Policy based routing support for multiple Kubernetes clusters
When you are using a single NetScaler to load balance multiple Kubernetes clusters, the NetScaler Ingress Controller adds pod CIDR networks through static routes. These routes establish networking connectivity between Kubernetes pods and NetScaler. However, when the pod CIDRs overlap there may be route conflicts. NetScaler supports policy based routing (PBR) to address the networking conflicts in such scenarios. In PBR, decisions are taken based on the criteria that you specify. Typically, a next hop is specified where you send the selected packets. In a multi-cluster Kubernetes environment, PBR is implemented by reserving a subnet IP address (SNIP) for each Kubernetes cluster or the NetScaler Ingress Controller. Using net profile, the SNIP is bound to all service groups created by the same NetScaler Ingress Controller. For all the traffic generated from service groups belonging to the same cluster, the source IP address is the same SNIP.
Following is a sample topology where PBR is configured for two Kubernetes clusters which are load balanced using a NetScaler VPX or MPX.
Configure PBR using the NetScaler Ingress Controller
To configure PBR, you need one SNIP or more per Kubernetes cluster. You can provide SNIP values either using the environment variable in the NetScaler Ingress Controller deployment YAML file during bootup or using ConfigMap.
Perform the following steps to deploy the NetScaler Ingress Controller and configure PBR using ConfigMap.
-
Download the
citrix-k8s-ingress-controller.yaml
using the following command:wget https://raw.githubusercontent.com/citrix/citrix-k8s-ingress-controller/master/deployment/baremetal/citrix-k8s-ingress-controller.yaml
-
Edit the NetScaler Ingress Controller YAML file:
- Specify the values of the environment variables as per your requirements. For more information on specifying the environment variables, see the [Deploy NetScaler Ingress Controller](/en-us/netscaler-k8s-ingress-controller/cic-yaml.html) documentation.
-
Deploy the NetScaler Ingress Controller using the edited YAML file with the following command on each cluster.
kubectl create -f citrix-k8s-ingress-controller.yaml
-
Create a YAML file
cic-configmap.yaml
with the required SNIP values in the ConfigMap.Following is an example for a ConfigMap with the SNIP values:
apiVersion: v1 kind: ConfigMap metadata: name: pbr-test namespace: default data: NS_SNIPS: '["192.0.2.2", "192.0.2.1"]'
-
Apply the ConfigMap.
kubectl create -f cic-configmap.yaml
You can also specify the SNIPs using the NS_SNIPS
environment variable in the NetScaler Ingress Controller deployment YAML file.
- name: "NS_SNIPS"
value: '["192.0.2.2", "192.0.2.1"]'
The following are the usage guidelines while using ConfigMap for configuring SNIP:
-
Only SNIPs can be added or removed via ConfigMap. The
feature-node-watch
argument can only be enabled during bootup. -
When you add a ConfigMap:
- If SNIPs are already provided using the environment variable during bootup and you want to retain them, those SNIPs should be specified in the ConfigMap along with the new SNIPs.
-
When you delete ConfigMap:
-
All PBRs generated by ConfigMap SNIPs are deleted. If SNIPs are provided via the environment variable, PBR for those IP addresses is added.
-
If SNIPs are not provided using the
NS_SNIPS
environment variable, static routes are added sincefeature-node-watch
is enabled.
-
Validate PBR configuration on a NetScaler after deploying the NetScaler Ingress Controller
This validation example uses a two node Kubernetes cluster with the NetScaler Ingress Controller deployed along with the following ConfigMap with two SNIPs.
You can verify that the NetScaler Ingress Controller adds the following configurations to the ADC:
-
An
IPset
of allNS_SNIPs
provided by the ConfigMap is added. -
A net profile is added with the
SrcIP
set to theIPset
. -
The service group added by the NetScaler Ingress Controller contains the net profile set.
-
Finally, the NetScaler Ingress Controller adds PBRs.
Here:
- The number of PBRs is equivalent to (number of SNIPs)
*
(number of Kubernetes nodes). In this case, it adds four(2*2) PBRs. - The
srcIP
of the PBR is theNS_SNIPS
provided to the NetScaler Ingress Controller by ConfigMap. ThedestIP
is the CNI overlay subnet range of the Kubernetes node. -
NextHop
is the IP address of the Kubernetes node.
- The number of PBRs is equivalent to (number of SNIPs)
-
You can also use the logs of the NetScaler Ingress Controller to validate the configuration.
Configure PBR using the node controller
You can configure PBR using the node controller for multiple Kubernetes clusters. When you are using a single NetScaler to load balance multiple Kubernetes clusters with node controller for networking, the static routes added by it to forward packets to the IP address of the VXLAN tunnel interface may cause route conflicts. To support PBR, node controller needs to works in conjunction with the NetScaler Ingress Controller to bind the net profile to the service group.
Perform the following steps to configure PBR using the node controller:
-
While starting the node controller, provide the
CLUSTER_NAME
as an environment variable. Specifying this variable indicates that it is a multi-cluster deployment and the node controller configures PBR instead of static routes.Example:
- name: CLUSTER_NAME value: "dev-cluster"
-
While deploying the NetScaler Ingress Controller, provide the
CLUSTER_NAME
as an environment variable. This value should be the same as the value provided in node controller.Example:
- name: CLUSTER_NAME value: "dev-cluster "
-
Specify the argument
--enable-cnc-pbr
asTrue
in the arguments section of the NetScaler Ingress Controller deployment YAML file. When you specify this argument, NetScaler Ingress Controller is aware that the node controller is configuring PBR on the NetScaler.Example:
args: - --enable-cnc-pbr True
Note:
The value provided for
CLUSTER_NAME
in the node controller and NetScaler Ingress Controller deployment files should match when they are deployed in the same Kubernetes cluster.The
CLUSTER_NAME
is used while creating the net profile entity and binding it to service groups on NetScaler VPX or MPX.
Validate PBR configuration on a NetScaler after deploying the node controller
This validation example uses a two node Kubernetes cluster with node controller and NetScaler Ingress Controller deployed.
You can verify that the following configurations are added to the ADC by node controller:
-
A net profile is added, with the value of
srcIP
set to the SNIP added by node controller while creating the VXLAN tunnel network between the NetScaler and Kubernetes nodes. -
NetScaler Ingress Controller binds the net profile to the service groups it creates.
-
Finally, node controller adds PBRs.
Here:
- The number of PBRs is equal to number of Kuberntes nodes. In this case, it adds two PBRs.
-
The
srcIP
of the PBR is theSNIP
added by node controller in tunnel network . ThedestIP
is the CNI overlay subnet range of the Kubernete node. TheNextHop
is the IP address of Kubernetes node’s VXLAN Tunnel interface.Note:
node controller adds PBRs instead of static routes. The rest of the configuration of the VXLAN and bridge table remains the same. For more information, see the node controller configuration.