Vertex Macro｜Amarathon 2025 Recap 09：Five Hard Lessons from Five Years of So-Called Serverless Databases

Five Hard Lessons from Five Years of So-Called Serverless Databases

● Serverless is not Serverless. Or Vice Versa?

Introduction of serverless databases on Amazon Web Services

● Aurora Serverless v1 GA (2018-08)

● DynamoDB On-Demand (2018-11)

● Amazon Timestream (2020-09)

● Aurora Serverless v2 Preview (2020-12)

● Aurora Serverless v2 GA (2022-04)

● Redshift Serverless (2022-07)

● Amazon Neptune Serverless (2022-10)

● ElastiCache Serverless (2023-11)

● Amazon DSQL (2025-05)

● DocumentDB Serverless (2025-07)

What about...

● Amazon S3

● Amazon SQS

● Amazon Route 53?

Storage is Underrated

● Aurora Serverless v2 scales instantly to support even the most demanding applications, delivering up to 90% cost savings compared to provisioning for peak capacity

● USD/Month

● 2022: 3336 CPU (RI), 1960 Storage (GP), 2274 Backup

● 2023: 4682 CPU (RI), 2387 Storage (GP), 3300 Backup

● 2024: 6006 CPU (RI), 2994 Storage (GP), 5840 Backup

● 2025: 6350 CPU (RI), 3755 Storage (GP), 6952 Backup

● Aurora Serverless: min 0.5 ACU

● Aurora Provisioned: db.r8g.large

● RDS for MySQL: 400 GB + db.r8g.large

How Long?

● Aurora Serverless: 3-4 seconds

● Aurora Provisioned: 2-3 seconds

● RDS for MySQL: 4-5 minutes

Predicting Costs is More Challenging

● "If you do not know how your workload performs on a serverless DB, forecasting ACU by using the baseline of a provisioned cluster is entirely useless."

● "Aurora DSQL is now Generally Available. What’s it cost? Nobody knows, especially Amazon Web Services. You get charged per DPU, which equates in the documentation to 'screw you, benchmark your workloads and find out for yourself.' Of all the pricing strategies from which to choose, I didn’t expect Amazon Web Services to pick 'completely give up.' My Aurora customers are... displeased." Corey Quinn

Compatibility Might Be a Challenge

● Compatibility?

● Wire (or Client Protocol)

● SQL / Query Language

● Feature / Behavior

● Operational / Ecosystem

● Version

● DSQL, Dynamo DB, Aurora Serverless (*)

● Both "Serverless" databases

● Horizontal vs Vertical is the Question

● Elasticity is not serverless

● "If the database still doesn't scale down to the minimum capacity configured, then stop and restart the database to reclaim any memory fragments that might have built up over time. Stopping and starting a database results in downtime, so we recommend doing this sparingly." (Amazon Web Services doc)

● Leverage serverless database, work on elasticity, and understand what runs under the hood