pssh

#!/usr/bin/perl

use English;
use strict;

my $maxParallelCount = 16;
my $waitTime = 0;

my $progname = $PROGRAM_NAME;
$progname =~ s/(.*\/)//;

#print "name $progname\n";

open HOSTFILE, $ENV{PSSH_HOSTS} or die "Cannot open PSSH_HOSTS $ENV{PSSH_HOSTS}";

my $maxCompleteTime = 10;
my %pids = ();
while (my $host = <HOSTFILE>) {
    $host =~ s/\s//g;
    next if ($host =~ /^\#/);
    next if ($host eq '');

    my $userhost = $ENV{PSSH_USER}.'@'.$host;
    my $cmd = '';
    if ($progname eq 'pssh') {
	$cmd = 'ssh '.$userhost." \'@ARGV\'";
    } elsif ($progname eq 'pscp') {
	$cmd = 'scp '.$ARGV[0]." $userhost:".$ARGV[1];
    } elsif ($progname eq 'prsync') {
	$cmd = 'rsync '.$ARGV[0]." ".$ARGV[1]." $userhost:".$ARGV[2];
	print "$cmd\n";
	exit (0);
#	$maxCompleteTime = 60;
    } elsif ($progname eq 'get-runlog') {
        #sbon/sbon-lefthand.eecs.harvard.edu.log
        #tcsh -c "get-runlog log runlog"
	$cmd = "scp $userhost:~/local/sbon/sbon-$host.log $ARGV[0]/";
	#print "$cmd\n";
	#exit (0);
	#$cmd = "scp $userhost:~/local/$ARGV[1] $ARGV[0]/$host.$ARGV[1]";
    }
    if (my $childpid = fork()) {
#	print "childpid = $childpid\n";
	$pids{$childpid} = $host;
    } else {
	printf (STDERR "$cmd\n");
	system ("$cmd");
	
	if ($? == -1) {
	    print "failed to execute: $!\n";
	}
	elsif ($? & 127) {
                       printf (STDERR "child died with signal %d, %s coredump\n",
		       ($? & 127),  ($? & 128) ? 'with' : 'without');
                   }
	elsif ($? == 0) {
	    print "success: $host\n";
	    ;
	} else {
	    printf (STDERR "child $host exited with value %d\n", $? >> 8);
	}
	exit ($?);
    }


    my $live_pids = $maxParallelCount;
    while ($live_pids >= $maxParallelCount) {
	my $printPids = 0;
	if ($waitTime > 10) {
	    $printPids = 1;
	}
	$live_pids = &count_children(\%pids, $printPids);
	#printf (STDERR "live pids = $live_pids\n");

	if ($live_pids >= $maxParallelCount) {
	    printf (STDERR "waiting for $waitTime\n");
	    $waitTime++;
	    sleep ($waitTime);
	} else {
	    $waitTime--;
	    if ($waitTime < 0) {
		$waitTime = 0;
	    }
	}
    }
}



############################################################

sub count_children {
    my ($pid_list, $printEm) = @_;
    my $count = 0;
    open (PS, "ps ax|") or die ("Cannot open ps");
    while (my $line = <PS>) {
#      print "line = $line\n";
	if ($line =~ /^\s*(\d+)\s+/) {
	    my $pid = $1;
#	    print "pid = $pid\n";
	    if (defined($pid_list->{$pid}) &&
		$line !~ /defunct/) {
		if ($printEm) {
		    printf (STDERR "$pids{$pid} ($pid) is running\n");
		}
#		$currentPids{$pid} = $pid_list->{$pid};
		$count++;
	    }
	} else {
	    #warn "line did not match $line\n";
	}
    }
    close PS;
    return $count;
}