CentOS - Jun 2011 - Restarting a Perl-script (socket daemon) from /etc/inittab

Hello fellow CentOS sysadmins,

I run a small multiplayer card game
with around 500 users at peak times.

The client is in Flash and the server is in Perl.

The Perl server binds to port 8080, i.e. only
1 instance of it can be started (important detail).

The Perl server poll()s TCP-sockets and forks
only once - at the startup by calling this method:

    sub daemonize {
        die "Can not fork: $!\n" unless defined (my $child = fork());
        # the parent should die
        exit 0 if $child;

        setsid();
        open(STDIN, '</dev/null');
        open(STDOUT, '>/tmp/pref.txt');
        open(STDERR, '>&STDOUT');
        chdir('/');
        umask(0);
    }

It runs on CentOS 5.6 Linux / 64 bit,
PostgreSQL 8.4.8 and Perl 5.8.8.

Because my budget is small and I had enough
troubles already, I want to use as little additional
software as possible - so that I can change hosters
or reinstall my cheapo server quickly. That is why
I for example just log to /tmp/pref.txt instead of
installing syslog-ng. And that is why I'd like to
use /etc/inittab for restarting my Perl daemon.

My Perl daemon runs mostly stable, but
approx. once a week it can crash with a

    May 29 11:06:46 myhost kernel: pref.pl[3113]:
    segfault at 00007fffa21e6fd8 rip 0000003cce274460
    rsp 00007fffa21e6fd0 error 6

Since I'm tired of restarting the server manually,
I've tried to add it to the /etc/inittab:

    pref:3:respawn:/bin/su -c '/usr/local/pref/pref.pl' nobody

(and I've added a nightly cronjob to
"pkill pref.pl" in the hope to refresh perl this way).

Unfortunately this does not work as expected -
in the /var/log/messages I see that the script
is being started again and again every 5 mins:

    Jun  2 18:55:56 myhost init: Id "pref"
    respawning too fast: disabled for 5 minutes

What am I doing wrong here? I was hoping to
being able to use /etc/inittab here, because
I remember using it for a similar situation at work
few years ago (also with a Perl daemon)
and it worked well then...

Thank you!
Alex

P.S. I've also posted my question at
http://serverfault.com/questions/276428/restarting-a-perl-script-socket-daemon-from-etc-inittab

Alexander Farber wrote:
> Hello fellow CentOS sysadmins,
>
> I run a small multiplayer card game
> with around 500 users at peak times.
>
> The client is in Flash and the server is in Perl.
<snip>
> My Perl daemon runs mostly stable, but
> approx. once a week it can crash with a
>
>     May 29 11:06:46 myhost kernel: pref.pl[3113]:
>     segfault at 00007fffa21e6fd8 rip 0000003cce274460
>     rsp 00007fffa21e6fd0 error 6
<snip>
Ok, here's the more significant question: why's it SEGV'ing?
What's at
line 3113 that could segv?

         mark, was a programmer longer than I've been a sysadmin

On 6/2/2011 2:46 PM, Alexander Farber wrote:
>
> The Perl server poll()s TCP-sockets and forks
> only once - at the startup by calling this method:
>
>      sub daemonize {
>          die "Can not fork: $!\n" unless defined (my $child =
fork());
>          # the parent should die
>          exit 0 if $child;
[....]
> Since I'm tired of restarting the server manually,
> I've tried to add it to the /etc/inittab:
>
>      pref:3:respawn:/bin/su -c '/usr/local/pref/pref.pl' nobody
>
> (and I've added a nightly cronjob to
> "pkill pref.pl" in the hope to refresh perl this way).
>
> Unfortunately this does not work as expected -
> in the /var/log/messages I see that the script
> is being started again and again every 5 mins:
>
>      Jun  2 18:55:56 myhost init: Id "pref"
>      respawning too fast: disabled for 5 minutes
>
> What am I doing wrong here?
It needs to not fork/exit on its own if you want init to respawn when it 
exits.

-- 
  Les Mikesell
    lesmikesell at gmail.com