Services or capabilities described in AWS documentation might vary by Region. To see the differences applicable to the AWS European Sovereign Cloud Region, see the AWS European Sovereign Cloud User Guide.

Scaling a development cluster

You scale a Ray cluster by changing the worker count directly, or by turning on autoscaling so Ray adds and removes workers with load.

Set a fixed worker count

In Studio, edit the cluster and change the replica count for a worker group. In a manifest, set replicas on the worker group.

workerGroupSpecs:
  - groupName: my-workers
    replicas: 4

Turn on autoscaling

Turn on the autoscaling toggle when you edit the cluster in Studio, then set a minimum and maximum for each worker group. In a manifest, set enableInTreeAutoscaling: true and the per-group bounds.

spec:
  enableInTreeAutoscaling: true
  workerGroupSpecs:
    - groupName: my-workers
      minReplicas: 1
      maxReplicas: 8

Ray scales each worker group between its minimum and maximum based on pending tasks. The minimum sets the idle footprint, and the maximum caps cost.