Create or version a compute config
This page provides an overview of creating and versioning reusable compute configs using the Anyscale console. You can use compute configs stored in your Anyscale cloud when deploying Anyscale workspace, jobs, or services. See Define a Ray cluster.
To prevent changing behaviors for existing workload definitions, you can't edit or update compute configs registered to Anyscale. The following table provides a high-level overview of options for registering compute configs:
Process | Description |
---|---|
Create a compute config | Create a new compute config when you have a new workload or use case and want to have a compute config for use across multiple Ray clusters. |
Create a new version of a compute config | Version an existing compute config for small updates or changes, such as during interactive development or troubleshooting production workloads. |
Modify and save a compute config for a workspace | Modify the compute settings for a workspace during interactive development, then optionally save this custom configuration to launch other Anyscale workloads. |
If you use the console to create a new Anyscale workspace, you can select an existing compute config or define a custom compute config. See Manage workspaces.
Your Anyscale cluster must have a head node. Worker nodes are optional. You configure worker nodes on Anyscale as groups of machines.
For an overview of Ray clusters, see the Ray docs on key Ray cluster concepts.
You can also create and modify compute configs using the CLI or SDK. See the following pages:
- CLI: Compute Config CLI
- SDK:
anyscale.compute_config.create
- Compute config model:
ComputeConfig
Create a compute config
Complete the following steps to create a new compute config:
- Log in to the Anyscale console.
- Click Configuration > Compute configs.
- Click + Create.
- In the Name field, enter a name for your compute config.
- To configure your head node, click the default entry under Head node, for example 8CPU-32GB.
- Use the Instance type field to change the instance type.
- Click Advanced config to set additional JSON-formatted options. See Node config.
- Click + Add worker nodes to define a group of worker nodes. A default worker group configuration appears, for example 8CPU-32GB.
- To configure your worker group, click the default entry.
- Use the Instance type field to specify the instance type for the group.
- Set the Autoscaling field to Enabled to configure scaling behaviors. See Worker nodes scaling config.
- Click Advanced config to set additional JSON-formatted options. See Node config.
- Add and configure additional worker groups as desired.
- Click Advanced settings to view cluster-wide configuration options. See Cluster config advanced settings.
- Click Create.
Create a new version of a compute config
Complete the following steps to create a new version of an existing compute config:
- Log in to the Anyscale console.
- Click Configuration > Compute configs.
- Click the name of the desired compute config.
- Click + Create new version to start with the most recent version, or click the desired version from the list.
- Modify the compute config settings as desired.
- Click Create.
The options to create a new version are identical to those for creating a new compute config, but you can't change the name.
Modify and save a compute config for a workspace
You can use the Anyscale console to modify the compute config for a workspace and optionally save it as a new compute config. Complete the following steps:
Modifying an existing compute config always results in a Custom configuration, which you can optionally save. You can't create a new version of an existing compute config using this flow.
- Navigate to the Anyscale console.
- Click Workspaces.
- Click the name of a workspace.
- In the top right corner, click ... and select Manage cluster.
The Compute resources panel slides open from the right side and displays the current compute config for your workspace. You can modify, add, or delete compute configurations by completing any of the following actions:
- To change the compute config for your workspace, click the edit icon
next to the name of your compute config.
- To change the settings for the head node or worker nodes in your cluster, click the edit icon
for the desired node. Click Save to commit changes to your current workspace cluster.
- You must disable Auto-select worker nodes to edit worker group configurations.
- To delete a worker group from your cluster, click the X for the desired worker group.
- To add new worker node groups, click the + Add worker nodes button. Click Save to commit changes to your current workspace cluster.
- To view and modify cluster-wide settings, click Advanced settings. Click Save to commit changes to your current workspace cluster.
Click the to save your current configuration to make it available to all users and workloads in your Anyscale cloud. You must provide a unique name for your compute config.