Changeset 81 for trunk/oarutils

Timestamp:

Jun 1, 2012, 9:54:12 AM (12 years ago)

Author:

g7moreau

Message:

• New async method for signal transmit

File:

• trunk/oarutils/oar-parexec (modified) (3 diffs)

trunk/oarutils/oar-parexec

@@ -153 +153 @@
       if $verbose;
    $oar_checkpoint->up();
-   kill $sig_checkpoint => keys %scheduled if $sig_transmit;
-   print "warning: transmit signal $sig_checkpoint to: " . join(' ', keys %scheduled) if $sig_transmit and $verbose;
    };
+
+# asynchrone notify job
+async {
+   while () {
+      next if $oar_checkpoint->count() == 0;
+
+      # only notify with transmit flag
+      if (not $sig_transmit) {
+         cede;
+         next;
+         }
+
+      for my $job_pid (keys %scheduled) {
+         my $job_name     = $scheduled{$job_pid}->{name};
+         my $job_pidfile  = $scheduled{$job_pid}->{pidfile};
+         my $node_connect = $scheduled{$job_pid}->{node_connect};
+
+         my $fh = IO::File->new();
+         $fh->open("| $oarsh $node_connect >/dev/null 2>&1")
+            or die "error: can't notify subjob: $!";
+
+         $fh->autoflush;
+         $fh = unblock $fh;
+
+         $fh->print("kill -$sig_checkpoint \$(cat $job_pidfile)");
+         $fh->print("exit\n");
+
+         print "warning: transmit signal $sig_checkpoint"
+            . " to $job_name on $node_connect.\n"
+            if $verbose;
+         
+         close $fh;
+         cede;
+         }
+      }
+   }
 
 # asynchrone start job block
 async {
-        JOB:
+   JOB:
    for my $job (@job) {
       my $job_name = $job->{name};
@@ -193 +227 @@
       $fh = unblock $fh;
 
+      my $msg = sprintf "start job %${job_name_maxlen}s / %5i at %s on node %s\n",
+         $job_name, $job_pid, time, $job_ressource;
+      $log_h->print($msg) if $logtrace;
+      print($msg) if $verbose;
+
+      my ($job_stdout, $job_stderr);
+      $job_stdout = ">  $stdout-$job_name.stdout" if $stdout ne '' and $switchio;
+      $job_stderr = "2> $stderr-$job_name.stderr" if $stderr ne '' and $switchio;
+
+      my $job_nodefile = "/tmp/oar-parexec-$ENV{LOGNAME}-$job_name";
+      my $job_pidfile  = "/tmp/oar-parexec-$ENV{LOGNAME}-$job_name.pid";
+
       $scheduled{$job_pid} = {
          fh           => $fh,
          node_connect => $node_connect,
          ressource    => $job_ressource,
-         name         => $job_name
+         name         => $job_name,
+         pidfile      => $job_pidfile,
          };
 
-      my $msg = sprintf "start job %${job_name_maxlen}s / %5i at %s on node %s\n",
-         $job_name, $job_pid, time, $job_ressource;
-      $log_h->print($msg) if $logtrace;
-      print($msg) if $verbose;
-
-      my ($job_stdout, $job_stderr);
-      $job_stdout = ">  $stdout-$job_name.stdout" if $stdout ne '' and $switchio;
-      $job_stderr = "2> $stderr-$job_name.stderr" if $stderr ne '' and $switchio;
-
-      my $job_nodefile = "/tmp/oar-parexec-$ENV{LOGNAME}-$job_name";
-
-     # set job environment, run it and clean
+      # set job environment, run it and clean
       if ($job_np > 1) {
          $fh->print("printf \""
@@ -223 +259 @@
          }
       $fh->print("cd $current_dir\n");
-      $fh->print("$job_cmd $job_stdout $job_stderr\n");
+      if ($sig_transmit) {
+         $fh->print("trap 'kill -$sig_checkpoint \$(jobs -p)' $sig_checkpoint\n");
+         $fh->print("echo \$\$ > $job_pidfile\n");
+         $fh->print("$job_cmd $job_stdout $job_stderr &\n");
+         $fh->print("while [ \$(jobs -p | wc -l) -gt 0 ]\n");
+         $fh->print("do\n");
+         $fh->print("   wait\n");
+         $fh->print("done\n");
+         $fh->print("rm -f $job_pidfile\n");
+         }
+      else {
+         $fh->print("$job_cmd $job_stdout $job_stderr\n");
+         }
       $fh->print("rm -f $job_nodefile\n") if $job_np > 1;
       $fh->print("exit\n");