Outcome Overview

This feature aims to integrate OpenShift’s implementation of Multus CNI with support for a Kubernetes DRA/NRI-based implementation of the upstream Kubernetes Network Driver (KND) model. The goal is to enable Kubernetes-native, Dynamic Resource Allocation (DRA)–based network attachment while preserving compatibility with existing Multus-based secondary networking workflows in OpenShift.

Success Criteria

The feature will be considered successful when the following criteria, subject to change based upon spike outcomes, are met:

Functional Success Criteria

• Pods running on OpenShift can successfully attach secondary networks provisioned via DRA/KND using Kubernetes DRA ResourceClaims.

• Multus correctly interoperates with DRA-allocated network interfaces without requiring custom patches or nonstandard Pod annotations.

• A single Pod can consume both legacy Multus/CNI-based networks and DRA/KND-based networks concurrently.

• Network interfaces allocated via DRA are correctly configured inside the Pod network namespace and remain stable for the Pod’s lifecycle.

• High-performance networking configurations (e.g., RDMA-capable interfaces) function as expected when provisioned via DRA.

Compatibility and Stability Criteria

• Existing Multus-based workloads and NetworkAttachmentDefinitions continue to function without regression.

• Cluster upgrades do not disrupt Pods using either legacy or DRA-based networking.

• Failure scenarios (e.g., resource exhaustion, node unavailability) result in predictable scheduling or admission failures consistent with DRA semantics.

Kubernetes and Upstream Alignment Criteria

• The integration aligns with upstream Kubernetes APIs, including Dynamic Resource Allocation and emerging Pod Network API concepts, without introducing OpenShift-only APIs.

• The implementation does not depend on deprecated or stalled CNI-DRA mechanisms and remains compatible with ongoing upstream DRA/KND development.

• The solution can be demonstrated using upstream DRA/KND components with minimal OpenShift-specific modifications.

Operational and Usability Criteria

• Cluster administrators can deploy and manage the solution using standard OpenShift installation and lifecycle workflows.

• Network resources advertised via DRA are observable and debuggable using standard Kubernetes tooling (e.g., kubectl describe, oc get resourceclaims).

• Clear documentation exists describing configuration, troubleshooting, and recommended usage patterns.

Performance and Scalability Criteria

• Network attachment and Pod startup times remain within acceptable bounds compared to existing Multus-based solutions.

• The system scales to large clusters with many nodes and network resources without introducing centralized bottlenecks.

• Resource utilization and scheduling behavior reflect the intended efficiency gains of DRA-based networking.

Expected Results (what, how, when)

• Kubernetes-native networking model – Enables adoption of the upstream KND/DRA approach, reducing reliance on legacy CNI chaining and custom attachment mechanisms.

• High-performance networking enablement – Unlocks efficient access to hardware-backed interfaces (e.g., RDMA, SR-IOV) for AI/ML and other demanding workloads.

• Backward compatibility and coexistence – Preserves existing Multus-based workflows while allowing gradual adoption of DRA-based networking.

• Improved consistency and extensibility – Aligns secondary networking with standard Kubernetes resource semantics, simplifying integration with scheduling and future networking APIs.

• Future-proof OpenShift networking – Positions OpenShift to track upstream Kubernetes networking evolution and reduce long-term technical debt.

Post Completion Review – Actual Results

After completing the work (as determined by the "when" in Expected Results above), list the actual results observed / measured during Post Completion review(s).