Back to docs
Recipe: Aurora multi-AZ cluster design
A production-grade Amazon Aurora cluster spanning three Availability Zones with writer/reader separation, automated failover, and connection pooling for high-throughput workloads.
3
Availability Zones with cross-AZ replication under 20ms
<1s
Automated failover with RPO of zero
2x
Read throughput via reader endpoint load balancing
Architecture
Deploy a single Aurora writer instance in AZ-a with two read replicas in AZ-b and AZ-c. The cluster volume is striped across all three AZs with six copies of data (two per AZ).
Applications connect via the cluster writer endpoint for writes and the reader endpoint for reads. RDS Proxy sits in front of both endpoints to manage connection pooling and reduce failover disruption.
Key decisions
- db.r6g.xlarge — Graviton2 for price/performance on write-heavy OLTP
- RDS Proxy eliminates connection storms during failover and reduces Lambda cold-start overhead
- Backtrack enabled with 24-hour window for rapid point-in-time recovery without restore
- Custom cluster parameter group— slow_query_log to 200ms, long_query_time to 0.5s
Cost estimate
Writer (on-demand)
$438/mo
2 Readers (on-demand)
$876/mo
RDS Proxy
$64/mo
Storage (100 GB)
$10/mo
Total ~$1,388/mo. Reserved instances reduce this by ~40%.