Sorry for the fuss - we solved it. We still not know why it worked in the previous version though.
After reviewing our haproxy configuration and routing the direct TCP traffic (instead of upgrading to SSL manually) the issue disappears. Closing this.