CPU measurement after autoscaling

Hi, we have a cluster that is constrained between M10-M30 but which is scaling (nearly every day) between M20 and M30. It seems that starting up the scaled-down cluster causes heavy CPU usage that causes the cluster to scale back up.

Example timestamps:
03:39 - Scaled to M20
03:50 - Alert “System: CPU (User) % has gone above 95”
03:57 - Alert resolved
04:55 - Scaled to M30

Looking at the metrics (in the Atlas interface and the exported Prometheus metrics), it’s pretty clear that the high CPU usage is from one of the instances starting up (i.e. there is a clear new instance started right at 03:50).

Is there any way to prevent the autoscaling from scaling the cluster up as a direct result of the initial CPU usage of scaling down?