This is an overview of actions in Cortex Fabric.
Starting with v6.3.0 development of actions independently from Skills is no longer supported as the best practice. CLI "actions" commands described on this page are only available for Actions developed separately from Skills and therefore, will no longer work for Skills created going forward. Likewise, the Console "Actions" tab in the Agents section only displays actions created independently.
At the core of every Skill is the action it takes to process data when it is invoked. Cortex supports two types of actions:
- Jobs: Used to process high volumes of data that would normally consume long-term memory if run in the foreground, as well as for running programs that require little to no user interaction and are required to run on a regular basis.
- Daemons: Web servers typically used for ML predictions and serving inquiries. Once started, a daemon runs indefinitely.
Jobs and daemons are deployed to Cortex and executed inside of Docker images.
NOTE: Use options
--project projectId or
--profile profileName if you are not authenticated to a specific project or profile context.
To identify the actions that have been deployed independently from Skills for a project run:
A list of actions is returned. Actions that are deployed as part of a Skill are NOT listed. You can use these names to execute other action monitoring commands.
Get Action details
To fetch the details of an action including its status run:
Actions deleted from CLI are removed from deployment and the catalog, but they remain in your system database. The
delete command sets the
isDeleted flag for all versions of the action.
If you delete an action that is being used by a Skill/Agent, the Skill/Agent will no longer run.
To delete an action run the following CLI command:
Define and Deploy Actions
Actions are defined as part of the Skill Definition process described here.
After they have been defined, Actions are saved/deployed as part of the Skill Definition described here in step 4.
The podspec patch file is created for Kubernetes jobs and daemons. It provides full customization of the deployed action resources, including allocation of memory and vcpu resources and support for reserved GPU nodes. The podspec patch is merged with and overwrites the Kuberenetes job or daemon spec created by Fabric during action deployment. The podspec file is saved locally as a json file.
To use the podspec option when you save/deploy an action, append the command with
Deploy with Additional K8S Resources
Optionally additional Kubernetes resource can be deployed along with a Skill's action.
When you save/deploy an action, use
-k8sResource option to pull in JSON or yaml files that define additional Kubernetes Resources for that are deployed with Skills using the Action.
The K8s option must be the last option specified in the command or the list of files must be followed with with "--".
Kubernetes Resources that you may configure and upload as .JSON or .YAML files that are saved locally.
Explanation, instructions and examples can be found at the links provided.
Types of K8S Resources
- Horizontal Pod Autoscaler (HPA)
- Pod Disruption Budget (PDB)
- Virtual Service
- Destination Rule
- Traffic Management
- Circuit breakers