Aurora is designed for high availability, fault tolerance, and elastic read scalability.
Each Aurora DB cluster has a single writer instance (handles all writes) and up to 15 read replicas for horizontal read scaling. All instances share the same distributed storage volume, so there is no replication lag between writer and readers.
The storage layer automatically stores 6 copies of your data across 3 Availability Zones (two copies per AZ). It requires 4/6 copies for writes and 3/6 for reads, ensuring durability even in case of multiple failures. The storage is striped across hundreds of volumes for performance and self-heals automatically through peer-to-peer replication.
Endpoints simplify connection management:
- Writer Endpoint always points to the current primary instance for writes.
- Reader Endpoint load-balances read requests across all read replicas.
- When read traffic increases, Aurora can auto-scale the number of read replicas based on metrics such as CPU, connections, or latency. New replicas are automatically registered behind the Reader Endpoint, and Aurora can scale them down when demand decreases.
The storage volume auto-expands in 10 GB increments up to 128 TB, providing low-latency, high-throughput access from all replicas.
If the writer instance fails, Aurora performs an automatic failover to a replica in under 30 seconds, maintaining availability. Cross-region replication is also supported for disaster recovery and latency reduction in global applications.