Open
Description
Description
Overprovisioning can be accomplished by running low-priority pods; when a new API replica is requested, it will replace the placeholder pods, which will then be pending and trigger autoscaling. See cluster-autoscaler FAQ.
Motivation
Currently, the cluster autoscaler doesn't add nodes until there is a pending, unschedulable pod. Therefore, there will be a delay of a few minutes until the requested pod will be live.