#!/usr/local/bin/perl 

# rjob: run a job on another machine on cluster in background.

# Usage: 
# rjob [-h -host host] [-dir workdir] [-s -show] command

# if no host specified, one with lowest load is selected (unless you
# have submitted a job there in the last $waittime minutes).
# if no working directory specified, work is done in current directory.
# -s or -show will show the command to be run and prompt you whether to run it.

# Caveats:
#   You must enclose command in double quotes if you have >, <, |, >>, & etc,
#   OR you can escape them with backslashes, \

# Examples:

#      rjob "a.out <  input >  output"       (preferred)
#      rjob  a.out \< input \> output 
#      rjob "a.out \< input \> output"       (not reco., for backward compat.)
#      rjob -host wren "a.out < input > output"
#      rjob -host wren -dir $HOME/work "hostname >>& file.out"
#      rjob "ps | grep a.out > file.out"
#

$waittime = '10';	# minutes waited to resubmit to same host

$Nice = 'nice '; 	# we try to be...

# Set following to your situation:

# remote shell command
#$RSH = 'rsh';		# BSD-ish

$RSH = 'remsh';		# SVR4-ish

$helpline = '37';
$waittime = $waittime/(24.0*60.0); 	# days!

$machinedir = "$ENV{'HOME'}".'/bin/cluster/';	# location of recent usage data
if ( ! -d $machinedir ) { system "mkdir $machinedir"; }

LOOP:			# process args
while (@ARGV) {
        $_ = shift @ARGV;
        CASE: {
                /^-h$/ && ( $host  = shift @ARGV, last CASE);
                /^-host$/ && ( $host = shift @ARGV, last CASE);
                /^-help$/ && ((system "head -$helpline $0"), exit 0, last CASE);
                /^-timefile$/ && ( (print "$machinedir $waittime\n"), exit 0, last CASE);
                /^-dir/ && ( $WD = shift @ARGV, last CASE);
                /^-s$/ && ( $show = 'True', last CASE);
                /^-show$/ && ( $show = 'True', last CASE);
		unshift(ARGV,$_);	# put back if done with flags
		last LOOP;
        }
}

if ( $#ARGV == 0 ) {		# if old form "a.out \< in \> out"
	$ARGV[0] =~ s/\\</</g;	# remove backslashes
	$ARGV[0] =~ s/\\>/>/g;
	$ARGV[0] =~ s/\\\|/\|/g;
	$ARGV[0] =~ s/\\&/&/g;
	$ARGV[0] =~ s/\\;/;/g;
}

unshift(ARGV,"$Nice"); 		# put nice onto command

if( ! $WD) { $WD = $ENV{cwd};} 	# we come to present directory (default),
if( ! $WD ) {chop($WD = `echo \$cwd`);}
if( ! $WD ) {chop($WD = `pwd`);}

if ($WD) {
	unshift(ARGV,"cd $WD",';'); # push cd cwd onto command
}

if ( ! $host ) {	# try to grab least loaded...
	open(LIST,"ruptime | awk '{print \$1, \$7}' | sort -n +1 |");
	while (<LIST>) {
		chop;
		($host_try,$load) = split(' ',$_);
		next if $host_try =~ /swan/;		# bad hosts, clean up
		next if $host_try =~ /thrush/;
		$machinefile = "$machinedir"."$host_try";
#
#		print "machinefile $machinefile\n";
#		$wash =  ( -M $machinefile );
#		print "$wash $waittime \n";
#
		print "try $host_try...";		# check if old enough
		if(! -f $machinefile || -M $machinefile > $waittime) {
			$host = $host_try;
			print "OK\n";
			last;
		} else {
			print "nope\n";
		}
		
	}
}

if ( ! $host ) {	# if still no host, select least loaded one I guess..
	chop($host = `ruptime | awk '{print \$1, \$7}' | sort -n +1 | head -1`);
	$load = ''; ($host,$load) = split(' ',$host);
}

if ( ! $host ) {	# if still no host, select present one!
	chop($host = `hostname`);
}

if(! $machinefile) { $machinefile = "$machinedir"."$host"; }

push(ARGV,'&'); # push `&' to end of command

#	Debugging
#	print "\n"; foreach (@ARGV) { print; print "\n"; } print "\n";

#	print "hostname $host\n";
	print "\nWill run:\n";
	$| = 1;
	$ErrorFile = "$ENV{'HOME'}/.errorWad";	# file to direct error...
	system "touch $ErrorFile";		# try to make exist
	if ( ! -f $ErrorFile ) { $ErrorFile = '/dev/null'; }

	print "$RSH $host -n \"( @ARGV ) >>& $ErrorFile \"\n";

	if($show) {				# show command and ask
		print  "\nRun it? ([y]/n) ";
		chop($response = <STDIN>);
		if ( $response ne 'y' && $response ne '' && $response ne 'Y' ) {print "Didn't run it\n"; exit 0;}
		else {print "OK, I run it\n";}
	}
	print "Errors and undirected output appended to $ErrorFile\n";
	system "$RSH $host -n \"( @ARGV ) >>& $ErrorFile \" ";	# run it
	if($machinefile) {
		system "rm -f $machinefile";	# update recent usage data
		system "touch $machinefile";
	}
	print "\n";
	system "(myjobs -nr $host; echo) &";