kgctl connect improve availability #369
Description
Hi!
We are utilizing 'kgctl connect' to connect a machine to the Services CIDR inside a cloud k8s cluster running kilo as a daemonset, which works great.
I looked at the code and saw that the first ready node is assigned the 'allowed ips' in its peer section. When this node is not available anymore we have a downtime until the next sync.
We are trying to mitigate the effects by sing a shorter resync period and some modification in the kgctl connect code.
We are using cluster api to manage the clusters and one of the first things cluster api does when deleting a node is to set spec.unschedulable to true. We only need access to the services CIDR from this machine, so we exclude all nodes which have spec.unschedulable=true. (https://github.com/squat/kilo/blob/main/cmd/kgctl/connect_linux.go#L305)
Is there any better solution?