Disabling PCI passthrough and vGPU support

If you want to stop using PCI devices on a node, you need to remove them from the configuration file, and then use this file to reconfigure the compute cluster.

Prerequisites

  • The compute cluster is reconfigured for PCI passthrough or vGPU support, as described in Enabling PCI passthrough and vGPU support.
  • Before disabling PCI passthrough or vGPU support for a host device, stop using this device inside virtual machines. To do this, stop the VMs that are currently using this device, and then change their flavor to the one without PCI passthrough and vGPU-related properties.

To disable PCI passthrough and vGPU for all devices on a node

  1. Create the configuration file with an empty device list:

    # cat config-empty.yaml
    - node_id: c3b2321a-7c12-8456-42ce-8005ff937e12
      devices: []
    - node_id: 1d6481c2-1fd5-406b-a0c7-330f24bd0e3d
      devices: []
  2. Reconfigure the compute cluster by using the new configuration file. For example:

    # vinfra service compute set --pci-passthrough-config config-empty.yaml

To disable PCI passthrough or vGPU for a particular device on a node

  1. Remove the information about the host device that you want to stop using from the configuration file. For example, to disable passthrough of the physical GPU with the 1b36:0100 VID and PID, remove these lines from config.yaml:

    - device_type: generic
      device: 1b36:0100
      alias: gpu
  2. Reconfigure the compute cluster by using the updated configuration file. For example:

    # vinfra service compute set --pci-passthrough-config config.yaml