In my scenario, I have used custom metrics auto scaling the pods but the issue am facing is when the pod is re scaling the connections are lost and I think k8s doesn't consider the live connections in a pod while re-scaling the pods.

Example:

Currently am having 2 pods and I have set the threshold to 100 socket connection. If I push 300 socket connections to the pods the pods will auto scale to 4 and consider each pod having more that 50 connections but the issue is when re-scaling, the pods will re-scale to 2 after the cool down but the connection will be terminated in the extra two pods which spin up. Is there any solution for the same.