Cluster-autoscaler: provide a mechanism to grow node groups simultaneously.

[mwielgus]

We are receiving request to provide an option to grow and shrink node groups more evenly - in many cases the node groups are in different zones/subregions but contain the same type of machines. Having the inbalance in the node groups increase the max severity of a node group outage.

[mumoshu]

@mwielgus I'd like to work on this.

AFAIK, CA as of today already randomly choose an ASG from candidates(=ASGs which serves currently unschedulble pods hence to be scaled out).
If that's true, what you meant is to make CA first tries to choose the ASG with the fewest nodes, right?

[mumoshu]

Assuming my assumption above is correct, I'm going to implement the balance expander.
It will try to balance number of nodes among target node groups by returning an expander.Option with the least NodeCount.

[mumoshu]

This issue had been moved to kubernetes/autoscaler#47 and then resolved there.

[mwielgus]

Closing in favor of kubernetes/autoscaler#47