ARP

ARP (sometimes referred to as layer 2 as it's updating the underlying network topology) is a simplistic protocol that is used to update the underlying network that in order to reach a certain IP address traffic should be sent to a specific piece of hardware. ARP will typically broadcast updates to the entire network that update the IP to Hardware (MAC) mapping this ensures traffic is sent to the correct physical or virtual network nic.

Control plane

When using ARP with the control plane feature, then a leader election will take place in order to determine the leader. This leader will then host the control place IP address and use ARP (through a gratuitous ARP broadcast) to update the network accordingly.

Services

Services can be exposed in two ways:

  • leaderElection for all services
  • leaderElection per service

leaderElection for all services

In this mode (default) all kube-vip pods will elect a leader and this leader will be in charge of exposing all service addresses through ARP. This mode can become a bottleneck as traffic for all services will come to a single node.

leaderElection per service

In this mode kube-vip will perform an election every time a new Kubernetes service is created allowing service addresses to be spread across all nodes where a kube-vip pod is running.