Serverless Relational Databases¶
TL;DR / when to use this: For hobby projects, serverless databases cost far less than an always-on RDS instance. The right pick depends on workload shape: variable workloads favor usage-based billing; intermittent workloads favor auto-pause; always-on favors committed use.
Comparison¶
| Service | Compute pricing | Storage pricing | Ideal for |
|---|---|---|---|
| Amazon Aurora Serverless | Per ACU per second | $0.10 / GB-month | Variable workloads, MySQL/PostgreSQL compatibility |
| Azure SQL Database Serverless | Per vCore per second (~$0.52/vCore-hour Gen5) | Per GB | Intermittent workloads with auto-scaling and pausing |
| Google Cloud SQL | On-demand / committed use (~$0.0825/hour for 1 vCPU) | $0.17 / GB-month SSD | Managed relational DBs with automated scaling |
| PlanetScale | Free tier; usage-based beyond | Included in usage pricing | MySQL-compatible with serverless scaling and modern dev features |
| CockroachDB Serverless | Free tier; usage-based beyond | Included in usage pricing | PostgreSQL-compatible distributed SQL, high availability, global scaling |
Notes from running these¶
- PlanetScale was the go-to for hobby projects — free tier plus MySQL compatibility, though branching workflows matter more than raw pricing.
- CockroachDB Serverless is the pick when you want Postgres semantics with HA without managing nodes.
- Watch the cold-start/pause behavior on Azure and Aurora — a paused instance bills nothing but adds latency on wake.
See also¶
- Oracle Always Free Tier — a free VPS that often beats serverless DB pricing for home projects
- Blog post: Serverless relational databases