Fargate Is Not Firecracker (2024)

I know this is a few years old but it is interesting. We use ECS on Fargate for a lot of internal services since we wanted containerized workloads in AWS but didn't want to spend for EKS. At the time ECS was new and shiny and we put just about everything into it. This was a mistake.

If you have a stateless, simple service application that you want containerized - ECS on Fargate is a great choice. We have a handful of these and the Fargate just works for us. But if you have a complex web application like say - Magento 2 - then ECS is the wrong tool and will end up costing you upwards of thousands of dollars more a month to run. Our migration back to EC2 may have increased our build times, but our monthly spend on compute plummeted.

I think we could have used ECS better but instead we were jamming supervisord, nginx and php-fpm all into a single container, and deploying them at 4vcpu and 8gb of RAM each. I will not do that again.

What's funny to me about this whole thing is that Firecracker wasn't ever once considered during this whole endeavor. We didn't choose ECS because of some marketing material implying it ran on Firecracker or was fast - we chose it because it was managed containerization.