Consider running multiple instances without overhead

[linki]

in order to make the daemons HA we could deploy multiple copies of it. though, only one of them should be active at any point in time.

each instance could try to grab a lock in etcd (or use config maps?) in order to elect a leader.

this is also how kubernetes' controller manager and scheduler work.

[linki]

leader election itself can be easily implemented with: https://github.com/coreos/coreos-kubernetes/blob/fb30f43c4ad01a987ae8b5dcc97ba031d4ff33e9/multi-node/generic/controller-install.sh#L428-L434

doesn't require etcd, uses kube apiserver directly

[ideahitme]

I suggest this delayed until this PR is merged kubernetes/kubernetes#39173

Currently k8s.io/kubernetes/... packages have internal conflicts with official Go CLI. Upon the merge I expect much easier integration of leaderelection mechanism.