Sharding
Splitting one dataset across several databases (shards) by a key, such as a hash of the user ID, so each shard stores and serves only part of the data. It scales writes and storage past one machine, but a query that spans shards must ask all of them, and one hot key can overload its shard while the others idle.
Data
In Quorum#
The Sharding research exists, but today its only effect is to unlock Leaderless replication: you cannot shard the Database. The closest mechanic is the Hot key Incident, where 30% of the reads to a Cache or Wide-column Store ask for one key on one shard, so the Node's capacity falls to about 1 / (0.3 × shards + 0.7). With Hybrid fan-out you can split hot keys across the Node's shards.
Read next#
Hot key, Consistent hashing, Leaderless replication, Fan-out