K8S Launch Kit Clean

Clean a Network Operator Deployment

Read k8s-launch-kit-shared first for binary discovery, JSON output, and error handling.

Safety Boundary

Treat cleanup as an irreversible live-cluster mutation. Require explicit user authority to remove the deployment. Do not infer cleanup authority from a request to inspect, validate, troubleshoot, or explain the command.

Before running it:

1. Resolve the intended kubeconfig and show its current context.
2. Determine the intended Network Operator namespace.
3. Explain that all namespaced custom resources in that namespace and every known cluster-scoped Network Operator CR will be deleted.
4. Confirm whether the Helm release should be uninstalled or retained.

Use read-only checks when the target is not already explicit:

kubectl --kubeconfig <PATH> config current-context
helm --kubeconfig <PATH> list --all-namespaces --filter '^network-operator$'

Stop and ask for direction if the kubeconfig, cluster, namespace, or Helm retention choice remains ambiguous.

Run Cleanup

An AI agent uses JSON mode after the user authorizes the exact target:

l8k clean \
  --kubeconfig <PATH> \
  --network-operator-namespace <NAMESPACE> \
  --output json 2>/dev/null | jq .

Omit --network-operator-namespace only when the user expects l8k to resolve it from --user-config, ./cluster-config.yaml, an explicit --config-dir, or the nvidia-network-operator default. l8k does not trust in-cluster objects to select a destructive cleanup target. The same trusted local config supplies networkOperator.skipHelmChart; when true, l8k treats the Helm release as externally owned and retains it.

To retain the installed release and its chart-managed resources:

l8k clean \
  --kubeconfig <PATH> \
  --network-operator-namespace <NAMESPACE> \
  --keep-helm-chart \
  --output json 2>/dev/null | jq .

JSON mode auto-confirms. Do not invoke it until the read-only target checks and authorization are complete. l8k clean has no dry-run mode.

Deletion Semantics

The command:

• Deletes every namespaced custom-resource instance in the resolved operator namespace.
• Deletes all HostDeviceNetwork, IPoIBNetwork, MacvlanNetwork, NicNodePolicy, and NicClusterPolicy instances cluster-wide.
• Sends deletion requests to the complete namespaced and cluster-scoped set before monitoring any CR for finalizer completion, then uses the same delete-all-then-wait ordering for re-sweeps.
• Keeps controllers installed until CR deletion and finalizers complete.
• Uninstalls the network-operator Helm release last unless networkOperator.skipHelmChart or --keep-helm-chart retains it.
• Preserves the namespace, CRDs, Secrets not owned by Helm, local files, and namespaced custom resources elsewhere. Helm metadata and chart-managed resources are removed with the release.
• Treats already-missing resources and Helm releases as successful no-ops.

Do not manually strip finalizers as part of the normal workflow. If cleanup waits indefinitely, inspect the affected CR, operator pods, and events; request separate authority before forcing finalizer removal.

Verify the Result

Require .success == true, then report:

• .cleanup.namespace
• .cleanup.customResourcesDeleted
• .cleanup.helmReleaseRemoved
• .cleanup.keepHelmChart

cleanup.keepHelmChart is the effective policy from config plus the explicit flag, not merely whether --keep-helm-chart appeared on the command line.

If the Helm release was meant to be removed, verify it is absent with the same read-only helm list command. If it was kept, verify it remains deployed.