olm-operator-plugins.md

OLM operator plugins development

Existing plugins

• AMD GPU Operator (AMDGPU)
• Authorino
• OpenShift Virtualization (CNV)
• Fence Agents Remediation (FAR)
• Kernel Module Management (KMM)
• Local Storage Operator (LSO)
• Logical Volume Manager (LVM)
• Multi-Cluster Engine (MCE)
• Migration Toolkit for Virtualization (MTV)
• NMState
• Node Feature Discovery
• Node Health Check Operator(NHO)
• Node Maintenance
• NVIDIA GPU Operator
• OpenShift Data Foundation (ODF)
• OpenShift AI
• OpenShift Service Mesh (OSC)
• OpenShift Pipelines
• Self Node Remediation (SNR)
• OpenShift Serverless
• OpenShift Service Mesh

How to implement a new OLM operator plugin

To implement support for a new OLM operator plugin you need to make following changes:

1. Introduce new validation IDs for the new operator in the swagger specification:
   • for host validation:

     host-validation-id:
       type: string
       enum:
         - 'connected'
         ...
         - 'lso-requirements-satisfied'
         - 'cnv-requirements-satisfied'
         - 'odf-requirements-satisfied'
         - 'lvm-requirements-satisfied'

   • for cluster validation:

     cluster-validation-id:
       type: string
       enum:
         - 'machine-cidr-defined'
         ...
         - 'lso-requirements-satisfied'
         - 'cnv-requirements-satisfied'
         - 'odf-requirements-satisfied'
         - 'lvm-requirements-satisfied'

2. Introduce new feature support ID in the swagger specification:

feature-support-level-id:
  type: string
  enum:
    - 'SNO'
    ...
    - 'LVM'
    - 'ODF'
    - 'LSO'
    - 'CNV'

3. Add the operator's name to the enum list for /v2/supported-operators endpoint swagger specification:

  type: array
  items:
    type: string
    enum:
    - 'amd-gpu'
    - 'lso'
    - 'mtv'
    - 'openshift-ai'
    ...
    - 'osc'
    - 'servicemesh'

4. Regenerate code by running

skipper make generate

5. Create and add the new validation IDs to proper category - "operators":

   • for cluster validation:

     func (v validationID) category() (string, error) {
     ...
       case IsCnvRequirementsSatisfied, IsLsoRequirementsSatisfied, IsOdfRequirementsSatisfied, IsLvmRequirementsSatisfied:
          return "operators", nil

   • for host validaton:

     func (v validationID) category() (string, error) {
     ...
       case AreLsoRequirementsSatisfied, AreCnvRequirementsSatisfied, AreOdfRequirementsSatisfied, AreLvmRequirementsSatisfied:
     		return "operators", nil

6. Modify the installation state machine by adding the new validationIDs to the list of required checks:

   • for cluster:

     var requiredForInstall = stateswitch.And(...,
        ..., If(IsLsoRequirementsSatisfied), If(IsCnvRequirementsSatisfied), If(IsOdfRequirementsSatisfied), If(IsLvmRequirementsSatisfied))

   • for host:

     	var isSufficientForInstall = stateswitch.And(...,
     		...,
     		If(AreLsoRequirementsSatisfied), If(AreCnvRequirementsSatisfied), If(AreOdfRequirementsSatisfied), If(AreLvmRequirementsSatisfied))

7. Add the new feature to the OLM operators list and implement the SupportLevelFeature interface

   • add feature to the support level list:

         // Olm Operators features
         models.FeatureSupportLevelIDLVM:                  (&LvmFeature{}).New(),
       ...
         models.FeatureSupportLevelIDMCE:                  (&MceFeature{}).New(),
         models.FeatureSupportLevelIDODF:                  (&OdfFeature{}).New(),
         models.FeatureSupportLevelIDMTV:                  (&MtvFeature{}).New(),
         models.FeatureSupportLevelIDOSC:                  (&OscFeature{}).New(),

   • implement the interface at the end of olm operators Your custom code shall reflect the requirements of this new feature by means of Openshift version and target platform (getSupportLevel), CPU architecture (getIncompatibleArchitectures), and compatibility with existing features (getIncompatibleFeatures).
   • in case of any such incompatibilities, add the feature also to the getIncompatibleFeatures methods of all the affected types:
     • for misc features
     • for networking features
     • for other existing operators
     • for platforms

8. Implement the Operator interface

9. Plug the new Operator implementation in the OperatorManager constructor:

func NewManager(log logrus.FieldLogger) Manager {
  return NewManagerWithOperators(log, lso.NewLSOperator(), cnv.NewCnvOperator(log), odf.NewOdfOperator(log), lvm.NewLvmOperator(log))
}

Notes about the Operator interface

Manifests generation

A plugin can generate two distinct set of manifests, specified by the two return values of the GenerateManifests(*common.Cluster) (map[string][]byte, []byte, error) method. They are required to:

1. Install the operator
2. Configure the operator

The first return value could be used to specify a set of manifests that will be applied directly to the control plane during the bootstrap phase. This set is usually composed by a manifest for creating a new namespace, a new subscription and a new operator group CR for the involved operator.

The second return value it's a manifest used to configure the freshly installed operator, and it will be applied by the assisted-installer-controller job, only after the cluster have been successfully created and the OLM operators are all ready (currently the assisted-installer-controller retrieves the whole list of configurations by downloading the custom_manifests.json file fetched from the Assisted Service).