Product Documentation
NetScaler ingress controller
View PDF

Multi-cluster ingress

August 21, 2024
Contributed by: 
C

Introduction

Multi-cluster Kubernetes solutions are ideal for distributing the workload across multiple clusters. NetScaler multi-cluster ingress solution enables NetScaler to load balance applications distributed across clusters using a single front-end IP address. The load-balanced applications can be either the same application, different applications of the same domain, or entirely different applications.

Earlier, to load balance applications in multiple clusters, a dedicated content switching virtual server was required on NetScaler for each instance of NetScaler Ingress Controller (NSIC) running in the clusters. With NetScaler multi-cluster ingress solution, multiple ingress controllers can share a content switching virtual server. Therefore, applications deployed across clusters can be load balanced using the same content switching virtual server IP (VIP) address.

To summarize, the multi-cluster ingress solution optimizes the use of load-balancing resources, thereby reducing operational costs.

Note:

The multi-cluster ingress solution is supported starting from NSIC version 2.0.x.

Deployment topology

The following diagram describes the multi-cluster ingress deployment topology for two Kubernetes clusters in a data center/site. Here, NetScaler load balances applications distributed across clusters using a single front-end IP address.

multi-cluster ingress

  • An NSIC is deployed in both the clusters. Both NSIC instances configure the same NetScaler.

  • Depending on the nature of applications deployed in the clusters, we typically have the following use cases:

    • A different application of the same domain (company.website.com) deployed in each Kubernetes cluster: App-B in cluster1 and App-C in cluster2.

      Here, a content switching virtual server on NetScaler load balances traffic across App-B and App-C. Separate content switching policies are configured for each application.

    • The same application deployed across both Kubernetes clusters.

      Here, a content switching virtual server on NetScaler load balances traffic for the same application App-A deployed in both the clusters. Only one content switching policy is created for App-A. which is used by both running instances (endpoints) of the application.

Let’s understand how the same frontend IP address is used to load balance traffic to applications App-B and App-c deployed across cluster1 and cluster2 with the following diagram.

load balancing in multi-cluster setup

Configuring the following resources enable the multi-cluster ingress solution: Listener and Ingress.

Listener

  • The listener resource deployed in a cluster creates a content switching virtual server on NetScaler using the VIP specified in the listener YAML. In the multi-cluster ingress setup, the same listener resource is deployed in all the clusters.

    Listener resource in the multi-cluster ingress solution handles the front-end traffic management. You must provide all secrets, ciphers, and front-end profile configuration required for the content switching virtual server. For more information on listener CRD, see Listener. Details on deploying listener resources for the multi-cluster ingress setup are available in the Steps to deploy a multi-cluster ingress solution section.

Ingress

  • The ingress resources deployed in each cluster to route traffic to an application in the multi-cluster setup refer to the same listener resource with the shared VIP address. Therefore, NSIC refers to the same content virtual server that exists and creates only the content switching policies on NetScaler that are later bound to the content switching virtual server.

In this case, the content switching policy for each application, that is, CSPOL-App-B and CSPOL-App-C, are bound to the same content switching virtual server. When a request for example, company.website/App-B is sent to the content switching virtual server IP (VIP) address, the content switching virtual server applies the CSPOL-App-B policy and routes the request to the LBvserver-App-B, which sends it to the App-B service.

Multi-cluster ingress deployment

In this procedure, we will deploy the same HTTPS application in two clusters, deploy the required listener and ingress resources in both the clusters, such that NetScaler VPX or NetScaler MPX load balances these applications using a single front-end IP address.

Prerequisites

  • Kubernetes cluster in cloud or on-premises. The Kubernetes cluster in the cloud can be a managed Kubernetes (for example: GKE, EKS, or AKS) or a custom created Kubernetes deployment. Kubernetes clusters must be properly configured and operational. Ensure the network connectivity between NetScaler and the cluster’s pod network.
  • Version 1.21 and later in the Kubernetes environment.
  • Version 4.11 and later in the OpenShift platform.
  • You have installed Helm version 3.x or later. To install Helm, see here.
  • NetScaler MPX/VPX is deployed in a standalone, HA, or cluster setup depending on your specific needs.
  • Determine the NSIP (NetScaler IP) address using which NetScaler Ingress Controller communicates with NetScaler. The IP address might be any one of the following IP addresses depending on the type of NetScaler deployment:

    • NSIP (for standalone appliances) - The management IP address of a standalone NetScaler appliance. For more information, see IP Addressing in NetScaler.

    • SNIP (for appliances in High Availability mode) - The subnet IP address. For more information, see IP Addressing in NetScaler.

    • CLIP (for appliances in Cluster mode) - The cluster management IP (CLIP) address for a cluster NetScaler deployment. For more information, see IP addressing for a cluster.

  • A user account in NetScaler VPX or NetScaler MPX. NetScaler Ingress Controller uses a system user account in NetScaler to configure NetScaler MPX or NetScaler VPX. For instructions to create the system user account on NetScaler, see Create System User Account for NetScaler Ingress Controller in NetScaler.

    You can directly pass the user name and password or use Kubernetes secrets. If you want to use Kubernetes secrets, create a secret for the user name and password using the following command:

     kubectl create secret generic nslogin --from-literal=username=<username> --from-literal=password=<password>

Steps to deploy a multi-cluster ingress solution

Follow these steps to deploy a multi-cluster ingress solution to expose the Cloud Native Networking (CNN) application deployed across two Kubernetes clusters using NetScaler and a shared front-end IP address.

Note:

Repeat the following steps in both the clusters to set up a multi-cluster ingress solution. There are exceptions added in some steps highlighting what needs to be done in a particular cluster; perform the steps accordingly.

  1. Deploy the NetScaler CNN application by using the following YAML.

    The CNN application is an HTTP-based application that lists the solutions offered under the NetScaler cloud native portfolio.

    kubectl apply -f - <<EOF 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: cnn-website
  labels:
    name: cnn-website     
    app: cnn-website    
spec:
  selector:
    matchLabels:
      app: cnn-website    
  replicas: 2
  template:
    metadata:
      labels:
        name: cnn-website    
        app: cnn-website    
    spec:
      containers:
      - name: cnn-website    
        image: quay.io/sample-apps/cnn-website:v1.0.0    
        ports:
        - name: http-80
          containerPort: 80
        - name: https-443
          containerPort: 443
---
apiVersion: v1
kind: Service
metadata:
  name: cnn-website    
  labels:
    app: cnn-website    
spec:
  type: NodePort
  ports:
  - name: http-80
    port: 80
    targetPort: 80
  - name: https-443
    port: 443
    targetPort: 443
  selector:
    name: cnn-website
EOF
<!--NeedCopy-->

  2. Add the NetScaler Helm chart repository to your local registry using the following command.

    helm repo add netscaler https://netscaler.github.io/netscaler-helm-charts/   

<!--NeedCopy-->

  3. Update values.yaml to configure NetScaler Ingress Controller as described as following.

    Example values.yaml for cluster1

      adcCredentialSecret: nslogin # K8s Secret Name
  nsIP: <x.x.x> # CLIP (for appliances in Cluster mode), SNIP (for appliances in High Availability mode) , NSIP (for standalone appliances)
  entityPrefix: cluster1 # unique for each NSIC instance.
  clusterName: cluster1
  ingressClass: ['nsic-vpx'] # ingress class used in the ingress resources
  multiClusterPrefix: mc # Multi-cluster prefix for the NSIC instance. Same value must be specified for a set of NSIC instances configuring NetScaler in multi-cluster setup.
  <!--NeedCopy-->

    Example values.yaml for cluster2

    
  adcCredentialSecret: nslogin # K8s Secret Name
  nsIP: <x.x.x> # CLIP (for appliances in Cluster mode), SNIP (for appliances in High Availability mode) , NSIP (for standalone appliances)
  entityPrefix: cluster2 # unique for each NSIC instance.
  clusterName: cluster2
  ingressClass: ['nsic-vpx'] # ingress class used in the ingress resources
  multiClusterPrefix: mc # Multi-cluster prefix for the NSIC instance. Same value must be specified for a set of NSIC instances configuring NetScaler in multi-cluster setup.
  <!--NeedCopy-->

    For information about the mandatory and optional parameters that you can configure during NSIC installation, see Configuration.

  4. Deploy NetScaler Ingress Controller with the modified values.yaml.

    helm install nsic netscaler-ingress-controller -f values.yaml
<!--NeedCopy-->

  5. Deploy the following listener CRD resource.

    kubectl apply -f - <<EOF 
apiVersion: citrix.com/v1
kind: Listener
metadata:
  name: mc-listener
spec:
  multicluster: True
  vip: <Provide the shared front-end IP address>
  certificates:  
  #  you need to specify either secret name or pre-configured cert-keyname
    - secret:
        name: <Provide k8s secret name>  # provide k8s secret containing cert-key of the application 
      default: true
    - preconfigured: <ADC-certkeyname> # provide pre-configured cert-key from ADC. 
      protocol: 'https'
EOF
<!--NeedCopy-->

    • Update the vip with the virtual IP address that is used to expose the application deployed in both the clusters.

    • multicluster must be set as True.

    • Either update a preconfigured secret with the ssl certificate-key name created on NetScaler or update the name of Kubernetes TLS Secret in the secret.name section. Refer Listener.certificates.

    In this example, the listener resource mc-listener specifies the front-end configuration such as VIP and secrets. It creates a content switching virtual server in NetScaler on port 443 for HTTPS traffic.

  6. Deploy the following ingress resource to expose the CNN application.

    kubectl apply -f - <<EOF 
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: frontend-ingress
  annotations:
    ingress.citrix.com/listener: mc-listener
spec:
  ingressClassName: nsic-vpx
  rules:
    - http:
        paths:
          - path: /
            pathType: Prefix
            backend:
              service:
                name: cnn-website
                port:
                  number: 80
---
apiVersion: networking.k8s.io/v1
kind: IngressClass
metadata:
  name: nsic-vpx
spec:
  controller: citrix.com/ingress-controller
EOF
<!--NeedCopy-->
    • The listener resource created in the previous step must be passed using ingress.citrix.com/listener annotation. In this example, the ingress resource frontend-ingress refers to the Listener resource mc-listener (created in step 6).
    • The ingress class is mentioned as nsic-vpx.

    Ingress creates a content switching policy, a load balancing virtual server, a service group and binds the application pod IP addresses as service group members.

    Notes:

    • When load balancing different applications in the multi-cluster ingress setup, separate content switching policies are created for each application. In such cases, if you require a particular sequence for policy binding, you must assign a priority number to the content switching policies by using the ingress.citrix.com/multicluster-policy-priority-order annotation. For more information, see Policy bindings.
    • For information about listener and ingress resources when load balancing different applications in the multi-cluster ingress setup, see Multi-cluster ingress setup with different applications.

  7. To validate the configuration, replace <VIP> with the virtual IP address provided in the listener resource in step 6.

    curl -kv https://<VIP>
<!--NeedCopy-->

Advanced use cases

When the same application is deployed across multiple clusters, diverse deployment scenarios with varying traffic distribution requirements arise. The default behavior, known as Active-Active, involves all instances of the application receiving traffic based on the load balancing method. Let’s see the advanced use cases considering the same application deployed in two clusters as described here.

Active-backup mode

In this mode, one application is always active, which receives traffic all the time. The other application receives traffic only when this cluster or application is down. You can define which application acts as a backup using ingress.citrix.com/multicluster-backup-order ingress annotation.

Note:

The same listener resource must be deployed in both the clusters. For an example listener resource, see step 6 in the Steps to deploy a multi-cluster ingress solution section.

  Cluster 1 (Active) Cluster2 (Backup)
Ingress annotation ingress.citrix.com/multicluster-backup-order: “1” ingress.citrix.com/multicluster-backup-order: “2”
Information The default is active (1). Annotation can be skipped Annotation is mandatory for the application to be considered as a backup.
  • Application deployed in cluster1 actively receives traffic, whereas cluster2 acts as a backup. The application in cluster2 receives traffic only when cluster1 or application in cluster1 is down.

Canary mode

Note:

For the following deployments, the same listener resource must be deployed in each cluster. For an example listener resource, see step 6 in the Steps to deploy a multi-cluster ingress solution section.

We support canary deployments with two distinct strategies: “Canary by weight” and “Canary by header”.

Canary deployment by weight

You can implement canary deployment by assigning different weights to instances of an application. For instance, if there are two clusters and a new version of the application is being rolled out, you can allocate a certain percentage of traffic to the new version (canary) and the remaining traffic to the existing version. This canary mode allows gradual rollout and monitoring of the new version’s performance in a controlled manner.

  Cluster1 (Version1) Cluster2 (Version2)
Ingress annotation NA ingress.citrix.com/multicluster-canary-weight: “10”
Information Annotation is not required Annotation is mandatory. This application receives 10% of client traffic.
Canary deployment by header

You can implement a canary deployment based on the specific HTTP headers. For example, a specific header in the HTTP request such as “X-Canary-Version” can be used to control the routing of traffic. In this case, requests with the “X-Canary-Version” header are directed to the canary version of the app, while others are routed to the stable version. This canary mode provides a more granular control over canary deployments, allowing you to target specific subsets of users or requests for testing purposes.

  Cluster1 (Version1) Cluster2 (Version2)
Annotation NA ingress.citrix.com/multicluster-canary-by-header: “version2”
Information Annotation is not required Annotation is mandatory for the canary application. All client traffic with the version2 header is sent to cluster2.

Policy bindings

If you require a particular sequence for policy binding, you must assign a priority number to the content switching policies by using the ingress.citrix.com/multicluster-policy-priority-order annotation. Lower the number, the higher the priority.

Let’s consider the following ingress resource example to understand policy bindings.

kubectl apply -f - <<EOF 
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: mc-ing
  annotations:
    ingress.citrix.com/listener: mc-listener
    ingress.citrix.com/multicluster-policy-priority-order: '{"frontend": {"80": "3", "9443": "1"}, "backend": "2"}'
spec:
  ingressClassName: nsic-vpx
  rules:
    - http:
        paths:
          - path: /
            pathType: Prefix
            backend:
              service:
                name: frontend
                port:
                  number: 80
          - path: /abc
            pathType: Prefix
            backend:
              service:
                name: frontend
                port:
                  number: 9443
          - path: /xyz
            pathType: Prefix
            backend:
              service:
                name: backend
                port:
                  number: 80
EOF
<!--NeedCopy-->

This example resource includes ingress.citrix.com/multicluster-policy-priority-order annotation, which defines the priority order for content switching policies. Rather than assigning priorities randomly, NetScaler Ingress Controller uses the priority values you have provided to bind content switching policies to the content switching virtual server.

For the ingress.citrix.com/multicluster-policy-priority-order: ‘{“frontend”: {“80”: “3”, “9443”: “1”}, “backend”: “2”}’ annotation, the priority is assigned as following:

  • The content switching policy associated with the frontend:80 service is bound to the content switching virtual server with a priority of 3.
  • The content switching policy associated with the frontend:9443 service is bound to the content switching virtual server with a priority of 1.
  • The content switching policy associated with the backend:80 service is bound to the content switching virtual server with a priority of 2.

Warning:

When a single service is mentioned in the ingress resource with different ports, you must explicitly specify a priority number for each port. Otherwise, random priorities are assigned.

Multi-cluster ingress setup with different applications

In the multi-cluster ingress setup for load balancing different applications, the listener resource deployed in each cluster must be the same. The ingress resource in each cluster must refer to the same listener. For the example described in the deployment topology section, mc-listener is the listener resource deployed in both cluster1 and cluster2. The following sample ingress resources deployed in cluster1 and cluster2 refers to mc-listener.

Ingress resource to expose App-B in cluster1:

  apiVersion: networking.k8s.io/v1
  kind: Ingress
  metadata:
    name: App-B
    annotations:
      ingress.citrix.com/listener: mc-listener 
  spec:
    ingressClassName: cluster1
    rules:
    - http:
        paths:
        - path: /App-B
          pathType: Prefix
          backend:
            service:
              name: App-B-svc
              port:
                number: 80
  <!--NeedCopy-->

Ingress resource to expose App-C in cluster2:

  apiVersion: networking.k8s.io/v1
  kind: Ingress
  metadata:
    name: App-C
    annotations:
      ingress.citrix.com/listener: mc-listener
  spec:
    ingressClassName: cluster2
    rules:
    - http:
        paths:
        - path: /App-C
          pathType: Prefix
          backend:
            service:
              name: App-C-svc
              port:
                number: 80
  <!--NeedCopy-->
Multi-cluster ingress
August 21, 2024
Contributed by: 
C
August 21, 2024
Contributed by: 
C

In this article

Site feedback | Your privacy choices footer linkYour Privacy Choices | Privacy and legal terms | Cookie preferences | Consent settings | docs.cloud.com
© 1999- Cloud Software Group, Inc. All rights reserved.