SHARD DESIGN DOESN'T START FROM YOUR TABLES

Summary from the TLDR DevOps newsletter

Neki "shard groups" let users co-locate related tables like customers and orders on the same physical cluster to keep joins and transactions local. Reference tables replicate lookup data to every shard, while global secondary indexes support queries that lack the shard key.