Changeset 34 for trunk/oarutils/oar-parexec

Timestamp:

Dec 1, 2011, 11:26:43 PM (13 years ago)

Author:

g7moreau

Message:

• Parallel sub-job !

File:

• trunk/oarutils/oar-parexec (modified) (11 diffs)

trunk/oarutils/oar-parexec

@@ -18 +18 @@
 my $file = '';
 my $verbose;
+my $job_np = 1;
 my $nodefile = $ENV{OAR_NODE_FILE} || '';
 my $masterio;
@@ -29 +30 @@
    'help'       => \$help,
    'oarsh=s'    => \$oarsh,
+   'jobnp=i'    => \$job_np,
    'nodefile=s' => \$nodefile,
    'masterio=s' => \$masterio,
@@ -46 +48 @@
 close JOB_LIST;
 
+my @ressources = ();
+open( NODE_FILE, '<', "$nodefile" )
+   or die "can't open $nodefile: $!";
+while (<NODE_FILE>) {
+   chomp;
+   next if m/^#/;
+   next if m/^\s*$/;
+   push @ressources, $_ ;
+   }
+close NODE_FILE;
+
+my $ressource_size = scalar(@ressources);
+die "not enought ressources jobnp $job_np > ressources $ressource_size" if not $job_np > $ressource_size;
+
+my $current_dir = getcwd();
+
 my $stderr = $ENV{OAR_STDERR} || '';
 $stderr =~ s/\.stderr$//;
@@ -53 +71 @@
 $stdout = $masterio if $masterio;
 
-my $current_dir = getcwd();
 
 my $finished = new Coro::Signal;
@@ -60 +77 @@
 
 my $ressources = new Coro::Channel;
-open( NODE_FILE, '<', "$nodefile" )
-   or die "can't open $nodefile: $!";
-while (<NODE_FILE>) {
-   chomp;
-   next if m/^#/;
-   next if m/^\s*$/;
-   $ressources->put($_);
-   }
-close NODE_FILE;
+for my $slot (1 .. int($ressource_size / $job_np)) {
+   $ressources->put( [ @ressources[(($slot - 1) * $job_np) .. (($slot * $job_np) - 1)] ] );
+   }
+
 
 my $job_num   = 0;
@@ -75 +87 @@
 async {
    for my $job (@job) {
-      my $node = $ressources->get;
+      my $nodes = $ressources->get;
 
       $job_num++;
 
+      my $node_connect = $nodes->[0];
       my $fh      = IO::File->new();
-      my $job_pid = $fh->open("| $oarsh $node >/dev/null 2>&1")
+      my $job_pid = $fh->open("| $oarsh $node_connect >/dev/null 2>&1")
          or die "don't start subjob: $!";
 
@@ -86 +99 @@
       $fh = unblock $fh;
 
-      $scheduled{$job_pid} = { fh => $fh, node => $node, num => $job_num };
+      $scheduled{$job_pid} = { fh => $fh, node => $node_connect, num => $job_num };
 
       printf "start job %5i / %5i on node %s at %s\n",
-         $job_num, $job_pid, $node, time
+         $job_num, $job_pid, $node_connect, time
          if $verbose;
 
@@ -96 +109 @@
       $job_stderr = "2> $stderr-$job_num.stderr" if $stderr ne '' and $switchio;
 
+      my $job_nodefile = "/tmp/oar-parexec-$ENV{USERNAME}-$job_num";
+
+      if ($job_np > 1) {
+         $fh->print("printf \".join('\n',@{$nodes}).\" > $job_nodefile\n");
+         $fh->print("OAR_NODE_FILE=$job_nodefile\n");
+         $fh->print("OAR_NP=$job_np\n");
+         $fh->print("export OAR_NODE_FILE\n");
+         $fh->print("export OAR_NP\n");
+         $fh->print("unset OAR_MSG_NODEFILE\n");
+         }
       $fh->print("cd $current_dir\n");
       $fh->print("$job $job_stdout $job_stderr\n");
+      $fh->print("rm -f $job_nodefile\n") if $job_np > 1;
       $fh->print("exit\n");
       cede;
@@ -136 +160 @@
 =head1 SYNOPSIS
 
- oar-parexec --file filecommand [--verbose]  [--nodefile filenode] [--masterio basefileio] [--switchio] [--oarsh sssh]
+ oar-parexec --file filecommand [--verbose] [--jobnp integer] [--nodefile filenode] [--masterio basefileio] [--switchio] [--oarsh sssh]
  oar-parexec --help
 
@@ -149 +173 @@
 
 Small job will be launch in the same folder as the master job.
+Two environment variable are define for each small job
+only in case of parallel small job (option C<--jobnp> > 1).
+
+ OAR_NODE_FILE - file that list node for parallel computing
+ OAR_NP        - number of processor affected
+
+The file define by OAR_NODE_FILE is created on the node before launching
+the small job in /tmp and will be delete after...
+C<oar-parexec> is a simple script,
+OAR_NODE_FILE will not be deleted in case of crash of the master job.
 
 
@@ -161 +195 @@
 =item B<-v|--verbose>
 
-=item B<-n|nodefile filenode>
+=item B<-j|--jobnp integer>
+
+Number of processor to allocated for each small job.
+1 by default.
+
+=item B<-n|--nodefile filenode>
 
 File name that list all the node to launch job.