Using netstat/lsof/tcp_dump from inside the containers did not help unfortunately. Eventual next step was to check nodes and kernel logs revealed the issue rightaway.