Right-sizing managed databases across RDS, Cloud SQL and Azure
7 min read · June 17, 2026 · TurboFinOps
Database instance sizing is sticky. Teams pick a class during planning — usually generous, “to be safe” — and never revisit it. Months later the workload is well understood and far smaller than the instance, but the bill never adjusted.
The under-utilization signal
Two metrics tell the story: CPU utilization and active connections. A managed database sitting at single-digit CPU with a handful of connections over a sustained window is over-provisioned for its real demand.
This holds across providers — AWS RDS, GCP Cloud SQL, and Azure SQL, PostgreSQL and MySQL flexible servers all expose the same shape of signal under different metric names.
Right-size, schedule or consolidate
For a steadily underutilized instance, step down one size and watch the headroom. For non-production databases, schedule them off outside business hours — a dev database rarely needs to run at 2am on a Sunday.
For a sprawl of tiny databases, consider consolidating onto fewer instances where isolation requirements allow.
Mind the safety rails
Databases are stateful, so change carefully: confirm storage and IOPS headroom, plan for the brief failover some resizes require, and stage non-production first.
The goal is to match the instance to observed demand plus a sensible buffer — not to run hot, just to stop paying for capacity that has never been used.
Frequently asked questions
- Is CPU enough to decide?
- Pair it with connections and, where you can, memory and IOPS. Low CPU with low connections over a representative window is a reliable right-sizing signal; a single spike is not.
- Does this cover Azure PostgreSQL and MySQL?
- Yes. Azure SQL, PostgreSQL and MySQL flexible servers, plus RDS and Cloud SQL, are all evaluated by the same underutilized-database rule.
See your own cloud waste in minutes
Connect AWS, Azure or GCP and get a read-only scan of your top savings opportunities — with verified savings receipts when you fix them.
Run a free cloud waste scan