Add support for taint until kubeip assigns IP *or* exit

[maxenced]

I'm running two different workloads on an autoscaling group (on gke). The autoscaling group can grow up to 30 nodes, but we only have 10 IPs that can be used (mostly because our workloads are heavily dynamic so we can go from 1 up to the 30 nodes, but unused IPs cost a lot).

For workloads which requires a known public IP, the taint provided by kubeip works as expected : when kubeip assigns an ip, taint is removed and workload can be scheduled.

However, I also have a lot of workloads which do not require known public IP, so they have a toleration. The issue is that the workload can be assigned to a node and then kubeip change node's IP, resulting in some connection error in the workload.

It would help a lot to have , for example, support for a second taint, which is removed by kubeip when either of this happens :

• kubeip has assigned an ip (same as existing taint then)
• kubeip exit / stops to try assigning an IP (real time depends on kubeip configuration)

If you think about another solution for this kind of workloads ?