We are trying to troubleshoot a networking issue between the TCP service we have deployed in an Autopilot clusters and our IoT devices. I suspect that we are running into a TCP reset issue in the K8s environment. We can't reproduce the issues running the server application on a Compute Engine VM.
So far, we have had no luck in capturing packets in the Autopilot cluster. Is it possible to run a packet capture in an Autopilot cluster?
Side question - is it possible to allocate additional RAM to the cluster nodes in Autopilot mode (as I've read that adding RAM to the nodes may alleviate conntrack congestion that I suspect is resulting in the TCP reset packets being sent to the devices)?
It is not possible to run a packet capture, since there is no SSH access to the nodes in an Autopilot cluster.
As for the additional resources, autopilot will allocate them as needed unless they are specified in your pod's definition. Take a look at the Resource Management for Pods and Containers documentation.