Shared control plane (single and multiple networks)

10 minute read

Setup a multicluster Istio service mesh across multiple clusters with a shared control plane. In this configuration, multiple Kubernetes clusters running a remote configuration connect to a shared Istio control plane running in a main cluster. Clusters may be on the same network or different networks than other clusters in the mesh. Once one or more remote Kubernetes clusters are connected to the Istio control plane, Envoy can then form a mesh.

Istio mesh spanning multiple Kubernetes clusters with direct network access to remote pods over VPN

Prerequisites

Two or more clusters running a supported Kubernetes version (1.15, 1.16, 1.17, 1.18).
All Kubernetes control plane API servers must be routable to each other.
Clusters on the same network must be an RFC1918 network, VPN, or an alternative more advanced network technique meeting the following requirements:
- Individual cluster Pod CIDR ranges and service CIDR ranges must be unique across the network and may not overlap.
- All pod CIDRs in the same network must be routable to each other.
Clusters on different networks must have istio-ingressgateway services which are accessible from every other cluster, ideally using L4 network load balancers (NLB). Not all cloud providers support NLBs and some require special annotations to use them, so please consult your cloud provider’s documentation for enabling NLBs for service object type load balancers. When deploying on platforms without NLB support, it may be necessary to modify the health checks for the load balancer to register the ingress gateway.

Preparation

Certificate Authority

Generate intermediate CA certificates for each cluster’s CA from your organization’s root CA. The shared root CA enables mutual TLS communication across different clusters. For illustration purposes, the following instructions use the certificates from the Istio samples directory for both clusters.

Run the following commands on each cluster in the mesh to install the certificates. See Certificate Authority (CA) certificates for more details on configuring an external CA.

$ kubectl create namespace istio-system
$ kubectl create secret generic cacerts -n istio-system \
    --from-file=@samples/certs/ca-cert.pem@ \
    --from-file=@samples/certs/ca-key.pem@ \
    --from-file=@samples/certs/root-cert.pem@ \
    --from-file=@samples/certs/cert-chain.pem@

Cross-cluster control plane access

Decide how to expose the main cluster’s Istiod discovery service to the remote clusters. Pick one of the two options:

Option (1) - Use the istio-ingressgateway gateway shared with data traffic.
Option (2) - Use a cloud provider’s internal load balancer on the Istiod service. For additional requirements and restrictions that may apply when using an internal load balancer between clusters, see Kubernetes internal load balancer documentation and your cloud provider’s documentation.

Cluster and network naming

Determine the name of the clusters and networks in the mesh. These names will be used in the mesh network configuration and when configuring the mesh’s service registries. Assign a unique name to each cluster. The name must be a DNS label name. In the example below the main cluster is called main0 and the remote cluster is remote0.

$ export MAIN_CLUSTER_CTX=<...>
$ export REMOTE_CLUSTER_CTX=<...>

$ export MAIN_CLUSTER_NAME=main0
$ export REMOTE_CLUSTER_NAME=remote0

If the clusters are on different networks, assign a unique network name for each network.

$ export MAIN_CLUSTER_NETWORK=network1
$ export REMOTE_CLUSTER_NETWORK=network2

If clusters are on the same network, the same network name is used for those clusters.

$ export MAIN_CLUSTER_NETWORK=network1
$ export REMOTE_CLUSTER_NETWORK=network1

Deployment

Main cluster

Create the main cluster’s configuration. Pick one of the two options for cross-cluster control plane access.

cat <<EOF> istio-main-cluster.yaml
apiVersion: install.istio.io/v1alpha1
kind: IstioOperator
spec:
  values:
    global:
      multiCluster:
        clusterName: ${MAIN_CLUSTER_NAME}
      network: ${MAIN_CLUSTER_NETWORK}

      # Mesh network configuration. This is optional and may be omitted if
      # all clusters are on the same network.
      meshNetworks:
        ${MAIN_CLUSTER_NETWORK}:
          endpoints:
          - fromRegistry:  ${MAIN_CLUSTER_NAME}
          gateways:
          - registry_service_name: istio-ingressgateway.istio-system.svc.cluster.local
            port: 443

        ${REMOTE_CLUSTER_NETWORK}:
          endpoints:
          - fromRegistry: ${REMOTE_CLUSTER_NAME}
          gateways:
          - registry_service_name: istio-ingressgateway.istio-system.svc.cluster.local
            port: 443

      # Use the existing istio-ingressgateway.
      meshExpansion:
        enabled: true
EOF

cat <<EOF> istio-main-cluster.yaml
apiVersion: install.istio.io/v1alpha1
kind: IstioOperator
spec:
  values:
    global:
      multiCluster:
        clusterName: ${MAIN_CLUSTER_NAME}
      network: ${MAIN_CLUSTER_NETWORK}

      # Mesh network configuration. This is optional and may be omitted if
      # all clusters are on the same network.
      meshNetworks:
        ${MAIN_CLUSTER_NETWORK}:
          endpoints:
          - fromRegistry:  ${MAIN_CLUSTER_NAME}
          gateways:
          - registry_service_name: istio-ingressgateway.istio-system.svc.cluster.local
            port: 443

        ${REMOTE_CLUSTER_NETWORK}:
          endpoints:
          - fromRegistry: ${REMOTE_CLUSTER_NAME}
          gateways:
          - registry_service_name: istio-ingressgateway.istio-system.svc.cluster.local
            port: 443

  # Change the Istio service `type=LoadBalancer` and add the cloud provider specific annotations. See
  # https://kubernetes.io/docs/concepts/services-networking/service/#internal-load-balancer for more
  # information. The example below shows the configuration for GCP/GKE.
  components:
    pilot:
      k8s:
        service:
          type: LoadBalancer
        service_annotations:
          cloud.google.com/load-balancer-type: Internal
EOF

Apply the main cluster’s configuration.

$ istioctl install -f istio-main-cluster.yaml --context=${MAIN_CLUSTER_CTX}

Wait for the control plane to be ready before proceeding.

$ kubectl get pod -n istio-system --context=${MAIN_CLUSTER_CTX}
NAME                                    READY   STATUS    RESTARTS   AGE
istio-ingressgateway-7c8dd65766-lv9ck   1/1     Running   0          136m
istiod-f756bbfc4-thkmk                  1/1     Running   0          136m
prometheus-b54c6f66b-q8hbt              2/2     Running   0          136m

Set the ISTIOD_REMOTE_EP environment variable based on which remote control plane configuration option was selected earlier.

$ export ISTIOD_REMOTE_EP=$(kubectl get svc -n istio-system --context=${MAIN_CLUSTER_CTX} istio-ingressgateway -o jsonpath='{.status.loadBalancer.ingress[0].ip}')
$ echo "ISTIOD_REMOTE_EP is ${ISTIOD_REMOTE_EP}"

$ export ISTIOD_REMOTE_EP=$(kubectl get svc -n istio-system --context=${MAIN_CLUSTER_CTX} istiod -o jsonpath='{.status.loadBalancer.ingress[0].ip}')
$ echo "ISTIOD_REMOTE_EP is ${ISTIOD_REMOTE_EP}"

Remote cluster

Create the remote cluster’s configuration.

cat <<EOF> istio-remote0-cluster.yaml
apiVersion: install.istio.io/v1alpha1
kind: IstioOperator
spec:
  values:
    global:
      # The remote cluster's name and network name must match the values specified in the
      # mesh network configuration of the main cluster.
      multiCluster:
        clusterName: ${REMOTE_CLUSTER_NAME}
      network: ${REMOTE_CLUSTER_NETWORK}

      # Replace ISTIOD_REMOTE_EP with the the value of ISTIOD_REMOTE_EP set earlier.
      remotePilotAddress: ${ISTIOD_REMOTE_EP}

  ## The istio-ingressgateway is not required in the remote cluster if both clusters are on
  ## the same network. To disable the istio-ingressgateway component, uncomment the lines below.
  #
  # components:
  #  ingressGateways:
  #  - name: istio-ingressgateway
  #    enabled: false
EOF

Apply the remote cluster configuration.

$ istioctl install -f istio-remote0-cluster.yaml --context ${REMOTE_CLUSTER_CTX}

Wait for the remote cluster to be ready.

$ kubectl get pod -n istio-system --context=${REMOTE_CLUSTER_CTX}
NAME                                    READY   STATUS    RESTARTS   AGE
istio-ingressgateway-55f784779d-s5hwl   1/1     Running   0          91m
istiod-7b4bfd7b4f-fwmks                 1/1     Running   0          91m
prometheus-c6df65594-pdxc4              2/2     Running   0          91m

Cross-cluster load balancing

Configure ingress gateways

Cross-network traffic is securely routed through each destination cluster’s ingress gateway. When clusters in a mesh are on different networks you need to configure port 443 on the ingress gateway to pass incoming traffic through to the target service specified in a request’s SNI header, for SNI values of the local top-level domain (i.e., the Kubernetes DNS domain). Mutual TLS connections will be used all the way from the source to the destination sidecar.

Apply the following configuration to each cluster.

cat <<EOF> cluster-aware-gateway.yaml
apiVersion: networking.istio.io/v1alpha3
kind: Gateway
metadata:
  name: cluster-aware-gateway
  namespace: istio-system
spec:
  selector:
    istio: ingressgateway
  servers:
  - port:
      number: 443
      name: tls
      protocol: TLS
    tls:
      mode: AUTO_PASSTHROUGH
    hosts:
    - "*.local"
EOF

$ kubectl apply -f cluster-aware-gateway.yaml --context=${MAIN_CLUSTER_CTX}
$ kubectl apply -f cluster-aware-gateway.yaml --context=${REMOTE_CLUSTER_CTX}

Configure cross-cluster service registries

To enable cross-cluster load balancing, the Istio control plane requires access to all clusters in the mesh to discover services, endpoints, and pod attributes. To configure access, create a secret for each remote cluster with credentials to access the remote cluster’s kube-apiserver and install it in the main cluster. This secret uses the credentials of the istio-reader-service-account in the remote cluster. --name specifies the remote cluster’s name. It must match the cluster name in main cluster’s IstioOperator configuration.

$ istioctl x create-remote-secret --name ${REMOTE_CLUSTER_NAME} --context=${REMOTE_CLUSTER_CTX} | \
    kubectl apply -f - --context=${MAIN_CLUSTER_CTX}

Deploy an example service

Deploy two instances of the helloworld service, one in each cluster. The difference between the two instances is the version of their helloworld image.

Deploy helloworld v2 in the remote cluster

Create a sample namespace with a sidecar auto-injection label:

$ kubectl create namespace sample --context=${REMOTE_CLUSTER_CTX}
$ kubectl label namespace sample istio-injection=enabled --context=${REMOTE_CLUSTER_CTX}

Deploy helloworld v2:

$ kubectl create -f @samples/helloworld/helloworld.yaml@ -l app=helloworld -n sample --context=${REMOTE_CLUSTER_CTX}
$ kubectl create -f @samples/helloworld/helloworld.yaml@ -l version=v2 -n sample --context=${REMOTE_CLUSTER_CTX}

Confirm helloworld v2 is running:

$ kubectl get pod -n sample --context=${REMOTE_CLUSTER_CTX}
NAME                             READY     STATUS    RESTARTS   AGE
helloworld-v2-7dd57c44c4-f56gq   2/2       Running   0          35s

Deploy helloworld v1 in the main cluster

Create a sample namespace with a sidecar auto-injection label:

$ kubectl create namespace sample --context=${MAIN_CLUSTER_CTX}
$ kubectl label namespace sample istio-injection=enabled --context=${MAIN_CLUSTER_CTX}

Deploy helloworld v1:

$ kubectl create -f @samples/helloworld/helloworld.yaml@ -l app=helloworld -n sample --context=${MAIN_CLUSTER_CTX}
$ kubectl create -f @samples/helloworld/helloworld.yaml@ -l version=v1 -n sample --context=${MAIN_CLUSTER_CTX}

Confirm helloworld v1 is running:

$ kubectl get pod -n sample --context=${MAIN_CLUSTER_CTX}
NAME                            READY     STATUS    RESTARTS   AGE
helloworld-v1-d4557d97b-pv2hr   2/2       Running   0          40s

Cross-cluster routing in action

To demonstrate how traffic to the helloworld service is distributed across the two clusters, call the helloworld service from another in-mesh sleep service.

Deploy the sleep service in both clusters:

$ kubectl apply -f @samples/sleep/sleep.yaml@ -n sample --context=${MAIN_CLUSTER_CTX}
$ kubectl apply -f @samples/sleep/sleep.yaml@ -n sample --context=${REMOTE_CLUSTER_CTX}

Wait for the sleep service to start in each cluster:

$ kubectl get pod -n sample -l app=sleep --context=${MAIN_CLUSTER_CTX}
sleep-754684654f-n6bzf           2/2     Running   0          5s

$ kubectl get pod -n sample -l app=sleep --context=${REMOTE_CLUSTER_CTX}
sleep-754684654f-dzl9j           2/2     Running   0          5s

Call the helloworld.sample service several times from the main cluster:

$ kubectl exec -it -n sample -c sleep --context=${MAIN_CLUSTER_CTX} $(kubectl get pod -n sample -l app=sleep --context=${MAIN_CLUSTER_CTX} -o jsonpath='{.items[0].metadata.name}') -- curl helloworld.sample:5000/hello

Call the helloworld.sample service several times from the remote cluster:

$ kubectl exec -it -n sample -c sleep --context=${REMOTE_CLUSTER_CTX} $(kubectl get pod -n sample -l app=sleep --context=${REMOTE_CLUSTER_CTX} -o jsonpath='{.items[0].metadata.name}') -- curl helloworld.sample:5000/hello

If set up correctly, the traffic to the helloworld.sample service will be distributed between instances on the main and remote clusters resulting in responses with either v1 or v2 in the body:

Hello version: v2, instance: helloworld-v2-758dd55874-6x4t8
Hello version: v1, instance: helloworld-v1-86f77cd7bd-cpxhv

You can also verify the IP addresses used to access the endpoints with istioctl proxy-config.

$ kubectl get pod -n sample -l app=sleep --context=${MAIN_CLUSTER_CTX} -o name | cut -f2 -d'/' | \
    xargs -I{} istioctl -n sample --context=${MAIN_CLUSTER_CTX} proxy-config endpoints {} --cluster "outbound|5000||helloworld.sample.svc.cluster.local"
ENDPOINT             STATUS      OUTLIER CHECK     CLUSTER
10.10.0.90:5000      HEALTHY     OK                outbound|5000||helloworld.sample.svc.cluster.local
192.23.120.32:443    HEALTHY     OK                outbound|5000||helloworld.sample.svc.cluster.local

In the main cluster, the endpoints are the gateway IP of the remote cluster (192.23.120.32:443) and the helloworld pod IP in the main cluster (10.10.0.90:5000).

$ kubectl get pod -n sample -l app=sleep --context=${REMOTE_CLUSTER_CTX} -o name | cut -f2 -d'/' | \
    xargs -I{} istioctl -n sample --context=${REMOTE_CLUSTER_CTX} proxy-config endpoints {} --cluster "outbound|5000||helloworld.sample.svc.cluster.local"
ENDPOINT             STATUS      OUTLIER CHECK     CLUSTER
10.32.0.9:5000       HEALTHY     OK                outbound|5000||helloworld.sample.svc.cluster.local
192.168.1.246:443    HEALTHY     OK                outbound|5000||helloworld.sample.svc.cluster.local

In the remote cluster, the endpoints are the gateway IP of the main cluster (192.168.1.246:443) and the pod IP in the remote cluster (10.32.0.9:5000).

Congratulations!

You have configured a multi-cluster Istio mesh, installed samples and verified cross cluster traffic routing.

Additional considerations

Automatic injection

The Istiod service in each cluster provides automatic sidecar injection for proxies in its own cluster. Namespaces must be labeled in each cluster following the automatic sidecar injection guide

Access services from different clusters

Kubernetes resolves DNS on a cluster basis. Because the DNS resolution is tied to the cluster, you must define the service object in every cluster where a client runs, regardless of the location of the service’s endpoints. To ensure this is the case, duplicate the service object to every cluster using kubectl. Duplication ensures Kubernetes can resolve the service name in any cluster. Since the service objects are defined in a namespace, you must define the namespace if it doesn’t exist, and include it in the service definitions in all clusters.

Security

The Istiod service in each cluster provides CA functionality to proxies in its own cluster. The CA setup earlier ensures proxies across clusters in the mesh have the same root of trust.

Uninstalling the remote cluster

To uninstall the remote cluster, run the following command:

$ istioctl x create-remote-secret --name ${REMOTE_CLUSTER_NAME} --context=${REMOTE_CLUSTER_CTX} | \
    kubectl delete -f - --context=${MAIN_CLUSTER_CTX}
$ istioctl manifest generate -f istio-remote0-cluster.yaml --context=${REMOTE_CLUSTER_CTX} | \
    kubectl delete -f - --context=${REMOTE_CLUSTER_CTX}
$ kubectl delete namespace sample --context=${REMOTE_CLUSTER_CTX}
$ unset REMOTE_CLUSTER_CTX REMOTE_CLUSTER_NAME REMOTE_CLUSTER_NETWORK
$ rm istio-remote0-cluster.yaml

To uninstall the main cluster, run the following command:

$ istioctl manifest generate -f istio-main-cluster.yaml --context=${MAIN_CLUSTER_CTX} | \
    kubectl delete -f - --context=${MAIN_CLUSTER_CTX}
$ kubectl delete namespace sample --context=${MAIN_CLUSTER_CTX}
$ unset MAIN_CLUSTER_CTX MAIN_CLUSTER_NAME MAIN_CLUSTER_NETWORK ISTIOD_REMOTE_EP
$ rm istio-main-cluster.yaml cluster-aware-gateway.yaml