Perl fork and kill - kill(0, $pid) always returns 1, and can't kill the child

Question

I am running a perl script on a Linux host. I'm trying to write a script that forks, where the child starts a program that takes forever and the parent times out after 5 seconds, killing the child. Here is what I have:

my $start = time();
my $timeOut = 5;

my $childPid = fork();
if ($childPid) {
    # I am the parent, and $childPid is my child
    while (kill(0, $childPid)) {
        if (time() > $start + $timeOut) {
            $numKilled = kill(SIGTERM, $childPid);
            printf("numKilled: %s\n", $numKilled);
        }
        sleep 1;
    }
}
else {
    # I am the child - do something that blocks forever
    `adb -s 410bf6c1 logcat`;
    exit;
}

Output:

aschirma@graphics9-lnx:~$ perl perl-fork-test.pl 
numKilled: 1
numKilled: 1
numKilled: 1
numKilled: 1
numKilled: 1
...

The behavior I expect is that I see "numKilled: 1" exactly once, and the child process (and any of its children) is killed after roughly 5 seconds. But I see from experiment that neither the child nor its children are getting killed. The kill(SIGTERM, $childPid) appears to do nothing.

How can I actually kill the child?

Solution 1

From perldoc fork:

If you fork without ever waiting on your children, you will accumulate zombies. On some systems, you can avoid this by setting $SIG{CHLD} to "IGNORE" .

I was able to get the desired behavior when adding $SIG{CHLD} = 'IGNORE'; to the top of your code.

[ben@imac ~]$ perl test.pl
numKilled: 1
[ben@imac ~]$ 

Alternatively, adding waitpid($childPid, 0); after kill did the trick as well.

Solution 2

Should be something like this. This isn't following best practices, but it should help you with your problem...

#!/usr/bin/perl

use warnings;
use strict;
use POSIX qw(:sys_wait_h);

my $timeOut = 5;
$SIG{ALRM} = \&timeout;
$SIG{CHLD} = 'IGNORE',
alarm($timeOut);

my $childPid = fork();
if ($childPid) {
    while(1) {
        print "[$$]: parent...\n"; 
        sleep(2); 
    }
}else {
    # I am the child - do something that blocks forever
    while(1){
        print "[$$]: child...\n";
        sleep(2);
    }
    exit;
}

sub timeout {
    print "killing $childPid\n";
    print "###\n" . `ps -ef | grep -v grep | grep perl` . "###\n";
    if ( ! (waitpid($childPid, WNOHANG)) ) {
        print "killing $childPid...\n";
        kill 9, $childPid;
        die "[$$]: exiting\n";
    }
}

OUTPUT:

$ forktest.pl
[24118]: parent...
[24119]: child...
[24118]: parent...
[24119]: child...
[24118]: parent...
[24119]: child...
killing 24119
###
cblack   24118 12548  0 14:12 pts/8    00:00:00 /usr/bin/perl ./forktest.pl
cblack   24119 24118  0 14:12 pts/8    00:00:00 /usr/bin/perl ./forktest.pl
###
killing 24119...
[24118]: exiting

Solution 3

Try killing the process group (use -$pid):

kill TERM => -$pid;

See perlipc.

Solution 4

I am also seeing this behaviour, that kill 0 returns true even after the process was gone; I suspected you might be missing a call to waitpid (often done in a SIGCHLD handler) which would result in this, but even after adding it, the kill 0 continues to return true.

I suggest you use non-blocking waitpid instead of kill (and verify that the process actually does die on a SIGTERM signal - some may not, at least immediately). You may also want to try SIGINT after a while if SIGTERM didn't work, and in the last resort SIGKILL.

View more solutions

Comments

[friedo]

Are you importing the SIGTERM constant from somewhere? I bet it evaluates to 0.

[melpomene]

Are you using use warnings; use strict;? Why not?

[antred]

Slightly off-topic, but consider checking if ( defined( $childPid ) ) before doing anything with your child process handle. forkreturns undefif it fails to create the process.

[PodTech.io]

kill(0, $childPid) is basically saying if the process is running then continue.. subsequently he is sending a SIGTERM