Changeset 38 for trunk/oarutils/oar-parexec

Timestamp:

Dec 4, 2011, 11:04:19 PM (13 years ago)

Author:

g7moreau

Message:

• Add logfile and restart fonctionality

File:

• trunk/oarutils/oar-parexec (modified) (7 diffs)

trunk/oarutils/oar-parexec

@@ -17 +17 @@
 
 my $file = '';
+my $logfile = '';
 my $verbose;
 my $job_np = 1;
@@ -27 +28 @@
 Getopt::Long::GetOptions(
    'file=s'     => \$file,
+   'logfile=s'  => \$logfile,
    'verbose'    => \$verbose,
    'help'       => \$help,
@@ -38 +40 @@
 pod2usage( -verbose => 2 ) if not -e $file;
 
+my %state;
+my $log_h = IO::File->new();
+if (-e $logfile) {
+        $log_h->open("< $logfile")
+         or die "can't read log file: $!";
+   while (<$log_h>) {
+                $state{$1} = 'start' if m/^start\sjob\s(\d+)/;
+                $state{$1} = 'end'   if m/^end\sjob\s(\d+)/;
+           }
+   $log_h->close();
+   }
+if ($logfile) {
+   $log_h->open(">> $logfile")
+         or die "can't open log file: $!";
+   $log_h = unblock $log_h;
+   }
+
 my @job = ();
 open( JOB_LIST, '<', "$file" ) or die "can't open $file: $!";
@@ -87 +106 @@
 async {
    for my $job (@job) {
+      $job_num++;
+                if (exists $state{$job_num} and $state{$job_num} eq 'end') {
+                        delete $state{$job_num};
+                        cede;
+                        next;
+                   }
+                if (exists $state{$job_num} and $state{$job_num} eq 'start') {
+                        print "warning: job $job_num was not finished, relaunching...\n" if $verbose;
+                   }            
+
       my $job_ressource = $ressources->get;
-
-      $job_num++;
 
       my ($node_connect) = split ',', $job_ressource;
@@ -101 +128 @@
       $scheduled{$job_pid} = { fh => $fh, node_connect => $node_connect, ressource => $job_ressource, num => $job_num };
 
+      $log_h->printf("start job %5i at %s\n", $job_num, time) if $logfile;
       printf "start job %5i / %5i at %s on node %s\n",
          $job_num, $job_pid, time, $job_ressource
@@ -133 +161 @@
       for my $job_pid ( keys %scheduled ) {
          if ( waitpid( $job_pid, WNOHANG ) ) {
+            $log_h->printf("end job %5i at %s\n", $job_num, time) if $logfile;
             printf "end   job %5i / %5i at %s on node %s\n",
                $scheduled{$job_pid}->{num},
@@ -155 +184 @@
 $finished->wait;
 
+$log_h->close if $logfile;
+
 __END__