC2 Auto Scaling groups work together. Scaling rules are set independently on the two, resulting in situations where ECS tries to start more tasks than can be run with the current cluster size. I’m not quite sure how scaling down works in the EC2 Auto Scaling group, either, as there is no obvious communication between ECS and EC2 Auto Scaling.
But I have a feeling that the ECS Capacity providers may improve the situation, but I haven’t had time to look into it yet.
Nevertheless, I am running a system in production with ECS and EC2 Auto Scaling. We go from 2 to 8 EC2 instances, depending on traffic to our system. Each EC2 instance runs a single task of our Elixir application. The application then figures out which nodes should run what services. This is really running smoothly.
Having said that, it would be really interesting to know of right and wrong ways to do things like this. There are thousand of ways to set something like this up. For instance ECS vs EKS
