I realize this is a long shot but here goes. For Oracle databases you can add (ENABLE=BROKEN) in your connect string to work in coordination with OS TCP Keepalive settings. We have used this with AWS Gateway Load Balancer which has a 350 second unchangeable idle timeout. By setting tcp_keepalive_time to less than 350 in Linux and using (ENABLE=BROKEN) an Oracle connection from an AWS EC2 to an on premises Oracle database through the GWLB will not timeout. Without (ENABLE=BROKEN) the Linux tcp_keepalive_time setting does not seem to do much if anything.
Anyway, we have GoldenGate between AWS and on premises and I would like to find the equivalent of (ENABLE=BROKEN) to cause GoldenGate to use the TCP keepalive functionality. We have set tcp_keepalive_time to 60 seconds in the EC2s that have GoldenGate but that by itself did nothing. We have weird unexplainable issues with all our GoldenGate connections between AWS and on premises and Oracle support has not been able to find anything so I thought it would be worth trying to get the keep alive features working for GGS. Our on premises GGS works like a charm so it is just the ground to cloud that is messed up.
Anyone out there have any ideas?