As part of the NVIDIA DPF Integration project, we are using Flannel as the primary CNI in HyperShift clusters. For Flannel to correctly handle pod CIDR allocation, it requires the kube-controller-manager (KCM) to be launched with the --allocate-node-cidrs=true flag.

Currently, HyperShift does not expose an API to customize arguments for the kube-controller-manager. While a comprehensive configuration API exists in standard OpenShift, our immediate and specific need for the DPF integration is to enable this single flag.

This feature is required to support Flannel CNI for DPF workloads.

• Target release: OpenShift v4.20

• Related work: Standard OCP KubeControllerManager API (link: https://docs.redhat.com/en/documentation/openshift_container_platform/4.12/html/operator_apis/kubecontrollermanager-operator-openshift-io-v1)

User Stories

• As a cluster service consumer (using DPF workers), I want to create HyperShift clusters with Flannel CNI, so that the kube-controller-manager correctly allocates node CIDRs and my pods receive functional networking.

• As a cluster service provider, I want to enable the --allocate-node-cidrs=true KCM flag for DPF-specific HyperShift clusters, so that I can support the DPF integration's requirement for Flannel.