Forking subprocesses in Perl unit tests stops prove; Test::Harness exiting

Question

I have been trying to use the Perl utility/module "prove" as a test harness for some unit tests. The unit tests are a little more "system" than "unit" as I need to fork off some background processes as part of the test, Using the following...

sub SpinupMonitor{
   my $base_dir = shift;
   my $config = shift;

   my $pid = fork();
   if($pid){
      return $pid;
   }else{
      my $cmd = "$base_dir\/..\/bin\/monitor_real.pl -config $config -test";
      close STDOUT;

      exec ($cmd) or die "cannot exec test code [$cmd]\n";
   }
}

sub KillMonitor{

   my $pid = shift;

   print "Killing monitor [$pid]\n";
   kill(1,$pid);
}

However for some reason when I have my .t file spin up some extra processes it causes the test harness to hang at the end of the first .t file after all the tests have finished, rather than going on to the next file, or exiting if there is only one.

At first I wondered if it might be because I was killing of my sub-processes and leaving them defunct. So I added..

$SIG{CHLD} = \&REAPER;
sub REAPER {
   my $pid = wait;
   $SIG{CHLD} = \&REAPER;
}

To the code. But that doesn't help. In fact on closed examination it turns out that my perl test file has exited and is now a defunct process and it is the prove wrapper script that has not reaped its child. In fact when I added a die() call at the end of my test script I got...

# Looks like your test died just after 7.

So my script exited but for some reason the harness isn't unraveling.

I did confirm that it is definitely my sub-processes that are upsetting it as when I disabled them while the tests failed the harness exited properly.

Is there anything I am doing wrong with the way I am starting up my processes that might upset the harness in some way?

Answer 1

Note that you don't test whether fork() failed. You need to make sure $pid is defined before assuming that "false" means "child."

Because your $cmd contains shell metacharacters (spaces), Perl is actually using a shell when you call exec(). While your monitor is running, there's (1) Perl, (2) a child sh -c, and (3) a grandchild Perl running monitor_real.pl. What that means in particular is when you call KillMonitor, you're only killing the shell (because that's the PID you have) and not the monitor.

You might also be interested in How do I fork a daemon process? from the Perl FAQ.