Problem
Suppose you have produced a simple prediction model that has been containerised and deployed on infrastructure like Kubernetes (K8S), configured to autoscale your service. As part of your model lifecycle, you wish to capture all predictions made when users interact with the service. You are currently storing these data to a sharded NoSQL technology (say MongoDB for the sake of this question), and are using range partitioning on the timestamp to distribute your data.
• What happens if your service gains in popularity? Is this sharding solution still viable?