Deployment Options¶
These options are configured using pool-member-type parameter in CIS deployment.
NodePort¶
This section documents K8S with integration of CIS and BIG-IP using NodePort configuration. Benefits of NodePort are:
- It works in any environment (no requirement for SDN)
- No persistence/visibility to backend Pod
- Can be deployed for “static” workloads (not ideal)
Similar to Docker, BIG-IP communicates with an ephemeral port, but in this case the kube-proxy keeps track of the backend Pod (container). This works well, but the downside is that you have an additional layer of load balancing with the kube-proxy.
When using NodePort, pool members represent the kube-proxy service on the node. BIG-IP needs a local route to the nodes. There is no need for VXLAN tunnels or Calico. BIG-IP can dynamically ARP for the Kube-proxy running on node.
ClusterIP¶
This section documents K8S with integration of CIS and BIG-IP using clusterIP configuration. Benefits of clusterIP are:
- Requires ability to route to Pod
- Flannel VXLAN, OpenShift VXLAN
- Alternately Pod routable through network, for example:
- Calico BGP
- Public Cloud network
- Static Routing Mode (CIS adds the static routes for Kubernetes nodes on BIG-IP, managing them for node updates). Currently supported with antrea, ovn-k8s, cilium-k8s and flannel CNIS.
The BIG-IP CIS also supports a cluster mode where Ingress traffic bypasses the Kube-proxy and routes traffic directly to the pod. This requires that the BIG-IP have the ability to route to the pod. This could be by using an overlay network that F5 supports (Flannel VXLAN, or OpenShift VXLAN). Leave the kube-proxy intact (no changes to underlying Kubernetes infrastructure).
NodePortLocal¶
This section documents K8S with integration of CIS and BIG-IP using the NodePortLocal feature provided by Antrea CNI. Benefits of NodePortLocal are:
- CIS consumes NPL port mappings published by the Antrea Agent (as K8s Pod annotations). So unlike NodePort mode, it bypasses the Kube-proxy and routes traffic directly to the pod removing the secondary hop of load balancing carried by kube-proxy.
- Exposes the port on node where pod is running. It decreases the port range requirements, unlike NodePort which requires a range of ports to be exposed in all Kubernetes nodes.
In NodePortLocal, rather than using kube-proxy for node to pod routing, there is a single iptables entry per pod which means direct connectivity to each pod.
Auto¶
This section documents K8S with integration of CIS and BIG-IP using the “auto” configuration.
Auto mode is a combination of cluster and nodeport mode. In “auto” mode, CIS will learn the service type and populates the BIG-IP with the respective pool members.
- If the service type is ClusterIP, then pod IP’s are populated on the BIGIP for the respective pool.
- If the service type NodePort, then node IP’s are populated on the BIGIP for the respective pool.
For detailed documentation, see: Pool Member Type - auto.
Note
To provide feedback on Container Ingress Services or this documentation, please file a GitHub Issue.