A sharded database assigns different parts of its data to different partitions, commonly according to a key such as an account identifier. Queries can be routed to the relevant partition, spreading work beyond the capacity of one database server.
Sharding adds coordination requirements and does not replace fault tolerance. The system must manage uneven load, cross-partition operations and the movement of data while retaining reliable access. Redundancy protects against failures; partitioning primarily distributes capacity.
ELI5
Sharding means dividing a large collection of data into smaller groups that separate machines can handle. It helps avoid sending every request to one machine.
For example, a service could place customer accounts into several groups and send each account's queries to its assigned group. Each group still needs protection against failure; splitting the data is not the same as making a backup.
Is sharding the same as replication?
No. Sharding distributes different portions of data, while replication maintains additional copies for availability or other purposes.
Does sharding remove every scaling limit?
No. Uneven traffic, cross-partition queries and coordination overhead can remain bottlenecks.
