We are working with nomad 1.0.4 and the 0.3.7 version of the autoscaler and we are observing with a very similar setup to the on-demand batch example that after an allocation is schedule on a client the autoscaler immediately set the node to drain. I believe this is a possible outcome but I wanted to verify this with the community. Is this actually possible? Is there a way to minimize this?
One difference with the example is that the job uses all the resources of a client and will not schedule more than one job. So the scale out happens when the are jobs pending and scale in when the job finishes.