Get started with VPP networking

Big picture

Install Calico and enable the beta release of the VPP dataplane.

Get started with VPP networking - 图1caution

The VPP dataplane is in beta and should not be used in production clusters. It has had lots of testing and is pretty stable. However, chances are that some bugs are still lurking around (please report these on the Calico Users slack or Github). In addition, it still does not support all the features of Calico.

Value

The VPP dataplane mode has several advantages over standard Linux networking pipeline mode:

  • Scales to higher throughput, especially with WireGuard encryption enabled
  • Further improves encryption performance with IPsec
  • Native support for Kubernetes services without needing kube-proxy, which:
    • Reduces first-packet latency for packets to services
    • Preserves external client source IP addresses all the way to the pod

The VPP dataplane is entirely compatible with the other Calico dataplanes, meaning you can have a cluster with VPP-enabled nodes along with regular nodes. This makes it possible to migrate a cluster from Linux or eBPF networking to VPP networking.

In addition, the VPP dataplane offers some specific features for network-intensive applications, such as providing memif userspace packet interfaces to the pods (instead of regular Linux network devices), or exposing the VPP Host Stack to run optimized L4+ applications in the pods.

Trying out the beta will give you a taste of these benefits and an opportunity to give feedback to the VPP dataplane team.

Features

This how-to guide uses the following Calico features:

  • calico/node
  • VPP dataplane

Concepts

VPP

The Vector Packet Processor (VPP) is a high-performance, open-source userspace network dataplane written in C, developed under the fd.io umbrella. It supports many standard networking features (L2 switching, L3 routing, NAT, encapsulations), and is easily extensible using plugins. The VPP dataplane uses plugins to efficiently implement Kubernetes services load balancing and Calico policies.

Operator based installation

This guide uses the Tigera operator to install Calico. The operator provides lifecycle management for Calico exposed via the Kubernetes API defined as a custom resource definition. While it is also technically possible to install Calico and configure it for VPP using manifests directly, only operator based installations are supported at this stage.

How to

This guide details two ways to install Calico with the VPP dataplane:

  • On a managed EKS cluster. This is the option that requires the least configuration
  • On a managed EKS cluster with the DPDK interface driver. This options is more complex to setup but provides better performance
  • On any Kubernetes cluster

In all cases, here are the details of what you will get:

PolicyIPAMCNIOverlayRoutingDatastore
  • Install on EKS
  • Install on EKS with DPDK
  • Install on any cluster

Install Calico with the VPP dataplane on an EKS cluster

Requirements

For these instructions, we will use eksctl to provision the cluster. However, you can use any of the methods in Getting Started with Amazon EKS

Before you get started, make sure you have downloaded and configured the necessary prerequisites

Provision the cluster

  1. First, create an Amazon EKS cluster without any nodes.

    1. eksctl create cluster --name my-calico-cluster --without-nodegroup
  2. Since this cluster will use Calico for networking, you must delete the aws-node DaemonSet to disable the default AWS VPC networking for the pods.

    1. kubectl delete daemonset -n kube-system aws-node

Install and configure Calico with the VPP dataplane

  1. Now that you have an empty cluster configured, you can install the Tigera operator.

    1. kubectl create -f https://raw.githubusercontent.com/projectcalico/calico/v3.25.0/manifests/tigera-operator.yaml

    Get started with VPP networking - 图2note

    Due to the large size of the CRD bundle, kubectl apply might exceed request limits. Instead, use kubectl create or kubectl replace.

  2. Then, you need to configure the Calico installation for the VPP dataplane. The yaml in the link below contains a minimal viable configuration for EKS. For more information on configuration options available in this manifest, see the installation reference.

    Get started with VPP networking - 图3note

    Before applying this manifest, read its contents and make sure its settings are correct for your environment. For example, you may need to specify the default IP pool CIDR to match your desired pod network CIDR.

    1. kubectl create -f https://raw.githubusercontent.com/projectcalico/vpp-dataplane/master/yaml/calico/installation-eks.yaml
  3. Now is time to install the VPP dataplane components.

    1. kubectl create -f https://raw.githubusercontent.com/projectcalico/vpp-dataplane/master/yaml/generated/calico-vpp-eks.yaml
  4. Finally, add nodes to the cluster.

    1. eksctl create nodegroup --cluster my-calico-cluster --node-type t3.medium --node-ami auto --max-pods-per-node 50

    Get started with VPP networking - 图4tip

    The —max-pods-per-node option above, ensures that EKS does not limit the number of pods based on node-type. For the full set of node group options, see eksctl create nodegroup --help.

Install Calico with the VPP dataplane on an EKS cluster with the DPDK driver

Requirements

DPDK provides better performance compared to the standard install but it requires some additional customisations (hugepages, for instance) in the EKS worker instances. We have a bash script, init_eks.sh, which takes care of applying the required customizations and we make use of the preBootstrapCommands property of eksctl configuration file to execute the script during the worker node creation. These instructions require the latest version of eksctl.

Provision the cluster

  1. First, create an Amazon EKS cluster without any nodes.

    1. eksctl create cluster --name my-calico-cluster --without-nodegroup
  2. Since this cluster will use Calico for networking, you must delete the aws-node DaemonSet to disable the default AWS VPC networking for the pods.

    1. kubectl delete daemonset -n kube-system aws-node

Install and configure Calico with the VPP dataplane

  1. Now that you have an empty cluster configured, you can install the Tigera operator.

    1. kubectl create -f https://raw.githubusercontent.com/projectcalico/calico/v3.25.0/manifests/tigera-operator.yaml

    Get started with VPP networking - 图5note

    Due to the large size of the CRD bundle, kubectl apply might exceed request limits. Instead, use kubectl create or kubectl replace.

  2. Then, you need to configure the Calico installation for the VPP dataplane. The yaml in the link below contains a minimal viable configuration for EKS. For more information on configuration options available in this manifest, see the installation reference.

    Get started with VPP networking - 图6note

    Before applying this manifest, read its contents and make sure its settings are correct for your environment. For example, you may need to specify the default IP pool CIDR to match your desired pod network CIDR.

    1. kubectl create -f https://raw.githubusercontent.com/projectcalico/vpp-dataplane/master/yaml/calico/installation-eks.yaml
  3. Now is time to install the VPP dataplane components.

    1. kubectl create -f https://raw.githubusercontent.com/projectcalico/vpp-dataplane/master/yaml/generated/calico-vpp-eks-dpdk.yaml
  4. Finally, time to add nodes to the cluster. Since we need to customize the nodes for DPDK, we will use an eksctl config file with the preBootstrapCommands property to create the worker nodes. The following command will create a managed nodegroup with 2 t3.large worker nodes in the cluster:

    1. cat <<EOF | eksctl create nodegroup -f -
    2. apiVersion: eksctl.io/v1alpha5
    3. kind: ClusterConfig
    4. metadata:
    5. name: my-calico-cluster
    6. region: us-east-2
    7. managedNodeGroups:
    8. - name: my-calico-cluster-ng
    9. desiredCapacity: 2
    10. instanceType: t3.large
    11. labels: {role: worker}
    12. preBootstrapCommands:
    13. - sudo curl -o /tmp/init_eks.sh "https://raw.githubusercontent.com/projectcalico/vpp-dataplane/master/scripts/init_eks.sh"
    14. - sudo chmod +x /tmp/init_eks.sh
    15. - sudo /tmp/init_eks.sh
    16. EOF

    Please edit the cluster name, region and other fields as appropriate for your cluster. In case you want to enable ssh access to the EKS worker instances, add the following to the above config file:

    1. ssh:
    2. publicKeyPath: <path to public key>

    For details on ssh access refer to Amazon EC2 key pairs and Linux instances.

Install Calico with the VPP dataplane on any Kubernetes cluster

Requirements

The VPP dataplane has the following requirements:

Required

  • A blank Kubernetes cluster, where no CNI was ever configured.

  • These base requirements, except those related to the management of cali*, tunl* and vxlan.calico interfaces.

    Get started with VPP networking - 图7note

    If you are using kubeadm to create the cluster please make sure to specify the pod network CIDR using the --pod-network-cidr command-line argument, i.e., sudo kubeadm init --pod-network-cidr=192.168.0.0/16. If 192.168.0.0/16 is already in use within your network you must select a different pod network CIDR.

Optional For some hardware, the following hugepages configuration may enable VPP to use more efficient drivers:

  • At least 512 x 2MB-hugepages are available (grep HugePages_Free /proc/meminfo)

  • The vfio-pci (vfio_pci on centos) or uio_pci_generic kernel module is loaded. For example:

    1. echo "vfio-pci" > /etc/modules-load.d/95-vpp.conf
    2. modprobe vfio-pci
    3. echo "vm.nr_hugepages = 512" >> /etc/sysctl.conf
    4. sysctl -p
    5. restart kubelet to take the changes into account
    6. you may need to use a different command depending on how kubelet was installed
    7. systemctl restart kubelet

Install Calico and configure it for VPP

  1. Start by installing the Tigera operator on your cluster.

    1. kubectl create -f https://raw.githubusercontent.com/projectcalico/calico/v3.25.0/manifests/tigera-operator.yaml

    Get started with VPP networking - 图8note

    Due to the large size of the CRD bundle, kubectl apply might exceed request limits. Instead, use kubectl create or kubectl replace.

  2. Then, you need to configure the Calico installation for the VPP dataplane. The yaml in the link below contains a minimal viable configuration for VPP. For more information on configuration options available in this manifest, see the installation reference.

    Get started with VPP networking - 图9note

    Before applying this manifest, read its contents and make sure its settings are correct for your environment. For example, you may need to specify the default IP pool CIDR to match your desired pod network CIDR.

    1. kubectl create -f https://raw.githubusercontent.com/projectcalico/vpp-dataplane/master/yaml/calico/installation-default.yaml

Install the VPP dataplane components

Start by getting the appropriate yaml manifest for the VPP dataplane resources:

  1. If you have configured hugepages on your machines
  2. curl -o calico-vpp.yaml https://raw.githubusercontent.com/projectcalico/vpp-dataplane/master/yaml/generated/calico-vpp.yaml
  1. If not, or if you're unsure
  2. curl -o calico-vpp.yaml https://raw.githubusercontent.com/projectcalico/vpp-dataplane/master/yaml/generated/calico-vpp-nohuge.yaml

Then locate the calico-vpp-config ConfigMap in this yaml manifest and configure it as follows.

Required Configuration

  • CALICOVPP_INTERFACES contains a dictionary with parameteres specific to interfaces in calicovpp. The field uplinkInterfaces contains a list of interfaces and their configuration, with the first element being the primary interface, and the rest (if any) being the secondary host interfaces.
  1. CALICOVPP_INTERFACES: |-
  2. {
  3. "uplinkInterfaces": [ { "interfaceName": "eth0" } ]
  4. }

The name of the used interface must be the name of a Linux interface, up and configured with an address. The address configured on this interface must be the node address in Kubernetes (kubectl get nodes -o wide).

Configuration options

CALICOVPP_INTERFACES

FieldDescriptionType
maxPodIfSpecspec containing max values for pod interfaces configInterfaceSpec
defaultPodIfSpecspec containing default values for pod interfaces configInterfaceSpec
vppHostTapSpecspec containing config for host tap interface in vppInterfaceSpec
uplinkInterfaceslist of host interfaces in vppList of UplinkInterfaceSpec

InterfaceSpec

FieldDescriptionTypeDefault
rxNumber of RX queuesint1
txNumber of TX queuesint1
rxqszRX queue sizeint1024
txqszTX queue sizeint1024
isl3Defines the interface mode (L2/L3)booleantrue for tuntap ; false for memif
rxModeRX modestring among “interrupt”, “adaptive”, or “polling”adaptive

UplinkInterfaceSpec

FieldDescriptionTypeDefault
rxNumber of RX queuesint1
txNumber of TX queuesint1
rxqszRX queue sizeint1024
txqszTX queue sizeint1024
isl3Defines the interface mode (L2/L3) for drivers that support itbooleantrue
rxModeRX modestring among “interrupt”, “adaptive”, or “polling”adaptive
InterfaceNameinterface namestringunset
vppDriverdriver to use in vppstringunset
newDriverlinux driver to use before passing the interface to VPPstringunset
mtuthe interface’s mtuintuse the existing MTU in linux
  • service_prefix is the Kubernetes service CIDR. You can retrieve it by running:
  1. kubectl cluster-info dump | grep -m 1 service-cluster-ip-range

If this command doesn’t return anything, you can leave the default value of 10.96.0.0/12.

Optional

  • To configure how VPP drives the physical interface, use vppDriver field for uplinkInterfaces elements in CALICOVPP_INTERFACES.

The supported values will depend on the interface type. Available values are:

  • "" : will automatically select and try drivers based on interface type and available resources, starting with the fastest
  • af_xdp : use an AF_XDP socket to drive the interface (requires kernel 5.4 or newer)
  • af_packet : use an AF_PACKET socket to drive the interface (not optimized but works everywhere)
  • avf : use the VPP native driver for Intel 700-Series and 800-Series interfaces (requires hugepages)
  • vmxnet3 : use the VPP native driver for VMware virtual interfaces (requires hugepages)
  • virtio : use the VPP native driver for Virtio virtual interfaces (requires hugepages)
  • rdma : use the VPP native driver for Mellanox CX-4 and CX-5 interfaces (requires hugepages)
  • dpdk : use the DPDK interface drivers with VPP (requires hugepages, works with most interfaces)
  • none : do not configure connectivity automatically. This can be used when configuring the interface manually

Legacy options

We maintain legacy support for the CALICOVPP_INTERFACE and CALICOVPP_NATIVE_DRIVER environment variables:

CALICOVPP_INTERFACE -> uplinkInterfaces[0].interfaceName

CALICOVPP_NATIVE_DRIVER -> uplinkInterfaces[0].vppDriver

If CALICOVPP_INTERFACES is unspecified, CALICOVPP_INTERFACE is the primary interface to be used. In that case, use CALICOVPP_NATIVE_DRIVER instead of vppDriver.

So either patch CALICOVPP_INTERFACES with the suitable interface in uplinkInterfaces, or delete CALICOVPP_INTERFACES and use CALICOVPP_INTERFACE instead.

Example

  1. kind: ConfigMap
  2. apiVersion: v1
  3. metadata:
  4. name: calico-config
  5. namespace: calico-vpp-dataplane
  6. data:
  7. service_prefix: 10.96.0.0/12
  8. vpp_dataplane_interface: eth1
  9. vpp_uplink_driver: ""
  10. ...

Apply the configuration

To apply the configuration, run:

  1. kubectl create -f calico-vpp.yaml

This will install all the resources required by the VPP dataplane in your cluster.

Next steps

After installing Calico with the VPP dataplane, you can benefit from the features of the VPP dataplane, such as fast IPsec or Wireguard encryption.

Tools

Security