Sharding
also called Horizontal Partitioning
Splitting one dataset across multiple independent databases by a partition key, so that each holds a disjoint subset.
Sharding is what you do when one database cannot hold the data or serve the writes, and it is a one-way door: the partition key is close to unchangeable once data is distributed and application queries assume it.
Choosing the key is the whole decision. It must appear in almost every query, or every query becomes a scatter-gather across all shards. It must distribute evenly, or one shard becomes hot — sharding a multi-tenant system by tenant ID works until one tenant is a hundred times larger than the rest. It should keep data that is queried together on the same shard, because cross-shard joins and cross-shard transactions are the two things sharding makes genuinely painful.
Exhaust the alternatives first: read replicas, caching, archiving cold data, a bigger machine. They are all reversible; this is not.