--- loncom/loncron	2000/10/26 22:21:02	1.6
+++ loncom/loncron	2000/12/06 21:11:35	1.7
@@ -303,17 +303,35 @@ print $fh "</pre>";
 
 my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
 
-if (-e $londfile) {
+my $restartflag=1;
+if (-e $londfile) {    
    my $lfh=IO::File->new("$londfile");
    my $londpid=<$lfh>;
    chomp($londpid);
    if (kill 0 => $londpid) {
       print $fh "<h3>lond at pid $londpid responding</h3>";
+      $restartflag=0;
    } else {
       $errors++; $errors++;
       print $fh "<h3>lond at pid $londpid not responding</h3>";
+      # Intelligently handle this.
+      # Possibility #1: there is no process
+      # Solution: remove .pid file and restart
+      if (getpgrp($londpid)==-1) {
+	  unlink($londfile);
+	  $restartflag=1;
+      }
+      # Possibility #2: there is a zombie process
+      # Possibility #3: there is a live process that is not responding
+      #                 for an unknown reason
+      # Solution: kill process, remove .pid and restart
+      kill 2 => $londpid;
+      unlink($londfile);
+      $restartflag=1;
+      print $fh "<h3>Deciding to clean up stale .pid file and restart lond</h3>";
    }
-} else {
+} 
+if ($restartflag==1) {
    $errors++;
    print $fh "<h3>lond not running, trying to start</h3>";
    system("$perlvar{'lonDaemons'}/lond");
@@ -383,6 +401,7 @@ print $fh "</pre>";
 
 my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
 
+$restartflag=1;
 if (-e $loncfile) {
    my $lfh=IO::File->new("$loncfile");
    my $loncpid=<$lfh>;
@@ -390,11 +409,28 @@ if (-e $loncfile) {
    if (kill 0 => $loncpid) {
       print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";
       kill USR1 => $loncpid;
+      $restartflag=0;
    } else {
       $errors++; $errors++;
       print $fh "<h3>lonc at pid $loncpid not responding</h3>";
+      # Intelligently handle this.
+      # Possibility #1: there is no process
+      # Solution: remove .pid file and restart
+      if (getpgrp($loncpid)==-1) {
+	  unlink($loncfile);
+	  $restartflag=1;
+      }
+      # Possibility #2: there is a zombie process
+      # Possibility #3: there is a live process that is not responding
+      #                 for an unknown reason
+      # Solution: kill process, remove .pid and restart
+      kill 2 => $loncpid;
+      unlink($loncfile);
+      $restartflag=1;
+      print $fh "<h3>Deciding to clean up stale .pid file and restart lonc</h3>";
    }
-} else {
+} 
+if ($restartflag==1) {
    $errors++;
    print $fh "<h3>lonc not running, trying to start</h3>";
    system("$perlvar{'lonDaemons'}/lonc");