v0.6.0

New Features

• nv per mem support in GPU Driver #191
• scale up support
• [helm chart] Images configuration #105

Bug Fixes

• Failure to deploy mofed on a single node in cluster will halt deployment progress #184
• Network Operator fails to update node labels #182
• When deploying mofed pod interface names change
• When deploying mofed pod interfaces become in down state

Known Limitations

• network-operator will deploy nvPeerDriver POD on a node only if NVIDIA GPU driver version < 465. Starting from v465 NVIDIA GPU driver includes a built-in nvidia_peermem module which is a replacement for nv_peer_mem module. NVIDIA GPU operator manages nvidia_peermem module loading.

v0.5.0

New Features

• Use upstream NFD chart #160
• Add OpenShift deployment support #94
• Update nv-peer-mem-driver to 1.1-0

Bug Fixes

• Change order of networking components rendering #159
• Add notes about compatibility with gpu-operator #161
• Increase terminationGracePeriodSeconds for OFED POD #164
• Fix typo in an example file name #163
• Network Operator deploys networking components (Multus CNI, CNI Plugins and Whereabout) before MOFED
• Use Kubernetes-specific image for SR-IOV Network operator by default #167
• Network Operator components work as expected after node reboot
• Fix pods deployment order after cluster scale-up

Known Limitations

• Network Operator is compatible with NVIDIA GPU Operator v1.5.2 and above

v0.4.0

New Features

• Modify network-operator namespaces #122-
• Rename 'devicePlugin' to 'rdmaSharedDevicePlugin' #135
• Print Age in CR status of all resources #136
• CRDs are not synced #134
• Stop using SelfLink #133
• Add linters for helm charts and Dockerfiles #150
• Add HostDeviceNetwork basic support #126
• Update images for v0.4.0 #153
• Add 'prepare-release.sh' script #149
• Add examples how to configure and deploy SR-IOV network #154

Bug Fixes

• fix typo in README.md #132
• [helm] fix nfd worker tolerations value #144
• Add additional nodeAffinity for nv-peer-mem-driver POD #152
• Add support for host OFED to peer-mem POD #155
• update image OFED container image to end of march release

v0.3.0

New Features

• Support deploying SR-IOV network operator #58
• Update go code to operator-sdk 1.3.0 #86
• Add status printer column for Custom Resources #96

Bug Fixes

• Change multus image to v3.6 #101
• Update helm chart kubeVersion to '>= 1.17.0' #110
• Use the NFD instance to v0.6.0-233-g3e00bfb #108
• Update gogo/protobuf to v1.3.2 - CVE-2021-3121 #97

Known Limitations

• deploying network operator before gpu operator may cause nv peer mem container to crash
• when deploying mofed pod interface names may change
• mlx_compat module isn't unloaded after kubernetes nicclusterpolicy teardown
• nv-peer-mem container is deployed at master if a GPU is present on it
• mofed container may reset VF node guid which breaks RoCE

0.2.0-beta

Main Features

• Deployment of Networking components for macvlan secondary network in a Kubernetes cluster
  • multus-cni for secondary network
  • whereabouts IPAM CNI plugin
  • containernetworking CNI plugins
• Helm chart Support
• Update MOFED driver container to version 5.2-1
• Update rdma shared device plugin version to v1.1.0
• Support HTTP proxy setting for Network-Operator

0.1.0-aplha

Main Features

• Deployment of OFED driver container
• Deployment of k8s-rdma-shared-device-plugin
• Deployment of NV Peer Memory driver container