Scale to zero

[Y-Less]

Is it possible to use this to scale the nodes down to 0, then back up again? I did some testing and got the cluster down to 0, but then it just stayed there, even when there were pods queued for scheduling. Am I doing something else wrong, or will this just not come up from 0?

[hjacobs]

@Y-Less where do you run the autoscaler? If it runs as a normal pod on worker nodes it won't run when you scale to zero :smirk:

[Y-Less]

I run the autoscaler on the controller node, but scale the workers only.

[hjacobs]

@Y-Less you are right, the autoscaler only works if at least one node exists. Reason for this is that it iterates over all nodes and only considers these ASGs, i.e. if no node to iterate over => no ASG to set capacity. This could probably be fixed by changing the logic and getting the ASG name from some static configuration or similar (I would want to avoid that).