This approach, for us, proved fastest when we scaled it using Kubernetes compared to the others we tried.