#!/usr/bin/perl
# Copyright 2012  Johns Hopkins University (Author: Daniel Povey).
# Apache 2.0.
use File::Basename;
use Cwd;

# Modifie pour l'usage au LIA, avec mosrun

# queue.pl has the same functionality as run.pl, except that
# it runs the job in question on the queue (Sun GridEngine).
# This version of queue.pl uses the task array functionality
# of the grid engine.  Note: it's different from the queue.pl
# in the s4 and earlier scripts.

$qsub_opts = "";
$sync = 0;

for ($x = 1; $x <= 3; $x++) { # This for-loop is to 
  # allow the JOB=1:n option to be interleaved with the
  # options to qsub.
  while (@ARGV >= 2 && $ARGV[0] =~ m:^-:) {
    $switch = shift @ARGV;
    if ($switch eq "-V") {
      $qsub_opts .= "-V ";
    } else {
      $option = shift @ARGV;
      if ($switch eq "-sync" && $option =~ m/^[yY]/) {
        $sync = 1;
      }
      $qsub_opts .= "$switch $option ";
      if ($switch eq "-pe") { # e.g. -pe smp 5
        $option2 = shift @ARGV;
        $qsub_opts .= "$option2 ";
      }
    }
  }
  if ($ARGV[0] =~ m/^([\w_][\w\d_]*)+=(\d+):(\d+)$/) {
    $jobname = $1;
    $jobstart = $2;
    $jobend = $3;
    shift;
    if ($jobstart > $jobend) {
      die "queue.pl: invalid job range $ARGV[0]";
    }
  } elsif ($ARGV[0] =~ m/^([\w_][\w\d_]*)+=(\d+)$/) { # e.g. JOB=1.
    $jobname = $1;
    $jobstart = $2;
    $jobend = $2;
    shift;
  } elsif ($ARGV[0] =~ m/.+\=.*\:.*$/) {
    print STDERR "Warning: suspicious first argument to queue.pl: $ARGV[0]\n";
  }
}

### pas d'option dans mosrun dans cmd.sh => les options de mosix sont passés en durs dans ce script
### print "---=debug=--> $0 : $jobname : $jobstart : $jobend : $qsub_opts : $option : $option2\n";



if (@ARGV < 2) {
  print STDERR
   "Usage: queue.pl [options to qsub] [JOB=1:n] log-file command-line arguments...\n" .
   "e.g.: queue.pl foo.log echo baz\n" .
   " (which will echo \"baz\", with stdout and stderr directed to foo.log)\n" .
   "or: queue.pl -q all.q\@xyz foo.log echo bar \| sed s/bar/baz/ \n" .
   " (which is an example of using a pipe; you can provide other escaped bash constructs)\n" .
   "or: queue.pl -q all.q\@qyz JOB=1:10 foo.JOB.log echo JOB \n" .
   " (which illustrates the mechanism to submit parallel jobs; note, you can use \n" .
   "  another string other than JOB)\n" .
   "Note: if you pass the \"-sync y\" option to qsub, this script will take note\n" .
   "and change its behavior.  Otherwise it uses qstat to work out when the job finished\n";
  exit 1;
}

$cwd = getcwd();
$logfile = shift @ARGV;


## recuperation du chemin vers les fichiers de log
#print "------=debug=------> $cwd \n";
#print "------=debug=------> $cwd $logfile\n";

#== verif syntaxe de la comande ==#
if (defined $jobname && $logfile !~ m/$jobname/ && $jobend > $jobstart) {
  print STDERR "run.pl: you are trying to run a parallel job but "
    . "you are putting the output into just one log file ($logfile)\n";
  exit(1);
}

# Work out the command; quote escaping is done here.
# Note: the rules for escaping stuff are worked out pretty arbitrarily, based on what we want it to do.  Some things that
# we pass as arguments to queue.pl, such as "|", we want to be  interpreted by bash, so we don't escape them.  Other things,
# such as archive specifiers like 'ark:gunzip -c foo.gz|', we want  to be passed, in quotes, to the Kaldi program.  Our heuristic
# is that stuff with spaces in should be quoted.  This doesn't  always work.

$cmd = "";
foreach $x (@ARGV) { 
  if ($x =~ m/^\S+$/) { $cmd .= $x . " "; } # If string contains no spaces, take
                                            # as-is.
  elsif ($x =~ m:\":) { $cmd .= "'\''$x'\'' "; } # else if no dbl-quotes, use single
  else { $cmd .= "\"$x\" "; }  # else use double.
}

# pattern de definition de la commande a repliquer
#print"------=debug=---------> $cmd\n";

#
# Work out the location of the script file, and open it for writing.
#
$dir = dirname($logfile);
$base = basename($logfile);
$qdir = "$dir/q";
$qdir =~ s:/(log|LOG)/*q:/q:; # If qdir ends in .../log/q, make it just .../q.
$queue_logfile = "$qdir/$base";

#print "---=debug=---> $dir \n";
#print "---=debug=---> $base \n";
#print "---=debug=---> $qdir \n";
#print "---=debug=---> $queue_logfile\n";


if (!-d $dir) { system "mkdir $dir 2>/dev/null"; } # another job may be doing this...
if (!-d $dir) { die "Cannot make the directory $dir\n"; }






# make a directory called "q",
# where we will put the log created by qsub... normally this doesn't contain
# anything interesting, evertyhing goes to $logfile.
if (! -d "$qdir") { 
  system "mkdir $qdir 2>/dev/null";
  sleep(2); 
  #sleep(5); ## This is to fix an issue we encountered in denominator lattice creation,
  ## where if e.g. the exp/tri2b_denlats/log/15/q directory had just been
  ## created and the job immediately ran, it would die with an error because nfs
  ## had not yet synced.  I'm also decreasing the acdirmin and acdirmax in our
  ## NFS settings to something like 5 seconds.
} 

##### if (defined $jobname) { # It's an array job.
#####  $queue_array_opt = "-t $jobstart:$jobend"; 
#####  $logfile =~ s/$jobname/\$SGE_TASK_ID/g; # This variable will get 
#####  # replaced by qsub, in each job, with the job-id.
#####  $cmd =~ s/$jobname/\$SGE_TASK_ID/g; # same for the command...
#####  $queue_logfile =~ s/\.?$jobname//; # the log file in the q/ subdirectory
#####  # is for the queue to put its log, and this doesn't need the task array subscript
#####  # so we remove it.
#####}
# print "---=debug 1 =----> $jobname :  $queue_array_opt :  $logfile :  $cmd :  $queue_logfile\n";

# queue_scriptfile is as $queue_logfile [e.g. dir/q/foo.log] but
# with the suffix .sh.

$queue_scriptfile = $queue_logfile;
($queue_scriptfile =~ s/\.[a-zA-Z]{1,5}$/.sh/) || ($queue_scriptfile .= ".sh");
if ($queue_scriptfile !~ m:^/:) {
  $queue_scriptfile = $cwd . "/" . $queue_scriptfile; # just in case.
}


## BB harmonisation des chemins vers les fichiers
## BB : N fichiers make_mfcc.N.sh
## tentative numero1 : $nb fichies bash et un lanceur bash
#print "---=debug 2=---> $queue_scriptfile\n";
#print "---=debug 3=---> $logfile\n";


## BB : preparation d'un script appelé avec mosrun
my $cmd_lanceur = "$cwd/$qdir/lanceurMosix.sh"  ;
print "$cwd/$qdir/lanceurMosix.sh\n";

open(L, ">$cwd/$qdir/lanceurMosix.sh") or die "unable to open $cwd/$qdir/lanceurMosix_make_mfcc.sh\n";
print L "#!/bin/bash\n";	


for (my $j=$jobstart; $j <= $jobend; $j++){

	$current_queue_scriptfile = $queue_scriptfile;
	$current_logfile	  = $logfile;
	$current_cmd		  = $cmd;
	$current_queue_logfile	  = $queue_logfile;
	$syncfile = "$qdir/done.$$.$j";

	if (defined $jobname){
		$current_queue_scriptfile =~ s/$jobname/$j/g; # This variable will get 
		$current_logfile =~ s/$jobname/$j/g; # This variable will get 
		$current_cmd =~ s/$jobname/$j/g; # same for the command...
		$current_queue_logfile =~ s/\.?$jobname//; # the log file in the q/ subdirectory

	}

	#print "===---debug---===> $current_queue_scriptfile :: $current_logfile :: $current_queue_logfile ::  $syncfile \n";
	system("rm $current_queue_logfile $syncfile 2>/dev/null");
	
	open(Q, ">$current_queue_scriptfile") || die "Failed to write to $current_queue_scriptfile";
	

	# ecriture de la comande dnas le lanceur mosix
	print L "bash $current_queue_scriptfile  \n";


	print Q "#!/bin/bash\n";
	print Q "cd $cwd\n";
	print Q ". EXP_NEW/path.sh\n";
	print Q "( echo '#' Running on \`hostname\`\n";
	print Q "  echo '#' Started at \`date\`\n";
	print Q "  echo -n '# '; cat <<EOF\n";
	print Q "$current_cmd\n"; # this is a way of echoing the command into a comment in the log file,
	print Q "EOF\n"; # without having to escape things like "|" and quote characters.
	print Q ") >$current_logfile\n";
	print Q " ($current_cmd ) 2>>$current_logfile >>$current_logfile\n";
	print Q "ret=\$?\n";
	print Q "echo '#' Finished at \`date\` with status \$ret >>$current_logfile\n";
	if (!defined $jobname) { # not an array job
		print Q "touch $syncfile\n"; # so we know it's done.
	} else {
		print Q "touch $syncfile\n"; # touch a bunch of sync-files.
	}
	print Q "exit \$[\$ret ? 1 : 0]\n"; # avoid status 100 which grid-engine
	#print Q "## submitted with:\n";       # treats specially.
	#print Q "# $qsub_cmd\n";
	if (!close(Q)) { # close was not successful... || die "Could not close script file $shfile";
		die "Failed to close the script file (full disk?)";
	}

	
}

if (!close(L)) { # close was not successful... || die "Could not close script file $shfile";
	die "Failed to close the script file (full disk?)";
}	


#### exit;
# We'll write to the standard input of "qsub" (the file-handle Q),
# the job that we want it to execute.
# Also keep our current PATH around, just in case there was something
# in it that we need (although we also source ./path.sh)
##### $syncfile = "$qdir/done.$$";
##### print "---=debug 3=---> $syncfile\n";
##### system("rm $queue_logfile $syncfile 2>/dev/null");
#
# Write to the script file, and then close it.
#
##### open(Q, ">$queue_scriptfile") || die "Failed to write to $queue_scriptfile";
##### print Q "#!/bin/bash\n";
##### print Q "cd $cwd\n";
##### print Q ". ./path.sh\n";
##### print Q "( echo '#' Running on \`hostname\`\n";
##### print Q "  echo '#' Started at \`date\`\n";
##### print Q "  echo -n '# '; cat <<EOF\n";
##### print Q "$cmd\n"; # this is a way of echoing the command into a comment in the log file,
##### print Q "EOF\n"; # without having to escape things like "|" and quote characters.
##### print Q ") >$logfile\n";
##### print Q " ( $cmd ) 2>>$logfile >>$logfile\n";
##### print Q "ret=\$?\n";
##### print Q "echo '#' Finished at \`date\` with status \$ret >>$logfile\n";
##### if (!defined $jobname) { # not an array job
#####   print Q "touch $syncfile\n"; # so we know it's done.
##### } else {
#####   print Q "touch $syncfile.\$SGE_TASK_ID\n"; # touch a bunch of sync-files.
##### }
##### print Q "exit \$[\$ret ? 1 : 0]\n"; # avoid status 100 which grid-engine
##### print Q "## submitted with:\n";       # treats specially.
##### print Q "# $qsub_cmd\n";
##### if (!close(Q)) { # close was not successful... || die "Could not close script file $shfile";
#####   die "Failed to close the script file (full disk?)";
##### }

#print "system (qsub -S /bin/bash -v PATH -cwd -j y $qsub_opts $queue_array_opt $queue_scriptfile >> $queue_logfile );\n";
#print "system (mosrun -w  -b -m4000 -M $cwd  $cmd_lanceur );\n";
print "mosrun -q -S16 -e -b -m4000 $cmd_lanceur\n";
$ret = system ("mosrun -q -S16 -e  -b -m4000 $cmd_lanceur");
exit ;
$ret = system ("qsub -S /bin/bash -v PATH -cwd -j y -o $queue_logfile $qsub_opts $queue_array_opt $queue_scriptfile >>$queue_logfile 2>&1");

if ($ret != 0) {
  if ($sync && $ret == 256) { # this is the exit status when a job failed (bad exit status)
    if (defined $jobname) { $logfile =~ s/\$SGE_TASK_ID/*/g; }
    print STDERR "queue.pl: job writing to $logfile failed\n";
  } else {
    print STDERR "queue.pl: error submitting jobs to queue (return status was $ret)\n";
    print STDERR `tail $queue_logfile`;
  }
  exit(1);
}

if (! $sync) { # We're not submitting with -sync y, so we
  # need to wait for the jobs to finish.  We wait for the
  # sync-files we "touched" in the script to exist.
  @syncfiles = ();
  if (!defined $jobname) { # not an array job.
    push @syncfiles, $syncfile;
  } else {
    for ($jobid = $jobstart; $jobid <= $jobend; $jobid++) {
      push @syncfiles, "$syncfile.$jobid";
    }
  }
  $wait = 0.1;
  foreach $f (@syncfiles) {
    # wait for them to finish one by one.
    while (! -f $f) {
      sleep($wait);
      $wait *= 1.2;
      if ($wait > 1.0) {
        $wait = 1.0; # never wait more than 1 second.
      }  
    }
  }
  $all_syncfiles = join(" ", @syncfiles);
  system("rm $all_syncfiles 2>/dev/null");
}

# OK, at this point we are synced; we know the job is done.
# But we don't know about its exit status.  We'll look at $logfile for this.
# First work out an array @logfiles of file-locations we need to
# read (just one, unless it's an array job).
@logfiles = ();
if (!defined $jobname) { # not an array job.
  push @logfiles, $logfile;
} else {
  for ($jobid = $jobstart; $jobid <= $jobend; $jobid++) {
    $l = $logfile; 
    $l =~ s/\$SGE_TASK_ID/$jobid/g;
    push @logfiles, $l;
  }
}

$num_failed = 0;
foreach $l (@logfiles) {
  @wait_times = (0.1, 0.2, 0.2, 0.3, 0.5, 0.5, 1.0, 2.0, 5.0, 5.0, 5.0, 10.0, 25.0);
  for ($iter = 0; $iter <= @wait_times; $iter++) {
    $line = `tail -1 $l 2>/dev/null`;
    if ($line =~ m/with status (\d+)/) {
      $status = $1;
      last;
    } else {
      if ($iter < @wait_times) {
        sleep($wait_times[$iter]);
      } else {
        if (! -f $l) {
          print STDERR "Log-file $l does not exist.\n";
        } else {
          print STDERR "The last line of log-file $l does not seem to indicate the "
            . "return status as expected\n";
        }
        exit(1);                # Something went wrong with the queue, or the
        # machine it was running on, probably.
      }
    }
  }
  # OK, now we have $status, which is the return-status of
  # the command in the job.
  if ($status != 0) { $num_failed++; }
}
if ($num_failed == 0) { exit(0); }
else { # we failed.
  if (@logfiles == 1) {
    if (defined $jobname) { $logfile =~ s/\$SGE_TASK_ID/$jobstart/g; }
    print STDERR "queue.pl: job failed with status $status, log is in $logfile\n";
    if ($logfile =~ m/JOB/) {
      print STDERR "queue.pl: probably you forgot to put JOB=1:\$nj in your script.\n";
    }
  } else {
    if (defined $jobname) { $logfile =~ s/\$SGE_TASK_ID/*/g; }
    $numjobs = 1 + $jobend - $jobstart;
    print STDERR "queue.pl: $num_failed / $numjobs failed, log is in $logfile\n";
  }
  exit(1);
}