Skip to content

README.md

Latest commit

 

History

History
133 lines (117 loc) · 6.9 KB

README.md

File metadata and controls

133 lines (117 loc) · 6.9 KB
Raw

Windows Customize Pipeline

This Pipeline clones the DataVolume of a basic and generalized Windows 10, 11 or Server 2k22 installation and runs arbitrary customization commands through an unattend.xml after startup of the VirtualMachine. As an example a ConfigMap which installs Microsoft SQL Server Express and generalizes the VirtualMachine after (windows-sqlserver) and a ConfigMap that installs VSCode (windows-vs-code) are included. For basic setup after the first start of a customized VirtualMachine an example unattend.xml is included in the Pipeline's ConfigMap windows10-unattend, or windows11-unattend.

This example Pipeline can be used for running Windows 10, 11 or Server 2k22 (or others - not tested!). Always adjust Pipeline parameters for the Windows version you are currently using (e.g. differe name, different base image name, etc.). It is possible to use windows-sqlserver ConfigMap for Windows 11 or Server 2k22 and vice versa (windows-vs-code for Windows 10 or Server 2k22).

The provided reference ConfigMap (windows-sqlserver) boots Windows 10, 11 or Windows Server 2k22 into Audit mode, applies the customizations as part of Powershell script (ran by SynchronousCommand) and then generalizes the VirtualMachine again. The Powershell script can be adapted as desired to apply other customizations.

Prerequisites

  • KubeVirt >= v1.0.0
  • Tekton Pipelines >= v0.43.0

Prepare unattend.xml ConfigMap

  1. Supply, generate or use the default unattend.xml. For information on answer files see Startup Scripts - KubeVirt User Guide.
  2. Create a new ConfigMap with the unattend.xml
  3. Pass the name of the new ConfigMap to the PipelineRun with the parameter customizeConfigMapName.

Pipeline Description

  import-unattend-configmaps --- copy-vm-root-disk --- create-vm --- wait-for-vmi-status --- create-datasource-root-disk --- cleanup-vm --- delete-imported-configmaps
  1. import-unattend-configmaps imports ConfigMap with unattend.xml needed for automated customization of Windows.
  2. copy-vm-root-disk Task copies PVC defined in sourceDiskImageName and sourceDiskImageNamespace parameters.
  3. create-vm Task creates a VirtualMachine called windows-customize-* from the base DataVolume and with the customize ConfigMap attached as a CD-ROM (Pipeline parameter customizeConfigMapName). The VirtualMachine has to be created in the same namespace as the source DataVolume.
  4. wait-for-vmi-status Task waits until the VirtualMachine shuts down.
  5. create-datasource-root-disk Task creates a DataSource object, which is used by UI for discovering bootable volumes and links PVC created in copy-vm-root-disk step.
  6. cleanup-vm deletes the installer VirtualMachine (also in case of failure of the previous Tasks).
  7. The output artifact will be the win*-customized DataVolume with the customized Windows installation. It will boot into the Windows OOBE and needs to be setup further before it can be used (depends on the applied customizations).
  8. The windows11-unattend ConfigMap can be used to boot the VirtualMachine into the Desktop (depends on the applied customizations).
  9. delete-imported-configmaps deletes imported ConfigMaps.

How to run

The pipeline uses a ConfigMap containing an unattend.xml file for automated customization of Windows. Example ConfigMaps are deployed within the Pipeline. In case you would like to use a different ConfigMap, specify a different URL in the unattendXMLConfigMapsURL parameter and adjust customizeConfigMapName parameter with correct the ConfigMap name. Examples of ConfigMaps can be found here.

Note

By default, the Pipeline requires the ServiceAccount pipeline to exist. Tekton does not create this ServiceAccount in namespaces which name starts with openshift or kube. In case you would like to run this Pipeline in a namespace which starts with openshift or kube, you have to create the pipeline ServiceAccount manually or specify a different ServiceAccount in the PipelineRun.

Pipeline runs with resolvers:

oc create -f - <<EOF
apiVersion: tekton.dev/v1
kind: PipelineRun
metadata:
    generateName: windows11-customize-run-
spec:
    pipelineRef:
        params:
        -   name: catalog
            value: kubevirt-tekton-pipelines
        -   name: type
            value: artifact
        -   name: kind
            value: pipeline
        -   name: name
            value: windows-customize
        -   name: version
            value: v0.25.0
        resolver: hub
EOF
oc create -f - <<EOF
apiVersion: tekton.dev/v1
kind: PipelineRun
metadata:
    generateName: windows2k22-customize-run-
spec:
    params:
    -   name: sourceDiskImageName
        value: win2k22
    -   name: baseDvName
        value: win2k22-customized
    -   name: preferenceName
        value: windows.2k22.virtio
    -   name: customizeConfigMapName
        value: windows-sqlserver
    pipelineRef:
        params:
        -   name: catalog
            value: kubevirt-tekton-pipelines
        -   name: type
            value: artifact
        -   name: kind
            value: pipeline
        -   name: name
            value: windows-customize
        -   name: version
            value: v0.25.0
        resolver: hub
EOF
oc create -f - <<EOF
apiVersion: tekton.dev/v1
kind: PipelineRun
metadata:
    generateName: windows10-customize-run-
spec:
    params:
    -   name: sourceDiskImageName
        value: win10
    -   name: baseDvName
        value: win10-customized
    -   name: preferenceName
        value: windows.10.virtio
    pipelineRef:
        params:
        -   name: catalog
            value: kubevirt-tekton-pipelines
        -   name: type
            value: artifact
        -   name: kind
            value: pipeline
        -   name: name
            value: windows-customize
        -   name: version
            value: v0.25.0
        resolver: hub
EOF

Cancelling/Deleting PipelineRuns

When running the example Pipelines, they create temporary objects (DataVolumes, VirtualMachines, etc.). Each Pipeline has its own clean up system which should keep the cluster clean from leftovers. In case user hard deletes or cancels running PipelineRun, the PipelineRun will not clean temporary objects and objects will stay in the cluster. To prevent this behaviour, cancel the PipelineRun gracefully. It triggers special Tasks, which remove temporary objects and keep only result DataSource/DataVolume/PVC.

Each object created by the Pipeline has OwnerReference to the Pod which created them (result DataVolume does not have it). In the case that the clean up steps are not triggered, by deleting the PipelineRun all leftover objects created by the Pipeline will be deleted.