You will not be billed if your database is configured with 0 vCPU as a minimum and is in idle status.
Serverless SQL Databases
You are billed both for the compute resources provisioned and the storage used. You can see the cost estimate in the Database creation wizard.
Compute: You are billed based on compute resources (vCPU and RAM) provisioned to handle your workload. Provisioned resources evolve according to the autoscaling parameter and stay between the minimum and maximum thresholds defined for your database. For each vCPU provisioned, 4 GB of RAM will be allocated.
Storage: You are billed for the storage consumed by your database. Storage price is based on the size of your database (in GB) and includes 7 days of automated backups at no additional costs. Storage is always billed even if your database is in an active state or idle state.
The 7 days of backups are free. Backup creation and restore operations are considered regular database activity (including the automatic daily backups), so you will be billed according to the resources consumed during the restore operation.
For example, if the restore operation lasts one hour, you will be billed for one hour of RAM and vCPU usage.
Backup creation and restore operation duration depend on several factors, such as the total storage used, and index number and size.
Backup and restore operations for databases that are less than 10 GB usually last a couple of minutes, but they can take longer if you rely heavily on indexes or constraints, or other time-consuming restore operations.
Serverless SQL Database is compatible with most PostgreSQL-compatible clients and tools. Due to connection pooling support, some minor features are disabled or will not work as expected. Refer to the Known differences for more information.
Serverless SQL Database handles connection pooling seamlessly. The number of maximum simultaneous connections will scale with CPU and RAM. Refer to the Serverless SQL Databases overview for more information on technical specifications.
Refer to How am I billed? for more information.
Scaling an active database or waking up an idle database will be seamless for clients connected to the database as the connections will be maintained, but queries performed during these operations may experience a few seconds of delay.
You can avoid cold starts by provisioning a minimum of 1 vCPU or more.
The following features are currently not supported and will be implemented at a later time:
- Manual backups (you can still perform manual backups using SQL queries or via SQL administration tools such as pgAdmin)
- Several SQL commands detailed in the Known differences with default PostgreSQL documentation