The auto downscale checks for the average of usage for the last 4h. But, sometimes we have a cluster with little usage for almost all the 4 hours, but at the last 30, 15 minutes, it starts to consume more CPU (something higher than 60% or even higher).

As the autoscale only checks at the last 4h it sometimes happen that a cluster does a auto downscale when it shouldn't. A litler time after it needs to scale back up because the usage goes higher than 90% (or at least higher than 75%)

It would be much better if the downscale would check not only the avarage of the last 4h but also the consumption of the time (or at least, the avg of the last 5 minutes)

We had to disabled the downscale for some of our clusters because of this king of problem.