ALT A group of one or more nodes that serve a single dataset is called a cluster. Each shard of a Amazon MemoryDB dataset contains a primary node and up to five optional replica nodes. A replica only handles read requests; a primary node handles both write and read requests. A replica node can be promoted to the new primary node for that shard when a primary node fails over to it. MemoryDB uses either Redis OSS or Valkey as its database engine, and you may choose which version to use when you build a cluster. The AWS Management Console, the MemoryDB API, or the AWS CLI can all be used to create and change a cluster.