Changeset 122

Timestamp:

Apr 3, 2017, 5:31:53 PM (7 years ago)

Author:

g7moreau

Message:

• Change log format to version 2

File:

• trunk/oarutils/oar-parexec (modified) (8 diffs)

trunk/oarutils/oar-parexec

@@ -55 +55 @@
 chomp $oar_version;
 
+# global time
+my $global_time_atstart    = time;
+my $global_time_total      = 0;
+my $global_time_cumulative = 0;
+
 # re-run, keep trace of job already done
 my %state;
@@ -62 +67 @@
       or die "error: can't read log file: $!";
    while (<$log_h>) {
+      # log version 1
       $state{$1} = 'start' if m/^start\s+job\s+([^\s]+)\s/;
       $state{$1} = 'end'   if m/^end\s+job\s+([^\s]+)\s/;
+      # log version 2
+      $state{$1} = 'start' if m/^start\s+subjob\s+([^\s]+)\s/;
+      $state{$1} = 'end'   if m/^end\s+subjob\s+([^\s]+)\s/;
+      ($global_time_total, $global_time_cumulative) = ($1, $2) if m/^global-time\s.*total\s+(\d+)\s+cumulative\s+(\d+)/;
       }
    $log_h->close();
@@ -260 +270 @@
       $fh = unblock $fh;
 
-      my $msg = sprintf "start job %${job_name_maxlen}s / %5i at %s oar job %i on node %s\n",
-         $job_name, $job_pid, time, $ENV{OAR_JOB_ID}, $job_ressource;
+      my $begin_at = time;
+      #my $msg = sprintf "start job %${job_name_maxlen}s / %5i at %s oar job %i on node %s\n",
+      my $msg = sprintf "start subjob %${job_name_maxlen}s pid %5i at %s oarjob %i onnode %s\n",
+         $job_name, $job_pid, $begin_at, $ENV{OAR_JOB_ID}, $job_ressource;
       $log_h->print($msg) if $logtrace;
       print($msg) if $verbose;
@@ -279 +291 @@
          name         => $job_name,
          pidfile      => $job_pidfile,
+         begin_at     => $begin_at,
          };
 
@@ -328 +341 @@
             #print "ERREUR0 $job_pid $job_retcode0\n" if $job_retcode0;
 
-            my $msg = sprintf "end   job %${job_name_maxlen}s / %5i at %s oar job %i on node %s\n",
+            # job time
+            my $end_at = time;
+            my $duration = $end_at - $scheduled{$job_pid}->{begin_at};
+            my $global_time_cumulative += $duration;
+
+            #my $msg = sprintf "end   job %${job_name_maxlen}s / %5i at %s oar job %i on node %s\n",
+            my $msg = sprintf "end     subjob %${job_name_maxlen}s pid %5i at %s oarjob %i onnode %s duration %i status %i\n",
                $scheduled{$job_pid}->{name},
-               $job_pid, time, $ENV{OAR_JOB_ID}, $scheduled{$job_pid}->{ressource};
+               $job_pid, $end_at, $ENV{OAR_JOB_ID}, $scheduled{$job_pid}->{ressource},
+               $duration, $job_retcode0;
 
             # Job error
-            $msg =~ s/^end\s+job/error:$job_retcode0 job/
+            $msg =~ s/^end\s+subjob/error   subjob/
                if $job_retcode0 > 0 and $job_retcode0 != 99;
 
             # Job non finish, just suspend if received checkpoint signal
-            $msg =~ s/^end\s+job/suspend job/
+            $msg =~ s/^end\s+subjob/suspend subjob/
                if $sig_transmit and $oar_checkpoint->count() > 0;
 
@@ -364 +384 @@
 $finished->wait;
 
+# global time
+$global_time_total += (time - $global_time_atstart);
+$log_h->print("global-time total $global_time_total cumulative $global_time_cumulative\n") if $logtrace;
+print("global-time total $global_time_total cumulative $global_time_cumulative\n") if $verbose;
+
 # close log trace file
 $log_h->close() if $logtrace;
@@ -696 +721 @@
  error   subjob  1 pid 101468 at 1450482556 oarjob 71725 onnode cl7n001 duration 657 status 0
  suspend subjob  1 pid 101468 at 1450482556 oarjob 71725 onnode cl7n001 duration 657 status 0
+ global-time total 555 cumulative 44444
 
 =item B<Version 1>
@@ -702 +728 @@
  start job 1 / 101468 at 1450482228 oar job 71725 on node cl7n001
  end   job 1 / 101468 at 1450482556 oar job 71725 on node cl7n001
+ end   job 1 / 101468 at 1450482556 oar job 71725 on node cl7n001
+ error:retcode job 1 / 101468 at 1450482556 oar job 71725 on node cl7n00
 
 =back
+
 
 =head1 SEE ALSO