I was using a db.t3.medium Aurora instance for my development environment, but with CPU utilization hovering around 10%, it was significantly underutilized.
After comparing instance costs, I discovered that if the Aurora Capacity Unit (ACU) usage stayed near the minimum capacity (0.5), Aurora Serverless v2 would be more cost-effective. So, I decided to switch to Serverless v2.
Instance Pricing (Aurora Standard, Tokyo):
-
db.t3.medium: $0.125/hour - Serverless v2: $0.15/ACU
Unexpected Cost Increase
After switching instances, I monitored the ServerlessDatabaseCapacity metric to track ACU usage. Initially, it spiked close to 2 ACUs but soon stabilized around 0.5 ACU, just as expected.
"Great, the costs should be lower!" I thought, but when I checked Cost Explorer, the daily cost had mysteriously increased by about $7.
Identifying the Cause
Grouping costs by "Usage Type" in Cost Explorer revealed that APN1-RDS:Proxy-ASv2-Usage accounted for nearly $10/day.
Before switching to Serverless v2, this item appeared as APN1-RDS:ProxyUsage. After the change, it showed up as APN1-RDS:Proxy-ASv2-Usage.
Checking the official documentation, I found that RDS Proxy for Aurora Serverless v2 incurs a minimum charge for 8 ACUs.
Source: Amazon Relational Database Service Proxy pricing
At the time, we were transitioning from a "Lambda + RDS Proxy + Aurora" setup to an "ECS + Aurora" architecture and had kept the RDS Proxy running. This was the root cause of the cost spike.
Cost Comparison
-
APN1-InstanceUsage:db.t3.medium: ~$3.00/day -
APN1-Aurora:ServerlessV2Usage: ~$2.21/day -
APN1-RDS:Proxy-ASv2-Usage: ~$10.00/day
Comparing instance costs alone, Serverless v2 was about $1 cheaper. However, after removing the RDS Proxy and running Aurora Serverless v2 independently, we optimized the costs significantly.
Bonus Tip: Further Cost Optimization with Aurora Versions
We were using Aurora MySQL version 3.07.0, where the minimum ACU is 0.5. Starting from version 3.08.0, the minimum ACU can be set to 0, allowing for even greater cost savings during idle periods.
Top comments (0)