#Uninstall

The sections are organized by product: Alauda Build of HAMi, Alauda Build of HAMi Ascend Device Plugin, and optional HAMi WebUI. For HAMi-managed Ascend, complete the Ascend Device Plugin section before executing the HAMi core removal steps.

Before removing a component that owns device allocation, stop or migrate workloads that use its resource keys. Delete or change controller-managed resources through their owning ACP plugin, OperatorHub subscription, or custom resource. Do not delete only a generated Pod or DaemonSet and assume the product is uninstalled.

#Uninstall Alauda Build of HAMi

For HAMi-managed Ascend, complete the Alauda Build of HAMi Ascend Device Plugin section below before executing these HAMi core removal steps.

1. Stop or migrate workloads that request HAMi resource keys. Existing Pods do not block the uninstall and might continue running, but restarted, recreated, or scaled workloads cannot obtain HAMi resources after the corresponding device plugin is removed.

2. Go to Administrator -> Marketplace -> Cluster Plugins, switch to the target cluster, find Alauda Build of HAMi, and uninstall the cluster plugin.

3. Confirm that the HAMi scheduler, webhook, and device-plugin workloads created by the plugin are removed.

4. Remove HAMi selector labels from nodes that no longer run HAMi.

   For HAMi NVIDIA:

   kubectl label nodes <nodeid> gpu-

   For HAMi Ascend backends:

   kubectl label nodes <nodeid> ascend-

5. Delete ACP resource metadata ConfigMaps only when no installed accelerator product or workload still uses the corresponding resource keys.

   For HAMi NVIDIA:

   kubectl -n kube-public delete configmap cf-crl-hami-nvidia-gpualloc
   kubectl -n kube-public delete configmap cf-crl-hami-nvidia-gpucores
   kubectl -n kube-public delete configmap cf-crl-hami-nvidia-gpumem
   kubectl -n kube-public delete configmap cf-crl-hami-config

   For HAMi Ascend vNPU:

   kubectl -n kube-public delete configmap \
     cf-crl-hami-ascend-vnpu \
     cf-crl-hami-ascend-vnpu-memory \
     cf-crl-hami-ascend-vnpu-config

   Delete only the ConfigMaps created for this deployment. Their names can differ when custom resource metadata names were used.

#Uninstall Alauda Build of HAMi Ascend Device Plugin

If the cluster uses HAMi-managed Ascend allocation, uninstall Alauda Build of HAMi Ascend Device Plugin before uninstalling Alauda Build of HAMi.

1. Stop or migrate workloads that request HAMi-managed Ascend resources.

2. Delete the HAMiAscendDevicePlugin custom resource instance. The default instance name is hami-ascend-device-plugin.

   kubectl -n <hami-namespace> delete hamiascenddeviceplugin.hami.io hami-ascend-device-plugin

   The instance uses the following resource identity:

   apiVersion: hami.io/v1alpha1
   kind: HAMiAscendDevicePlugin
   metadata:
     name: hami-ascend-device-plugin
     namespace: <hami-namespace>

3. Confirm that the device-plugin Pod created by Alauda Build of HAMi Ascend Device Plugin has been removed from the target Ascend nodes.

4. Go to Administrator -> Marketplace -> OperatorHub, switch to the target cluster, find Alauda Build of HAMi Ascend Device Plugin, and uninstall the operator bundle.

5. Confirm that the operator workload has been removed:

   kubectl get pods -A | grep hami-ascend-device-plugin

Do not delete only the generated hami-ascend-device-plugin DaemonSet. The HAMiAscendDevicePlugin custom resource can cause the operator to recreate it.

If the native Ascend Device Plugin was previously stopped on these nodes, restore it only when the nodes should return to the direct Ascend NPU allocation path. Follow Switch the Ascend Device Plugin Owner; do not restore it by applying a standalone DaemonSet from old HAMi documentation.

#Uninstall Alauda Build of HAMi-WebUI

1. Go to Administrator -> Marketplace -> Cluster Plugins, switch to the target cluster, find Alauda Build of HAMi-WebUI, and uninstall the cluster plugin.

2. Confirm that the HAMi-WebUI Pod and Service are removed from the namespace selected by the package. The standard package uses cpaas-system:

   kubectl -n cpaas-system get pods,service | grep hami-webui

Removing HAMi-WebUI does not remove the HAMi scheduler, device plugins, or running HAMi workloads. It removes only the optional WebUI and its access path.

Do not delete the platform monitoring BasicAuth Secret used to generate the WebUI Authorization value. The Secret belongs to the monitoring service and can be shared by other monitoring clients.