Changeset 40

Timestamp:

Dec 5, 2011, 11:19:27 AM (13 years ago)

Author:

g7moreau

Message:

• Small bug in logfile implementation

File:

• trunk/oarutils/oar-parexec (modified) (6 diffs)

trunk/oarutils/oar-parexec

@@ -43 +43 @@
 my $log_h = IO::File->new();
 if (-e $logfile) {
-        $log_h->open("< $logfile")
-         or die "can't read log file: $!";
+   $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+)/;
+                $state{$1} = 'start' if m/^start\s+job\s+(\d+)\s/;
+                $state{$1} = 'end'   if m/^end\s+job\s+(\d+)\s/;
            }
    $log_h->close();
@@ -53 +53 @@
 if ($logfile) {
    $log_h->open(">> $logfile")
-         or die "can't open log file: $!";
+         or die "can't append log file $logfile: $!";
+   $log_h->autoflush;
    $log_h = unblock $log_h;
    }
 
 my @job = ();
-open( JOB_LIST, '<', "$file" ) or die "can't open $file: $!";
+open( JOB_LIST, '<', "$file" ) or die "can't open job file $file: $!";
 while (<JOB_LIST>) {
    chomp;
@@ -110 +111 @@
    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;
-                   }            
+
+      if (exists $state{$job_num}) {
+         if ($state{$job_num} eq 'start') {
+            print "warning: job $job_num was not finished, relaunching...\n" if $verbose;
+            }
+         elsif ($state{$job_num} eq 'end') {
+            delete $state{$job_num};
+            $job_todo->down;
+            print "warning: job $job_num already done\n" if $verbose;
+            cede;
+            next;
+            }
+         }              
 
       my $job_ressource = $ressources->get;
 
-                last if $oar_checkpoint->count > 0;
+      last if $oar_checkpoint->count() > 0;
 
       my ($node_connect) = split ',', $job_ressource;
@@ -166 +172 @@
       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;
+            $log_h->printf("end   job %5i at %s\n",
+               $scheduled{$job_pid}->{num},
+               time) if $logfile;
             printf "end   job %5i / %5i at %s on node %s\n",
                $scheduled{$job_pid}->{num},
@@ -180 +188 @@
          }
 
-                $finished->send  if $oar_checkpoint->count > 0 and keys(%scheduled) == 0;
+      $finished->send if $oar_checkpoint->count > 0 and keys(%scheduled) == 0;
 
       $finished->send if $job_todo->count == 0;
@@ -191 +199 @@
 $finished->wait;
 
-$log_h->close if $logfile;
+$log_h->close() if $logfile;
+
 
 __END__