The storage layer for Amazon Aurora is custom-built to provide a high degree of durability and availability. A total of six copies of your data are spread across three AZs, and snapshot backups are sent to S3 to give you the ability to do a point-in-time restore if your data becomes corrupted. Read replicas in Aurora share the underlying storage with the primary instance, so the replica lag typically associated with reading replicas is much shorter, and there is little to no performance impact on the primary.

Failover for Aurora works a bit differently than with a normal RDS database. Since the storage and compute layers are separate, and storage is spread out over three different AZs, failures that can occur with relational databases due to storage issues are much less likely. If an underlying compute instance needs to be restarted due to a crash, nothing needs to be done to correct or repair the data, and the buffer cache is separate from database processes, so your application will be back up and running at full speed quickly. Another benefit is if you have a cross-region replica set up, you can also fail the database over to the other region.