Skip to navigation
Development

NVIDIA Run:ai CRD Reference

View as Markdown

The cluster plane is CRD-driven. The agent reconciles control-plane state into cluster CRDs; cluster-sync watches those CRDs and reports status back. All CRDs listed here use the run.ai API group.

For an overview of how CRDs fit into the platform architecture, see the Architecture and Components. For the services that manage these CRDs, see the Service Reference.

Workload Types

KindAPI groupDescription
ExternalWorkloadrun.ai/v1alpha1A workload originating outside NVIDIA Run:ai (e.g. a plain Kubernetes Deployment) that the platform tracks for quota and visibility.
WorkloadRunnerrun.ai/v1Internal lower-level resource that materializes a workload’s final pod spec; produced by workload controllers and consumed by the KAI Scheduler.

Cluster Configuration and Access

KindAPI groupDescription
RunaiConfigrun.ai/v1Top-level operator configuration object; controls settings for every platform component in the cluster.
Clusterrun.ai/v1Singleton that registers the cluster with the control plane and reports observed health conditions.
AccessRulerun.ai/v1alpha1Binds a subject (user or group) to a named role within a scope (cluster, department, or project).
DataVolumerun.ai/v1alpha1Cluster-scoped reference to a source PVC that can be shared across multiple project namespaces as a reusable storage asset.
Applicationrun.ai/v1alpha1An OAuth2 client application registered with the cluster’s identity manager.