Replicated control planes

11 minute read

Follow this guide to install an Istio multicluster deployment with replicated control plane instances in every cluster and using gateways to connect services across clusters.

Instead of using a shared Istio control plane to manage the mesh, in this configuration each cluster has its own Istio control plane installation, each managing its own endpoints. All of the clusters are under a shared administrative control for the purposes of policy enforcement and security.

A single Istio service mesh across the clusters is achieved by replicating shared services and namespaces and using a common root CA in all of the clusters. Cross-cluster communication occurs over the Istio gateways of the respective clusters.

Istio mesh spanning multiple Kubernetes clusters using Istio Gateway to reach remote pods

Prerequisites

Two or more Kubernetes clusters with versions: 1.15, 1.16, 1.17, 1.18.
Authority to deploy the Istio control plane on each Kubernetes cluster.
The IP address of the istio-ingressgateway service in each cluster must be 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.
A Root CA. Cross cluster communication requires mutual TLS connection between services. To enable mutual TLS communication across clusters, each cluster’s Istio CA will be configured with intermediate CA credentials generated by a shared root CA. For illustration purposes, you use a sample root CA certificate available in the Istio installation under the samples/certs directory.

Deploy the Istio control plane in each cluster

Generate intermediate CA certificates for each cluster’s Istio 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. In real world deployments, you would likely use a different CA certificate for each cluster, all signed by a common root CA.
Run the following commands in every cluster to deploy an identical Istio control plane configuration in all of them.
- Create a Kubernetes secret for your generated CA certificates using a command similar to the following. See Certificate Authority (CA) certificates for more details.
  The root and intermediate certificate from the samples directory are widely distributed and known. Do not use these certificates in production as your clusters would then be open to security vulnerabilities and compromise.
  ZipZipZipZip
```
$ 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@
```
- Install Istio:
```
$ istioctl install \
    -f manifests/examples/multicluster/values-istio-multicluster-gateways.yaml
```
For further details and customization options, refer to the installation instructions.

Setup DNS

Providing DNS resolution for services in remote clusters will allow existing applications to function unmodified, as applications typically expect to resolve services by their DNS names and access the resulting IP. Istio itself does not use the DNS for routing requests between services. Services local to a cluster share a common DNS suffix (e.g., svc.cluster.local). Kubernetes DNS provides DNS resolution for these services.

To provide a similar setup for services from remote clusters, you name services from remote clusters in the format <name>.<namespace>.global. Istio also ships with a CoreDNS server that will provide DNS resolution for these services. In order to utilize this DNS, Kubernetes’ DNS must be configured to stub a domain for .global.

Create one of the following ConfigMaps, or update an existing one, in each cluster that will be calling services in remote clusters (every cluster in the general case):

$ kubectl apply -f - <<EOF
apiVersion: v1
kind: ConfigMap
metadata:
  name: kube-dns
  namespace: kube-system
data:
  stubDomains: |
    {"global": ["$(kubectl get svc -n istio-system istiocoredns -o jsonpath={.spec.clusterIP})"]}
EOF

$ kubectl apply -f - <<EOF
apiVersion: v1
kind: ConfigMap
metadata:
  name: coredns
  namespace: kube-system
data:
  Corefile: |
    .:53 {
        errors
        health
        kubernetes cluster.local in-addr.arpa ip6.arpa {
           pods insecure
           upstream
           fallthrough in-addr.arpa ip6.arpa
        }
        prometheus :9153
        proxy . /etc/resolv.conf
        cache 30
        loop
        reload
        loadbalance
    }
    global:53 {
        errors
        cache 30
        proxy . $(kubectl get svc -n istio-system istiocoredns -o jsonpath={.spec.clusterIP})
    }
EOF

$ kubectl apply -f - <<EOF
apiVersion: v1
kind: ConfigMap
metadata:
  name: coredns
  namespace: kube-system
data:
  Corefile: |
    .:53 {
        errors
        health
        kubernetes cluster.local in-addr.arpa ip6.arpa {
           pods insecure
           upstream
           fallthrough in-addr.arpa ip6.arpa
        }
        prometheus :9153
        forward . /etc/resolv.conf
        cache 30
        loop
        reload
        loadbalance
    }
    global:53 {
        errors
        cache 30
        forward . $(kubectl get svc -n istio-system istiocoredns -o jsonpath={.spec.clusterIP}):53
    }
EOF

$ kubectl apply -f - <<EOF
apiVersion: v1
kind: ConfigMap
metadata:
  name: coredns
  namespace: kube-system
data:
  Corefile: |
    .:53 {
        errors
        health
        ready
        kubernetes cluster.local in-addr.arpa ip6.arpa {
           pods insecure
           upstream
           fallthrough in-addr.arpa ip6.arpa
        }
        prometheus :9153
        forward . /etc/resolv.conf
        cache 30
        loop
        reload
        loadbalance
    }
    global:53 {
        errors
        cache 30
        forward . $(kubectl get svc -n istio-system istiocoredns -o jsonpath={.spec.clusterIP}):53
    }
EOF

Configure application services

Every service in a given cluster that needs to be accessed from a different remote cluster requires a ServiceEntry configuration in the remote cluster. The host used in the service entry should be of the form <name>.<namespace>.global where name and namespace correspond to the service’s name and namespace respectively.

To demonstrate cross cluster access, configure the sleep service running in one cluster to call the httpbin service running in a second cluster. Before you begin:

Choose two of your Istio clusters, to be referred to as cluster1 and cluster2.

You can use the kubectl command to access both the cluster1 and cluster2 clusters with the --context flag, for example kubectl get pods --context cluster1. Use the following command to list your contexts:

$ kubectl config get-contexts
CURRENT   NAME       CLUSTER    AUTHINFO       NAMESPACE
*         cluster1   cluster1   user@foo.com   default
          cluster2   cluster2   user@foo.com   default

Store the context names of your clusters in environment variables:
```
$ export CTX_CLUSTER1=$(kubectl config view -o jsonpath='{.contexts[0].name}')
$ export CTX_CLUSTER2=$(kubectl config view -o jsonpath='{.contexts[1].name}')
$ echo CTX_CLUSTER1 = ${CTX_CLUSTER1}, CTX_CLUSTER2 = ${CTX_CLUSTER2}
CTX_CLUSTER1 = cluster1, CTX_CLUSTER2 = cluster2
```
If you have more than two clusters in the context list and you want to configure your mesh using clusters other than the first two, you will need to manually set the environment variables to the appropriate context names.

Configure the example services

Deploy the sleep service in cluster1.

$ kubectl create --context=$CTX_CLUSTER1 namespace foo
$ kubectl label --context=$CTX_CLUSTER1 namespace foo istio-injection=enabled
$ kubectl apply --context=$CTX_CLUSTER1 -n foo -f @samples/sleep/sleep.yaml@
$ export SLEEP_POD=$(kubectl get --context=$CTX_CLUSTER1 -n foo pod -l app=sleep -o jsonpath={.items..metadata.name})

Deploy the httpbin service in cluster2.

$ kubectl create --context=$CTX_CLUSTER2 namespace bar
$ kubectl label --context=$CTX_CLUSTER2 namespace bar istio-injection=enabled
$ kubectl apply --context=$CTX_CLUSTER2 -n bar -f @samples/httpbin/httpbin.yaml@

Export the cluster2 gateway address:
```
$ export CLUSTER2_GW_ADDR=$(kubectl get --context=$CTX_CLUSTER2 svc --selector=app=istio-ingressgateway \
    -n istio-system -o jsonpath='{.items[0].status.loadBalancer.ingress[0].ip}')
```
This command sets the value to the gateway’s public IP, but note that you can set it to a DNS name instead, if you have one.
If cluster2 is running in an environment that does not support external load balancers, you will need to use a nodePort to access the gateway. Instructions for obtaining the IP to use can be found in the Control Ingress Traffic guide. You will also need to change the service entry endpoint port in the following step from 15443 to its corresponding nodePort (i.e., kubectl --context=$CTX_CLUSTER2 get svc -n istio-system istio-ingressgateway -o=jsonpath='{.spec.ports[?(@.port==15443)].nodePort}').
Create a service entry for the httpbin service in cluster1.
To allow sleep in cluster1 to access httpbin in cluster2, we need to create a service entry for it. The host name of the service entry should be of the form <name>.<namespace>.global where name and namespace correspond to the remote service’s name and namespace respectively.
For DNS resolution for services under the *.global domain, you need to assign these services an IP address.
Each service (in the .global DNS domain) must have a unique IP within the cluster.
If the global services have actual VIPs, you can use those, but otherwise we suggest using IPs from the class E addresses range 240.0.0.0/4. Application traffic for these IPs will be captured by the sidecar and routed to the appropriate remote service.
Multicast addresses (224.0.0.0 ~ 239.255.255.255) should not be used because there is no route to them by default. Loopback addresses (127.0.0.0/8) should also not be used because traffic sent to them may be redirected to the sidecar inbound listener.
```
$ kubectl apply --context=$CTX_CLUSTER1 -n foo -f - <<EOF
apiVersion: networking.istio.io/v1alpha3
kind: ServiceEntry
metadata:
  name: httpbin-bar
spec:
  hosts:
  # must be of form name.namespace.global
  - httpbin.bar.global
  # Treat remote cluster services as part of the service mesh
  # as all clusters in the service mesh share the same root of trust.
  location: MESH_INTERNAL
  ports:
  - name: http1
    number: 8000
    protocol: http
  resolution: DNS
  addresses:
  # the IP address to which httpbin.bar.global will resolve to
  # must be unique for each remote service, within a given cluster.
  # This address need not be routable. Traffic for this IP will be captured
  # by the sidecar and routed appropriately.
  - 240.0.0.2
  endpoints:
  # This is the routable address of the ingress gateway in cluster2 that
  # sits in front of sleep.foo service. Traffic from the sidecar will be
  # routed to this address.
  - address: ${CLUSTER2_GW_ADDR}
    ports:
      http1: 15443 # Do not change this port value
EOF
```
The configurations above will result in all traffic in cluster1 for httpbin.bar.global on any port to be routed to the endpoint $CLUSTER2_GW_ADDR:15443 over a mutual TLS connection.
The gateway for port 15443 is a special SNI-aware Envoy preconfigured and installed when you deployed the Istio control plane in the cluster. Traffic entering port 15443 will be load balanced among pods of the appropriate internal service of the target cluster (in this case, httpbin.bar in cluster2).
Do not create a Gateway configuration for port 15443.

Verify that httpbin is accessible from the sleep service.

$ kubectl exec --context=$CTX_CLUSTER1 $SLEEP_POD -n foo -c sleep -- curl -I httpbin.bar.global:8000/headers

Send remote traffic via an egress gateway

If you want to route traffic from cluster1 via a dedicated egress gateway, instead of directly from the sidecars, use the following service entry for httpbin.bar instead of the one in the previous section.

If $CLUSTER2_GW_ADDR is an IP address, use the $CLUSTER2_GW_ADDR - IP address option. If $CLUSTER2_GW_ADDR is a hostname, use the $CLUSTER2_GW_ADDR - hostname option.

Export the cluster1 egress gateway address:

$ export CLUSTER1_EGW_ADDR=$(kubectl get --context=$CTX_CLUSTER1 svc --selector=app=istio-egressgateway \
    -n istio-system -o yaml -o jsonpath='{.items[0].spec.clusterIP}')

Apply the httpbin-bar service entry:

$ kubectl apply --context=$CTX_CLUSTER1 -n foo -f - <<EOF
apiVersion: networking.istio.io/v1alpha3
kind: ServiceEntry
metadata:
  name: httpbin-bar
spec:
  hosts:
  # must be of form name.namespace.global
  - httpbin.bar.global
  location: MESH_INTERNAL
  ports:
  - name: http1
    number: 8000
    protocol: http
  resolution: STATIC
  addresses:
  - 240.0.0.2
  endpoints:
  - address: ${CLUSTER2_GW_ADDR}
    network: external
    ports:
      http1: 15443 # Do not change this port value
  - address: ${CLUSTER1_EGW_ADDR}
    ports:
      http1: 15443
EOF

If the ${CLUSTER2_GW_ADDR} is a hostname, you can use resolution: DNS for the endpoint resolution:

$ kubectl apply --context=$CTX_CLUSTER1 -n foo -f - <<EOF
apiVersion: networking.istio.io/v1alpha3
kind: ServiceEntry
metadata:
  name: httpbin-bar
spec:
  hosts:
  # must be of form name.namespace.global
  - httpbin.bar.global
  location: MESH_INTERNAL
  ports:
  - name: http1
    number: 8000
    protocol: http
  resolution: DNS
  addresses:
  - 240.0.0.2
  endpoints:
  - address: ${CLUSTER2_GW_ADDR}
    network: external
    ports:
      http1: 15443 # Do not change this port value
  - address: istio-egressgateway.istio-system.svc.cluster.local
    ports:
      http1: 15443
EOF

Cleanup the example

Execute the following commands to clean up the example services.

Cleanup cluster1:

$ kubectl delete --context=$CTX_CLUSTER1 -n foo -f @samples/sleep/sleep.yaml@
$ kubectl delete --context=$CTX_CLUSTER1 -n foo serviceentry httpbin-bar
$ kubectl delete --context=$CTX_CLUSTER1 ns foo

Cleanup cluster2:

$ kubectl delete --context=$CTX_CLUSTER2 -n bar -f @samples/httpbin/httpbin.yaml@
$ kubectl delete --context=$CTX_CLUSTER2 ns bar

Cleanup environment variables:

$ unset SLEEP_POD CLUSTER2_GW_ADDR CLUSTER1_EGW_ADDR CTX_CLUSTER1 CTX_CLUSTER2

Version-aware routing to remote services

If the remote service has multiple versions, you can add labels to the service entry endpoints. For example:

$ kubectl apply --context=$CTX_CLUSTER1 -n foo -f - <<EOF
apiVersion: networking.istio.io/v1alpha3
kind: ServiceEntry
metadata:
  name: httpbin-bar
spec:
  hosts:
  # must be of form name.namespace.global
  - httpbin.bar.global
  location: MESH_INTERNAL
  ports:
  - name: http1
    number: 8000
    protocol: http
  resolution: DNS
  addresses:
  # the IP address to which httpbin.bar.global will resolve to
  # must be unique for each service.
  - 240.0.0.2
  endpoints:
  - address: ${CLUSTER2_GW_ADDR}
    labels:
      cluster: cluster2
    ports:
      http1: 15443 # Do not change this port value
EOF

You can then create virtual services and destination rules to define subsets of the httpbin.bar.global service using the appropriate gateway label selectors. The instructions are the same as those used for routing to a local service. See multicluster version routing for a complete example.

Uninstalling

Uninstall Istio by running the following commands on every cluster:

$ istioctl manifest generate \
    -f manifests/examples/multicluster/values-istio-multicluster-gateways.yaml \
    | kubectl delete -f -

Summary

Using Istio gateways, a common root CA, and service entries, you can configure a single Istio service mesh across multiple Kubernetes clusters. Once configured this way, traffic can be transparently routed to remote clusters without any application involvement. Although this approach requires a certain amount of manual configuration for remote service access, the service entry creation process could be automated.