Question

I have a unstable program that kept crashing randomly, I could use a bash script to keep it running, but is there a way to monitor it’s state ? Not by polling, but event-driven.

And and optional question, what if the monitor process isn’t the parent process of it, is this still possible ?

Answer 1

You can write a simple wrapper which forks and execs the crashing process, and just wakes up when it gets SIGCHLD and starts all over again; conceptual code below:

volatile sig_atomic_t gotchld = 0;

void
sigchld_handler(int sig)
{
    if (sig == SIGCHLD)
        gotchld = 1;
}

signal(SIGCHLD, sigchld_handler);

for (;;) {
    switch ((pid = fork()) {
    case -1:
        err(1, "fork");
    case 0:
        exec_my_crashy_program_here();
        errx(1, "came back from exec!");    /* paranoia */
    default:
        printf("sleeping for pid %u\n", pid);
        while (!gotchld)
            select(0, NULL, NULL, NULL, NULL);  /* sleep forever waiting for signal */
        /* possibly do fancy stuff with the wait() family of functions */
        gotchild = 0;
        continue;    
    }
}

You don’t need anything fancy, just a basic knowledge of how child and parent processes interact via signals.

Please also consider that it’s been a while since I’ve dealt with signal stuff directly, so double-check that the code there is correct.