Scaling Render Services

You can run multiple instances of a web service, private service, or background worker to handle additional load. For services that receive incoming traffic, Render load-balances that traffic evenly across all running instances:

Each instance of a scaled service uses the same compute plan and is billed accordingly.

Scaling limits

Render Workflows uses a different scaling mechanism with much higher limits.

One workflow can orchestrate thousands of concurrent long-running tasks, with a per-minute compute limit for new runs. Learn more.

Each service's maximum instance count is determined by its compute plan:

If you need to scale a service beyond these limits, please reach out.

Scaling methods

Render supports two scaling methods: manual scaling and autoscaling:

Manual scaling

You can manually scale your service to any fixed number of instances, up to its maximum instance count.

1. In the Render Dashboard, open your service's Compute page and scroll down to the Manual Scaling section.

   

2. Drag the slider to the desired number of instances, or enter a value in the text box.

3. Click Save Changes.

Render immediately provisions or deprovisions instances as needed to match the new instance count.

Manual scaling events appear in the timeline on your service's Events page:

Autoscaling

Autoscaling requires a Pro workspace plan or higher.

For details, see the pricing page.

Render can automatically scale your service up and down based on CPU and/or memory utilization targets that you specify. This helps you handle periods of high traffic while also minimizing compute costs.

Enable autoscaling for your service from its Compute page in the Render Dashboard:

1. Use the slider to set your desired minimum and maximum instance count, or enter a value in each text box.

   • Render always keeps your instance count within the specified range, even if resource utilization is significantly below or above your specified target.

2. Scroll down to set your target CPU and/or memory utilization:

   

   Enable one or both of the toggles and set your target utilization percentage(s).

   If you enable neither toggle, autoscaling is disabled for the service.

3. Click Save Changes.

Render begins monitoring resource utilization and automatically scales your service up or down as needed based on your specified targets.

Autoscaling events appear in the timeline on your service's Events page:

How autoscaling works

Render periodically calculates average resource utilization across all instances of your autoscaled service. Using that value (current_util), Render determines whether to scale your service based on the following formula:

If new_instances doesn't equal current_instances, Render scales your service up or down to the new instance count.

Render waits a few minutes before scaling a service down.

If utilization rises again during this period, Render does not scale the service down. This minimizes unnecessary scaling actions during periods of "spiky" usage.

Render always immediately scales a service up to handle increased load.

Example 1: Scaling up

In this scenario, Render immediately scales the service up from 2 instances to 3.

Example 2: Scaling down

In this scenario, Render waits a few minutes, then scales the service down from 5 instances to 2 if memory utilization remains low.

If you set targets for both CPU and memory utilization, Render calculates new_instances based on each and uses the larger result.

Billing for scaled services

Billing for a scaled service is based entirely on compute usage, which is prorated by the second. There is no additional cost for performing a scaling action.

Here are some example scenarios:

See your exact compute usage for the month on your Billing page. You can also review your invoice history.

Application considerations

• Services with an attached persistent disk cannot scale to multiple instances.

• You can update your service's scaling configuration programmatically via the Render API.

  • Separate endpoints are available for manual scaling and autoscaling.

• If you configure both manual scaling and autoscaling for a service, Render enables autoscaling and ignores the manual scaling configuration.

Horizontal vs. vertical scaling

The sections above describe Render's support for horizontal scaling, where you adjust a service's number of running instances.

In contrast, vertical scaling refers to adjusting a service's compute resources (CPU and RAM). You vertically scale a service by changing its compute plan.

When to use each

• Scale horizontally to handle a higher number of simultaneous operations (such as incoming requests).
• Scale vertically if each single operation requires additional RAM or CPU to run efficiently.
  • For particularly resource-intensive operations, offload to a workflow or background worker to keep your web services responsive.

Horizontal scaling usually occurs much more frequently than vertical scaling. An autoscaled service might change its instance count multiple times per day, whereas you might upgrade its compute plan once in a year.