Uploaded image for project: 'OpenShift Request For Enhancement'
  1. OpenShift Request For Enhancement
  2. RFE-8658

Enable Hybrid Bare Metal and KubeVirt Providers Within a Single Hosted Cluster

XMLWordPrintable

    • None
    • Product / Portfolio Work
    • None
    • False
    • Hide

      None

      Show
      None
    • None
    • None
    • None
    • None
    • None
    • None
    • None
    • None

      1. Proposed title of this feature request

      Enable Hybrid Bare Metal and KubeVirt Providers Within a Single Hosted Cluster

      2. What is the nature and description of the request?

      In the Hosted Control Planes (HyperShift) model, the infrastructure provider choice for NodePools is effectively fixed at cluster creation time.

      A HostedCluster created with:

      • Agent-based Bare Metal NodePools cannot later introduce KubeVirt-based NodePools, and
      • A HostedCluster created with KubeVirt NodePools cannot later introduce Bare Metal NodePools

      This mutual exclusivity significantly limits the ability to:

      • Reassess infrastructure requirements after workload onboarding
      • Mix physical and virtual capacity in the same hosted cluster
      • Fully leverage the decoupled control plane and worker model that HyperShift is designed to provide

      Customer Use Cases Impacted

      1. Post-deployment Infrastructure Reassessment in Hosted Clusters

      In several Hosted Control Plane deployments:

      • Clusters were initially provisioned with Agent-based Bare Metal NodePools based on projected workload requirements.
      • After real workload behavior was observed, it became clear that:
        • Capacity was overestimated, and
        • Virtualized NodePools (KubeVirt) would have been more cost-effective and operationally suitable.

      Because the HostedCluster was created with the Agent provider:

      • KubeVirt NodePools could not be added
      • Customers were forced to migrate workloads to a new HostedCluster

      This undermines a core HyperShift value proposition:
      the ability to evolve worker infrastructure independently of the control plane.

      The reverse scenario is also common:

      • HostedClusters created with KubeVirt NodePools for rapid onboarding
      • Later requirements for Bare Metal GPU nodes (AI/ML, HPC, low-latency workloads)
      • Again requiring full cluster recreation

      3. Why does the customer need this? (List the business requirements here)

      Many customers require:

      • Bare Metal GPU-backed NodePools for training or inference workloads
      • Virtualized (KubeVirt) NodePools for:
        • Preprocessing
        • CI/CD
        • Supporting and auxiliary services

      Currently:

      • HyperShift does not allow mixing Agent-based Bare Metal and KubeVirt NodePools within the same HostedCluster
      • Customers must deploy multiple HostedClusters, increasing:
        • Operational complexity
        • Cost
        • Control plane sprawl

      This is contrary to the design goals of Hosted Control Planes as a multi-tenant, flexible consumption model.

      4. List any affected packages or components.

      Hypershift

              racedoro@redhat.com Ramon Acedo
              rhn-support-chdeshpa Chinmay Deshpande
              None
              Votes:
              1 Vote for this issue
              Watchers:
              4 Start watching this issue

                Created:
                Updated:
                None
                None