Yes, you are right about the init stopping if the process respawns too quickly. I forgot that, it's true.

You can stop the script manually instaed of waiting for init to decide, but that's no big deal. More important is that a well chosen sleep time between restarts will keep your system responsive anyway. That would also be a crude workaround if the cause for the programs death is temporary (say, a missing resource like a nfs share etc) and the program just "die"s instead of doing a wait-retry-cycle itself. Finally, logging and/or an alert mechanism can easily be implemented.

To prevent a misunderstanding: one could (and maybe should) put that code into the original program, so the admin can leave the watch-respwan work to init.

Andreas