Current situation

• Python UDP server running in Kubernetes in Google Cloud.
• The cluster has 5 virtual machines.
• The autoescaling is enable from 1 POD to 10 POD.
• The container resources are limited: CPU=100m and Memory=256MB
• The cluster has a LoadBalence

The problem

When the UDP server start to receive the data as the load grows the cluster begin to create several PODs. However this new PODs does not receive any data.

This causes that only the first POD process all the load and reach his limits very quick, starting restar and finally dead.

Due the other PODs does not process the data, the server stop to receive the incoming data.

What could be wrong? Is not suppose that the load balance will distribute the load between all PODs?

Note:

In the load balance I have tried with and without health check

Actual situation:

After talking with Google's engineer and make a deeper test we have seen the problem is that each POD receive the data from the same IP, so if the data are send through the same IP, all data will go to the same POD (even if you have a load balance)