With our backend being integrated rapidly by packs across the globe and traffic surging exponentially, we had to think of cost-effective solutions that would allow us to scale beyond anything we had ever been confronted with before. This talk describes our journey from normal on-demand AWS instances as our primary k8s workers, to creating a truly elastic and cost efficient cluster with spot instances – and the challenges we had to overcome in the process. These include roadblocks we had with KOPS (which we use to deploy and maintain this particular cluster), and core kubernetes components such as the cluster autoscaler that were driven out of their own comfort zone. As an added bonus, we will talk about how we initially used Thumbor as the core component for this application, and later built Dali our open source Rust-based image processor to drive up scale, and drive down cost!