Diff for /loncom/loncron between versions 1.13 and 1.46

version 1.13, 2001/01/10 17:39:36 version 1.46, 2004/05/11 19:12:50
Line 9 Line 9
 #  #
 # 7/14,7/15,7/19,7/21,7/22,11/18,  # 7/14,7/15,7/19,7/21,7/22,11/18,
 # 2/8 Gerd Kortemeyer  # 2/8 Gerd Kortemeyer
 # Dec 00 Scott Harrison  
 # 12/23 Gerd Kortemeyer  # 12/23 Gerd Kortemeyer
   # YEAR=2001
   # 09/04,09/06,11/26 Gerd Kortemeyer
   
   $|=1;
   
   use lib '/home/httpd/lib/perl/';
   use LONCAPA::Configuration;
   
 use IO::File;  use IO::File;
 use IO::Socket;  use IO::Socket;
   use strict;
   #globals
   use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount);
   
   my $statusdir="/home/httpd/html/lon-status";
   
   
 # -------------------------------------------------- Non-critical communication  # -------------------------------------------------- Non-critical communication
 sub reply {  sub reply {
Line 32  sub reply { Line 44  sub reply {
   
 # --------------------------------------------------------- Output error status  # --------------------------------------------------------- Output error status
   
   sub log {
       my $fh=shift;
       if ($fh) { print $fh @_  }
   }
   
 sub errout {  sub errout {
    my $fh=shift;     my $fh=shift;
    print $fh (<<ENDERROUT);     &log($fh,(<<ENDERROUT));
      <p><table border=2 bgcolor="#CCCCCC">       <p><table border=2 bgcolor="#CCCCCC">
      <tr><td>Notices</td><td>$notices</td></tr>       <tr><td>Notices</td><td>$notices</td></tr>
      <tr><td>Warnings</td><td>$warnings</td></tr>       <tr><td>Warnings</td><td>$warnings</td></tr>
Line 43  sub errout { Line 60  sub errout {
 ENDERROUT  ENDERROUT
 }  }
   
 # ================================================================ Main Program  sub start_daemon {
       my ($fh,$daemon,$pidfile) = @_;
 # ------------------------------------------------------------ Read access.conf      my $progname=$daemon;
 {      if ($daemon eq 'lonc' && $ARGV[0] eq 'new') {
     my $config=IO::File->new("/etc/httpd/conf/access.conf");   $progname='loncnew'; 
    print "new ";
     while (my $configline=<$config>) {  
         if ($configline =~ /PerlSetVar/) {  
    my ($dummy,$varname,$varvalue)=split(/\s+/,$configline);  
            $perlvar{$varname}=$varvalue;  
         }  
     }      }
       system("$perlvar{'lonDaemons'}/$progname 2>>$perlvar{'lonDaemons'}/logs/${daemon}_errors");
       sleep 2;
       if (-e $pidfile) {
    &log($fh,"Seems like it started ...<p>");
    my $lfh=IO::File->new("$pidfile");
    my $daemonpid=<$lfh>;
    chomp($daemonpid);
    sleep 2;
    if (kill 0 => $daemonpid) {
       return 1;
    } else {
       return 0;
    }
       }
       &log($fh,"Seems like that did not work!<p>");
       $errors++;
       return 0;
 }  }
   
 # ----------------------------- Make sure this process is running from user=www  sub checkon_daemon {
 my $wwwid=getpwnam('www');      my ($fh,$daemon,$maxsize,$sendusr1)=@_;
 if ($wwwid!=$<) {  
     print("User ID mismatch.  This program must be run as user 'www'\n") unless $noprint;  
    $emailto="$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      &log($fh,'<hr><a name="'.$daemon.'"><h2>'.$daemon.'</h2><h3>Log</h3><pre>');
 {      printf("%-10s ",$daemon);
     my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");      if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
    open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
    while (my $line=<DFH>) { 
       &log($fh,"$line");
       if ($line=~/INFO/) { $notices++; }
       if ($line=~/WARNING/) { $notices++; }
       if ($line=~/CRITICAL/) { $warnings++; }
    };
    close (DFH);
       }
       &log($fh,"</pre>");
       
       my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid";
       
       my $restartflag=1;
       my $daemonpid;
       if (-e $pidfile) {
    my $lfh=IO::File->new("$pidfile");
    $daemonpid=<$lfh>;
    chomp($daemonpid);
    if (kill 0 => $daemonpid) {
       &log($fh,"<h3>$daemon at pid $daemonpid responding");
       if ($sendusr1) { &log($fh,", sending USR1"); }
       &log($fh,"</h3>");
       if ($sendusr1) { kill USR1 => $daemonpid; }
       $restartflag=0;
       print "running\n";
    } else {
       $errors++;
       &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
       $restartflag=1;
       &log($fh,"<h3>Decided to clean up stale .pid file and restart $daemon</h3>");
    }
       }
       if ($restartflag==1) {
    $simplestatus{$daemon}='off';
    $errors++;
    &log($fh,'<br><font color="red">Killall '.$daemon.': '.
       `killall $daemon 2>&1`.' - ');
    sleep 2;
    &log($fh,unlink($pidfile).' - '.
       `killall -9 $daemon 2>&1`.
       '</font><br>');
    &log($fh,"<h3>$daemon not running, trying to start</h3>");
   
    if (&start_daemon($fh,$daemon,$pidfile)) {
       &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
       $simplestatus{$daemon}='restarted';
       print "started\n";
    } else {
       $errors++;
       &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
       &log($fh,"Give it one more try ...<p>");
       print " ";
       if (&start_daemon($fh,$daemon,$pidfile)) {
    &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
    $simplestatus{$daemon}='restarted';
    print "started\n";
       } else {
    print " failed\n";
    $simplestatus{$daemon}='failed';
    $errors++; $errors++;
    &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
    &log($fh,"Unable to start $daemon<p>");
       }
    }
   
     while (my $configline=<$config>) {   if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
        my ($id,$domain,$role,$name,$ip)=split(/:/,$configline);      &log($fh,"<p><pre>");
        $hostname{$id}=$name;      open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|");
        $hostdom{$id}=$domain;      while (my $line=<DFH>) { 
        $hostrole{$id}=$role;   &log($fh,"$line");
        $hostip{$id}=$ip;   if ($line=~/WARNING/) { $notices++; }
        if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) {   if ($line=~/CRITICAL/) { $notices++; }
    $libserv{$id}=$name;      };
        }      close (DFH);
       &log($fh,"</pre>");
    }
       }
       
       my $fname="$perlvar{'lonDaemons'}/logs/$daemon.log";
       
       my ($dev,$ino,$mode,$nlink,
    $uid,$gid,$rdev,$size,
    $atime,$mtime,$ctime,
    $blksize,$blocks)=stat($fname);
       
       if ($size>$maxsize) {
    &log($fh,"Rotating logs ...<p>");
    rename("$fname.2","$fname.3");
    rename("$fname.1","$fname.2");
    rename("$fname","$fname.1");
     }      }
   
       &errout($fh);
 }  }
   
 # ------------------------------------------------------ Read spare server file  # --------------------------------------------------------------------- Machine
 {  sub log_machine_info {
     my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");      my ($fh)=@_;
       &log($fh,'<hr><a name="machine"><h2>Machine Information</h2>');
       &log($fh,"<h3>loadavg</h3>");
   
       open (LOADAVGH,"/proc/loadavg");
       my $loadavg=<LOADAVGH>;
       close (LOADAVGH);
       
       &log($fh,"<tt>$loadavg</tt>");
       
       my @parts=split(/\s+/,$loadavg);
       if ($parts[1]>4.0) {
    $errors++;
       } elsif ($parts[1]>2.0) {
    $warnings++;
       } elsif ($parts[1]>1.0) {
    $notices++;
       }
   
     while (my $configline=<$config>) {      &log($fh,"<h3>df</h3>");
        chomp($configline);      &log($fh,"<pre>");
        if (($configline) && ($configline ne $perlvar{'lonHostID'})) {  
           $spareid{$configline}=1;      open (DFH,"df|");
        }      while (my $line=<DFH>) { 
    &log($fh,"$line"); 
    @parts=split(/\s+/,$line);
    my $usage=$parts[4];
    $usage=~s/\W//g;
    if ($usage>90) { 
       $warnings++;
       $notices++; 
    } elsif ($usage>80) {
       $warnings++;
    } elsif ($usage>60) {
       $notices++;
    }
    if ($usage>95) { $warnings++; $warnings++; $simplestatus{'diskfull'}++; }
     }      }
 }      close (DFH);
       &log($fh,"</pre>");
   
 # ---------------------------------------------------------------- Start report  
   
 $statusdir="/home/httpd/html/lon-status";      &log($fh,"<h3>ps</h3>");
       &log($fh,"<pre>");
       my $psproc=0;
   
 $errors=0;      open (PSH,"ps -aux --cols 140 |");
 $warnings=0;      while (my $line=<PSH>) { 
 $notices=0;   &log($fh,"$line"); 
    $psproc++;
       }
       close (PSH);
       &log($fh,"</pre>");
   
       if ($psproc>200) { $notices++; }
       if ($psproc>250) { $notices++; }
   
 $now=time;      &errout($fh);
 $date=localtime($now);  }
   
 {  sub start_logging {
 my $fh=IO::File->new(">$statusdir/newstatus.html");      my ($hostdom,$hostrole,$hostname,$spareid)=@_;
       my $fh=IO::File->new(">$statusdir/newstatus.html");
       my %simplestatus=();
       my $now=time;
       my $date=localtime($now);
       
   
 print $fh (<<ENDHEADERS);      &log($fh,(<<ENDHEADERS));
 <html>  <html>
 <head>  <head>
 <title>LON Status Report $perlvar{'lonHostID'}</title>  <title>LON Status Report $perlvar{'lonHostID'}</title>
Line 128  print $fh (<<ENDHEADERS); Line 276  print $fh (<<ENDHEADERS);
 <li><a href="#lonsql">lonsql</a>  <li><a href="#lonsql">lonsql</a>
 <li><a href="#lond">lond</a>  <li><a href="#lond">lond</a>
 <li><a href="#lonc">lonc</a>  <li><a href="#lonc">lonc</a>
   <li><a href="#lonhttpd">lonhttpd</a>
 <li><a href="#lonnet">lonnet</a>  <li><a href="#lonnet">lonnet</a>
 <li><a href="#connections">Connections</a>  <li><a href="#connections">Connections</a>
 <li><a href="#delayed">Delayed Messages</a>  <li><a href="#delayed">Delayed Messages</a>
 <li><a href="#errcount">Error Count</a>  <li><a href="#errcount">Error Coindex.html.unt</a>
 </ol>  </ol>
 <hr>  <hr>
 <a name="configuration">  <a name="configuration">
Line 140  print $fh (<<ENDHEADERS); Line 289  print $fh (<<ENDHEADERS);
 <table border=2>  <table border=2>
 ENDHEADERS  ENDHEADERS
   
 foreach $varname (keys %perlvar) {      foreach my $varname (sort(keys(%perlvar))) {
     print $fh "<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n";   &log($fh,"<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n");
 }      }
 print $fh "</table><h3>Hosts</h3><table border=2>";      &log($fh,"</table><h3>Hosts</h3><table border=2>");
 foreach $id (keys %hostname) {      foreach my $id (sort(keys(%{$hostname}))) {
 print $fh    &log($fh,
     "<tr><td>$id</td><td>$hostdom{$id}</td><td>$hostrole{$id}</td>";      "<tr><td>$id</td><td>".$hostdom->{$id}.
 print $fh "<td>$hostname{$id}</td><td>$hostip{$id}</td></tr>\n";      "</td><td>".$hostrole->{$id}.
 }      "</td><td>".$hostname->{$id}."</td></tr>\n");
 print $fh "</table><h3>Spare Hosts</h3><ol>";      }
 foreach $id (keys %spareid) {      &log($fh,"</table><h3>Spare Hosts</h3><ol>");
     print $fh "<li>$id\n";      foreach my $id (sort(keys(%{$spareid}))) {
    &log($fh,"<li>$id\n");
       }
       &log($fh,"</ol>\n");
       return $fh;
 }  }
   
 print $fh "</ol>\n";  
   
 # --------------------------------------------------------------------- Machine  
   
 print $fh '<hr><a name="machine"><h2>Machine Information</h2>';  
 print $fh "<h3>loadavg</h3>";  
   
 open (LOADAVGH,"/proc/loadavg");  
 $loadavg=<LOADAVGH>;  
 close (LOADAVGH);  
   
 print $fh "<tt>$loadavg</tt>";  
   
 @parts=split(/\s+/,$loadavg);  
 if ($parts[1]>4.0) {  
     $errors++;  
 } elsif ($parts[1]>2.0) {  
     $warnings++;  
 } elsif ($parts[1]>1.0) {  
     $notices++;  
 }  
   
 print $fh "<h3>df</h3>";  
 print $fh "<pre>";  
   
 open (DFH,"df|");  
 while ($line=<DFH>) {   
    print $fh "$line";   
    @parts=split(/\s+/,$line);  
    $usage=$parts[4];  
    $usage=~s/\W//g;  
    if ($usage>90) {   
       $warnings++;   
    } elsif ($usage>80) {  
       $warnings++;  
    } elsif ($usage>60) {  
       $notices++;  
    }  
    if ($usage>95) { $warnings++; $warnings++ }  
 }  
 close (DFH);  
 print $fh "</pre>";  
 &errout($fh);  
   
 # --------------------------------------------------------------- clean out tmp  # --------------------------------------------------------------- clean out tmp
 print $fh '<hr><a name="tmp"><h2>Temporary Files</h2>';  sub clean_tmp {
 $cleaned=0;      my ($fh)=@_;
 while ($fname=<$perlvar{'lonDaemons'}/tmp/*>) {      &log($fh,'<hr><a name="tmp"><h2>Temporary Files</h2>');
                           my ($dev,$ino,$mode,$nlink,      my $cleaned=0;
                               $uid,$gid,$rdev,$size,      my $old=0;
                               $atime,$mtime,$ctime,      while (my $fname=<$perlvar{'lonDaemons'}/tmp/*>) {
                               $blksize,$blocks)=stat($fname);   my ($dev,$ino,$mode,$nlink,
                           $now=time;      $uid,$gid,$rdev,$size,
                           $since=$now-$mtime;      $atime,$mtime,$ctime,
                           if ($since>$perlvar{'lonExpire'}) {      $blksize,$blocks)=stat($fname);
                               $cleaned++;   my $now=time;
                               unlink("$fname");   my $since=$now-$mtime;
                           }   if ($since>$perlvar{'lonExpire'}) {
           my $line='';
       if (open(PROBE,$fname)) {
    $line=<PROBE>;
    close(PROBE);
       }
       unless ($line=~/^CHECKOUTTOKEN\&/) {
    $cleaned++;
    unlink("$fname");
       } else {
    if ($since>365*$perlvar{'lonExpire'}) {
       $cleaned++;
       unlink("$fname");
    } else { $old++; }
       }
    }
       }
       &log($fh,"Cleaned up ".$cleaned." files (".$old." old checkout tokens).");
 }  }
 print $fh "Cleaned up ".$cleaned." files.";  
   
 # ------------------------------------------------------------ clean out lonIDs  # ------------------------------------------------------------ clean out lonIDs
 print $fh '<hr><a name="tokens"><h2>Session Tokens</h2>';  sub clean_lonIDs {
 $cleaned=0;      my ($fh)=@_;
 $active=0;      &log($fh,'<hr><a name="tokens"><h2>Session Tokens</h2>');
 while ($fname=<$perlvar{'lonIDsDir'}/*>) {      my $cleaned=0;
                           my ($dev,$ino,$mode,$nlink,      my $active=0;
                               $uid,$gid,$rdev,$size,      while (my $fname=<$perlvar{'lonIDsDir'}/*>) {
                               $atime,$mtime,$ctime,   my ($dev,$ino,$mode,$nlink,
                               $blksize,$blocks)=stat($fname);      $uid,$gid,$rdev,$size,
                           $now=time;      $atime,$mtime,$ctime,
                           $since=$now-$mtime;      $blksize,$blocks)=stat($fname);
                           if ($since>$perlvar{'lonExpire'}) {   my $now=time;
                               $cleaned++;   my $since=$now-$mtime;
                               print $fh "Unlinking $fname<br>";   if ($since>$perlvar{'lonExpire'}) {
                               unlink("$fname");      $cleaned++;
                           } else {      &log($fh,"Unlinking $fname<br>");
                               $active++;      unlink("$fname");
                           }   } else {
           $active++;
    }
       }
       &log($fh,"<p>Cleaned up ".$cleaned." stale session token(s).");
       &log($fh,"<h3>$active open session(s)</h3>");
 }  }
 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>';  
   
 open (DFH,"tail -n40 /etc/httpd/logs/access_log|");  
 while ($line=<DFH>) { print $fh "$line" };  
 close (DFH);  
   
 print $fh "</pre><h3>Error Log</h3><pre>";  
   
 open (DFH,"tail -n50 /etc/httpd/logs/error_log|");  
 while ($line=<DFH>) {   
    print $fh "$line";  
    if ($line=~/\[error\]/) { $notices++; }   
 };  
 close (DFH);  
 print $fh "</pre>";  
 &errout($fh);  
   
   # ----------------------------------------------------------------------- httpd
   sub check_httpd_logs {
       my ($fh)=@_;
       &log($fh,'<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>');
       
       open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
       while (my $line=<DFH>) { &log($fh,"$line") };
       close (DFH);
   
       &log($fh,"</pre><h3>Error Log</h3><pre>");
   
       open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
       while (my $line=<DFH>) { 
    &log($fh,"$line");
    if ($line=~/\[error\]/) { $notices++; } 
       }
       close (DFH);
       &log($fh,"</pre>");
       &errout($fh);
   }
   
 # ---------------------------------------------------------------------- lonsql  # ---------------------------------------------------------------------- lonnet
 #  
 # Do not run for now  
 #  
 if ($perlvar{'lonRole'} eq "library" && 1==0) {  
   
     print $fh '<hr><a name="lonsql"><h2>lonsql</h2><h3>Log</h3><pre>';  sub rotate_logs {
           my ($fh)=@_;
     if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){      &log($fh,'<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>');
  open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");      print "checking logs\n";
  while ($line=<DFH>) {       if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
     print $fh "$line";   open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
     if ($line=~/INFO/) { $notices++; }   while (my $line=<DFH>) { 
     if ($line=~/WARNING/) { $notices++; }      &log($fh,"$line");
     if ($line=~/CRITICAL/) { $warnings++; }   }
  };  
  close (DFH);   close (DFH);
     }      }
     print $fh "</pre>";      &log($fh,"</pre><h3>Perm Log</h3><pre>");
           
     my $lonsqlfile="$perlvar{'lonDaemons'}/logs/lonsql.pid";      if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
        open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
     if (-e $lonsqlfile) {   while (my $line=<DFH>) { 
  my $lfh=IO::File->new("$lonsqlfile");      &log($fh,"$line");
  my $lonsqlpid=<$lfh>;  
  chomp($lonsqlpid);  
  if (kill 0 => $lonsqlpid) {  
     print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";  
  } else {  
     $errors++; $errors++;  
     print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";  
  }  
     } else {  
  $errors++;  
  print $fh "<h3>lonsql not running, trying to start</h3>";  
  system("$perlvar{'lonDaemons'}/lonsql");  
  sleep 120;  
  if (-e $lonsqlfile) {  
     print $fh "Seems like it started ...<p>";  
     my $lfh=IO::File->new("$lonsqlfile");  
     my $lonsqlpid=<$lfh>;  
     chomp($lonsqlpid);  
     sleep 30;  
     if (kill 0 => $lonsqlpid) {  
  print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";  
     } else {  
  $errors++; $errors++;  
  print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";  
  print $fh "Give it one more try ...<p>";  
  system("$perlvar{'lonDaemons'}/lonsql");  
  sleep 120;  
     }  
  } else {  
     print $fh "Seems like that did not work!<p>";  
     $errors++;  
  }   }
  if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){   close (DFH);
     print $fh "<p><pre>";      } else { &log($fh,"No perm log\n") }
     open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");  
     while ($line=<DFH>) {   
  print $fh "$line";  
  if ($line=~/WARNING/) { $notices++; }  
  if ($line=~/CRITICAL/) { $notices++; }  
     };  
     close (DFH);  
     print $fh "</pre>";  
  }  
     }  
   
     $fname="$perlvar{'lonDaemons'}/logs/lonsql.log";      my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
   
     my ($dev,$ino,$mode,$nlink,      my ($dev,$ino,$mode,$nlink,
  $uid,$gid,$rdev,$size,   $uid,$gid,$rdev,$size,
Line 336  if ($perlvar{'lonRole'} eq "library" && Line 418  if ($perlvar{'lonRole'} eq "library" &&
  $blksize,$blocks)=stat($fname);   $blksize,$blocks)=stat($fname);
   
     if ($size>40000) {      if ($size>40000) {
  print $fh "Rotating logs ...<p>";   &log($fh,"Rotating logs ...<p>");
  rename("$fname.2","$fname.3");   rename("$fname.2","$fname.3");
  rename("$fname.1","$fname.2");   rename("$fname.1","$fname.2");
  rename("$fname","$fname.1");   rename("$fname","$fname.1");
     }      }
   
       &log($fh,"</pre>");
     &errout($fh);      &errout($fh);
 }  }
 # ------------------------------------------------------------------------ lond  
   
 print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>';  # ----------------------------------------------------------------- Connections
   sub test_connections {
       my ($fh,$hostname)=@_;
       &log($fh,'<hr><a name="connections"><h2>Connections</h2>');
       print "testing connections\n";
       &log($fh,"<table border=2>");
       foreach my $tryserver (sort(keys(%{$hostname}))) {
    print(".");
    my $result;
    my $answer=reply("pong",$tryserver);
    if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
       $result="<b>ok</b>";
    } else {
       $result=$answer;
       $warnings++;
       if ($answer eq 'con_lost') { $warnings++; }
    }
    if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }
    &log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n");
       }
       &log($fh,"</table>");
   
 if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){      &errout($fh);
 open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");  
 while ($line=<DFH>) {   
    print $fh "$line";  
    if ($line=~/INFO/) { $notices++; }  
    if ($line=~/WARNING/) { $notices++; }  
    if ($line=~/CRITICAL/) { $warnings++; }  
 };  
 close (DFH);  
 }  
 print $fh "</pre>";  
   
 my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";  
   
 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++;  
       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 $fh   
   "<h3>Deciding to clean up stale .pid file and restart lond</h3>";  
    }  
 }   
 if ($restartflag==1) {  
    $errors++;  
    print $fh "<h3>lond not running, trying to start</h3>";  
    system("$perlvar{'lonDaemons'}/lond");  
    sleep 120;  
    if (-e $londfile) {  
        print $fh "Seems like it started ...<p>";  
        my $lfh=IO::File->new("$londfile");  
        my $londpid=<$lfh>;  
        chomp($londpid);  
        sleep 30;  
        if (kill 0 => $londpid) {  
           print $fh "<h3>lond at pid $londpid responding</h3>";  
        } else {  
           $errors++; $errors++;  
           print $fh "<h3>lond at pid $londpid not responding</h3>";  
           print $fh "Give it one more try ...<p>";  
           system("$perlvar{'lonDaemons'}/lond");  
           sleep 120;  
        }  
    } else {  
        print $fh "Seems like that did not work!<p>";  
        $errors++;  
    }  
    if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){  
     print $fh "<p><pre>";  
     open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");  
     while ($line=<DFH>) {   
       print $fh "$line";  
       if ($line=~/WARNING/) { $notices++; }  
       if ($line=~/CRITICAL/) { $notices++; }  
     };  
     close (DFH);  
     print $fh "</pre>";  
    }  
 }  }
   
 $fname="$perlvar{'lonDaemons'}/logs/lond.log";  
   
                           my ($dev,$ino,$mode,$nlink,  # ------------------------------------------------------------ Delayed messages
                               $uid,$gid,$rdev,$size,  sub check_delayed_msg {
                               $atime,$mtime,$ctime,      my ($fh)=@_;
                               $blksize,$blocks)=stat($fname);      &log($fh,'<hr><a name="delayed"><h2>Delayed Messages</h2>');
       print "checking buffers\n";
 if ($size>40000) {      
     print $fh "Rotating logs ...<p>";      &log($fh,'<h3>Scanning Permanent Log</h3>');
     rename("$fname.2","$fname.3");  
     rename("$fname.1","$fname.2");  
     rename("$fname","$fname.1");  
 }  
   
 &errout($fh);  
 # ------------------------------------------------------------------------ lonc  
   
 print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>';  
   
 if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){  
 open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");  
 while ($line=<DFH>) {   
    print $fh "$line";  
    if ($line=~/INFO/) { $notices++; }  
    if ($line=~/WARNING/) { $notices++; }  
    if ($line=~/CRITICAL/) { $warnings++; }  
 };  
 close (DFH);  
 }  
 print $fh "</pre>";  
   
 my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";  
   
 $restartflag=1;  
 if (-e $loncfile) {  
    my $lfh=IO::File->new("$loncfile");  
    my $loncpid=<$lfh>;  
    chomp($loncpid);  
    if (kill 0 => $loncpid) {  
       print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";  
       kill USR1 => $loncpid;  
       $restartflag=0;  
    } else {  
       $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 $fh   
   "<h3>Deciding to clean up stale .pid file and restart lonc</h3>";  
    }  
 }   
 if ($restartflag==1) {  
    $errors++;  
    print $fh "<h3>lonc not running, trying to start</h3>";  
    system("$perlvar{'lonDaemons'}/lonc");  
    sleep 120;  
    if (-e $loncfile) {  
        print $fh "Seems like it started ...<p>";  
        my $lfh=IO::File->new("$loncfile");  
        my $loncpid=<$lfh>;  
        chomp($loncpid);  
        sleep 30;  
        if (kill 0 => $loncpid) {  
           print $fh "<h3>lonc at pid $loncpid responding</h3>";  
        } else {  
           $errors++; $errors++;  
           print $fh "<h3>lonc at pid $loncpid not responding</h3>";  
           print $fh "Give it one more try ...<p>";  
           system("$perlvar{'lonDaemons'}/lonc");  
           sleep 120;  
        }  
    } else {  
        print $fh "Seems like that did not work!<p>";  
        $errors++;  
    }  
    if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") {  
     print $fh "<p><pre>";  
     open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");  
     while ($line=<DFH>) {   
       print $fh "$line";  
       if ($line=~/WARNING/) { $notices++; }  
       if ($line=~/CRITICAL/) { $notices++; }  
     };  
     close (DFH);  
     print $fh "</pre>";  
    }  
 }  
   
 $fname="$perlvar{'lonDaemons'}/logs/lonc.log";      my $unsend=0;
   
                           my ($dev,$ino,$mode,$nlink,      my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
                               $uid,$gid,$rdev,$size,      while (my $line=<$dfh>) {
                               $atime,$mtime,$ctime,   my ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
                               $blksize,$blocks)=stat($fname);   if ($sdf eq 'F') { 
       my $local=localtime($time);
       &log($fh,"<b>Failed: $time, $dserv, $dcmd</b><br>");
       $warnings++;
    }
    if ($sdf eq 'S') { $unsend--; }
    if ($sdf eq 'D') { $unsend++; }
       }
   
 if ($size>40000) {      &log($fh,"Total unsend messages: <b>$unsend</b><p>\n");
     print $fh "Rotating logs ...<p>";      $warnings=$warnings+5*$unsend;
     rename("$fname.2","$fname.3");  
     rename("$fname.1","$fname.2");  
     rename("$fname","$fname.1");  
 }  
   
          if ($unsend) { $simplestatus{'unsend'}=$unsend; }
 &errout($fh);      &log($fh,"<h3>Outgoing Buffer</h3>");
 # ---------------------------------------------------------------------- lonnet  
   
 print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>';      open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
 if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){      while (my $line=<DFH>) { 
 open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");   &log($fh,"$line<br>");
 while ($line=<DFH>) {       }
     print $fh "$line";      close (DFH);
 };  
 close (DFH);  
 }  
 print $fh "</pre><h3>Perm Log</h3><pre>";  
   
 if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {  
     open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");  
 while ($line=<DFH>) {   
    print $fh "$line";  
 };  
 close (DFH);  
 } else { print $fh "No perm log\n" }  
   
 $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";  
   
                           my ($dev,$ino,$mode,$nlink,  
                               $uid,$gid,$rdev,$size,  
                               $atime,$mtime,$ctime,  
                               $blksize,$blocks)=stat($fname);  
   
 if ($size>40000) {  
     print $fh "Rotating logs ...<p>";  
     rename("$fname.2","$fname.3");  
     rename("$fname.1","$fname.2");  
     rename("$fname","$fname.1");  
 }  }
   
 print $fh "</pre>";  sub finish_logging {
 &errout($fh);      my ($fh)=@_;
 # ----------------------------------------------------------------- Connections      &log($fh,"<a name=errcount>\n");
       $totalcount=$notices+4*$warnings+100*$errors;
       &errout($fh);
       &log($fh,"<h1>Total Error Count: $totalcount</h1>");
       my $now=time;
       my $date=localtime($now);
       &log($fh,"<hr>$date ($now)</body></html>\n");
       print "lon-status webpage updated\n";
       $fh->close();
   
       if ($errors) { $simplestatus{'errors'}=$errors; }
       if ($warnings) { $simplestatus{'warnings'}=$warnings; }
       if ($notices) { $simplestatus{'notices'}=$notices; }
       $simplestatus{'time'}=time;
   }
   
 print $fh '<hr><a name="connections"><h2>Connections</h2>';  sub log_simplestatus {
       rename ("$statusdir/newstatus.html","$statusdir/index.html");
       
       my $sfh=IO::File->new(">$statusdir/loncron_simple.txt");
       foreach (keys %simplestatus) {
    print $sfh $_.'='.$simplestatus{$_}.'&';
       }
       print $sfh "\n";
       $sfh->close();
   }
   
 print $fh "<table border=2>";  sub send_mail {
 foreach $tryserver (keys %hostname) {      print "sending mail\n";
       my $emailto="$perlvar{'lonAdmEMail'}";
       if ($totalcount>1000) {
    $emailto.=",$perlvar{'lonSysEMail'}";
       }
       my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; 
       system("metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");
   }
   
     $answer=reply("pong",$tryserver);  # ================================================================ Main Program
     if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {  sub main () {
  $result="<b>ok</b>";  # --------------------------------- Read loncapa_apache.conf and loncapa.conf
     } else {      my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
         $result=$answer;      %perlvar=%{$perlvarref};
         $warnings++;      undef $perlvarref;
         if ($answer eq 'con_lost') { $warnings++; }      delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
       delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
   
   # --------------------------------------- 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");
    my $emailto=$perlvar{'lonSysEMail'};
    my $hostname=`/bin/hostname`;
    chop $hostname;
    $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
    my $subj="LON: Unconfigured machine $hostname";
    system("echo 'Unconfigured machine $hostname.' |\
    mailto $emailto -s '$subj' > /dev/null");
    exit 1;
     }      }
     print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n";  
   
 }  # ----------------------------- Make sure this process is running from user=www
 print $fh "</table>";      my $wwwid=getpwnam('www');
       if ($wwwid!=$<) {
    print("User ID mismatch.  This program must be run as user 'www'\n");
    my $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
    my $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;
       }
   
 &errout($fh);  # ------------------------------------------------------------- Read hosts file
 # ------------------------------------------------------------ Delayed messages      my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
       
       my (%hostname,%hostdom,%hostrole,%spareid);
       while (my $configline=<$config>) {
    my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline);
    if ($id && $domain && $role && $name && $ip) {
       $hostname{$id}=$name;
       $hostdom{$id}=$domain;
       $hostrole{$id}=$role;
    } else {
       if ($configline) {
   # &logthis("Skipping hosts.tab line -$configline-");
       }
    }
       }
       undef $config;
   
 print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';  # ------------------------------------------------------ Read spare server file
       $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
       
       while (my $configline=<$config>) {
    chomp($configline);
    if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
       $spareid{$configline}=1;
    }
       }
       undef $config;
   
 print $fh '<h3>Scanning Permanent Log</h3>';  # ---------------------------------------------------------------- Start report
   
 $unsend=0;      $errors=0;
 {      $warnings=0;
     my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");      $notices=0;
     while ($line=<$dfh>) {  
  ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);      my $fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);
         if ($sdf eq 'F') {   
     $local=localtime($time);      &log_machine_info($fh);
             print "<b>Failed: $time, $dserv, $dcmd</b><br>";      &clean_tmp($fh);
             $warnings++;      &clean_lonIDs($fh);
         }      &check_httpd_logs($fh);
         if ($sdf eq 'S') { $unsend--; }      &checkon_daemon($fh,'lonsql',200000);
         if ($sdf eq 'D') { $unsend++; }      &checkon_daemon($fh,'lond',40000,1);
     }      &checkon_daemon($fh,'lonc',40000,1);
 }      &checkon_daemon($fh,'lonhttpd',40000);
 print $fh "Total unsend messages: <b>$unsend</b><p>\n";      
 $warnings=$warnings+5*$unsend;      &test_connections($fh,\%hostname);
       &check_delayed_msg($fh);
 print $fh "<h3>Outgoing Buffer</h3>";  
   
 open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");  
 while ($line=<DFH>) {   
     print $fh "$line<br>";  
 };  
 close (DFH);  
   
 # ------------------------------------------------------------------------- End  
 print $fh "<a name=errcount>\n";  
 $totalcount=$notices+4*$warnings+100*$errors;  
 &errout($fh);  
 print $fh "<h1>Total Error Count: $totalcount</h1>";  
 $now=time;  
 $date=localtime($now);  
 print $fh "<hr>$date ($now)</body></html>\n";  
   
       &finish_logging($fh);
       &log_simplestatus();
   
       if ($totalcount>200) { &send_mail(); }
 }  }
   
 rename ("$statusdir/newstatus.html","$statusdir/index.html");  &main();
   
 if ($totalcount>200) {  
    $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";  
    $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";   
    system(  
  "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");  
 }  
 1;  1;
   
   

Removed from v.1.13  
changed lines
  Added in v.1.46


FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>