--- loncom/lond	2001/12/22 21:46:02	1.62
+++ loncom/lond	2002/02/05 18:05:47	1.66
@@ -2,7 +2,7 @@
 # The LearningOnline Network
 # lond "LON Daemon" Server (port "LOND" 5663)
 #
-# $Id: lond,v 1.62 2001/12/22 21:46:02 www Exp $
+# $Id: lond,v 1.66 2002/02/05 18:05:47 www Exp $
 #
 # Copyright Michigan State University Board of Trustees
 #
@@ -46,7 +46,8 @@
 # 11/26,11/27 Gerd Kortemeyer
 # 12/20 Scott Harrison
 # 12/22 Gerd Kortemeyer
-#
+# YEAR=2002
+# 01/20/02,02/05 Gerd Kortemeyer
 ###
 
 # based on "Perl Cookbook" ISBN 1-56592-243-3
@@ -84,6 +85,10 @@ sub catchexception {
     die($error);
 }
 
+sub timeout {
+    &logthis("<font color=ref>CRITICAL: TIME OUT ".$$."</font>");
+    &catchexception('Timeout');
+}
 # -------------------------------- Set signal handlers to record abnormal exits
 
 $SIG{'QUIT'}=\&catchexception;
@@ -187,6 +192,7 @@ sub checkchildren {
     &initnewstatus();
     &logstatus();
     &logthis('Going to check on the children');
+    $docdir=$perlvar{'lonDocRoot'};
     foreach (sort keys %children) {
 	sleep 1;
         unless (kill 'USR1' => $_) {
@@ -194,6 +200,19 @@ sub checkchildren {
             &logstatus($$.' is dead');
         } 
     }
+    sleep 5;
+    foreach (sort keys %children) {
+        unless (-e "$docdir/lon-status/londchld/$_.txt") {
+	    &logthis('Child '.$_.' did not respond');
+            kill 9 => $_;
+   $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
+   $subj="LON: $perlvar{'lonHostID'} killed lond process $_";
+   system("echo 'Killed lond process $_.' |\
+ mailto $emailto -s '$subj' > /dev/null");
+	    $execdir=$perlvar{'lonDaemons'};
+            system("cp $execdir/logs/lond.log $execdir/logs/lond.log.".$_);
+        }
+    }
 }
 
 # --------------------------------------------------------------------- Logging
@@ -212,8 +231,16 @@ sub logthis {
 
 sub logstatus {
     my $docdir=$perlvar{'lonDocRoot'};
+    {
     my $fh=IO::File->new(">>$docdir/lon-status/londstatus.txt");
     print $fh $$."\t".$status."\t".$lastlog."\n";
+    $fh->close();
+    }
+    {
+	my $fh=IO::File->new(">$docdir/lon-status/londchld/$$.txt");
+        print $fh $status."\n".$lastlog."\n".time;
+        $fh->close();
+    }
 }
 
 sub initnewstatus {
@@ -222,6 +249,11 @@ sub initnewstatus {
     my $now=time;
     my $local=localtime($now);
     print $fh "LOND status $local - parent $$\n\n";
+    opendir(DIR,"$docdir/lon-status/londchld");
+    while ($filename=readdir(DIR)) {
+        unlink("$docdir/lon-status/londchld/$filename");
+    }
+    closedir(DIR);
 }
 
 # -------------------------------------------------------------- Status setting
@@ -431,6 +463,7 @@ sub make_new_child {
         # Child can *not* return from this subroutine.
         $SIG{INT} = 'DEFAULT';      # make SIGINT kill us as it did before
         $SIG{USR1}= \&logstatus;
+        $SIG{ALRM}= \&timeout;
         $lastlog='Forked ';
         $status='Forked';
 
@@ -499,6 +532,7 @@ sub make_new_child {
                 chomp($userinput);
                 &status('Processing '.$hostid{$clientip}.': '.$userinput);
                 my $wasenc=0;
+                alarm(120);
 # ------------------------------------------------------------ See if encrypted
 		if ($userinput =~ /^enc/) {
 		  if ($cipher) {
@@ -677,7 +711,7 @@ sub make_new_child {
                                $fpnow.='/'.$fpparts[$i]; 
                                unless (-e $fpnow) {
 				   unless (mkdir($fpnow,0777)) {
-                                      $fperror="error:$!\n";
+                                      $fperror="error:$!";
                                    }
                                }
                            }
@@ -1335,6 +1369,7 @@ sub make_new_child {
                        print $client "unknown_cmd\n";
                    }
 # -------------------------------------------------------------------- complete
+		   alarm(0);
                    &status('Listening to '.$hostid{$clientip});
 	       }
 # --------------------------------------------- client unknown or fishy, refuse