start and kill a process in a local shell

[danmcb]

I want to run a program on a local machine that does not exit itself. I want to capture it's output, and then send it a CTRL-C to stop it. What's the easiest way?

[shmem]

See open. Opening a pipe to a program returns that programs PID.

$pid = open $pipe, "/some/program |";
while (<$pipe>) {
    ... # /some/program's output in $_
}
close $pipe;         # program gets a sigpipe
if (kill 0, $pid) {  # still there?
   kill 15, $pid;
}

[oha]

check IPC::Open2, it will open a pipe to a command, and return the PID, with the pid you can send a kill

Oha

[apl]

I'm assuming this is a daemon running in the background on a *nix box.

To capture it's output, have crontab or whatever starts it redirect to some standard directory with a filename containing the date and time the process started.

As far as stopping it is concerned, don't send a CTRL-C, use a kill -9 <id> where <id> is the process-number of the program.

[shmem]

Signal KILL (9) is last resort and should be used as such. A process killed that way has no chance to clean up resources (e.g. flush buffers, close filehandles).

Signals 1, 2, 3 or 15 should be used, an 9 only when the process is still running after those.

[apl]

You are correct. But when I saw a program (...) that does not exit itself, I made the unwarranted assumption that the original writer of the program didn't seem particularly interested in cleaning up.

[cdarke]

> then send it a CTRL-C to stop it.

CTRL-C is a SIGINT, so it is kill INT,pid;

[cmv]

I recently had to do something similar, and stumbled into a rats nest of problems revolving around SIGSTOP, and process groups.

In my scenario, I had a perl script starting up and managing a unix process on a Sun workstation. Anytime I hit ctl-z at my shell prompt (the shell that started the perl script), both the perl script and the background unix process stopped. This is not what I wanted at all.

After much gnashing of teeth and tearing out hair (I should have asked here), I finally figured out that the SIGSTOP was being sent to everyone in my process group. To solve the problem, I needed to create the unix process in its own process group. This is how I learned about the wonders of the perl "setpgrp" command, and process groups in general.

So here is my code:

# Fork a unix process...
my $pid;
if($pid = fork) {
    print STDERR "parent pid=$$ child pid=$pid\n";
}elsif (defined $pid) {
    print STDERR "child pid=$$ pid=$pid\n";
    # Must do setpgrp as child for nohup and job control stuff...
    setpgrp(0, $$);
    exec $unixcmd || die "Bad exec $!";
}
kill 'INT', $pid;

Hope that helps.

-Craig

UPDATE:

oha opened my mind, by letting me know that the setpgrp could be done from either the parent or the child process. oha++ Thanks!