Disk autoscaling

Cloud providers like AWS and GCP limit how frequently network-attached disks can be resized. In both cases, there is a multi-hour cooldown period between resizing operations. These volumes also typically do not support shrinking. PlanetScale disk autoscaling handles the automatic increasing and decreasing of disk size beyond these AWS and GCP limitations. Disk autoscaling is enabled by default upon database creation. We provide three scaling modes designed to optimize cost and performance while maintaining data availability:

• In-place growth mode — This is the default scaling mode that expands storage capacity by resizing existing volumes directly, without requiring failovers or connection disruption. This method leverages AWS EBS’s native resize capability.
• Surge growth mode — Surge growth creates new volumes with larger capacity and orchestrates failover to the new storage, circumventing AWS EBS resize limitations.
• Shrink mode — In shrink mode, the autoscaler reduces storage capacity for underutilized volumes. This can help to optimize costs after a surge event.

Autoscaling will grow the cluster’s storage without requiring you to make any configuration changes. The new additional space will become available as soon as the scaling action has completed.

​

Autoscaling thresholds and behaviors

When enabled, disk surge and shrink autoscaling will kick in when your disk utilization reaches the following thresholds:

​

Surge mode thresholds

Automatic disk growth activates when disk utilization reaches these thresholds:

• 70% for disks smaller than 4 TiB
• 25% for disks larger than or equal to 4 TiB

​

Shrink mode thresholds

Automatic shrinking activates when disk utilization falls below these thresholds:

• 12.5% for disks smaller than 1 TiB
• 15% for disks between 1 TiB and 2 TiB
• 25% for disks larger than 2 TiB

For example, if you have a 200 GiB disk allocation and are only using 20 GiB, we mark the disk for shrinking because it’s below the existing 12.5% threshold. We wait until the 3 day cooldown period has passed, and then we shrink it.

​

Key disk autoscaling behavior:

• Cluster storage can only scale once in a multi-hour period
• Cluster storage scales proportionally based on current size
  • Smaller disks receive larger percentage increases, while larger disks receive smaller percentage increases
• All disks grow by a minimum of 50% when autoscaling occurs
• No shrink is triggered if a disk was recently autoscaled. We wait for up to 3 days until we start observing shrink conditions.
• If you need to scale cluster storage by more than 200% within 24 hours, manually scale disk size ahead of time
• Autoscaling will not scale past your configured Storage limits

​

Surge growth mode

Our surge growth creates new volumes with larger capacity and orchestrates failover to the new storage, circumventing AWS EBS resize limitations. When our disk auto-scaler is able to spread out disk scale-up sufficiently, no downtime is required to scale the disks. When data growth is rapid, the auto-scaler may need to complete a surge resize to support the writes. In this case, PlanetScale creates brand new, larger network-attached storage volumes to replace the old ones. Surge growth causes a brief failover event that severs existing database connections. Applications must handle connection recovery. If the surge autoscaler is able to complete the resize before your disk fills, downtime will be minimal for growing the disk (seconds). If your disk fills before the new disks are ready, you will experience a longer period of downtime. We make every effort to keep your network-attached storage disk from filling, but it’s important for the database administrators to pay close attention to storage and take manual intervention when necessary.

​

Shrink mode

In shrink mode, the disk autoscaler reduces storage capacity for underutilized volumes through surge operations, as AWS EBS does not support in-place volume shrinking. This mode optimizes costs by rightsizing storage while also preventing oscillation between grow and shrink operations. When the shrink condition is detected, we have a 3 day cooldown period before we start the actual shrinking action for automatic shrinking. If a database has a maintenance window, disk shrinking will only occur during that window. Shrink operations cause a brief failover event that severs existing database connections. Applications must handle connection recovery.

​

Enable or disable disk autoscaling

Disk autoscaling is enabled by default upon database creation. Both of these options can be configured by going to “Cluster configuration” > “Storage” > and clicking the “Enable autoscaling” checkbox. You also have the option to only automatically scale up.

​

Storage limits

The storage limit sets the maximum amount of storage that can be allocated to your database cluster through autoscaling. This acts as a ceiling to prevent unlimited storage growth and helps control costs. When autoscaling is enabled, your storage can grow from the minimum disk size up to the storage limit you specify. The storage limit should be set higher than your initial disk size to allow for growth while providing a reasonable upper bound for your storage costs.