Single Consul Datacenter in Multiple Kubernetes Clusters

This page describes deploying a single Consul datacenter in multiple Kubernetes clusters, with servers and clients running in one cluster and only clients in the rest of the clusters. This example uses two Kubernetes clusters, but this approach could be extended to using more than two.

Requirements

• Consul-Helm version v0.32.1 or higher
• This deployment topology requires that the Kubernetes clusters have a flat network for both pods and nodes so that pods or nodes from one cluster can connect to pods or nodes in another. In many hosted Kubernetes environments, this may have to be explicitly configured based on the hosting provider's network. Refer to the following documentation for instructions:
  • Azure AKS CNI
  • AWS EKS CNI
  • GKE VPC-native clusters.
• Either the Helm release name for each Kubernetes cluster must be unique, or global.name for each Kubernetes cluster must be unique to prevent collisions of ACL resources with the same prefix.

Prepare Helm release name ahead of installs

The Helm release name must be unique for each Kubernetes cluster. The Helm chart uses the Helm release name as a prefix for the ACL resources that it creates, such as tokens and auth methods. If the names of the Helm releases are identical, or if global.name for each cluster is identical, subsequent Consul on Kubernetes clusters will overwrite existing ACL resources and cause the clusters to fail.

Before proceeding with installation, prepare the Helm release names as environment variables for both the server and client install.

  $ export HELM_RELEASE_SERVER=server
  $ export HELM_RELEASE_CLIENT=client
  ...
  $ export HELM_RELEASE_CLIENT2=client2

Deploying Consul servers and clients in the first cluster

First, deploy the first cluster with Consul Servers and Clients with the example Helm configuration below.

global:
  datacenter: dc1
  tls:
    enabled: true
    enableAutoEncrypt: true
  acls:
    manageSystemACLs: true
  gossipEncryption:
    secretName: consul-gossip-encryption-key
    secretKey: key
connectInject:
  enabled: true
controller:
  enabled: true
ui:
  service:
    type: NodePort

Note that this will deploy a secure configuration with gossip encryption, TLS for all components and ACLs. In addition, this will enable the Consul Service Mesh and the controller for CRDs that can be used later to verify the connectivity of services across clusters.

The UI's service type is set to be NodePort. This is needed to connect to servers from another cluster without using the pod IPs of the servers, which are likely going to change.

To deploy, first generate the Gossip encryption key and save it as a Kubernetes secret.

$ kubectl create secret generic consul-gossip-encryption-key --from-literal=key=$(consul keygen)

Now install Consul cluster with Helm:

$ helm install ${HELM_RELEASE_SERVER} --values cluster1-config.yaml hashicorp/consul

Once the installation finishes and all components are running and ready, the following information needs to be extracted (using the below command) and applied to the second Kubernetes cluster.

• The Gossip encryption key created
• The CA certificate generated during installation
• The ACL bootstrap token generated during installation

$ kubectl get secret consul-gossip-encryption-key ${HELM_RELEASE_SERVER}-consul-ca-cert ${HELM_RELEASE_SERVER}-consul-bootstrap-acl-token --output yaml > cluster1-credentials.yaml

Deploying Consul clients in the second cluster

Switch to the second Kubernetes cluster where Consul clients will be deployed that will join the first Consul cluster.

$ kubectl config use-context <K8S_CONTEXT_NAME>

First, apply the credentials extracted from the first cluster to the second cluster:

$ kubectl apply --filename cluster1-credentials.yaml

To deploy in the second cluster, the following example Helm configuration will be used:

global:
  enabled: false
  datacenter: dc1
  acls:
    manageSystemACLs: true
    bootstrapToken:
      secretName: cluster1-consul-bootstrap-acl-token
      secretKey: token
  gossipEncryption:
    secretName: consul-gossip-encryption-key
    secretKey: key
  tls:
    enabled: true
    enableAutoEncrypt: true
    caCert:
      secretName: cluster1-consul-ca-cert
      secretKey: tls.crt
externalServers:
  enabled: true
  # This should be any node IP of the first k8s cluster
  hosts: ["10.0.0.4"]
  # The node port of the UI's NodePort service
  httpsPort: 31557
  tlsServerName: server.dc1.consul
  # The address of the kube API server of this Kubernetes cluster
  k8sAuthMethodHost: https://kubernetes.example.com:443
client:
  enabled: true
  join: ["provider=k8s kubeconfig=/consul/userconfig/cluster1-kubeconfig/kubeconfig label_selector=\"app=consul,component=server\""]
  extraVolumes:
    - type: secret
      name: cluster1-kubeconfig
      load: false
connectInject:
  enabled: true

Note the references to the secrets extracted and applied from the first cluster in ACL, gossip, and TLS configuration.

The externalServers.hosts and externalServers.httpsPort refer to the IP and port of the UI's NodePort service deployed in the first cluster. Set the externalServers.hosts to any Node IP of the first cluster, which can be seen by running kubectl get nodes --output wide. Set externalServers.httpsPort to the nodePort of the cluster1-consul-ui service. In our example, the port is 31557.

$ kubectl get service cluster1-consul-ui --context cluster1
NAME                 TYPE       CLUSTER-IP    EXTERNAL-IP   PORT(S)         AGE
cluster1-consul-ui   NodePort   10.0.240.80   <none>        443:31557/TCP   40h

Set the externalServer.tlsServerName to server.dc1.consul. This the DNS SAN (Subject Alternative Name) that is present in the Consul server's certificate. This is required because the connection to the Consul servers uses the node IP, but that IP isn't present in the server's certificate. To make sure that the hostname verification succeeds during the TLS handshake, set the TLS server name to a DNS name that is present in the certificate.

Next, set externalServers.k8sAuthMethodHost to the address of the second Kubernetes API server. This should be the address that is reachable from the first cluster, so it cannot be the internal DNS available in each Kubernetes cluster. Consul needs it so that consul login with the Kubernetes auth method will work from the second cluster. More specifically, the Consul server will need to perform the verification of the Kubernetes service account whenever consul login is called, and to verify service accounts from the second cluster, it needs to reach the Kubernetes API in that cluster. The easiest way to get it is from the kubeconfig by running kubectl config view and grabbing the value of cluster.server for the second cluster.

Lastly, set up the clients so that they can discover the servers in the first cluster. For this, Consul's cloud auto-join feature for the Kubernetes provider can be used.

This can be configured by saving the kubeconfig for the first cluster as a Kubernetes secret in the second cluster and referencing it in the clients.join value. Note that the secret is made available to the client pods by setting it in client.extraVolumes.

Now, proceed with the installation of the second cluster.

$ helm install ${HELM_RELEASE_CLIENT} --values cluster2-config.yaml hashicorp/consul

Verifying the Consul Service Mesh works

Now that the Consul cluster spanning across multiple k8s clusters is up and running, deploy two services in separate k8s clusters and verify that they can connect to each other.

First, deploy static-server service in the first cluster:

---
apiVersion: consul.hashicorp.com/v1alpha1
kind: ServiceIntentions
metadata:
  name: static-server
spec:
  destination:
    name: static-server
  sources:
    - name: static-client
      action: allow
---
apiVersion: v1
kind: Service
metadata:
  name: static-server
spec:
  type: ClusterIP
  selector:
    app: static-server
  ports:
    - protocol: TCP
      port: 80
      targetPort: 8080
---
apiVersion: v1
kind: ServiceAccount
metadata:
  name: static-server
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: static-server
spec:
  replicas: 1
  selector:
    matchLabels:
      app: static-server
  template:
    metadata:
      name: static-server
      labels:
        app: static-server
      annotations:
        "consul.hashicorp.com/connect-inject": "true"
    spec:
      containers:
        - name: static-server
          image: hashicorp/http-echo:latest
          args:
            - -text="hello world"
            - -listen=:8080
          ports:
            - containerPort: 8080
              name: http
      serviceAccountName: static-server

Note that defining a Service intention is required so that our services are allowed to talk to each other.

Next, deploy static-client in the second cluster with the following configuration:

apiVersion: v1
kind: Service
metadata:
  name: static-client
spec:
  selector:
    app: static-client
  ports:
    - port: 80
---
apiVersion: v1
kind: ServiceAccount
metadata:
  name: static-client
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: static-client
spec:
  replicas: 1
  selector:
    matchLabels:
      app: static-client
  template:
    metadata:
      name: static-client
      labels:
        app: static-client
      annotations:
        "consul.hashicorp.com/connect-inject": "true"
        "consul.hashicorp.com/connect-service-upstreams": "static-server:1234"
    spec:
      containers:
        - name: static-client
          image: curlimages/curl:latest
          command: [ "/bin/sh", "-c", "--" ]
          args: [ "while true; do sleep 30; done;" ]
      serviceAccountName: static-client

Once both services are up and running, try connecting to the static-server from static-client:

$ kubectl exec deploy/static-client -- curl --silent localhost:1234
"hello world"

A successful installation would return hello world for the above curl command output.