Yeah, last place I worked had transient loads we pretty much did everything with just ec2 t3a.medium spot instances (minus s3 and rds [which kind sucks with usi

Yeah, last place I worked had transient loads we pretty much did everything with just ec2 t3a.medium spot instances (minus s3 and rds [which kind sucks with using postgresql if you want to resize down or load from a pg_dump since you cant copy the sql to the rds instance and forced to backup over the network], I wanted to move to ec2 instances and from s3 to b2 since its cheaper for bandwidth with [we had cdn77 in front of cloudfront to lower costs outside of US/EU]). Using docker or k8s just doubled the costs over just using disk images and custom load-balancing daemon with nginx (with 3rd party modules for changing upstreams without restarts/reloads)/python, on top of adding more networking complexity (like to caching/search/rabbitmq instances). Probably around 20+ machines at peak when i left, but kept them all below 20% cpu. Sad to see so many places go in the docker/k8s direction.