- waitdemo: wait for a process to die
New problem: Let's wait for the process to die, OR 0.75 sec, whichever
comes first. This relates to the functionality of the timeout
command.
- waitlimit-poll
This polling solution works, but uses a lot of CPU.
- waitlimit-block
This blocking solution works without wasting CPU, relying on the fact
that signals—such as SIGCHLD, which is delivered to a parent when a
child exits—interrupt long-running system calls.
But does it really work? Remove the usleep() in the child and replace fork() with nfork(), then run it multiple times. You should eventually see a problem!
- waitlimit-block2
This blocking solution appears to solve the problem by using signals
in a different way. This time, SIGALRM interrupts waitpid.
This seems more reliable in practice. But does it really work in all situations?
- waitlimit-block3–waitlimit-block5
These blocking solutions really do work.