kube-vip
Gives the Kubernetes control plane a floating virtual IP on bare metal, so kubectl has an address to talk to even while the node behind it reboots.
A cluster with three control-plane nodes needs a shared endpoint, and in the cloud you get one free from the provider's load balancer. On your own hardware the traditional answer has been keepalived for the virtual address plus HAProxy for the distribution — two extra services to configure and keep alive before the cluster even exists. kube-vip does the same job as a static pod on the nodes: leader election runs through Kubernetes' own leases, and the address is advertised either with ARP from the elected leader or via BGP to the network's routers. The same component can also hand out addresses to LoadBalancer services, making MetalLB unnecessary in smaller setups.