The scaling configured according to cpu usage:
On the chart it's about 8 minutes between upscaling and 2 mins between downscalings.
We cannot upscale faster because an instance requires ~5min to launch, however I'd like to have downscaling with similar delay too. Other variables:
Default Cooldown 600
Health Check Grace Period 400
What's configured wrong here?
In your decrease logic, increase the value for "consecutive periods...".
That should induce delay in your ASG shrink actions.