Skip to main content

RDS: Migrate Aurora Serverless to Aurora Instance (Provisioned DB)

Updated this week

Difficulty

Advanced

Description

Aurora Serverless v2 scales automatically based on demand but becomes expensive for stable, predictable workloads. For databases consistently operating within a narrow ACU range, migrating to a provisioned Aurora instance covered by a 1-year Reserved Instance can reduce costs by up to 80% while maintaining equivalent performance.


Selection Criteria

  • Database has been running for at least 31 days.

  • Maximum ACU range is used to identify the right Aurora provisioned instance size.

  • Annualized savings vs equivalent Aurora Instance covered by a 1-year No Upfront RI exceed $100.

  • The target Aurora instance type can handle the maximum observed connection count.


Expected Saving

Aurora Serverless v2 costs $0.12/ACU/hour. A database consistently running at 8 ACUs costs ~$840/month. An equivalent provisioned db.r6g.large with a 1-year No Upfront RI costs ~$120/month — savings of over 85% for stable workloads. The minimum savings threshold for this recommendation is $100/year annualized.


Operational Impact

  • Downtime: Migration requires creating a new provisioned Aurora cluster from a snapshot and cutting over the database endpoint. Downtime is expected during the endpoint switch.

  • Breaking changes: The database endpoint changes. Application connection strings must be updated after the cutover.

  • Recovery: Keep the Serverless cluster running in parallel until the provisioned cluster is fully validated before decommissioning.

  • ⚠️ Warning — max_connections: Provisioned instances have lower default max_connections than Serverless v2 at peak ACUs. Verify the target instance type can handle the current maximum connection count before migrating (see ACU-to-instance mapping in Archive).

  • ⚠️ Warning — No auto-scaling: Provisioned instances cannot scale dynamically. This recommendation is only appropriate for workloads with stable, predictable usage patterns. Significant traffic spikes may cause resource contention.


References

Did this answer your question?