xcat-core/xCAT/postscripts/HPCbootstatus.aix

#!/usr/bin/perl

use IO::Socket;

# Sample postscript to check appstatus for GPFS,LAPI and LoadL on AIX system
# Report the appstatus to xcatd on MN.

my $timeout = 300;
# The number of items to check, currently including:
# gpfs-daemon, gpfs-quorum, gpfs-filesystem, lapi-pnsd, loadl-schedd, loadl-startd
my $itemnum = 6;
my $passed = 0;
# Set sleep time
my $interval = 15;

my $master;

# Get server IP from xcatinfo file
if (-f "/etc/xcatinfo") {
	# if this file exists assume it has the correct server name
	my $cmd = "cat /etc/xcatinfo | grep 'XCATSERVER'";
	my $SNline = `$cmd`;
	my ($junk, $servnode) =  split(/=/, $SNline);
	chomp $servnode; 
	$servnode =~ s/^\s*//;
	if ($servnode) {
		$master = $servnode;
	}
}

my $gpfsd = "hpcbootstatus gpfs-daemon=up";
my $gpfsq = "hpcbootstatus gpfs-quorum=achieved";

# add HPCbootstatus.aix to /etc/inittab for diskful node reboot
# see if it is already there
my $cmd = "/usr/sbin/lsitab hpcbootstatus > /dev/null 2>&1";
my $rc = system("$cmd") >>8;
if ($rc != 0)
{
    # add new entry
    my $mkcmd = '/usr/sbin/mkitab "hpcbootstatus:2:wait:/xcatpost/HPCbootstatus.aix > /dev/console 2>&1"';
    system("$mkcmd");
}

while (1)
{
    my $count= 0;
    
    # Check gpfs-daemon via port 1191.
    my $cmd = "netstat -nA|grep 1191";
    my $rc = `$cmd`;

    if($rc)
    {
        # send message to xcatd
        if (&updateflag($master,$gpfsd) != 0)
        {
            print "HPCbootstatus.aix: Failed to report $gpfsd to xcatd on $master!\n";
        }
        $count++;
    }

    # Check gpfs-quorum via /usr/lpp/mmfs/bin/mmgetstate
    $cmd = "/usr/lpp/mmfs/bin/mmgetstate -s|grep achieved";
    $rc = `$cmd`;
    
    if($rc)
    {
        # send message to xcatd
        if (&updateflag($master,$gpfsq) != 0)
        {
            print "HPCbootstatus.aix: Failed to report $gpfsq to xcatd on $master!\n";
        }
        $count++;
    }

    # Check gpfs-filesystem via /usr/sbin/mount
    $cmd = "cat /tmp/mount |grep mmfs |awk '{print \$2}'";
    my @fs = `$cmd`;
    my $new = "";

    if(scalar @fs)
    {
        foreach my $line (@fs)
        {
            chomp $line;
            $new .= $line;
            $new .= "!";
        }

        # send message to xcatd
        my $gpfsf = "hpcbootstatus gpfs-filesystem=$new";
        if (&updateflag($master,$gpfsf) != 0)
        {
            print "HPCbootstatus.aix: Failed to report $gpfsf to xcatd on $master!\n";
        }
        $count++;
    }
    
    # Check LAPI via lssrc
    $cmd = "/usr/bin/lssrc -s pnsd |/usr/bin/awk '(NR==2){print \$3}'";
    $rc = `$cmd`;

    if($rc)
    {
        # send message to xcatd
        chomp $rc;
        my $lapi = "hpcbootstatus lapi-pnsd=$rc";
        if (&updateflag($master,$lapi) != 0)
        {
            print "HPCbootstatus.aix: Failed to report $lapi to xcatd on $master!\n";
        }
        $count++;
    }

    # Check loadl-schedd via llrstatus
    $cmd = "/usr/bin/llrstatus -h `hostname` -r %sca %scs";
    $rc = `$cmd`;

    if($rc)
    {
        # send message to xcatd
        chomp $rc;
        my $llsc = "hpcbootstatus loadl-schedd=$rc";
        if (&updateflag($master,$llsc) != 0)
        {
            print "HPCbootstatus.aix: Failed to report $llsc to xcatd on $master!\n";
        }
        $count++;
    }

    # Check loadl-startd via llrstatus
    $cmd = "/usr/bin/llrstatus -h `hostname` -r %sta %sts";
    $rc = `$cmd`;

    if($rc)
    {
        # send message to xcatd
        chomp $rc;
        my $llst = "hpcbootstatus loadl-startd=$rc";
        if (&updateflag($master,$llst) != 0)
        {
            print "HPCbootstatus.aix: Failed to report $llst to xcatd on $master!\n";
        }
        $count++;
    }

    # All is done or timeout?
    if (($count == $itemnum ) || ($passed >= $timeout))
    {
        last;
    }

    $passed += $interval;
    sleep $interval;
}

############################################################
#
#   updateflag
#     Tells xCAT on the server about the appstatus. 
#     e.g. $status = "hpcbootstatus loadl-schedd=1!Avail";
#
############################################################
sub  updateflag {
    my $server = shift;
    my $status = shift;

    my $port = "3002";  #by default
    
    my $servip = `host $server | awk '{print \$3}'`;
    chomp($servip);
    $servip =~ s/,$//; # remove trailing comma
    my $remote = IO::Socket::INET->new( Proto => "tcp", PeerAddr  => $servip, PeerPort  => $port, Timeout => 0);

    my $try = 10;
    while (!$remote && $try-- > 0) {
        sleep 1;
        $remote = IO::Socket::INET->new( Proto => "tcp", PeerAddr  => $servip, PeerPort  => $port, Timeout => 0);
        print "HPCbootstatus.aix: Cannot connect to host \'$servip\'\n";
    }
    unless ($remote) { 
        print "HPCbootstatus.aix: Cannot connect to host \'$servip\'\n";
	return 1;
    }
    $remote->autoflush(1);

    my $line;
    while (defined ($line = <$remote>)) {
        chomp $line;
        if ($line eq "ready") {
           print $remote "$status\n";
        } elsif ($line eq "done") {
	    last;
        }
    }
    close $remote;
    return 0;
}