Configuration

The Kubeflow Pipelines operator can be configured with the following parameters:

Parameter name	Description	Example
`defaultExperiment`	Default Experiment name to be used for creating pipeline runs	`Default`
`defaultProvider`	Default provider name to be used (see Using Multiple Providers). Note: This is deprecated as of v1alpha6 and will be removed when support for v1alpha5 is dropped	`vertex-ai-europe`
`multiversion`	If enabled, it will support previous versions of the CRDs, only the latest otherwise	`true`
`workflowNamespace`	Namespace where operator Argo workflows should be running - defaults to the operator’s namespace	`kfp-operator-workflows`
`runCompletionTTL`	Duration string for how long to keep one-off runs after completion - a zero-length or negative duration will result in runs being deleted immediately after completion; defaults to empty (never delete runs)	`10m`
`runCompletionFeed`	Configuration of the service for the run completion feed back to KFP Operator
`frameworks`	Map of pipeline framework to docker image. Pipeline framework to match value in Pipeline resource	`default: kfp-operator-tfx-compiler:latest`
`defaultProviderValues`	Configuration of the deployment and service created for providers

Run Completion Feed Configuration

Parameter name	Description	Example
`port`	The port that the feed endpoint will listen on	`8082`
`endpoints`	Array of run completion event handler endpoints that should be called per feed message	`- host: run-completion-event-handler<br/>  path: /<br/>  port: 12000`

Provider Values Configuration

Parameter name	Description	Example
`replicas`	Number of replicas used within the deployment	`2`
`serviceContainerName`	Name of the container that will execute the provider image. Note: this must match the podTemplateSpec name	`provider-service`
`servicePort`	The port that should expose the service Note: this must match the podTemplateSpec ports	`8080`
`metricsPort`	The port that should expose the metrics service. Note: this must match the `metrics` port defined in the podTemplateSpec	`8081`
`labels`	A map of additional labels to add to the provider service Deployment and Service. The Deployment and Service will also have a label of `app: provider-<provider_name>`. To attach labels to a provider service pod, use `podTemplateSpec`.	`a: label`
`podTemplateSpec`	k8s pod template spec for the provider service deployment

An example configuration:

apiVersion: config.kubeflow.org/v1beta1
kind: KfpControllerConfig
controller:
  health:
    healthProbeBindAddress: :8081
  metrics:
    bindAddress: :8080
  webhook:
    port: 9443
  leaderElection:
    leaderElect: true
    resourceName: kfp-operator-lock
spec:
  multiversion: true
  workflowNamespace: kfp-operator-system
  workflowTemplatePrefix: kfp-operator-
  frameworks:
    default: kfp-operator-tfx-compiler:latest
  runCompletionFeed:
    port: 8082
    endpoints: []
  defaultProviderValues:
    replicas: 1
    serviceContainerName: provider-service
    servicePort: 8080
    metricsPort: 8081
    labels: 
      a: label
      another: label
    podTemplateSpec:
      metadata:
        labels: 
          a: label
      spec:
        volumes: []
        securityContext:
          runAsNonRoot: true
        containers:
          - name: provider-service
            ports:
              - containerPort: 8080
                name: http
                protocol: TCP
              - containerPort: 8081
                name: metrics
                protocol: TCP
            volumeMounts: []
            securityContext:
              allowPrivilegeEscalation: false
            resources:
              limits:
                cpu: 500m
                memory: 256Mi
              requests:
                cpu: 250m
                memory: 128Mi
            env:
              - name: POD_NAMESPACE
                valueFrom:
                  fieldRef:
                    fieldPath: metadata.namespace
              - name: OPERATORWEBHOOK
                value: http://kfp-operator-runcompletion-webhook-service.kfp-operator-system:80/events
              - name: SERVER_PORT
                value: '8080'
              - name: METRICS_PORT
                value: '8081'