Alauda GPU Management

Uninstallation

Uninstall Alauda GPU Management when GPUs no longer need to be managed on the cluster, or before migrating from a standalone plugin delivery.

WARNING

Uninstalling the operator removes the managed GPU stack. Confirm that no GPU workloads depend on it, and plan a maintenance window if the driver was operator-managed.

TOC

Uninstall Alauda GPU ManagementUninstall Alauda Build of DCGM-Exporter

Uninstall Alauda GPU Management

  1. Remove the ClusterPolicy instance first, so the operator can deprovision the managed components (driver, container toolkit, device plugin, GPU Feature Discovery, DCGM-Exporter):

    kubectl delete clusterpolicy <name>

  2. Go to Administrator > Marketplace > OperatorHub (or Installed Operators), switch to the target cluster, find Alauda GPU Management, and uninstall it.

  3. In precompiled or upstream driver mode, the driver is removed with the managed components. In preinstalled-driver mode (driver.enabled=false), the host driver is left untouched and remains under its own lifecycle.

  4. If you added an ImageWhiteList for the driver or operand images, remove the entries you no longer need.

Uninstall Alauda Build of DCGM-Exporter

DCGM-Exporter is the NVIDIA GPU metrics component. When it is managed by Alauda GPU Management, it is removed with the ClusterPolicy. When it was installed as a standalone plugin — for example to monitor a HAMi NVIDIA GPU backend — uninstall it separately:

  1. If a GPU metrics dashboard resource was created for it, delete that resource.

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

  3. Remove the node label used to schedule the exporter, if one was added.

INFO

Do not remove a shared DCGM-Exporter instance that other workloads still depend on. If DCGM-Exporter is shared between Alauda GPU Management path and a HAMi backend, uninstall it only when no remaining consumer needs GPU metrics.