--- loncom/loncron	2000/10/26 22:21:02	1.6
+++ loncom/loncron	2001/02/12 18:53:32	1.16
@@ -9,6 +9,9 @@
 #
 # 7/14,7/15,7/19,7/21,7/22,11/18,
 # 2/8 Gerd Kortemeyer
+# Dec 00 Scott Harrison
+# 12/23 Gerd Kortemeyer
+# 02/12/2001 Scott Harrison
 
 use IO::File;
 use IO::Socket;
@@ -43,7 +46,6 @@ ENDERROUT
 
 # ================================================================ Main Program
 
-
 # ------------------------------------------------------------ Read access.conf
 {
     my $config=IO::File->new("/etc/httpd/conf/access.conf");
@@ -56,6 +58,31 @@ ENDERROUT
     }
 }
 
+# --------------------------------------- Make sure that LON-CAPA is configured
+# I only test for one thing here (lonHostID).  This is just a safeguard.
+if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) {
+   print("Unconfigured machine.\n");
+   $emailto=$perlvar{'lonSysEMail'};
+   $hostname=`/bin/hostname`;
+   chop $hostname;
+   $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
+   $subj="LON: Unconfigured machine $hostname";
+   system("echo 'Unconfigured machine $hostname.' |\
+ mailto $emailto -s '$subj' > /dev/null");
+    exit 1;
+}
+
+# ----------------------------- Make sure this process is running from user=www
+my $wwwid=getpwnam('www');
+if ($wwwid!=$<) {
+   print("User ID mismatch.  This program must be run as user 'www'\n");
+   $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
+   $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
+   system("echo 'User ID mismatch.  loncron must be run as user www.' |\
+ mailto $emailto -s '$subj' > /dev/null");
+   exit 1;
+}
+
 # ------------------------------------------------------------- Read hosts file
 {
     my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
@@ -110,7 +137,10 @@ print $fh (<<ENDHEADERS);
 <ol>
 <li><a href="#configuration">Configuration</a>
 <li><a href="#machine">Machine Information</a>
+<li><a href="#tmp">Temporary Files</a>
+<li><a href="#tokens">Session Tokens</a>
 <li><a href="#httpd">httpd</a>
+<li><a href="#lonsql">lonsql</a>
 <li><a href="#lond">lond</a>
 <li><a href="#lonc">lonc</a>
 <li><a href="#lonnet">lonnet</a>
@@ -182,6 +212,48 @@ while ($line=<DFH>) {
 close (DFH);
 print $fh "</pre>";
 &errout($fh);
+
+# --------------------------------------------------------------- clean out tmp
+print $fh '<hr><a name="tmp"><h2>Temporary Files</h2>';
+$cleaned=0;
+while ($fname=<$perlvar{'lonDaemons'}/tmp/*>) {
+                          my ($dev,$ino,$mode,$nlink,
+                              $uid,$gid,$rdev,$size,
+                              $atime,$mtime,$ctime,
+                              $blksize,$blocks)=stat($fname);
+                          $now=time;
+                          $since=$now-$mtime;
+                          if ($since>$perlvar{'lonExpire'}) {
+                              $cleaned++;
+                              unlink("$fname");
+                          }
+    
+}
+print $fh "Cleaned up ".$cleaned." files.";
+
+# ------------------------------------------------------------ clean out lonIDs
+print $fh '<hr><a name="tokens"><h2>Session Tokens</h2>';
+$cleaned=0;
+$active=0;
+while ($fname=<$perlvar{'lonIDsDir'}/*>) {
+                          my ($dev,$ino,$mode,$nlink,
+                              $uid,$gid,$rdev,$size,
+                              $atime,$mtime,$ctime,
+                              $blksize,$blocks)=stat($fname);
+                          $now=time;
+                          $since=$now-$mtime;
+                          if ($since>$perlvar{'lonExpire'}) {
+                              $cleaned++;
+                              print $fh "Unlinking $fname<br>";
+                              unlink("$fname");
+                          } else {
+                              $active++;
+                          }
+    
+}
+print $fh "<p>Cleaned up ".$cleaned." stale session token(s).";
+print $fh "<h3>$active open session(s)</h3>";
+
 # ----------------------------------------------------------------------- httpd
 
 print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>';
@@ -202,11 +274,13 @@ print $fh "</pre>";
 &errout($fh);
 
 
-# ------------------------------------------------------------------------ lonsql
-
+# ---------------------------------------------------------------------- lonsql
+#
+# Do not run for now
+#
 if ($perlvar{'lonRole'} eq "library" && 1==0) {
 
-    print $fh '<hr><a name="lond"><h2>lonsql</h2><h3>Log</h3><pre>';
+    print $fh '<hr><a name="lonsql"><h2>lonsql</h2><h3>Log</h3><pre>';
     
     if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
 	open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
@@ -235,7 +309,8 @@ if ($perlvar{'lonRole'} eq "library" &&
     } else {
 	$errors++;
 	print $fh "<h3>lonsql not running, trying to start</h3>";
-	system("$perlvar{'lonDaemons'}/lonsql");
+	system(
+ "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors");
 	sleep 120;
 	if (-e $lonsqlfile) {
 	    print $fh "Seems like it started ...<p>";
@@ -249,7 +324,8 @@ if ($perlvar{'lonRole'} eq "library" &&
 		$errors++; $errors++;
 		print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
 		print $fh "Give it one more try ...<p>";
-		system("$perlvar{'lonDaemons'}/lonsql");
+		system(
+ "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors");
 		sleep 120;
 	    }
 	} else {
@@ -303,20 +379,41 @@ 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++;
+      $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);
+	  $restartflag=1;
+      }
+      else {
+      # Possibility #2: there is a live process that is not responding
+      #                 for an unknown reason
+      # Solution: kill parent and children processes, remove .pid and restart
+	  `killall -9 lond`;
+	  unlink($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");
+   system(
+     "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
    sleep 120;
    if (-e $londfile) {
        print $fh "Seems like it started ...<p>";
@@ -330,7 +427,8 @@ if (-e $londfile) {
           $errors++; $errors++;
           print $fh "<h3>lond at pid $londpid not responding</h3>";
           print $fh "Give it one more try ...<p>";
-          system("$perlvar{'lonDaemons'}/lond");
+	  system(
+ "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
           sleep 120;
        }
    } else {
@@ -383,6 +481,7 @@ print $fh "</pre>";
 
 my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
 
+$restartflag=1;
 if (-e $loncfile) {
    my $lfh=IO::File->new("$loncfile");
    my $loncpid=<$lfh>;
@@ -390,14 +489,34 @@ 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++;
+      $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);
+	  $restartflag=1;
+      }
+      else {
+      # Possibility #2: there is a live process that is not responding
+      #                 for an unknown reason
+      # Solution: kill parent and children processes, remove .pid and restart
+	  `killall -9 lonc`;
+	  unlink($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");
+	system(
+ "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
    sleep 120;
    if (-e $loncfile) {
        print $fh "Seems like it started ...<p>";
@@ -411,7 +530,8 @@ if (-e $loncfile) {
           $errors++; $errors++;
           print $fh "<h3>lonc at pid $loncpid not responding</h3>";
           print $fh "Give it one more try ...<p>";
-          system("$perlvar{'lonDaemons'}/lonc");
+ 	  system(
+ "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
           sleep 120;
        }
    } else {
@@ -457,7 +577,7 @@ while ($line=<DFH>) {
 };
 close (DFH);
 }
-print $fh "</pre><h3>Perm Log</h3>";
+print $fh "</pre><h3>Perm Log</h3><pre>";
 
 if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
     open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");