mcs

Multi-Cluster Service Operator on Azure

An operator for exposing services from multiple Kubernetes clusters by leveraging Azure cross-region load balancer.

• Status: DRAFT
• Issues: due to limitations of Global LoadBalancer, FloatingIP needs to be enabled on both Global LoadBalancer and regional LoadBalancer. And the the global VIP needs to be configured on each nodes to make FloatingIP work (this could be done by adding global VIP to each cluster's service.status.loadBalancer.ingress[].

How to deploy MCS operator

Create Azure service principal and then create the the following cloud-config file:

{
  "cloud": "AzurePublicCloud",
  "tenantId": "<tenantId>",
  "subscriptionId": "<subscriptionId>",
  "aadClientId": "<aadClientId>",
  "aadClientSecret": "<aadClientSecret>",
  "globalLoadBalancerName": "<glbName>",
  "globalVIPLocation": "<region>",
  "globalLoadBalancerResourceGroup": "<resourceGroup>"
}

Then create a secret based on this config file:

kubectl create secret generic azure-mcs-config --from-file=cloud-config

After that, deploy the MCS operator in MCS cluster (it could be any Kubernetes cluster):

kubectx mcs-cluster
make deploy

Samples

Create KubeCluster

Assume you have two Kubernetes clusters running on Azure, one is created via aks-engine and the other is created via AKS.

Switch kubeconfig to MCS cluster, and then create the following two KubeCluster:

# For cluster aks-engine
kubectl create secret generic aks-engine --from-file=kubeconfig
cat <<EOF | kubectl apply -f -
apiVersion: networking.aks.io/v1alpha1
kind: KubeCluster
metadata:
  name: aks-engine
  namespace: default
spec:
  kubeConfigSecret: aks-engine
EOF

# For cluster aks-cluster
kubectl create secret generic aks-cluster --from-file=kubeconfig
cat <<EOF | kubectl apply -f -
apiVersion: networking.aks.io/v1alpha1
kind: KubeCluster
metadata:
  name: aks-cluster
  namespace: default
spec:
  kubeConfigSecret: aks-cluster
EOF

Create GlobalService

Switch kubeconfig to MCS cluster and then create global service:

kubectx mcs-cluster

cat <<EOF | kubectl apply -f -
apiVersion: networking.aks.io/v1alpha1
kind: GlobalService
metadata:
  name: nginx
  namespace: default
spec:
  clusterSet:
  - aks-engine
  - aks-cluster
  ports:
  - name: http
    port: 80
    protocol: TCP
EOF

Deploy nginx service in both cluster

MCS operator assumes the service names are same as globalservice in all clusters.

kubectx aks-engine
kubectl create deployment nginx --image nginx --save-config
kubectl expose deploy nginx --port=80 --type=LoadBalancer
kubectl get service nginx

kubectx aks-cluster
kubectl create deployment nginx --image nginx --save-config
kubectl expose deploy nginx --port=80 --type=LoadBalancer
kubectl get service nginx

Verify Global VIP

Switch kubeconfig back to MCS cluster and then verify the VIP for the global service:

$ kubectl get globalservice nginx -o yaml
apiVersion: networking.aks.io/v1alpha1
kind: GlobalService
metadata:
  name: nginx
  namespace: default
  resourceVersion: "279410693"
  selfLink: /apis/networking.aks.io/v1alpha1/namespaces/default/globalservices/nginx
  uid: edb771f9-328e-4785-b94d-b70e13b10ffe
spec:
  ports:
  - name: http
    port: 80
    protocol: TCP
status:
  endpoints:
  - cluster: default/aks-engine
    ip: 52.184.9.136
    resourceGroup: "<rg1>"
  - cluster: default/aks-cluster
    ip: 20.198.185.119
    resourceGroup: "<rg2>"
  vip: 23.98.101.30

# verify the vip is available
$ curl 23.98.101.30