Kubernetes Cloud Controller Manager for Linode

The purpose of the CCM

The Linode Cloud Controller Manager (CCM) creates a fully supported Kubernetes experience on Linode.

• Load balancers, Linode NodeBalancers, are automatically deployed when a Kubernetes Service of type "LoadBalancer" is deployed. This is the most reliable way to allow services running in your cluster to be reachable from the Internet.
• Linode hostnames and network addresses (private/public IPs) are automatically associated with their corresponding Kubernetes resources, forming the basis for a variety of Kubernetes features.
• Nodes resources are put into the correct state when Linodes are shut down, allowing pods to be appropriately rescheduled.
• Nodes are annotated with the Linode region, which is the basis for scheduling based on failure domains.

Kubernetes Supported Versions

Kubernetes 1.9+

Usage

LoadBalancer Services

Kubernetes Services of type LoadBalancer will be served through a Linode NodeBalancer which the Cloud Controller Manager will provision on demand. For general feature and usage notes, refer to the Getting Started with Linode NodeBalancers guide.

Annotations

The Linode CCM accepts several annotations which affect the properties of the underlying NodeBalancer deployment.

All of the service annotation names listed below have been shortened for readability. Each annotation MUST be prefixed with service.beta.kubernetes.io/linode-loadbalancer-. The values, such as http, are case-sensitive.

Annotation (Suffix)	Values	Default	Description
throttle	0-20 (0 to disable)	20	Client Connection Throttle, which limits the number of subsequent new connections per second from the same client IP
default-protocol	tcp, http, https	tcp	This annotation is used to specify the default protocol for Linode NodeBalancer.
default-proxy-protocol	none, v1, v2	none	Specifies whether to use a version of Proxy Protocol on the underlying NodeBalancer.
port-*	json (e.g. { "tls-secret-name": "prod-app-tls", "protocol": "https", "proxy-protocol": "v2"})		Specifies port specific NodeBalancer configuration. See Port Specific Configuration. * is the port being configured, e.g. linode-loadbalancer-port-443
check-type	none, connection, http, http_body		The type of health check to perform against back-ends to ensure they are serving requests
check-path	string		The URL path to check on each back-end during health checks
check-body	string		Text which must be present in the response body to pass the NodeBalancer health check
check-interval	int		Duration, in seconds, to wait between health checks
check-timeout	int (1-30)		Duration, in seconds, to wait for a health check to succeed before considering it a failure
check-attempts	int (1-30)		Number of health check failures necessary to remove a back-end from the service
check-passive	bool	false	When true, 5xx status codes will cause the health check to fail
preserve	bool	false	When true, deleting a LoadBalancer service does not delete the underlying NodeBalancer. This will also prevent deletion of the former LoadBalancer when another one is specified with the nodebalancer-id annotation.
nodebalancer-id	string		The ID of the NodeBalancer to front the service. When not specified, a new NodeBalancer will be created. This can be configured on service creation or patching
hostname-only-ingress	bool	false	When true, the LoadBalancerStatus for the service will only contain the Hostname. This is useful for bypassing kube-proxy's rerouting of in-cluster requests originally intended for the external LoadBalancer to the service's constituent pod IPs.

Deprecated Annotations

These annotations are deprecated, and will be removed in a future release.

Annotation (Suffix)	Values	Default	Description	Scheduled Removal
proxy-protcol	none, v1, v2	none	Specifies whether to use a version of Proxy Protocol on the underlying NodeBalancer	Q4 2021

Annotation bool values

For annotations with bool value types, "1", "t", "T", "True", "true" and "True" are valid string representations of true. Any other values will be interpreted as false. For more details, see strconv.ParseBool.

Port Specific Configuration

These configuration options can be specified via the port-* annotation, encoded in JSON.

Key	Values	Default	Description
protocol	tcp, http, https	tcp	Specifies protocol of the NodeBalancer port. Overwrites default-protocol.
proxy-protocol	none, v1, v2	none	Specifies whether to use a version of Proxy Protocol on the underlying NodeBalancer. Overwrites default-proxy-protocol.
tls-secret-name	string		Specifies a secret to use for TLS. The secret type should be kubernetes.io/tls.

Example usage

kind: Service
apiVersion: v1
metadata:
  name: https-lb
  annotations:
    service.beta.kubernetes.io/linode-loadbalancer-throttle: "4"
    service.beta.kubernetes.io/linode-loadbalancer-default-protocol: "http"
    service.beta.kubernetes.io/linode-loadbalancer-port-443: |
      {
        "tls-secret-name": "example-secret",
        "protocol": "https"
      }
spec:
  type: LoadBalancer
  selector:
    app: nginx-https-example
  ports:
    - name: http
      protocol: TCP
      port: 80
      targetPort: http
    - name: https
      protocol: TCP
      port: 443
      targetPort: https

---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: nginx-https-deployment
spec:
  replicas: 2
  selector:
    matchLabels:
      app: nginx-https-example
  template:
    metadata:
      labels:
        app: nginx-https-example
    spec:
      containers:
      - name: nginx
        image: nginx
        ports:
          - name: http
            containerPort: 80
            protocol: TCP
          - name: https
            containerPort: 80
            protocol: TCP

See more in the examples directory

Why stickiness and algorithm annotations don't exist

As kube-proxy will simply double-hop the traffic to a random backend Pod anyway, it doesn't matter which backend Node traffic is forwarded-to for the sake of session stickiness. These annotations are not necessary to implement session stickiness, as kube-proxy will simply double-hop the packets to a random backend Pod. It would not make a difference to set a backend Node that would receive the network traffic in an attempt to set session stickiness.

How to use sessionAffinity

In Kubernetes, sessionAffinity refers to a mechanism that allows a client always to be redirected to the same pod when the client hits a service.

To enable sessionAffinity service.spec.sessionAffinity field must be set to ClientIP as the following service yaml:

apiVersion: v1
kind: Service
metadata:
  name: wordpress-lsmnl-wordpress
  namespace: wordpress-lsmnl
  labels:
    app: wordpress-lsmnl-wordpress
spec:
  type: LoadBalancer
  selector:
    app: wordpress-lsmnl-wordpress
  sessionAffinity: ClientIP

The max session sticky time can be set by setting the field service.spec.sessionAffinityConfig.clientIP.timeoutSeconds as below:

sessionAffinityConfig:
  clientIP:
    timeoutSeconds: 100

Generating a Manifest for Deployment

Use the script located at ./deploy/generate-manifest.sh to generate a self-contained deployment manifest for the Linode CCM. Two arguments are required.

The first argument must be a Linode APIv4 Personal Access Token with all permissions. (https://cloud.linode.com/profile/tokens)

The second argument must be a Linode region. (https://api.linode.com/v4/regions)

Example:

./deploy/generate-manifest.sh $LINODE_API_TOKEN us-east

This will create a file ccm-linode.yaml which you can use to deploy the CCM.

kubectl apply -f ccm-linode.yaml

Note: Your kubelets, controller-manager, and apiserver must be started with --cloud-provider=external as noted in the following documentation.

Upstream Documentation Including Deployment Instructions

Kubernetes Cloud Controller Manager.

Upstream Developer Documentation

Developing a Cloud Controller Manager.

Development Guide

Building the Linode Cloud Controller Manager

Some of the Linode Cloud Controller Manager development helper scripts rely on a fairly up-to-date GNU tools environment, so most recent Linux distros should work just fine out-of-the-box.

Setup Go

The Linode Cloud Controller Manager is written in Google's Go programming language. Currently, the Linode Cloud Controller Manager is developed and tested on Go 1.8.3. If you haven't set up a Go development environment, please follow these instructions to install Go.

On macOS, Homebrew has a nice package

brew install golang

Download Source

go get github.com/linode/linode-cloud-controller-manager
cd $(go env GOPATH)/src/github.com/linode/linode-cloud-controller-manager

Install Dev tools

To install various dev tools for Pharm Controller Manager, run the following command:

./hack/builddeps.sh

Build Binary

Use the following Make targets to build and run a local binary

$ make build
$ make run
# You can also run the binary directly to pass additional args
$ dist/linode-cloud-controller-manager

Dependency management

Linode Cloud Controller Manager uses Go Modules to manage dependencies. If you want to update/add dependencies, run:

go mod tidy

Building Docker images

To build and push a Docker image, use the following make targets.

# Set the repo/image:tag with the TAG environment variable
# Then run the docker-build make target
$ IMG=linode/linode-cloud-controller-manager:canary make docker-build

# Push Image
$ IMG=linode/linode-cloud-controller-manager:canary make docker-push

Then, to run the image

docker run -ti linode/linode-cloud-controller-manager:canary

Contribution Guidelines

Want to improve the linode-cloud-controller-manager? Please start here.

Join the Kubernetes Community

For general help or discussion, join us in #linode on the Kubernetes Slack. To sign up, use the Kubernetes Slack inviter.