#!/usr/local/bin/perl

# rload: formats output of "ruptime" info on load on cluster machines

# Usage: rload

# Examples:

#	rload
#	rload -help

$helpline = '11';
if ( $ARGV[0] =~ /^-h.*/ ) {
	system "head -$helpline $0; sleep 1";
} else {
	print "\n";
}

$ncol = '3';

$icol = '0';
$wad = '';

open(LIST,"ruptime | awk '{print \$1, \$7}' | sort -n +1 |");
$numhosts = '0';
$ngreater = '0';
$ntotal = '0';
while (<LIST>) {
	chop; chop;
	($name, $load) = split(' ',$_,2);
	$Hosts[$numhosts++] = $name;
#	print "name load $name $load\n";
	$wad = $wad .  $name . ' ' . $load . ' ';
	if ( $load >= 1.00 ) {
		$ngreater++;
	}
	$ntotal++;
	if ( ++$icol >= $ncol ) {
#		print "$wad\n";
		($a,$b,$c,$d,$e,$f) = split(' ',$wad); write;
		$wad = ''; $icol = 0;
	} 
}

$wad .= '--- 0 --- 0'; 


($a,$b,$c,$d,$e,$f) = split(' ',$wad); write;

# now clean out expired time files
chop($result = `rjob -timefile`);
($dir, $waittime) = split(' ', $result);
##print "got: $dir $waittime";
foreach (@Hosts) {
	$fname = "$dir"."$_";
	if( -f $fname && -M $fname > 2*$waittime) { system "rm -f $fname"; }
}

print "\nNumber of machines with load > 1.00 is: $ngreater out of $ntotal\n\n"; 
exit 0;

format STDOUT =
@<<<<<<<<<<   @##.###   @<<<<<<<<<<   @##.###   @<<<<<<<<<<   @##.###
$a,$b,$c,$d,$e,$f
.