Use cpu_load_hottest_node as metric for autoscaling

[franciscocpg]

Is there any plan to use cpu_load_hottest_node as a metric for autoscaling?

I'm asking because I see some cases where the cpu_load is ok (but cpu_load_hottest_node is high), bigtable-autoscaler scales down and errors count starts to increase immediately. The image below is an example (around 06:00).

If I manually scale up the bigtable instances again, the error count will disappear.