We are running a standard Aurora cluster, but our workloads are extremely spikey, causing us to either over-pay for idle capacity or suffer from performance lag during peaks. I’m looking at Aurora Serverless v2. How fast does it actually scale compared to v1, and can I use it in a mixed cluster?
3 answers
Aurora Serverless v2 is a massive improvement because it scales in "ACU" increments of 0.5, allowing for near-instant adjustment to traffic spikes. Unlike v1, v2 can be part of a standard Aurora cluster. You can have a provisioned primary for baseline traffic and a serverless reader that scales up during peak hours. This "Mixed Configuration" is excellent for cost-optimization. Regarding limitations, v2 supports Read Replicas, Multi-AZ, and Global Databases, making it a viable option for high-availability enterprise applications that previously relied on expensive, fixed-size provisioned instances.
How does the pricing of v2 compare to a small provisioned instance? Is there a "Minimum Capacity" that we have to pay for even when the database is completely idle at night?
You can set a "Maximum Capacity" limit on your v2 instances to prevent an unexpected traffic surge or a runaway query from blowing your monthly AWS database budget.
Setting that cap is essential for peace of mind. It allows us to benefit from the auto-scaling without the fear of an unlimited bill if something goes wrong with our application code.
There is a minimum capacity of 0.5 ACU, so you are always paying a small baseline fee. However, the value comes from the fact that it doesn't need to "re-provision" and reboot like v1 did. For us, the break-even point was reached because we no longer have to pay for an r6g.xlarge just to handle a two-hour daily spike. The v2 instance stays small most of the day and only scales up when needed. We've seen a 30% reduction in our total RDS spend while actually improving our application response times during heavy load.