Annotation of loncom/loncron, revision 1.52

1.1       albertel    1: #!/usr/bin/perl
                      2: 
1.47      albertel    3: # Housekeeping program, started by cron, loncontrol and loncron.pl
                      4: #
1.52    ! albertel    5: # $Id: loncron,v 1.51 2004/05/25 15:20:31 albertel Exp $
1.47      albertel    6: #
                      7: # Copyright Michigan State University Board of Trustees
                      8: #
                      9: # This file is part of the LearningOnline Network with CAPA (LON-CAPA).
                     10: #
                     11: # LON-CAPA is free software; you can redistribute it and/or modify
                     12: # it under the terms of the GNU General Public License as published by
                     13: # the Free Software Foundation; either version 2 of the License, or
                     14: # (at your option) any later version.
                     15: #
                     16: # LON-CAPA is distributed in the hope that it will be useful,
                     17: # but WITHOUT ANY WARRANTY; without even the implied warranty of
                     18: # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
                     19: # GNU General Public License for more details.
                     20: #
                     21: # You should have received a copy of the GNU General Public License
                     22: # along with LON-CAPA; if not, write to the Free Software
                     23: # Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
                     24: #
                     25: # /home/httpd/html/adm/gpl.txt
                     26: #
                     27: # http://www.lon-capa.org/
1.1       albertel   28: #
1.24      www        29: 
                     30: $|=1;
1.48      albertel   31: use strict;
1.1       albertel   32: 
1.26      harris41   33: use lib '/home/httpd/lib/perl/';
                     34: use LONCAPA::Configuration;
                     35: 
1.1       albertel   36: use IO::File;
                     37: use IO::Socket;
1.48      albertel   38: use HTML::Entities;
1.49      albertel   39: use Getopt::Long;
1.46      albertel   40: #globals
                     41: use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount);
                     42: 
                     43: my $statusdir="/home/httpd/html/lon-status";
                     44: 
1.1       albertel   45: 
                     46: # -------------------------------------------------- Non-critical communication
                     47: sub reply {
                     48:     my ($cmd,$server)=@_;
                     49:     my $peerfile="$perlvar{'lonSockDir'}/$server";
                     50:     my $client=IO::Socket::UNIX->new(Peer    =>"$peerfile",
                     51:                                      Type    => SOCK_STREAM,
                     52:                                      Timeout => 10)
                     53:        or return "con_lost";
                     54:     print $client "$cmd\n";
                     55:     my $answer=<$client>;
                     56:     chomp($answer);
                     57:     if (!$answer) { $answer="con_lost"; }
                     58:     return $answer;
                     59: }
                     60: 
                     61: # --------------------------------------------------------- Output error status
                     62: 
1.46      albertel   63: sub log {
                     64:     my $fh=shift;
                     65:     if ($fh) {	print $fh @_  }
                     66: }
                     67: 
1.1       albertel   68: sub errout {
                     69:    my $fh=shift;
1.46      albertel   70:    &log($fh,(<<ENDERROUT));
1.48      albertel   71:      <table border="2" bgcolor="#CCCCCC">
1.1       albertel   72:      <tr><td>Notices</td><td>$notices</td></tr>
                     73:      <tr><td>Warnings</td><td>$warnings</td></tr>
                     74:      <tr><td>Errors</td><td>$errors</td></tr>
1.48      albertel   75:      </table><p><a href="#top">Top</a></p>
1.1       albertel   76: ENDERROUT
                     77: }
                     78: 
1.42      albertel   79: sub start_daemon {
1.50      albertel   80:     my ($fh,$daemon,$pidfile,$args) = @_;
1.44      albertel   81:     my $progname=$daemon;
1.50      albertel   82:     if ($daemon eq 'lonc' && $args eq 'new') {
1.44      albertel   83: 	$progname='loncnew'; 
                     84: 	print "new ";
                     85:     }
1.51      albertel   86:     my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors";
                     87:     my $size=(stat($error_fname))[7];
                     88:     if ($size>40000) {
                     89: 	&log($fh,"<p>Rotating error logs ...</p>");
                     90: 	rename("$error_fname.2","$error_fname.3");
                     91: 	rename("$error_fname.1","$error_fname.2");
                     92: 	rename("$error_fname","$error_fname.1");
                     93:     }
                     94:     system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors");
1.42      albertel   95:     sleep 2;
                     96:     if (-e $pidfile) {
1.48      albertel   97: 	&log($fh,"<p>Seems like it started ...</p>");
1.42      albertel   98: 	my $lfh=IO::File->new("$pidfile");
                     99: 	my $daemonpid=<$lfh>;
                    100: 	chomp($daemonpid);
                    101: 	sleep 2;
                    102: 	if (kill 0 => $daemonpid) {
                    103: 	    return 1;
                    104: 	} else {
                    105: 	    return 0;
                    106: 	}
                    107:     }
1.48      albertel  108:     &log($fh,"<p>Seems like that did not work!</p>");
1.42      albertel  109:     $errors++;
                    110:     return 0;
                    111: }
                    112: 
                    113: sub checkon_daemon {
1.50      albertel  114:     my ($fh,$daemon,$maxsize,$sendusr1,$args)=@_;
1.42      albertel  115: 
1.48      albertel  116:     &log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>');
1.42      albertel  117:     printf("%-10s ",$daemon);
                    118:     if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
                    119: 	open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
1.46      albertel  120: 	while (my $line=<DFH>) { 
                    121: 	    &log($fh,"$line");
1.42      albertel  122: 	    if ($line=~/INFO/) { $notices++; }
                    123: 	    if ($line=~/WARNING/) { $notices++; }
                    124: 	    if ($line=~/CRITICAL/) { $warnings++; }
                    125: 	};
                    126: 	close (DFH);
                    127:     }
1.48      albertel  128:     &log($fh,"</tt></p>");
1.42      albertel  129:     
                    130:     my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid";
                    131:     
                    132:     my $restartflag=1;
1.46      albertel  133:     my $daemonpid;
1.42      albertel  134:     if (-e $pidfile) {
                    135: 	my $lfh=IO::File->new("$pidfile");
1.46      albertel  136: 	$daemonpid=<$lfh>;
1.42      albertel  137: 	chomp($daemonpid);
                    138: 	if (kill 0 => $daemonpid) {
1.46      albertel  139: 	    &log($fh,"<h3>$daemon at pid $daemonpid responding");
                    140: 	    if ($sendusr1) { &log($fh,", sending USR1"); }
                    141: 	    &log($fh,"</h3>");
1.42      albertel  142: 	    if ($sendusr1) { kill USR1 => $daemonpid; }
                    143: 	    $restartflag=0;
                    144: 	    print "running\n";
                    145: 	} else {
                    146: 	    $errors++;
1.46      albertel  147: 	    &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.42      albertel  148: 	    $restartflag=1;
1.46      albertel  149: 	    &log($fh,"<h3>Decided to clean up stale .pid file and restart $daemon</h3>");
1.42      albertel  150: 	}
                    151:     }
                    152:     if ($restartflag==1) {
                    153: 	$simplestatus{$daemon}='off';
                    154: 	$errors++;
1.46      albertel  155: 	&log($fh,'<br><font color="red">Killall '.$daemon.': '.
                    156: 	    `killall $daemon 2>&1`.' - ');
1.42      albertel  157: 	sleep 2;
1.46      albertel  158: 	&log($fh,unlink($pidfile).' - '.
1.42      albertel  159: 	    `killall -9 $daemon 2>&1`.
1.46      albertel  160: 	    '</font><br>');
                    161: 	&log($fh,"<h3>$daemon not running, trying to start</h3>");
1.42      albertel  162: 	
1.50      albertel  163: 	if (&start_daemon($fh,$daemon,$pidfile,$args)) {
1.46      albertel  164: 	    &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
1.42      albertel  165: 	    $simplestatus{$daemon}='restarted';
                    166: 	    print "started\n";
                    167: 	} else {
                    168: 	    $errors++;
1.46      albertel  169: 	    &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.48      albertel  170: 	    &log($fh,"<p>Give it one more try ...</p>");
1.42      albertel  171: 	    print " ";
1.50      albertel  172: 	    if (&start_daemon($fh,$daemon,$pidfile,$args)) {
1.46      albertel  173: 		&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
1.42      albertel  174: 		$simplestatus{$daemon}='restarted';
                    175: 		print "started\n";
                    176: 	    } else {
                    177: 		print " failed\n";
                    178: 		$simplestatus{$daemon}='failed';
                    179: 		$errors++; $errors++;
1.46      albertel  180: 		&log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.48      albertel  181: 		&log($fh,"<p>Unable to start $daemon</p>");
1.42      albertel  182: 	    }
                    183: 	}
                    184: 
                    185: 	if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
1.46      albertel  186: 	    &log($fh,"<p><pre>");
1.42      albertel  187: 	    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|");
1.46      albertel  188: 	    while (my $line=<DFH>) { 
                    189: 		&log($fh,"$line");
1.42      albertel  190: 		if ($line=~/WARNING/) { $notices++; }
                    191: 		if ($line=~/CRITICAL/) { $notices++; }
                    192: 	    };
                    193: 	    close (DFH);
1.48      albertel  194: 	    &log($fh,"</pre></p>");
1.42      albertel  195: 	}
                    196:     }
                    197:     
1.46      albertel  198:     my $fname="$perlvar{'lonDaemons'}/logs/$daemon.log";
1.42      albertel  199:     
                    200:     my ($dev,$ino,$mode,$nlink,
                    201: 	$uid,$gid,$rdev,$size,
                    202: 	$atime,$mtime,$ctime,
                    203: 	$blksize,$blocks)=stat($fname);
                    204:     
                    205:     if ($size>$maxsize) {
1.48      albertel  206: 	&log($fh,"<p>Rotating logs ...</p>");
1.42      albertel  207: 	rename("$fname.2","$fname.3");
                    208: 	rename("$fname.1","$fname.2");
                    209: 	rename("$fname","$fname.1");
                    210:     }
                    211: 
                    212:     &errout($fh);
                    213: }
1.1       albertel  214: 
1.46      albertel  215: # --------------------------------------------------------------------- Machine
                    216: sub log_machine_info {
                    217:     my ($fh)=@_;
1.48      albertel  218:     &log($fh,'<hr /><a name="machine" /><h2>Machine Information</h2>');
1.46      albertel  219:     &log($fh,"<h3>loadavg</h3>");
                    220: 	
                    221:     open (LOADAVGH,"/proc/loadavg");
                    222:     my $loadavg=<LOADAVGH>;
                    223:     close (LOADAVGH);
                    224:     
                    225:     &log($fh,"<tt>$loadavg</tt>");
                    226:     
                    227:     my @parts=split(/\s+/,$loadavg);
                    228:     if ($parts[1]>4.0) {
                    229: 	$errors++;
                    230:     } elsif ($parts[1]>2.0) {
                    231: 	$warnings++;
                    232:     } elsif ($parts[1]>1.0) {
                    233: 	$notices++;
                    234:     }
1.13      harris41  235: 
1.46      albertel  236:     &log($fh,"<h3>df</h3>");
                    237:     &log($fh,"<pre>");
1.14      harris41  238: 
1.46      albertel  239:     open (DFH,"df|");
                    240:     while (my $line=<DFH>) { 
1.48      albertel  241: 	&log($fh,&encode_entities($line,'<>&"')); 
1.46      albertel  242: 	@parts=split(/\s+/,$line);
                    243: 	my $usage=$parts[4];
                    244: 	$usage=~s/\W//g;
                    245: 	if ($usage>90) { 
                    246: 	    $warnings++;
                    247: 	    $notices++; 
                    248: 	} elsif ($usage>80) {
                    249: 	    $warnings++;
                    250: 	} elsif ($usage>60) {
                    251: 	    $notices++;
1.31      albertel  252: 	}
1.46      albertel  253: 	if ($usage>95) { $warnings++; $warnings++; $simplestatus{'diskfull'}++; }
1.1       albertel  254:     }
1.46      albertel  255:     close (DFH);
                    256:     &log($fh,"</pre>");
1.1       albertel  257: 
                    258: 
1.46      albertel  259:     &log($fh,"<h3>ps</h3>");
                    260:     &log($fh,"<pre>");
                    261:     my $psproc=0;
1.1       albertel  262: 
1.46      albertel  263:     open (PSH,"ps -aux --cols 140 |");
                    264:     while (my $line=<PSH>) { 
1.48      albertel  265: 	&log($fh,&encode_entities($line,'<>&"')); 
1.46      albertel  266: 	$psproc++;
                    267:     }
                    268:     close (PSH);
                    269:     &log($fh,"</pre>");
1.1       albertel  270: 
1.46      albertel  271:     if ($psproc>200) { $notices++; }
                    272:     if ($psproc>250) { $notices++; }
1.1       albertel  273: 
1.46      albertel  274:     &errout($fh);
                    275: }
1.1       albertel  276: 
1.46      albertel  277: sub start_logging {
                    278:     my ($hostdom,$hostrole,$hostname,$spareid)=@_;
1.43      albertel  279:     my $fh=IO::File->new(">$statusdir/newstatus.html");
                    280:     my %simplestatus=();
1.46      albertel  281:     my $now=time;
                    282:     my $date=localtime($now);
1.43      albertel  283:     
1.46      albertel  284: 
                    285:     &log($fh,(<<ENDHEADERS));
1.1       albertel  286: <html>
                    287: <head>
                    288: <title>LON Status Report $perlvar{'lonHostID'}</title>
                    289: </head>
1.3       www       290: <body bgcolor="#AAAAAA">
1.48      albertel  291: <a name="top" />
1.1       albertel  292: <h1>LON Status Report $perlvar{'lonHostID'}</h1>
                    293: <h2>$date ($now)</h2>
                    294: <ol>
1.48      albertel  295: <li><a href="#configuration">Configuration</a></li>
                    296: <li><a href="#machine">Machine Information</a></li>
                    297: <li><a href="#tmp">Temporary Files</a></li>
                    298: <li><a href="#tokens">Session Tokens</a></li>
                    299: <li><a href="#httpd">httpd</a></li>
                    300: <li><a href="#lonsql">lonsql</a></li>
                    301: <li><a href="#lond">lond</a></li>
                    302: <li><a href="#lonc">lonc</a></li>
                    303: <li><a href="#lonhttpd">lonhttpd</a></li>
                    304: <li><a href="#lonnet">lonnet</a></li>
                    305: <li><a href="#connections">Connections</a></li>
                    306: <li><a href="#delayed">Delayed Messages</a></li>
                    307: <li><a href="#errcount">Error Count</a></li>
1.1       albertel  308: </ol>
1.48      albertel  309: <hr />
                    310: <a name="configuration" />
1.1       albertel  311: <h2>Configuration</h2>
                    312: <h3>PerlVars</h3>
1.48      albertel  313: <table border="2">
1.1       albertel  314: ENDHEADERS
                    315: 
1.46      albertel  316:     foreach my $varname (sort(keys(%perlvar))) {
1.48      albertel  317: 	&log($fh,"<tr><td>$varname</td><td>".
                    318: 	     &encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n");
1.43      albertel  319:     }
1.48      albertel  320:     &log($fh,"</table><h3>Hosts</h3><table border='2'>");
1.46      albertel  321:     foreach my $id (sort(keys(%{$hostname}))) {
                    322: 	&log($fh,
                    323: 	    "<tr><td>$id</td><td>".$hostdom->{$id}.
                    324: 	    "</td><td>".$hostrole->{$id}.
                    325: 	    "</td><td>".$hostname->{$id}."</td></tr>\n");
                    326:     }
                    327:     &log($fh,"</table><h3>Spare Hosts</h3><ol>");
                    328:     foreach my $id (sort(keys(%{$spareid}))) {
1.48      albertel  329: 	&log($fh,"<li>$id\n</li>");
1.43      albertel  330:     }
1.46      albertel  331:     &log($fh,"</ol>\n");
                    332:     return $fh;
                    333: }
1.11      www       334: 
                    335: # --------------------------------------------------------------- clean out tmp
1.46      albertel  336: sub clean_tmp {
                    337:     my ($fh)=@_;
1.48      albertel  338:     &log($fh,'<hr /><a name="tmp" /><h2>Temporary Files</h2>');
1.46      albertel  339:     my $cleaned=0;
                    340:     my $old=0;
                    341:     while (my $fname=<$perlvar{'lonDaemons'}/tmp/*>) {
1.43      albertel  342: 	my ($dev,$ino,$mode,$nlink,
                    343: 	    $uid,$gid,$rdev,$size,
                    344: 	    $atime,$mtime,$ctime,
                    345: 	    $blksize,$blocks)=stat($fname);
1.46      albertel  346: 	my $now=time;
                    347: 	my $since=$now-$mtime;
1.43      albertel  348: 	if ($since>$perlvar{'lonExpire'}) {
1.46      albertel  349: 	    my $line='';
1.43      albertel  350: 	    if (open(PROBE,$fname)) {
                    351: 		$line=<PROBE>;
                    352: 		close(PROBE);
                    353: 	    }
                    354: 	    unless ($line=~/^CHECKOUTTOKEN\&/) {
                    355: 		$cleaned++;
                    356: 		unlink("$fname");
                    357: 	    } else {
                    358: 		if ($since>365*$perlvar{'lonExpire'}) {
                    359: 		    $cleaned++;
                    360: 		    unlink("$fname");
                    361: 		} else { $old++; }
                    362: 	    }
                    363: 	}
                    364:     }
1.46      albertel  365:     &log($fh,"Cleaned up ".$cleaned." files (".$old." old checkout tokens).");
                    366: }
1.11      www       367: 
                    368: # ------------------------------------------------------------ clean out lonIDs
1.46      albertel  369: sub clean_lonIDs {
                    370:     my ($fh)=@_;
1.48      albertel  371:     &log($fh,'<hr /><a name="tokens" /><h2>Session Tokens</h2>');
1.46      albertel  372:     my $cleaned=0;
                    373:     my $active=0;
                    374:     while (my $fname=<$perlvar{'lonIDsDir'}/*>) {
1.43      albertel  375: 	my ($dev,$ino,$mode,$nlink,
                    376: 	    $uid,$gid,$rdev,$size,
                    377: 	    $atime,$mtime,$ctime,
                    378: 	    $blksize,$blocks)=stat($fname);
1.46      albertel  379: 	my $now=time;
                    380: 	my $since=$now-$mtime;
1.43      albertel  381: 	if ($since>$perlvar{'lonExpire'}) {
                    382: 	    $cleaned++;
1.46      albertel  383: 	    &log($fh,"Unlinking $fname<br>");
1.43      albertel  384: 	    unlink("$fname");
                    385: 	} else {
                    386: 	    $active++;
                    387: 	}
1.46      albertel  388:     }
1.48      albertel  389:     &log($fh,"<p>Cleaned up ".$cleaned." stale session token(s).</p>");
1.46      albertel  390:     &log($fh,"<h3>$active open session(s)</h3>");
                    391: }
1.43      albertel  392: 
1.11      www       393: 
1.1       albertel  394: # ----------------------------------------------------------------------- httpd
1.46      albertel  395: sub check_httpd_logs {
                    396:     my ($fh)=@_;
1.48      albertel  397:     &log($fh,'<hr /><a name="httpd" /><h2>httpd</h2><h3>Access Log</h3><pre>');
1.43      albertel  398:     
                    399:     open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
1.48      albertel  400:     while (my $line=<DFH>) { &log($fh,&encode_entities($line,'<>&"')) };
1.43      albertel  401:     close (DFH);
1.46      albertel  402: 	
                    403:     &log($fh,"</pre><h3>Error Log</h3><pre>");
                    404: 	
1.43      albertel  405:     open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
1.46      albertel  406:     while (my $line=<DFH>) { 
                    407: 	&log($fh,"$line");
1.43      albertel  408: 	if ($line=~/\[error\]/) { $notices++; } 
1.46      albertel  409:     }
1.43      albertel  410:     close (DFH);
1.46      albertel  411:     &log($fh,"</pre>");
1.43      albertel  412:     &errout($fh);
1.46      albertel  413: }
1.1       albertel  414: 
                    415: # ---------------------------------------------------------------------- lonnet
                    416: 
1.48      albertel  417: sub rotate_lonnet_logs {
1.46      albertel  418:     my ($fh)=@_;
1.48      albertel  419:     &log($fh,'<hr /><a name="lonnet" /><h2>lonnet</h2><h3>Temp Log</h3><pre>');
1.43      albertel  420:     print "checking logs\n";
                    421:     if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
                    422: 	open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
1.46      albertel  423: 	while (my $line=<DFH>) { 
1.48      albertel  424: 	    &log($fh,&encode_entities($line,'<>&"'));
1.46      albertel  425: 	}
1.43      albertel  426: 	close (DFH);
                    427:     }
1.46      albertel  428:     &log($fh,"</pre><h3>Perm Log</h3><pre>");
1.43      albertel  429:     
                    430:     if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
                    431: 	open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
1.46      albertel  432: 	while (my $line=<DFH>) { 
1.48      albertel  433: 	    &log($fh,&encode_entities($line,'<>&"'));
1.46      albertel  434: 	}
1.43      albertel  435: 	close (DFH);
1.46      albertel  436:     } else { &log($fh,"No perm log\n") }
1.43      albertel  437: 
1.46      albertel  438:     my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
1.43      albertel  439: 
                    440:     my ($dev,$ino,$mode,$nlink,
                    441: 	$uid,$gid,$rdev,$size,
                    442: 	$atime,$mtime,$ctime,
                    443: 	$blksize,$blocks)=stat($fname);
                    444: 
                    445:     if ($size>40000) {
1.48      albertel  446: 	&log($fh,"<p>Rotating logs ...</p>");
1.43      albertel  447: 	rename("$fname.2","$fname.3");
                    448: 	rename("$fname.1","$fname.2");
                    449: 	rename("$fname","$fname.1");
                    450:     }
1.1       albertel  451: 
1.46      albertel  452:     &log($fh,"</pre>");
1.43      albertel  453:     &errout($fh);
1.46      albertel  454: }
                    455: 
1.43      albertel  456: # ----------------------------------------------------------------- Connections
1.46      albertel  457: sub test_connections {
                    458:     my ($fh,$hostname)=@_;
1.48      albertel  459:     &log($fh,'<hr /><a name="connections" /><h2>Connections</h2>');
1.43      albertel  460:     print "testing connections\n";
1.48      albertel  461:     &log($fh,"<table border='2'>");
1.49      albertel  462:     my ($good,$bad)=(0,0);
1.46      albertel  463:     foreach my $tryserver (sort(keys(%{$hostname}))) {
1.43      albertel  464: 	print(".");
1.46      albertel  465: 	my $result;
                    466: 	my $answer=reply("pong",$tryserver);
1.43      albertel  467: 	if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
                    468: 	    $result="<b>ok</b>";
1.49      albertel  469: 	    $good++;
1.43      albertel  470: 	} else {
                    471: 	    $result=$answer;
                    472: 	    $warnings++;
1.49      albertel  473: 	    if ($answer eq 'con_lost') {
                    474: 		$bad++;
                    475: 		$warnings++;
1.50      albertel  476: 	    } else {
                    477: 		$good++; #self connection
1.49      albertel  478: 	    }
1.43      albertel  479: 	}
                    480: 	if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }
1.46      albertel  481: 	&log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n");
1.1       albertel  482:     }
1.46      albertel  483:     &log($fh,"</table>");
1.49      albertel  484:     print "\n$good good, $bad bad connections\n";
1.43      albertel  485:     &errout($fh);
1.46      albertel  486: }
                    487: 
                    488: 
1.1       albertel  489: # ------------------------------------------------------------ Delayed messages
1.46      albertel  490: sub check_delayed_msg {
                    491:     my ($fh)=@_;
1.48      albertel  492:     &log($fh,'<hr /><a name="delayed" /><h2>Delayed Messages</h2>');
1.43      albertel  493:     print "checking buffers\n";
1.46      albertel  494:     
                    495:     &log($fh,'<h3>Scanning Permanent Log</h3>');
1.1       albertel  496: 
1.46      albertel  497:     my $unsend=0;
1.1       albertel  498: 
1.46      albertel  499:     my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
                    500:     while (my $line=<$dfh>) {
                    501: 	my ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
                    502: 	if ($sdf eq 'F') { 
                    503: 	    my $local=localtime($time);
                    504: 	    &log($fh,"<b>Failed: $time, $dserv, $dcmd</b><br>");
                    505: 	    $warnings++;
1.43      albertel  506: 	}
1.46      albertel  507: 	if ($sdf eq 'S') { $unsend--; }
                    508: 	if ($sdf eq 'D') { $unsend++; }
1.1       albertel  509:     }
1.46      albertel  510: 
1.48      albertel  511:     &log($fh,"<p>Total unsend messages: <b>$unsend</b></p>\n");
1.43      albertel  512:     $warnings=$warnings+5*$unsend;
1.1       albertel  513: 
1.43      albertel  514:     if ($unsend) { $simplestatus{'unsend'}=$unsend; }
1.48      albertel  515:     &log($fh,"<h3>Outgoing Buffer</h3>\n<pre>");
1.1       albertel  516: 
1.43      albertel  517:     open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
1.46      albertel  518:     while (my $line=<DFH>) { 
1.48      albertel  519: 	&log($fh,&encode_entities($line,'<>&"'));
1.46      albertel  520:     }
1.48      albertel  521:     &log($fh,"</pre>\n");
1.43      albertel  522:     close (DFH);
1.46      albertel  523: }
1.1       albertel  524: 
1.46      albertel  525: sub finish_logging {
                    526:     my ($fh)=@_;
1.48      albertel  527:     &log($fh,"<a name='errcount' />\n");
1.43      albertel  528:     $totalcount=$notices+4*$warnings+100*$errors;
                    529:     &errout($fh);
1.46      albertel  530:     &log($fh,"<h1>Total Error Count: $totalcount</h1>");
                    531:     my $now=time;
                    532:     my $date=localtime($now);
1.48      albertel  533:     &log($fh,"<hr />$date ($now)</body></html>\n");
1.43      albertel  534:     print "lon-status webpage updated\n";
                    535:     $fh->close();
1.46      albertel  536: 
                    537:     if ($errors) { $simplestatus{'errors'}=$errors; }
                    538:     if ($warnings) { $simplestatus{'warnings'}=$warnings; }
                    539:     if ($notices) { $simplestatus{'notices'}=$notices; }
                    540:     $simplestatus{'time'}=time;
1.1       albertel  541: }
                    542: 
1.46      albertel  543: sub log_simplestatus {
                    544:     rename ("$statusdir/newstatus.html","$statusdir/index.html");
                    545:     
1.43      albertel  546:     my $sfh=IO::File->new(">$statusdir/loncron_simple.txt");
                    547:     foreach (keys %simplestatus) {
                    548: 	print $sfh $_.'='.$simplestatus{$_}.'&';
                    549:     }
                    550:     print $sfh "\n";
                    551:     $sfh->close();
1.41      www       552: }
1.46      albertel  553: 
                    554: sub send_mail {
1.43      albertel  555:     print "sending mail\n";
1.46      albertel  556:     my $emailto="$perlvar{'lonAdmEMail'}";
1.43      albertel  557:     if ($totalcount>1000) {
                    558: 	$emailto.=",$perlvar{'lonSysEMail'}";
                    559:     }
1.46      albertel  560:     my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; 
1.52    ! albertel  561: 
        !           562:     my $result=system("metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html >& /dev/null");
        !           563:     if ($result != 0) {
        !           564: 	$result=system("mail -s '$subj' $emailto < $statusdir/index.html");
        !           565:     }
1.1       albertel  566: }
1.46      albertel  567: 
1.49      albertel  568: sub usage {
                    569:     print(<<USAGE);
                    570: loncron - housekeeping program that checks up on various parts of Lon-CAPA
                    571: 
                    572: Options:
                    573:    --help     Display help
                    574:    --oldlonc  When starting the lonc daemon use 'lonc' not 'loncnew'
                    575:    --noemail  Do not send the status email
                    576:    --justcheckconnections  Only check the current status of the lonc/d
                    577:                                 connections, do not send emails do not
                    578:                                 check if the daemons are running, do not
                    579:                                 generate lon-status
                    580:    --justcheckdaemons      Only check that all of the Lon-CAPA daemons are
                    581:                                 running, do not send emails do not
                    582:                                 check the lonc/d connections, do not
                    583:                                 generate lon-status
                    584:                            
                    585: USAGE
                    586: }
                    587: 
1.46      albertel  588: # ================================================================ Main Program
                    589: sub main () {
1.49      albertel  590:     my ($oldlonc,$help,$justcheckdaemons,$noemail,$justcheckconnections);
                    591:     &GetOptions("help"                 => \$help,
                    592: 		"oldlonc"              => \$oldlonc,
                    593: 		"justcheckdaemons"     => \$justcheckdaemons,
                    594: 		"noemail"              => \$noemail,
                    595: 		"justcheckconnections" => \$justcheckconnections
                    596: 		);
                    597:     if ($help) { &usage(); return; }
1.46      albertel  598: # --------------------------------- Read loncapa_apache.conf and loncapa.conf
                    599:     my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
                    600:     %perlvar=%{$perlvarref};
                    601:     undef $perlvarref;
                    602:     delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
                    603:     delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
                    604: 
                    605: # --------------------------------------- Make sure that LON-CAPA is configured
                    606: # I only test for one thing here (lonHostID).  This is just a safeguard.
                    607:     if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) {
                    608: 	print("Unconfigured machine.\n");
                    609: 	my $emailto=$perlvar{'lonSysEMail'};
                    610: 	my $hostname=`/bin/hostname`;
                    611: 	chop $hostname;
                    612: 	$hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
                    613: 	my $subj="LON: Unconfigured machine $hostname";
                    614: 	system("echo 'Unconfigured machine $hostname.' |\
                    615:  mailto $emailto -s '$subj' > /dev/null");
                    616: 	exit 1;
                    617:     }
                    618: 
                    619: # ----------------------------- Make sure this process is running from user=www
                    620:     my $wwwid=getpwnam('www');
                    621:     if ($wwwid!=$<) {
                    622: 	print("User ID mismatch.  This program must be run as user 'www'\n");
                    623: 	my $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
                    624: 	my $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
                    625: 	system("echo 'User ID mismatch.  loncron must be run as user www.' |\
                    626:  mailto $emailto -s '$subj' > /dev/null");
                    627: 	exit 1;
                    628:     }
                    629: 
                    630: # ------------------------------------------------------------- Read hosts file
                    631:     my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
                    632:     
                    633:     my (%hostname,%hostdom,%hostrole,%spareid);
                    634:     while (my $configline=<$config>) {
1.49      albertel  635: 	next if ($configline =~ /^(\#|\s*\$)/);
1.46      albertel  636: 	my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline);
                    637: 	if ($id && $domain && $role && $name && $ip) {
                    638: 	    $hostname{$id}=$name;
                    639: 	    $hostdom{$id}=$domain;
                    640: 	    $hostrole{$id}=$role;
                    641: 	}
                    642:     }
                    643:     undef $config;
                    644: 
                    645: # ------------------------------------------------------ Read spare server file
                    646:     $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
                    647:     
                    648:     while (my $configline=<$config>) {
                    649: 	chomp($configline);
                    650: 	if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
                    651: 	    $spareid{$configline}=1;
                    652: 	}
                    653:     }
                    654:     undef $config;
                    655: 
                    656: # ---------------------------------------------------------------- Start report
                    657: 
                    658:     $errors=0;
                    659:     $warnings=0;
                    660:     $notices=0;
                    661: 
                    662: 	
1.49      albertel  663:     my $fh;
                    664:     if (!$justcheckdaemons && !$justcheckconnections) {
                    665: 	$fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);
                    666: 
                    667: 	&log_machine_info($fh);
                    668: 	&clean_tmp($fh);
                    669: 	&clean_lonIDs($fh);
                    670: 	&check_httpd_logs($fh);
                    671: 	&rotate_lonnet_logs($fh);
                    672:     }
                    673:     if (!$justcheckconnections) {
                    674: 	&checkon_daemon($fh,'lonsql',200000);
                    675: 	&checkon_daemon($fh,'lond',40000,1);
1.50      albertel  676: 	my $args='new';
                    677: 	if ($oldlonc) { $args = ''; }
                    678: 	&checkon_daemon($fh,'lonc',40000,1,$args);
1.49      albertel  679: 	&checkon_daemon($fh,'lonhttpd',40000);
                    680:     }
                    681:     if (!$justcheckdaemons) {
                    682: 	&test_connections($fh,\%hostname);
                    683:     }
                    684:     if (!$justcheckdaemons && !$justcheckconnections) {
                    685: 	&check_delayed_msg($fh);
                    686: 	&finish_logging($fh);
                    687: 	&log_simplestatus();
                    688: 	
                    689: 	if ($totalcount>200 && !$noemail) { &send_mail(); }
                    690:     }
1.46      albertel  691: }
                    692: 
                    693: &main();
1.1       albertel  694: 1;
                    695: 
                    696: 
                    697: 
                    698: 
                    699: 
                    700: 
                    701: 
                    702: 

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