Applies to SUSE AI Factory with NVIDIA

6 How do I manage AI workloads? #

SUSE AI Factory with NVIDIA helps you manage the lifecycle of workloads that you install on your Kubernetes cluster.

Screenshot showing a list of AI workloads

Figure 6.1: List of AI workloads #

6.1 How do I manage application workloads? #

From the main left panel, select SUSE AI Factory › Workloads.
Select the Manage button for the application workload you want to manage.
In the Manage wizard, you can update the application version, destination cluster and Helm chart details.
Confirm with Save.

6.2 How do I upgrade application workloads? #

Select the Manage button for the application workload you want to upgrade.
On the Basic Information screen, change the target version number.
Finish the wizard and the workload upgrade starts automatically.
Figure 6.2: Upgrade of application workloads #

6.3 How do I upgrade blueprint workloads? #

From the main left panel, select SUSE AI Factory › Workloads.
Select the Upgrade button for the blueprint workload you want to upgrade. Once the rollout completes, the workload returns to the Running state at the new version.
Figure 6.3: Upgrade of blueprint workloads #

6.4 How do I roll back a blueprint workload? #

If an upgrade does not complete successfully, you can return the workload to the previous blueprint version.

From the main left panel, select SUSE AI Factory › Workloads.
Select the Roll Back button for the blueprint workload you want to revert.
Review the target version in the confirmation dialog and confirm the rollback.

Note: Rollback target

A blueprint version becomes a rollback target only after it has been fully rolled out across every target cluster of the workload. A workload that has never reached this state has no rollback target, and the Roll Back button is not available.

6.5 How do I retry a failed deployment? #

If a deployment fails for a transient reason, such as a registry that was temporarily unreachable, you can repeat it without changing the blueprint version.

From the main left panel, select SUSE AI Factory › Workloads.
Select the Retry button for the failed blueprint workload.

Note: Supported workload types

The Upgrade, Roll Back and Retry operations apply to blueprint workloads only. Application workloads reject these operations. To change an application workload, use Manage as described in Section 6.1, “How do I manage application workloads?”.

6.6 How do I inspect the status of a workload? #

A workload that is not in the Running state reports its status for each component and each pod, including while a deployment is still in progress or failing. Use this detail to identify the component that is blocking the rollout before you retry or roll back.

From the main left panel, select SUSE AI Factory › Workloads. You will see the underlying pod status in the status column in the table.
To inspect the underlying resources in Rancher Prime, select Open in Rancher for the workload. For an application workload, this opens the Helm release; for a blueprint workload, it opens the namespace. For a workload that targets multiple clusters, select the cluster first.

6.7 How do I delete workloads? #

From the main left panel, select SUSE AI Factory › Workloads.
Select the Delete button for the workload you want to uninstall, and confirm the deletion.

Note: Deleting workloads outside the UI

Deleting an AIWorkload resource with kubectl or through GitOps uninstalls the underlying Helm chart only when a Rancher Prime API token is configured in the Rancher API Access section of SUSE AI Factory › Settings.

Without a token, the deletion is intentionally held rather than leaving the Helm release behind, and the workload reports that it is waiting for the uninstall. To complete such a deletion, either configure a token or uninstall the application from the Rancher Prime Apps page.