Scaling

Set replicas, turn on autoscaling, and let idle services sleep to zero.

Last updated October 23, 2026

View as Markdown Open in Claude Open in ChatGPT

Scale a service by hand with replicas, automatically with autoscaling, or all the way down to zero when it’s idle.

Replicas & resources#

In the service's Settings, the Resources card sets the replica count and what each replica gets. Changes apply on the next deploy.

Replicas: How many copies run, from 1 to 10 (your plan caps the maximum).
Memory (MB): Memory per replica, in steps of 128 MB. Your plan sets the maximum.
CPU (vCPU): CPU per replica, in steps of 0.1 vCPU. Your plan sets the maximum.

Autoscaling#

On Pro and higher, turn on autoscaling to add and remove replicas based on CPU load. Set the minimum and maximum replicas and a CPU target from 10 to 95%. Gheima then adds or removes replicas within those bounds to hold the target.

Scale to zero (sleep)#

Set "Sleep after (minutes idle)" to scale a service to zero when it gets no traffic. It wakes on the next request. Set it to 0 to turn sleep off. On the Free plan, new web services sleep after 30 minutes idle by default.

Sleep suits staging and low-traffic services. A sleeping service runs no replicas, so it uses no CPU or memory.

Pause & resume#

To stop a service entirely, click "Pause" in the Danger zone. It stops every replica. Resume the service to bring it back with its previous replica count.

Limits at a glance#

Memory: 128 to 8192 MB per replica, up to your plan's limit.
CPU: 0.1 to 8 vCPU per replica, up to your plan's limit.
Replicas: 1 to 10, capped by your plan.
CPU target: 10 to 95% for autoscaling.
Sleep after: 0 to 1440 minutes (0 turns sleep off).

Autoscaling reacts to CPU load only. Memory and CPU per replica stay fixed at the values you set.

✦ Your service runs at the capacity you chose: fixed replicas, CPU-driven autoscaling, or zero while idle.