Replication and Sharding

Replication

Main Database          Replica
         synchronous updates
[           ]         [       ]
[           ] ----->  [       ]
[           ] ----->  [       ]
[           ] ----->  [       ]
[           ]         [       ]
         asynchronous updates

• When you write to the main database, you write to the replica as well.

• If writes fail to the replica, then rollback the transaction.

• This improves availability in some ways, but lowers performance.

• Asynchronous updates are more performant, but lower consistency, and have more difficult failure modes.

Sharding

• Split up the database into different servers.

Strategies:

• Location based sharding:

• Table based sharding:

• Row-based sharding:

  • Split some rows into some computers.

• Some shards might get more traffic because of this.

• You can use hashing to write and read to databases.

• You can use a reverse proxy to hash data to find the location of where to shard.

Prev: [replication-and-sharding](replication-and-sharding.md) Next: [leader-election](leader-election.md)