Aurora PostgreSQL Replica Pricing

With Aurora (PostgreSQL and MySQL), when you add a read replica instance to an existing cluster, you are billed for the instance hours based on the instance type you choose for the replica instance. You are NOT billed for additional storage - there is no additional storage for read replica instances in Aurora, as the read replica instances share access to the same Aurora storage volume as the primary read-write instance; as such, data is not replicated to the replica instances.

Thanks for being an Aurora PostgreSQL customer!

-Kevin J, PM for Aurora PostgreSQL

Here is the conceptual info about how Aurora storage works:

https://docs.aws.amazon.com/AmazonRDS/latest/AuroraUserGuide/Aurora.Overview.StorageReliability.html

You can think of the writer and reader instances as multiple servers all hooked up to the same hard drive. All the writes happen from one server. The reader instances only transfer data off the shared storage as needed to process queries. The shared storage does 6x duplication behind the scenes for durability. But you are only charged for the one copy of the data.

It's a very different setup than what you find with "traditional" database replication. In "old-style" architecture, all the data gets transferred and stored in duplicate format on multiple servers. There's overhead on the original server to transmit the data everywhere. There's overhead on each replica to process and write out a full extra copy of the data.

With Aurora, the reader instances basically determine which pages need to be evicted from the memory cache due to updates to the underlying storage. That results in less CPU overhead, less network I/O, and less disk writing. The fact that the reader instances are not doing "replication" per se is why the terminology was changed to "writer and reader instances", instead of "Aurora Replicas".