Skip to content

Uninstalling the RBLN NPU Operator

Removing the operator involves three steps: locate the Helm release, uninstall it, and then delete the CRDs scoped to the cluster and the namespace.

1. Locate the Helm release

List the releases in the operator namespace to confirm the exact release name. It may differ from rbln-npu-operator if you used --generate-name or supplied a custom name during installation:

1
2
3
$ helm list -n rbln-system
NAME                 NAMESPACE      REVISION    UPDATED                                 STATUS      CHART                            APP VERSION
rbln-npu-operator    rbln-system    1           2026-04-28 10:42:21.844639 +0900 KST    deployed    rbln-npu-operator-chart-0.3.4    v0.3.4

Use the value from the NAME column in the next step.

2. Uninstall the Helm release

1
$ helm uninstall <release-name> --namespace rbln-system

For the example above, run:

1
$ helm uninstall rbln-npu-operator --namespace rbln-system

The resources managed by the operator (DaemonSets, ConfigMaps, Services, RBAC objects) are tied to the parent CRs through Kubernetes ownerReferences. Therefore, when the RBLNClusterPolicy is deleted, related resources are cleaned up automatically. helm uninstall also removes the policy while uninstalling the release.

3. Delete the CRDs and namespace

CRDs are scoped to the cluster and remain after the Helm release is removed. Delete the CRDs and the operator namespace explicitly:

1
2
$ kubectl delete crd rblnclusterpolicies.rebellions.ai rblndrivers.rebellions.ai
$ kubectl delete namespace rbln-system

Warning

Deleting the CRDs also deletes every RBLNClusterPolicy and RBLNDriver custom resource in the cluster, and this action cannot be undone. Confirm that no other workloads still depend on these resources before running the command.