Thanks for your suggestion. The job recovery sample you mentioned demonstrates the manual recovery, as far as I understand. The client cannot automatically recover job since it cannot detect the broken connection without using heartbeats. Manual recovery is not suitable in some case.
Actually, server become the single point of failure of a JPPF cluster. If there would be a hot-standby server synchronizing server state with the primary server, the server fail-over could take place in the event of server failure.