If you're getting KILL, then forget about it. It can't be ignored or caught. How is it determining how old your process is? Would an occasional fork() (letting the parent die) help?

A better technique anyway might be to write an occasional checkpoint file out to disk and to run the job more frequently in cron. If the task is already running do nothing. But if the task is gone, and the checkpoint file is there, pick it up and continue processing.