Multiple Network Attachments with bridge CNI

Introduction

Over the last years the KubeVirt project has improved a lot regarding secondary interfaces networking configuration. Now it’s possible to do an end to end configuration from host networking to a VM using just the Kubernetes API with special Custom Resource Definitions. Moreover, the deployment of all the projects has been simplified by introducing KubeVirt hyperconverged cluster operator (HCO) and cluster network addons operator (CNAO) to install the networking components.

The following is the operator hierarchy list presenting the deployment responsibilities of the HCO and CNAO operators used in this blog post:

  • kubevirt-hyperconverged-cluster-operator (HCO)
    • cluster-network-addons-operator (CNAO)
      • multus
      • bridge-cni
      • kubemacpool
      • kubernetes-nmstate
    • KubeVirt

Introducing cluster-network-addons-operator

The cluster network addons operator manages the lifecycle (deploy/update/delete) of different Kubernetes network components needed to configure secondary interfaces, manage MAC addresses and defines networking on hosts for pods and VMs.

A Good thing about having an operator is that everything is done through the API and you don’t have to go over all nodes to install these components yourself and assures smooth updates.

In this blog post we are going to use the following components, explained in a greater detail later on:

  • multus: to start a secondary interface on containers in pods
  • linux bridge CNI: to use bridge CNI and connect the secondary interfaces from pods to a linux bridge at nodes
  • kubemacpool: to manage mac addresses
  • kubernetes-nmstate: to configure the linux bridge on the nodes

The list of components we want CNAO to deploy is specified by the NetworkAddonsConfig Custom Resource (CR) and the progress of the installation appears in the CR status field, split per component. To inspect this progress we can query the CR status with the following command:

kubectl get NetworkAddonsConfig cluster -o yaml

To simplify this blog post we are going to use directly the NetworkAddonsConfig from HCO, which by default installs all the network components, but just to illustrate CNAO configuration, the following is a NetworkAddonsConfig CR instructing to deploy multus, linuxBridge, nmstate and kubemacpool components:

apiVersion: networkaddonsoperator.network.kubevirt.io/v1
kind: NetworkAddonsConfig
metadata:
  name: cluster
spec:
  multus: {}
  linuxBridge: {}
  nmstate: {}
  imagePullPolicy: Always

Connecting Pods, VMs and Nodes over a single secondary network with bridge CNI

Although Kubernetes provides a default interface that gives connectivity to pods and VMs, it’s not easy to configure which NIC should be used for specific pods or VMs in a multi NIC node cluster. A Typical use case is to split control/traffic planes isolated by different NICs on nodes.

With linux bridge CNI + multus it’s possible to create a secondary NIC in pod containers and attach it to a L2 linux bridge on nodes. This will add container’s connectivity to a specific NIC on nodes if that NIC is part of the L2 linux bridge.

To ensure the configuration is applied only in pods on nodes that have the bridge, the k8s.v1.cni.cncf.io/resourceName label is added. This goes hand in hand with another component, bridge-marker which inspects nodes networking and if a new bridge pops up it will mark the node status with it.

This is an example of the results from bridge-marker on nodes where bridge br0 is already configured:

---
status:
  allocatable:
    bridge.network.kubevirt.io/br0: 1k
  capacity:
    bridge.network.kubevirt.io/br0: 1k

This is an example of NetworkAttachmentDefinition to expose the bridge available on the host to users:

apiVersion: k8s.cni.cncf.io/v1
kind: NetworkAttachmentDefinition
metadata:
  name: bridge-network
  annotations:
    k8s.v1.cni.cncf.io/resourceName: bridge.network.kubevirt.io/br0
spec:
  config: >
    {
        "cniVersion": "0.3.1",
        "name": "br0-l2",
        "plugins": [{
            "type": "bridge",
            "bridge": "br0",
            "ipam": {}
        }]
    }

Then adding the bridge secondary network to a pod is a matter of adding the following annotation to it:

annotations:
  k8s.v1.cni.cncf.io/networks: bridge-network

Setting up node networking with NodeNetworkConfigurationPolicy (aka nncp)

Changing Kubernetes cluster node networking can be done manually iterating over all the cluster nodes and making changes or using different automatization tools like ansible. However, using just another Kubernetes resource is more convenient. For this purpose the kubernetes-nmstate project was born as a cluster wide node network administrator based on Kubernetes CRs on top of nmstate.

It works as a Kubernetes DaemonSet running pods on all the cluster nodes and reconciling three different CRs:

Note

Project kubernetes-nmstate has a distributed architecture to reduce kube-apiserver connectivity dependency, this means that every pod will configure the networking on the node that it’s running without much interaction with kube-apiserver.

In case something goes wrong and the pod changing the node network cannot ping the default gateway, resolve DNS root servers or has lost the kube-apiserver connectivity it will rollback to the previous configuration to go back to a working state. Those errors can be checked by running kubectl get nnce. The command displays potential issues per node and nncp.

The desired state fields follow the nmstate API described at their awesome doc

Also for more details on kubernetes-nmstate there are guides covering reporting, configuration and troubleshooting. There are also nncp examples.

Demo: mixing it all together, VM to VM communication between nodes

With the following recipe we will end up with a pair of virtual machines pair on two different nodes with one secondary NICs, eth1 at vlan 100. They will be connected to each other using the same bridge on nodes that also have the external secondary NIC eth1 connected.

Demo environment setup

We are going to use a kubevirtci as Kubernetes ephemeral cluster provider.

To start it up with two nodes and one secondary NIC and install NetworkManager >= 1.22 (needed for kubernetes-nmstate) and dnsmasq follow these steps:

git clone https://github.com/kubevirt/kubevirtci
cd kubevirtci
# Pin to version working with blog post steps in case
# k8s-1.19 provider disappear in the future
git reset d5d8e3e376b4c3b45824fbfe320b4c5175b37171 --hard
export KUBEVIRT_PROVIDER=k8s-1.19
export KUBEVIRT_NUM_NODES=2
export KUBEVIRT_NUM_SECONDARY_NICS=1
make cluster-up
export KUBECONFIG=$(./cluster-up/kubeconfig.sh)

Installing components

To install KubeVirt we are going to use the operator kubevirt-hyper-converged-operator, this will install all the components needed to have a functional KubeVirt with all the features including the ones we are going to use: multus, linux-bridge, kubemacpool and kubernetes-nmstate.

curl https://raw.githubusercontent.com/kubevirt/hyperconverged-cluster-operator/master/deploy/deploy.sh | bash
kubectl wait hco -n kubevirt-hyperconverged kubevirt-hyperconverged --for condition=Available --timeout=500s

Now we have a Kubernetes cluster with all the pieces to startup a VM with bridge attached to a secondary NIC.

Creating the br0 on nodes with a port attached to secondary NIC eth1

First step is to create a L2 linux-bridge at nodes with one port on the secondary NIC eth1, this will be used later on by the bridge CNI.

cat <<EOF | kubectl apply -f -
apiVersion: nmstate.io/v1alpha1
kind: NodeNetworkConfigurationPolicy
metadata:
  name: br0-eth1
spec:
  desiredState:
    interfaces:
    - name: br0
      description: Linux bridge with eth1 as a port
      type: linux-bridge
      state: up
      bridge:
        options:
          stp:
            enabled: false
        port:
        - name: eth1
EOF

Now we wait for the bridge to be created checking nncp conditions:

kubectl wait nncp br0-eth1 --for condition=Available --timeout 2m

After the nncp becomes available, we can query the nncp resources in the cluster and see it listed with successful status.

kubectl get nncp
NAME       STATUS
br0-eth1   SuccessfullyConfigured

We can inspect the status of applying the policy to each node. For that there is the NodeNetworkConfigurationEnactment CR (nnce):

kubectl get nnce
NAME              STATUS
node01.br0-eth1   SuccessfullyConfigured
node02.br0-eth1   SuccessfullyConfigured

Note

In case of errors it is possible to retrieve the error dumped by nmstate running kubectl get nnce -o yaml the status will contain the error.

We can also inspect the network state on the nodes by retrieving the NodeNetworkState and checking if the bridge br0 is up using jsonpath

kubectl get nns node01 -o=jsonpath='{.status.currentState.interfaces[?(@.name=="br0")].state}'
kubectl get nns node02 -o=jsonpath='{.status.currentState.interfaces[?(@.name=="br0")].state}'

When inspecting the full currentState yaml we get the following interface configuration:

kubectl get nns node01 -o yaml
status:
  currentState:
    interfaces:
      - bridge:
          options:
            group-forward-mask: 0
            mac-ageing-time: 300
            multicast-snooping: true
            stp:
              enabled: false
              forward-delay: 15
              hello-time: 2
              max-age: 20
              priority: 32768
          port:
            - name: eth1
              stp-hairpin-mode: false
              stp-path-cost: 100
              stp-priority: 32
        description: Linux bridge with eth1 as a port
        ipv4:
          dhcp: false
          enabled: false
        ipv6:
          autoconf: false
          dhcp: false
          enabled: false
        mac-address: 52:55:00:D1:56:00
        mtu: 1500
        name: br0
        state: up
        type: linux-bridge

We can also check that the bridge-marker is working and check verify on nodes:

kubectl get node node01 -o yaml

The following should appear stating that br0 can be consumed on the node:

status:
  allocatable:
    bridge.network.kubevirt.io/br0: 1k
  capacity:
    bridge.network.kubevirt.io/br0: 1k

At this point we have an L2 linux bridge ready and connected to NIC eth1.

Configure network attachment with a L2 bridge and a vlan

In order to make the bridge a L2 bridge, we specify no IPAM (IP Address Management) since we are not going to configure any ip address for the bridge. To configure bridge vlan-filtering we add the vlan we want to use to isolate our VMs:

cat <<EOF | kubectl apply -f -
apiVersion: k8s.cni.cncf.io/v1
kind: NetworkAttachmentDefinition
metadata:
  name: br0-100-l2
  annotations:
    k8s.v1.cni.cncf.io/resourceName: bridge.network.kubevirt.io/br0
spec:
  config: >
    {
        "cniVersion": "0.3.1",
        "name": "br0-100-l2-config",
        "plugins": [
            {
                "type": "bridge",
                "bridge": "br0",
                "vlan": 100,
                "ipam": {}
            },
            {
                "type": "tuning"
            }
        ]
    }
EOF

Start a pair of VMs on different nodes using the multus configuration to connect a secondary interfaces to br0

Now it’s time to startup the VMs running on different nodes so we can check external connectivity of br0. They will also have a secondary NIC eth1 to connect to the other VM running at different node, so they go over the br0 at nodes.

The following picture illustrates the cluster:

bridge cluster_kubevirtci kubevirtci cluster cluster_node01 node01 cluster_vma vma cluster_node02 node02 cluster_vmb vmb nd_br1_kubevirtci br1 nd_br0_node01 br0 nd_eth1_node01 eth1 nd_br0_node01--nd_eth1_node01 nd_eth1_vma eth1 nd_br0_node01--nd_eth1_vma nd_eth1_node01--nd_br1_kubevirtci nd_br0_node02 br0 nd_eth1_node02 eth1 nd_br0_node02--nd_eth1_node02 nd_eth1_vmb eth1 nd_br0_node02--nd_eth1_vmb nd_eth1_node02--nd_br1_kubevirtci

First step is to install the virtctl command line tool to play with virtual machines:

curl -L -o virtctl https://github.com/kubevirt/kubevirt/releases/download/v0.33.0/virtctl-v0.33.0-linux-amd64
chmod +x virtctl
sudo install virtctl /usr/local/bin

Now let’s create two VirtualMachines on each node. They will have one secondary NIC connected to br0 using the multus configuration for vlan 100. We will also activate kubemacpool to be sure that mac addresses are unique in the cluster and install the qemu-guest-agent so IP addresses from secondary NICs are reported to VM and we can inspect them later on.

cat <<EOF | kubectl apply -f -
apiVersion: v1
kind: Namespace
metadata:
  name: default
  labels:
    mutatevirtualmachines.kubemacpool.io: allocate
---
apiVersion: kubevirt.io/v1alpha3
kind: VirtualMachine
metadata:
  name: vma
spec:
  running: true
  template:
    spec:
      nodeSelector:
        kubernetes.io/hostname: node01
      domain:
        devices:
          disks:
            - name: containerdisk
              disk:
                bus: virtio
            - name: cloudinitdisk
              disk:
                bus: virtio
          interfaces:
          - name: default
            masquerade: {}
          - name: br0-100
            bridge: {}
        machine:
          type: ""
        resources:
          requests:
            memory: 1024M
      networks:
      - name: default
        pod: {}
      - name: br0-100
        multus:
          networkName: br0-100-l2
      terminationGracePeriodSeconds: 0
      volumes:
        - name: containerdisk
          containerDisk:
            image: kubevirt/fedora-cloud-container-disk-demo
        - name: cloudinitdisk
          cloudInitNoCloud:
            networkData: |
              version: 2
              ethernets:
                eth1:
                  addresses: [ 10.200.0.1/24 ]
            userData: |-
              #!/bin/bash
              echo "fedora" |passwd fedora --stdin
              dnf -y install qemu-guest-agent
              sudo systemctl enable qemu-guest-agent
              sudo systemctl start qemu-guest-agent
---
apiVersion: kubevirt.io/v1alpha3
kind: VirtualMachine
metadata:
  name: vmb
spec:
  running: true
  template:
    spec:
      nodeSelector:
        kubernetes.io/hostname: node02
      domain:
        devices:
          disks:
            - name: containerdisk
              disk:
                bus: virtio
            - name: cloudinitdisk
              disk:
                bus: virtio
          interfaces:
          - name: default
            masquerade: {}
          - name: br0-100
            bridge: {}
        machine:
          type: ""
        resources:
          requests:
            memory: 1024M
      networks:
      - name: default
        pod: {}
      - name: br0-100
        multus:
          networkName: br0-100-l2
      terminationGracePeriodSeconds: 0
      volumes:
        - name: containerdisk
          containerDisk:
            image: kubevirt/fedora-cloud-container-disk-demo
        - name: cloudinitdisk
          cloudInitNoCloud:
            networkData: |
              version: 2
              ethernets:
                eth1:
                  addresses: [ 10.200.0.2/24 ]
            userData: |-
              #!/bin/bash
              echo "fedora" |passwd fedora --stdin
              dnf -y install qemu-guest-agent
              sudo systemctl enable qemu-guest-agent
              sudo systemctl start qemu-guest-agent
EOF

Wait for the two VMs to be ready. Eventually you will see something like this:

kubectl get vmi
NAME      AGE    PHASE     IP               NODENAME
vma      2m4s   Running   10.244.196.142   node01
vmb      2m4s   Running   10.244.140.86    node02

We can check that they have one secondary NIC without address assigned:

kubectl get vmi -o yaml
## vma
  interfaces:
  - interfaceName: eth0
    ipAddress: 10.244.196.144
    ipAddresses:
    - 10.244.196.144
    - fd10:244::c48f
    mac: 02:4a:be:00:00:0a
    name: default
  - interfaceName: eth1
    ipAddress: 10.200.0.1/24
    ipAddresses:
    - 10.200.0.1/24
    - fe80::4a:beff:fe00:b/64
    mac: 02:4a:be:00:00:0b
    name: br0-100
## vmb
  interfaces:
  - interfaceName: eth0
    ipAddress: 10.244.140.84
    ipAddresses:
    - 10.244.140.84
    - fd10:244::8c53
    mac: 02:4a:be:00:00:0e
    name: default
  - interfaceName: eth1
    ipAddress: 10.200.0.2/24
    ipAddresses:
    - 10.200.0.2/24
    - fe80::4a:beff:fe00:f/64
    mac: 02:4a:be:00:00:0f
    name: br0-100

Let’s finish this section by verifying connectivity between vma and vmb using ping. Open the console of vma virtual machine and use ping command with destination IP address 10.200.0.2, which is the address assigned to the secondary interface of vmb:

Note

The user and password for this VMs is fedora, it was configured at cloudinit userData

virtctl console vma
ping 10.200.0.2 -c 3
PING 10.200.0.2 (10.200.0.2): 56 data bytes
64 bytes from 10.200.0.2: seq=0 ttl=50 time=357.040 ms
64 bytes from 10.200.0.2: seq=1 ttl=50 time=379.742 ms
64 bytes from 10.200.0.2: seq=2 ttl=50 time=404.066 ms

--- 10.200.0.2 ping statistics ---
3 packets transmitted, 3 packets received, 0% packet loss
round-trip min/avg/max = 357.040/380.282/404.066 ms

Conclusion

In this blog post we used network components from KubeVirt project to connect two VMs on different nodes through a linux bridge connected to a secondary NIC. This illustrates how VM traffic can be directed to a specific NIC on a node using a secondary NIC on a VM.