Absorb demand spikes without manually resizing compute.
Variable production traffic
Neon · Const
Autoscaling range
An autoscaling range defines the minimum and maximum Compute Units for a Neon compute. Neon scales within those bounds as workload changes, subject to an 8 CU maximum spread.
How it works
An autoscaling range sets the minimum and maximum compute sizes for a Neon compute. Neon adjusts resources within those boundaries according to workload, while the difference between the maximum and minimum cannot exceed 8 CU.
Syntax
neon branches add-compute <branch> --cu <min>-<max>Parameters
| Name | Type | Description |
|---|---|---|
min | CU | Smallest compute size allowed by the range, measured in Compute Units. |
max | CU | Largest compute size allowed by the range, measured in Compute Units. |
Returns
A compute configuration that automatically scales between the specified minimum and maximum CU values.
Key points
- The range establishes the two compute-size boundaries used for automatic scaling.
- Scaling responds to workload but remains inside both configured limits.
- Supported autoscaling sizes run from 0.25 CU through 16 CU.
- Compute Units provide an approximate memory mapping of 4 GB RAM per CU.
- The spread between the configured minimum and maximum is capped at 8 CU.
Common mistakes
- The maximum-minus-minimum spread is limited to 8 CU; a 1 CU minimum therefore permits at most a 9 CU maximum.
- The range can be configured for both primary computes and read replicas.
Example: Configure an autoscaling range
neon branches add-compute main --cu 0.5-2Sets the main branch compute to autoscale from 0.5 CU to 2 CU.
When to use it
Use a small lower bound while permitting temporary growth.
Development and testing
Apply independent scaling boundaries to read replicas.
Read-heavy architectures
Interview questions
QWhat does an autoscaling range control?
An autoscaling range is the lower and upper compute-size boundary for a Neon compute. Neon scales between those values based on current workload, never going below the minimum or above the maximum.
Follow-up: What constraint applies to the two bounds? The maximum cannot exceed the minimum by more than 8 CU.
QWhat compute sizes and memory mapping does autoscaling use?
The documented autoscaling sizes span 0.25 CU through 16 CU. One CU corresponds to approximately 4 GB of RAM, so the selected range also determines the approximate memory envelope.
Follow-up: Can every range across that span be selected? No; the maximum-minus-minimum difference is limited to 8 CU.
QWhere can autoscaling be enabled?
Autoscaling applies to a primary compute or a read replica. It is configured by opening the compute's Edit compute drawer and setting the minimum and maximum sizes.
Follow-up: How can it be configured outside the console? The computes documentation shows updating the endpoint with minimum and maximum autoscaling CU fields via the Neon API.
Related
Autoscalingcomputeread replicascale to zero