Track · 2:30 · Liner note
Sharded but Whole
Sharding splits a large dataset across multiple servers by a shard key, so each server holds a portion while queries are routed or distributed so the data still behaves like one table.
One server eventually cannot hold or serve everything. Sharding splits the data across several of them by a shard key, each holding a slice, while a routing layer makes the whole behave like one table. Sharded but Whole is the goal: growth without the application needing to know where any row lives.
The shard key is the decision you live with. A skewed key creates hot shards, and a key unrelated to your queries turns every read into a fan-out to all of them. Choose for even spread and for how you actually filter, and plan how to rebalance.