Reputation: 265
I am trying to run an application inside a Perl script using system()
. The application I'm running gets stuck sometimes (it enters some kind of infinite loop). Is there a way I can know if this application is stuck and kill it to continue with the Perl script?
I'm trying to do something like this:
start testapp.exe;
if(stuck with testapp.exe) {
kill testapp.exe;
}
Upvotes: 2
Views: 6535
Reputation: 37734
Determining if "it is stuck in infinite loop" is called Halting Problem and is undecidable.
If you want to kill it, you will have to fork the application using fork
and then kill it from the other fork, if it is going for too long.
You can determine if the proccess is going for too long by this
use POSIX ":sys_wait_h";
waitpid($pid, WNOHANG)>0 #waitpid returns 0 if it still running
at least, according to this manual page
I am not sure how well it works on various systems, you can try it out.
Not a direct answer, but I can recommend using forks
module if you want to fork with ease, but it works only on UNIX systems (not windows).
OK, more helping code :) It works in UNIX, according to perlfork perldoc, it should work on Windows exactly the same way.
use warnings;
use strict;
use POSIX ":sys_wait_h";
my $exited_cleanly; #to this variable I will save the info about exiting
my $pid = fork;
if (!$pid) {
system("anything_long.exe"); #your long program
} else {
sleep 10; #wait 10 seconds (can be longer)
my $result = waitpid(-1, WNOHANG); #here will be the result
if ($result==0) { #system is still running
$exited_cleanly = 0; #I already know I had to kill it
kill('TERM', $pid); #kill it with TERM ("cleaner") first
sleep(1); #wait a bit if it ends
my $result_term = waitpid(-1, WNOHANG);
#did it end?
if ($result_term == 0) { #if it still didnt...
kill('KILL', $pid); #kill it with full force!
}
} else {
$exited_cleanly = 1; #it exited cleanly
}
}
#you can now say something to the user, for example
if (!$exited_cleanly) {...}
Upvotes: 2
Reputation: 14865
Here is one way you can handle the problem if you know that testapp should not take more than N seconds to do its thing, then you can use a timeout to kill the app by way of IPC::Run.
In the example below there is a timeout of 1 second which kills the sleep 10
command that takes too long (longer than the timeout of 1 second). If this doesn't do what you want, then you should provide more information about how you can detect that testapp.exe is "stuck".
#!/usr/bin/env perl
use IPC::Run qw( run timeout );
eval { # if (stuck with testapp.exe for more than N seconds)
@cmd = ('sleep', '10'); # this could be testapp.exe instead of sleep
run \@cmd, \$in, \$out, \$err, timeout( 1 ) or die "test"; # start testapp.exe
print "do stuff if cmd succeeds\n";
};
print "more stuff to do afterwards whether or not command fails or succeeds\n";
Upvotes: 1
Reputation: 386706
system("start testapp")
is short for
system("cmd", "/c", "start testapp")
Perl just knows about cmd
; it doesn't know anything about start
, much less about testapp
. system
is not the tool you want. That's the first problem.
The second problem is that you haven't defined what it means to be "stuck". If you want to monitor a program, it needs a heartbeat. A heartbeat is a periodic activity that can be externally examined. It can be writing to a pipe. It can be changing a file. Anything.
The monitoring program listens for this heartbeat, and presumes the program is dead if the heart stops beating, so to speak.
"Killing" is done using signals in unix, but it's done using TerminateProcess
in Windows. The third problem is that Perl core does not give you access to that function.
The solution to the first and third problem is Win32::Process. It allows you to launch a process in the background, and it also allows you to terminate it.
Creating a heartbeat is up to you.
Upvotes: 1
Reputation: 755074
You can't determine that the application is stuck if you execute it like that, because the system
statement won't return until the application terminates.
So, at least, you need to start the test application so it can run asynchronously from the Perl script that is to monitor it.
Having resolved that part of the problem, you have to establish a mechanism that will allow the monitoring Perl script to determine that the application is stuck. That is a non-trivial exercise, and likely system dependent, unless you adopt a simple expedient such as requiring the application to write a heart-beat indication somewhere, and the Perl script monitors for the heart-beat. For example (not necessarily a good example), the application could write the current time into a file identified by its PID, and the Perl script could monitor the file to see if the heart-beat is sufficiently recent. Of course, this assumes that the 'infinite loop' doesn't include code that writes to the heart-beat file.
Upvotes: 0