Skip to content

Latest commit

 

History

History
171 lines (148 loc) · 6.33 KB

File metadata and controls

171 lines (148 loc) · 6.33 KB

multus-dynamic-networks-controller

This project provides a Kubernetes controller granting the ability to plug/unplug network interfaces to / from running pods.

This controller extends the multus-cni functionality, by listening to pod's network selection elements (i.e. the pod k8s.v1.cni.cncf.io/networks annotation); whenever those change (adding / removing network selection elements), it will invoke the corresponding delegate effectively adding (or removing) a network interface to a running pod.

The controller provides a reconciliation loop between the desired state (the network selection elements), and the current state (the network status); in order for it to work, all the attachments featured in the desired state must feature interface names. If they don't, the user should patch up the pod's network selection elements, adding it the name of the interface for each respective attachment.

Please refer to the multus-cni docs for more information on how additional interfaces are added to a pod.

We've finally hit MVP. We would be extremely interested in getting your feedback.

Please open issues / RFEs if something does not meet your expectations.

Usage

Requirements

  • a running Kubernetes cluster
  • multus deployed (in thick-plugin mode) in the Kubernetes cluster
  • the kubectl binary

Installation

Use the provided manifest to install the controller on your cluster:

kubectl apply -f manifests/dynamic-networks-controller.yaml

Removal

Use kubectl to remove the controller from your cluster:

kubectl delete -f manifests/dynamic-networks-controller.yaml

Adding / removing network interfaces

To add (or remove...) network interfaces from a running pod, the user should simply edit the running pod network selection elements - i.e. the k8s.v1.cni.cncf.io/networks annotation.

Thus, if we had this running pod (and NetworkAttachmentDefinition):

---
apiVersion: "k8s.cni.cncf.io/v1"
kind: NetworkAttachmentDefinition
metadata:
  name: macvlan1-config
spec:
  config: '{
            "cniVersion": "0.4.0",
            "plugins": [
                {
                    "type": "macvlan",
                    "capabilities": { "ips": true },
                    "master": "eth1",
                    "mode": "bridge",
                    "ipam": {
                        "type": "static"
                    }
                }, {
                    "type": "tuning"
                } ]
        }'
---
apiVersion: v1
kind: Pod
metadata:
  name: macvlan1-worker1
  annotations:
    k8s.v1.cni.cncf.io/networks: '[
            {
                "name": "macvlan1-config",
                "ips": [ "10.1.1.11/24" ],
                "interface": "net1"
            }
    ]'
  labels:
    app: macvlan
spec:
  containers:
  - name: macvlan-worker1
    image: docker.io/library/alpine:latest
    command: ["/bin/sleep", "10000"]

NOTE the "interface": "net1" in the JSON networks annotations element is an optional parameter, however it's needed right now to hot-unplug this interface see issue

We would run this example yaml to add an interface from it:

---
apiVersion: v1
kind: Pod
metadata:
  name: macvlan1-worker1
  annotations:
    k8s.v1.cni.cncf.io/networks: '[
            {
                "name": "macvlan1-config",
                "ips": [ "10.1.1.11/24" ],
                "interface": "net1"
            },
            {
                "name": "macvlan1-config",
                "ips": [ "10.1.2.11/24" ],
                "interface": "ens4"
            }
    ]'
  labels:
    app: macvlan
spec:
  containers:
    - name: macvlan-worker1
      image: docker.io/library/alpine:latest
      command: ["/bin/sleep", "10000"]

And we would run this example yaml to remove an interface to it:

---
apiVersion: v1
kind: Pod
metadata:
  name: macvlan1-worker1
  annotations:
    k8s.v1.cni.cncf.io/networks: '[]' # this will remove all networks from the pod
  labels:
    app: macvlan
spec:
  containers:
    - name: macvlan-worker1
      image: docker.io/library/alpine:latest
      command: ["/bin/sleep", "10000"]

Configuration

The multus-dynamic-networks-controller configuration is encoded in JSON, and allows the following keys:

  • "criSocketPath": specify the path to the CRI socket. Defaults to /run/containerd/containerd.sock.
  • "multusSocketPath": specify the path to the multus socket. Defaults to /var/run/multus-cni/multus.sock.

The configuration is defined in a ConfigMap, which is defined in the installation manifest, and mounted into the pod.

The name of the ConfigMap is dynamic-networks-controller-config.

Developer Workflow

Below you can find information on how to push local code changes to a kind cluster.

  • change code :)
  • start up a kind cluster. I've tested using the multus repo e2e kind cluster.
  • build image: IMAGE_REGISTRY=localhost:5000/maiqueb OCI_BIN=podman make img-build. NOTE: this assumes podman is used. docker is the default OCI_BIN.
  • push image to local registry: podman push localhost:5000/maiqueb/multus-dynamic-networks-controller
  • update manifests to use the generated image: IMAGE_REGISTRY=localhost:5000/maiqueb make manifests
  • deploy the controller: kubectl apply -f manifests/dynamic-networks-controller.yaml

Mapping a container image to the code

To know which git commit ID is in a certain container image perform the following steps:

podman inspect ghcr.io/k8snetworkplumbingwg/multus-dynamic-networks-controller:latest-amd64 -f '{{index .Labels "multi.GIT_SHA"}}'
e1db8da3c6267b3c2a5aca72ef8dd6a10b0ec9fd

Known limitations

  • plug / unplug interfaces to networks requiring device-plugin interaction. We must investigate this further; an RFE may be opened once we have the required data.