Sorry to bump an old post, but was this ever resolved? We have the same issue on a cluster and can't pinpoint where the connection is being dropped.