Summary of All YAML Resources

The following manifests appear in this chapter:

Resource name

Resource type

Scope

Purpose

openshift-storage

Namespace, OperatorGroup, Subscription

openshift-storage

Installs the LVM Storage Operator.

lvmcluster

LVMCluster

openshift-storage

Pools spare disks and creates the default dynamic StorageClass.

openshift-local-storage

Namespace, OperatorGroup, Subscription

openshift-local-storage

Installs the Local Storage Operator, as an alternative to LVM Storage.

auto-discover-devices

LocalVolumeDiscovery

openshift-local-storage

Discovers available local disks automatically.

local-disks

LocalVolume

openshift-local-storage

Provisions static persistent volumes from dedicated local disks.

openshift-kueue-operator

Namespace, OperatorGroup, Subscription

openshift-kueue-operator

Installs the Red Hat build of Kueue Operator, required for the kueue, ray, and trainingoperator components.

rhods-operator

Namespace, OperatorGroup, Subscription

redhat-ods-operator

Installs the OpenShift AI Operator from the stable-3.x channel.

default-dsci

DSCInitialization (v2)

Cluster-scoped

Configures cluster-level RHOAI settings. Auto-created by the operator.

default-dsc

DataScienceCluster (v2)

Cluster-scoped

Enables the required RHOAI components.

intel-gpu-dra

HardwareProfile (infrastructure.opendatahub.io/v1)

redhat-ods-applications

Exposes DRA-managed Intel GPUs to RHOAI workloads.

intel-gpu-xe

HardwareProfile (infrastructure.opendatahub.io/v1)

redhat-ods-applications

Exposes device-plugin Intel GPUs to RHOAI workloads.

model-cache

PersistentVolumeClaim

User namespace

Stores downloaded models on persistent storage.

intel-vllm-runtime

ServingRuntime

User namespace

Defines the Intel vLLM serving runtime with the XPU backend.

intel-vllm-opt125m

InferenceService

User namespace

Deploys GPU-accelerated LLM inference in RawDeployment mode.

intel-vllm-opt125m-external

Service, Route

User namespace

Exposes the inference endpoint outside the cluster, because the KServe predictor service is headless.

vllm-gpu-claim

ResourceClaimTemplate

User namespace

Requests an Intel GPU through DRA on DRA-only clusters.

vllm-opt125m

Deployment, Service, Route

User namespace

Deploys vLLM with a DRA claim, for clusters without the device plugin.

Note

Models are served through KServe RawDeployment and the dashboard is routed through a Gateway API Gateway, so this chapter installs neither OpenShift Serverless nor OpenShift Service Mesh and defines no KnativeServing or ServiceMeshControlPlane resource.