Absorb workload spikes without permanently provisioning the peak size
Variable traffic
Neon · Const
Autoscaling
Neon Autoscaling adjusts compute resources up or down as database load changes. Set minimum and maximum CU bounds for a primary compute or read replica; scaling occurs without manual intervention or restarts.
How it works
Neon autoscaling changes a compute's allocated resources according to current workload. Configure a minimum and maximum CU for a primary compute or read replica; Neon adjusts capacity within those bounds without requiring a restart or manual resizing.
Key points
- Autoscaling increases or decreases the compute resources assigned to match current workload demand.
- Capacity changes happen without manually resizing or restarting the database.
- Both primary computes and read replicas can use autoscaling.
- The configured CU range establishes the lower and upper bounds for automatic resizing.
Common mistakes
- The difference between the configured maximum and minimum cannot exceed 8 CU.
Example: Configure an autoscaling range
neonctl compute update --autoscaling-min-cu 0.25 --autoscaling-max-cu 4Sets a compute's autoscaling range from 0.25 to 4 CU.
When to use it
Run at a smaller baseline and expand only when demand increases
Cost control
Let a read replica adjust capacity for changing query volume
Read scaling
Interview questions
QWhat does Neon autoscaling do?
Autoscaling changes compute capacity in response to current database load. A compute is configured with minimum and maximum CU boundaries, and Neon scales within that range without a restart.
Follow-up: How would you constrain cost while handling bursts? Set a low minimum for baseline usage and a maximum that covers expected spikes, while ensuring the range stays within Neon’s permitted spread.
QWhich compute types support autoscaling?
Autoscaling can be enabled for both primary computes and read replicas. The range is configured per compute through the Edit compute controls.
Follow-up: What should be configured separately for a read-heavy workload? Use a read replica with its own autoscaling range so read capacity can respond independently.
QWhat limits apply to an autoscaling range?
The supported compute-size range is 0.25 to 16 CU, and the maximum difference between the minimum and maximum is 8 CU. For example, 0.25–4 CU is valid, while 0.25–16 CU exceeds the permitted spread.
Follow-up: Can a compute use 16 CU as its maximum with autoscaling? Yes, provided its minimum is at least 8 CU, because the maximum-minus-minimum difference must not exceed 8 CU.
Related
computeread replicacompute sizescale to zeroEdit compute