We have a single instance C++ application running in a Pod that is receiving external UDP traffic. We are running a 3 node k8s/Rancher HA setup running on ESXi. After a while, the Pod will restart and the UDP packets no longer reach the new Pod (within the same node). However, we can see that the Node itself still receives the UDP packets via tcpdump.
Initially, we identified this problem in k8s v1.13.5 and we found an actual bug that might be related to our issue that was fixed in k8s v14.0.x. https://github.com/kubernetes/kubernetes/pull/71573
However, after we have updated k8s, we still see this same behaviour. Does anyone have any ideas of how we can solve it? We know the application is fine because we use this in production without issue in Docker, on hardware, and in VMS. However, in Rancher, the traffic just stops and the only way to start it again is to redeploy the container and to restart the external application sending the UDP traffic.