Enable Autoscaling in Neon

This guide demonstrates how to enable autoscaling in your Neon project and how to visualize your usage.

Enable autoscaling for a compute

You can edit an individual compute to alter the compute configuration, which includes autoscaling.

To edit a compute:

1. In the Neon Console, select Branches.

2. Select a branch.

3. On the Computes tab, identify the compute you want to configure and click Edit.

4. On the Edit compute drawer, select Autoscale and use the slider to specify a minimum and maximum compute size.

   Neon scales the compute size up and down within the specified range to meet workload demand. Autoscaling currently supports a range of 1/4 (.25) to 16 vCPUs. One vCPU has 4 GB of RAM, 2 vCPUs have 8 GB of RAM, and so on. The amount of RAM in GB is always 4 times the number of vCPUs. For an overview of available compute sizes, see Compute size and autoscaling configuration. Please note that when the autoscaling maximum is > 10, the autoscaling minimum must be ≥ (max / 8).

   note

   You can configure the scale to zero setting for your compute at the same time. For more, see Scale to Zero.

5. Click Save.

Configure autoscaling defaults for your project

You can configure autoscaling configuration defaults for your project so that newly created computes (including those created when you create a new branch or add read replica) are created with the same autoscaling configuration. This will save your from having to configure autoscaling each time, assuming you want the same settings for all of your computes.

To configure autoscaling defaults:

1. Navigate to your Project Dashboard and select Settings from the sidebar.
2. Select Compute.
3. Select Change to open the Change default compute settings modal.
4. Use the slider to specify a minimum and maximum compute size and Save your changes.

The next time you create a compute, these settings will be applied to it.

Monitor autoscaling

From the Neon Console, you can view how your vCPU and RAM usage have scaled for the past 24 hours. On the Project Dashboard page, navigate down the page to the Monitoring section.

Some key points about this Autoscaling graph:

• Allocated refers to the vCPU and memory size provisioned to handle current demand; autoscaling automatically adjusts this allocation, increasing or decreasing the allocated vCPU and memory size in a step-wise fashion as demand fluctuates, within your minimum and maximum limits.
• VCPU Usage is represented by the green line
• RAM usage is represented by the blue line.
• A re-activated compute scales up immediately to your minimum allocation, ensuring adequate performance for your anticipated demand.

Place your cursor anywhere in the graph to get more usage detail about that particular point in time.

See below for some rules of thumb on actions you might want to take based on trends you see in this view.

Start with a good minimum

Ideally, for smaller datasets, you want to keep as much of your dataset in memory (RAM) as possible. This improves performance by minimizing I/O operations. We recommend setting a large enough minimum limit to fit your full dataset in memory. For larger datasets and more sizing advice, see how to size your compute.

Setting your maximum

If your autoscaling graphs show regular spikes that hit your maximum setting, consider increasing your maximum. However, because these spikes plateau at the maximum setting, it can be difficult to determine your actual demand.

Another approach is to set a higher threshold than you need and monitor usage spikes to get a sense of where your typical maximum demand reaches; you can then throttle the maximum setting down closer to anticipated/historical demand. Either way, with autoscaling you only use what's necessary; a higher setting does not translate to increased usage unless there's demand for it.

The neon_utils extension

Another tool for understanding usage, the neon_utils extension provides a num_cpus() function that helps you monitor how the Autoscaling feature allocates compute resources in response to workload. For more information, see The neon_utils extension.