The system I used didn't have any notion of repair, just retry-forever. What did you use for that? I've written service tree management tools that do that sort of thing on a single host but not any kind of distributed system.