Annotation of loncom/loncron, revision 1.62

1.1       albertel    1: #!/usr/bin/perl
                      2: 
1.47      albertel    3: # Housekeeping program, started by cron, loncontrol and loncron.pl
                      4: #
1.62    ! albertel    5: # $Id: loncron,v 1.61 2005/10/06 20:35:50 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");
1.60      albertel   95:     sleep 1;
1.42      albertel   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($daemonpid);
1.62    ! albertel  101: 	if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
1.42      albertel  102: 	    return 1;
                    103: 	} else {
                    104: 	    return 0;
                    105: 	}
                    106:     }
1.48      albertel  107:     &log($fh,"<p>Seems like that did not work!</p>");
1.42      albertel  108:     $errors++;
                    109:     return 0;
                    110: }
                    111: 
                    112: sub checkon_daemon {
1.59      albertel  113:     my ($fh,$daemon,$maxsize,$send,$args)=@_;
1.42      albertel  114: 
1.48      albertel  115:     &log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>');
1.57      albertel  116:     printf("%-15s ",$daemon);
1.42      albertel  117:     if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
                    118: 	open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
1.46      albertel  119: 	while (my $line=<DFH>) { 
                    120: 	    &log($fh,"$line");
1.42      albertel  121: 	    if ($line=~/INFO/) { $notices++; }
                    122: 	    if ($line=~/WARNING/) { $notices++; }
                    123: 	    if ($line=~/CRITICAL/) { $warnings++; }
                    124: 	};
                    125: 	close (DFH);
                    126:     }
1.48      albertel  127:     &log($fh,"</tt></p>");
1.42      albertel  128:     
                    129:     my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid";
                    130:     
                    131:     my $restartflag=1;
1.46      albertel  132:     my $daemonpid;
1.42      albertel  133:     if (-e $pidfile) {
                    134: 	my $lfh=IO::File->new("$pidfile");
1.46      albertel  135: 	$daemonpid=<$lfh>;
1.42      albertel  136: 	chomp($daemonpid);
1.62    ! albertel  137: 	if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
1.46      albertel  138: 	    &log($fh,"<h3>$daemon at pid $daemonpid responding");
1.59      albertel  139: 	    if ($send) { &log($fh,", sending $send"); }
1.46      albertel  140: 	    &log($fh,"</h3>");
1.59      albertel  141: 	    if ($send eq 'USR1') { kill USR1 => $daemonpid; }
                    142: 	    if ($send eq 'USR2') { kill USR2 => $daemonpid; }
1.42      albertel  143: 	    $restartflag=0;
1.59      albertel  144: 	    if ($send eq 'USR2') {
                    145: 		print "reloaded\n";
                    146: 	    } else {
                    147: 		print "running\n";
                    148: 	    }
1.42      albertel  149: 	} else {
                    150: 	    $errors++;
1.46      albertel  151: 	    &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.42      albertel  152: 	    $restartflag=1;
1.46      albertel  153: 	    &log($fh,"<h3>Decided to clean up stale .pid file and restart $daemon</h3>");
1.42      albertel  154: 	}
                    155:     }
                    156:     if ($restartflag==1) {
                    157: 	$simplestatus{$daemon}='off';
                    158: 	$errors++;
1.57      albertel  159: 	my $kadaemon=$daemon;
                    160: 	if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; }
1.46      albertel  161: 	&log($fh,'<br><font color="red">Killall '.$daemon.': '.
1.57      albertel  162: 	    `killall $kadaemon 2>&1`.' - ');
1.60      albertel  163: 	sleep 1;
1.46      albertel  164: 	&log($fh,unlink($pidfile).' - '.
1.57      albertel  165: 	    `killall -9 $kadaemon 2>&1`.
1.46      albertel  166: 	    '</font><br>');
                    167: 	&log($fh,"<h3>$daemon not running, trying to start</h3>");
1.42      albertel  168: 	
1.50      albertel  169: 	if (&start_daemon($fh,$daemon,$pidfile,$args)) {
1.46      albertel  170: 	    &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
1.42      albertel  171: 	    $simplestatus{$daemon}='restarted';
                    172: 	    print "started\n";
                    173: 	} else {
                    174: 	    $errors++;
1.46      albertel  175: 	    &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.48      albertel  176: 	    &log($fh,"<p>Give it one more try ...</p>");
1.42      albertel  177: 	    print " ";
1.50      albertel  178: 	    if (&start_daemon($fh,$daemon,$pidfile,$args)) {
1.46      albertel  179: 		&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
1.42      albertel  180: 		$simplestatus{$daemon}='restarted';
                    181: 		print "started\n";
                    182: 	    } else {
                    183: 		print " failed\n";
                    184: 		$simplestatus{$daemon}='failed';
                    185: 		$errors++; $errors++;
1.46      albertel  186: 		&log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.48      albertel  187: 		&log($fh,"<p>Unable to start $daemon</p>");
1.42      albertel  188: 	    }
                    189: 	}
                    190: 
                    191: 	if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
1.46      albertel  192: 	    &log($fh,"<p><pre>");
1.42      albertel  193: 	    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|");
1.46      albertel  194: 	    while (my $line=<DFH>) { 
                    195: 		&log($fh,"$line");
1.42      albertel  196: 		if ($line=~/WARNING/) { $notices++; }
                    197: 		if ($line=~/CRITICAL/) { $notices++; }
                    198: 	    };
                    199: 	    close (DFH);
1.48      albertel  200: 	    &log($fh,"</pre></p>");
1.42      albertel  201: 	}
                    202:     }
                    203:     
1.46      albertel  204:     my $fname="$perlvar{'lonDaemons'}/logs/$daemon.log";
1.42      albertel  205:     
                    206:     my ($dev,$ino,$mode,$nlink,
                    207: 	$uid,$gid,$rdev,$size,
                    208: 	$atime,$mtime,$ctime,
                    209: 	$blksize,$blocks)=stat($fname);
                    210:     
                    211:     if ($size>$maxsize) {
1.48      albertel  212: 	&log($fh,"<p>Rotating logs ...</p>");
1.42      albertel  213: 	rename("$fname.2","$fname.3");
                    214: 	rename("$fname.1","$fname.2");
                    215: 	rename("$fname","$fname.1");
                    216:     }
                    217: 
                    218:     &errout($fh);
                    219: }
1.1       albertel  220: 
1.46      albertel  221: # --------------------------------------------------------------------- Machine
                    222: sub log_machine_info {
                    223:     my ($fh)=@_;
1.48      albertel  224:     &log($fh,'<hr /><a name="machine" /><h2>Machine Information</h2>');
1.46      albertel  225:     &log($fh,"<h3>loadavg</h3>");
                    226: 	
                    227:     open (LOADAVGH,"/proc/loadavg");
                    228:     my $loadavg=<LOADAVGH>;
                    229:     close (LOADAVGH);
                    230:     
                    231:     &log($fh,"<tt>$loadavg</tt>");
                    232:     
                    233:     my @parts=split(/\s+/,$loadavg);
                    234:     if ($parts[1]>4.0) {
                    235: 	$errors++;
                    236:     } elsif ($parts[1]>2.0) {
                    237: 	$warnings++;
                    238:     } elsif ($parts[1]>1.0) {
                    239: 	$notices++;
                    240:     }
1.13      harris41  241: 
1.46      albertel  242:     &log($fh,"<h3>df</h3>");
                    243:     &log($fh,"<pre>");
1.14      harris41  244: 
1.46      albertel  245:     open (DFH,"df|");
                    246:     while (my $line=<DFH>) { 
1.48      albertel  247: 	&log($fh,&encode_entities($line,'<>&"')); 
1.46      albertel  248: 	@parts=split(/\s+/,$line);
                    249: 	my $usage=$parts[4];
                    250: 	$usage=~s/\W//g;
                    251: 	if ($usage>90) { 
                    252: 	    $warnings++;
                    253: 	    $notices++; 
                    254: 	} elsif ($usage>80) {
                    255: 	    $warnings++;
                    256: 	} elsif ($usage>60) {
                    257: 	    $notices++;
1.31      albertel  258: 	}
1.46      albertel  259: 	if ($usage>95) { $warnings++; $warnings++; $simplestatus{'diskfull'}++; }
1.1       albertel  260:     }
1.46      albertel  261:     close (DFH);
                    262:     &log($fh,"</pre>");
1.1       albertel  263: 
                    264: 
1.46      albertel  265:     &log($fh,"<h3>ps</h3>");
                    266:     &log($fh,"<pre>");
                    267:     my $psproc=0;
1.1       albertel  268: 
1.53      albertel  269:     open (PSH,"ps aux --cols 140 |");
1.46      albertel  270:     while (my $line=<PSH>) { 
1.48      albertel  271: 	&log($fh,&encode_entities($line,'<>&"')); 
1.46      albertel  272: 	$psproc++;
                    273:     }
                    274:     close (PSH);
                    275:     &log($fh,"</pre>");
1.1       albertel  276: 
1.46      albertel  277:     if ($psproc>200) { $notices++; }
                    278:     if ($psproc>250) { $notices++; }
1.1       albertel  279: 
1.61      albertel  280:     &log($fh,"<h3>distprobe</h3>");
                    281:     &log($fh,"<pre>");
                    282:     open(DSH,"$perlvar{'lonDaemons'}/distprobe |");
                    283:     while (my $line=<DSH>) { 
                    284: 	&log($fh,&encode_entities($line,'<>&"')); 
                    285: 	$psproc++;
                    286:     }
                    287:     close(DSH);
                    288:     &log($fh,"</pre>");
                    289: 
1.46      albertel  290:     &errout($fh);
                    291: }
1.1       albertel  292: 
1.46      albertel  293: sub start_logging {
                    294:     my ($hostdom,$hostrole,$hostname,$spareid)=@_;
1.43      albertel  295:     my $fh=IO::File->new(">$statusdir/newstatus.html");
                    296:     my %simplestatus=();
1.46      albertel  297:     my $now=time;
                    298:     my $date=localtime($now);
1.43      albertel  299:     
1.46      albertel  300: 
                    301:     &log($fh,(<<ENDHEADERS));
1.1       albertel  302: <html>
                    303: <head>
                    304: <title>LON Status Report $perlvar{'lonHostID'}</title>
                    305: </head>
1.3       www       306: <body bgcolor="#AAAAAA">
1.48      albertel  307: <a name="top" />
1.1       albertel  308: <h1>LON Status Report $perlvar{'lonHostID'}</h1>
                    309: <h2>$date ($now)</h2>
                    310: <ol>
1.48      albertel  311: <li><a href="#configuration">Configuration</a></li>
                    312: <li><a href="#machine">Machine Information</a></li>
                    313: <li><a href="#tmp">Temporary Files</a></li>
                    314: <li><a href="#tokens">Session Tokens</a></li>
                    315: <li><a href="#httpd">httpd</a></li>
                    316: <li><a href="#lonsql">lonsql</a></li>
                    317: <li><a href="#lond">lond</a></li>
                    318: <li><a href="#lonc">lonc</a></li>
                    319: <li><a href="#lonhttpd">lonhttpd</a></li>
                    320: <li><a href="#lonnet">lonnet</a></li>
                    321: <li><a href="#connections">Connections</a></li>
                    322: <li><a href="#delayed">Delayed Messages</a></li>
                    323: <li><a href="#errcount">Error Count</a></li>
1.1       albertel  324: </ol>
1.48      albertel  325: <hr />
                    326: <a name="configuration" />
1.1       albertel  327: <h2>Configuration</h2>
                    328: <h3>PerlVars</h3>
1.48      albertel  329: <table border="2">
1.1       albertel  330: ENDHEADERS
                    331: 
1.46      albertel  332:     foreach my $varname (sort(keys(%perlvar))) {
1.48      albertel  333: 	&log($fh,"<tr><td>$varname</td><td>".
                    334: 	     &encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n");
1.43      albertel  335:     }
1.48      albertel  336:     &log($fh,"</table><h3>Hosts</h3><table border='2'>");
1.46      albertel  337:     foreach my $id (sort(keys(%{$hostname}))) {
                    338: 	&log($fh,
                    339: 	    "<tr><td>$id</td><td>".$hostdom->{$id}.
                    340: 	    "</td><td>".$hostrole->{$id}.
                    341: 	    "</td><td>".$hostname->{$id}."</td></tr>\n");
                    342:     }
                    343:     &log($fh,"</table><h3>Spare Hosts</h3><ol>");
                    344:     foreach my $id (sort(keys(%{$spareid}))) {
1.48      albertel  345: 	&log($fh,"<li>$id\n</li>");
1.43      albertel  346:     }
1.46      albertel  347:     &log($fh,"</ol>\n");
                    348:     return $fh;
                    349: }
1.11      www       350: 
                    351: # --------------------------------------------------------------- clean out tmp
1.46      albertel  352: sub clean_tmp {
                    353:     my ($fh)=@_;
1.48      albertel  354:     &log($fh,'<hr /><a name="tmp" /><h2>Temporary Files</h2>');
1.46      albertel  355:     my $cleaned=0;
                    356:     my $old=0;
                    357:     while (my $fname=<$perlvar{'lonDaemons'}/tmp/*>) {
1.43      albertel  358: 	my ($dev,$ino,$mode,$nlink,
                    359: 	    $uid,$gid,$rdev,$size,
                    360: 	    $atime,$mtime,$ctime,
                    361: 	    $blksize,$blocks)=stat($fname);
1.46      albertel  362: 	my $now=time;
                    363: 	my $since=$now-$mtime;
1.43      albertel  364: 	if ($since>$perlvar{'lonExpire'}) {
1.46      albertel  365: 	    my $line='';
1.43      albertel  366: 	    if (open(PROBE,$fname)) {
                    367: 		$line=<PROBE>;
                    368: 		close(PROBE);
                    369: 	    }
                    370: 	    unless ($line=~/^CHECKOUTTOKEN\&/) {
                    371: 		$cleaned++;
                    372: 		unlink("$fname");
                    373: 	    } else {
                    374: 		if ($since>365*$perlvar{'lonExpire'}) {
                    375: 		    $cleaned++;
                    376: 		    unlink("$fname");
                    377: 		} else { $old++; }
                    378: 	    }
                    379: 	}
                    380:     }
1.46      albertel  381:     &log($fh,"Cleaned up ".$cleaned." files (".$old." old checkout tokens).");
                    382: }
1.11      www       383: 
                    384: # ------------------------------------------------------------ clean out lonIDs
1.46      albertel  385: sub clean_lonIDs {
                    386:     my ($fh)=@_;
1.48      albertel  387:     &log($fh,'<hr /><a name="tokens" /><h2>Session Tokens</h2>');
1.46      albertel  388:     my $cleaned=0;
                    389:     my $active=0;
                    390:     while (my $fname=<$perlvar{'lonIDsDir'}/*>) {
1.43      albertel  391: 	my ($dev,$ino,$mode,$nlink,
                    392: 	    $uid,$gid,$rdev,$size,
                    393: 	    $atime,$mtime,$ctime,
                    394: 	    $blksize,$blocks)=stat($fname);
1.46      albertel  395: 	my $now=time;
                    396: 	my $since=$now-$mtime;
1.43      albertel  397: 	if ($since>$perlvar{'lonExpire'}) {
                    398: 	    $cleaned++;
1.46      albertel  399: 	    &log($fh,"Unlinking $fname<br>");
1.43      albertel  400: 	    unlink("$fname");
                    401: 	} else {
                    402: 	    $active++;
                    403: 	}
1.46      albertel  404:     }
1.48      albertel  405:     &log($fh,"<p>Cleaned up ".$cleaned." stale session token(s).</p>");
1.46      albertel  406:     &log($fh,"<h3>$active open session(s)</h3>");
                    407: }
1.43      albertel  408: 
1.11      www       409: 
1.1       albertel  410: # ----------------------------------------------------------------------- httpd
1.46      albertel  411: sub check_httpd_logs {
                    412:     my ($fh)=@_;
1.48      albertel  413:     &log($fh,'<hr /><a name="httpd" /><h2>httpd</h2><h3>Access Log</h3><pre>');
1.43      albertel  414:     
                    415:     open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
1.48      albertel  416:     while (my $line=<DFH>) { &log($fh,&encode_entities($line,'<>&"')) };
1.43      albertel  417:     close (DFH);
1.46      albertel  418: 	
                    419:     &log($fh,"</pre><h3>Error Log</h3><pre>");
                    420: 	
1.43      albertel  421:     open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
1.46      albertel  422:     while (my $line=<DFH>) { 
                    423: 	&log($fh,"$line");
1.43      albertel  424: 	if ($line=~/\[error\]/) { $notices++; } 
1.46      albertel  425:     }
1.43      albertel  426:     close (DFH);
1.46      albertel  427:     &log($fh,"</pre>");
1.43      albertel  428:     &errout($fh);
1.46      albertel  429: }
1.1       albertel  430: 
                    431: # ---------------------------------------------------------------------- lonnet
                    432: 
1.48      albertel  433: sub rotate_lonnet_logs {
1.46      albertel  434:     my ($fh)=@_;
1.48      albertel  435:     &log($fh,'<hr /><a name="lonnet" /><h2>lonnet</h2><h3>Temp Log</h3><pre>');
1.43      albertel  436:     print "checking logs\n";
                    437:     if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
                    438: 	open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
1.46      albertel  439: 	while (my $line=<DFH>) { 
1.48      albertel  440: 	    &log($fh,&encode_entities($line,'<>&"'));
1.46      albertel  441: 	}
1.43      albertel  442: 	close (DFH);
                    443:     }
1.46      albertel  444:     &log($fh,"</pre><h3>Perm Log</h3><pre>");
1.43      albertel  445:     
                    446:     if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
                    447: 	open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
1.46      albertel  448: 	while (my $line=<DFH>) { 
1.48      albertel  449: 	    &log($fh,&encode_entities($line,'<>&"'));
1.46      albertel  450: 	}
1.43      albertel  451: 	close (DFH);
1.46      albertel  452:     } else { &log($fh,"No perm log\n") }
1.43      albertel  453: 
1.46      albertel  454:     my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
1.43      albertel  455: 
                    456:     my ($dev,$ino,$mode,$nlink,
                    457: 	$uid,$gid,$rdev,$size,
                    458: 	$atime,$mtime,$ctime,
                    459: 	$blksize,$blocks)=stat($fname);
                    460: 
                    461:     if ($size>40000) {
1.48      albertel  462: 	&log($fh,"<p>Rotating logs ...</p>");
1.43      albertel  463: 	rename("$fname.2","$fname.3");
                    464: 	rename("$fname.1","$fname.2");
                    465: 	rename("$fname","$fname.1");
                    466:     }
1.1       albertel  467: 
1.46      albertel  468:     &log($fh,"</pre>");
1.43      albertel  469:     &errout($fh);
1.46      albertel  470: }
                    471: 
1.43      albertel  472: # ----------------------------------------------------------------- Connections
1.46      albertel  473: sub test_connections {
                    474:     my ($fh,$hostname)=@_;
1.48      albertel  475:     &log($fh,'<hr /><a name="connections" /><h2>Connections</h2>');
1.43      albertel  476:     print "testing connections\n";
1.48      albertel  477:     &log($fh,"<table border='2'>");
1.49      albertel  478:     my ($good,$bad)=(0,0);
1.46      albertel  479:     foreach my $tryserver (sort(keys(%{$hostname}))) {
1.43      albertel  480: 	print(".");
1.46      albertel  481: 	my $result;
1.55      albertel  482: 	my $answer=reply("ping",$tryserver);
1.43      albertel  483: 	if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
                    484: 	    $result="<b>ok</b>";
1.49      albertel  485: 	    $good++;
1.43      albertel  486: 	} else {
                    487: 	    $result=$answer;
                    488: 	    $warnings++;
1.49      albertel  489: 	    if ($answer eq 'con_lost') {
                    490: 		$bad++;
                    491: 		$warnings++;
1.50      albertel  492: 	    } else {
                    493: 		$good++; #self connection
1.49      albertel  494: 	    }
1.43      albertel  495: 	}
                    496: 	if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }
1.46      albertel  497: 	&log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n");
1.1       albertel  498:     }
1.46      albertel  499:     &log($fh,"</table>");
1.49      albertel  500:     print "\n$good good, $bad bad connections\n";
1.43      albertel  501:     &errout($fh);
1.46      albertel  502: }
                    503: 
                    504: 
1.1       albertel  505: # ------------------------------------------------------------ Delayed messages
1.46      albertel  506: sub check_delayed_msg {
                    507:     my ($fh)=@_;
1.48      albertel  508:     &log($fh,'<hr /><a name="delayed" /><h2>Delayed Messages</h2>');
1.43      albertel  509:     print "checking buffers\n";
1.46      albertel  510:     
                    511:     &log($fh,'<h3>Scanning Permanent Log</h3>');
1.1       albertel  512: 
1.46      albertel  513:     my $unsend=0;
1.1       albertel  514: 
1.46      albertel  515:     my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
                    516:     while (my $line=<$dfh>) {
                    517: 	my ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
                    518: 	if ($sdf eq 'F') { 
                    519: 	    my $local=localtime($time);
                    520: 	    &log($fh,"<b>Failed: $time, $dserv, $dcmd</b><br>");
                    521: 	    $warnings++;
1.43      albertel  522: 	}
1.46      albertel  523: 	if ($sdf eq 'S') { $unsend--; }
                    524: 	if ($sdf eq 'D') { $unsend++; }
1.1       albertel  525:     }
1.46      albertel  526: 
1.48      albertel  527:     &log($fh,"<p>Total unsend messages: <b>$unsend</b></p>\n");
1.43      albertel  528:     $warnings=$warnings+5*$unsend;
1.1       albertel  529: 
1.43      albertel  530:     if ($unsend) { $simplestatus{'unsend'}=$unsend; }
1.48      albertel  531:     &log($fh,"<h3>Outgoing Buffer</h3>\n<pre>");
1.1       albertel  532: 
1.43      albertel  533:     open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
1.46      albertel  534:     while (my $line=<DFH>) { 
1.48      albertel  535: 	&log($fh,&encode_entities($line,'<>&"'));
1.46      albertel  536:     }
1.48      albertel  537:     &log($fh,"</pre>\n");
1.43      albertel  538:     close (DFH);
1.46      albertel  539: }
1.1       albertel  540: 
1.46      albertel  541: sub finish_logging {
                    542:     my ($fh)=@_;
1.48      albertel  543:     &log($fh,"<a name='errcount' />\n");
1.43      albertel  544:     $totalcount=$notices+4*$warnings+100*$errors;
                    545:     &errout($fh);
1.46      albertel  546:     &log($fh,"<h1>Total Error Count: $totalcount</h1>");
                    547:     my $now=time;
                    548:     my $date=localtime($now);
1.48      albertel  549:     &log($fh,"<hr />$date ($now)</body></html>\n");
1.43      albertel  550:     print "lon-status webpage updated\n";
                    551:     $fh->close();
1.46      albertel  552: 
                    553:     if ($errors) { $simplestatus{'errors'}=$errors; }
                    554:     if ($warnings) { $simplestatus{'warnings'}=$warnings; }
                    555:     if ($notices) { $simplestatus{'notices'}=$notices; }
                    556:     $simplestatus{'time'}=time;
1.1       albertel  557: }
                    558: 
1.46      albertel  559: sub log_simplestatus {
                    560:     rename ("$statusdir/newstatus.html","$statusdir/index.html");
                    561:     
1.43      albertel  562:     my $sfh=IO::File->new(">$statusdir/loncron_simple.txt");
                    563:     foreach (keys %simplestatus) {
                    564: 	print $sfh $_.'='.$simplestatus{$_}.'&';
                    565:     }
                    566:     print $sfh "\n";
                    567:     $sfh->close();
1.41      www       568: }
1.46      albertel  569: 
                    570: sub send_mail {
1.43      albertel  571:     print "sending mail\n";
1.46      albertel  572:     my $emailto="$perlvar{'lonAdmEMail'}";
1.54      www       573:     if ($totalcount>2500) {
1.43      albertel  574: 	$emailto.=",$perlvar{'lonSysEMail'}";
                    575:     }
1.46      albertel  576:     my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; 
1.52      albertel  577: 
1.58      albertel  578:     my $result=system("metasend -b -S 4000000 -t $emailto -s '$subj' -f $statusdir/index.html -m text/html >& /dev/null");
1.52      albertel  579:     if ($result != 0) {
                    580: 	$result=system("mail -s '$subj' $emailto < $statusdir/index.html");
                    581:     }
1.1       albertel  582: }
1.46      albertel  583: 
1.49      albertel  584: sub usage {
                    585:     print(<<USAGE);
                    586: loncron - housekeeping program that checks up on various parts of Lon-CAPA
                    587: 
                    588: Options:
                    589:    --help     Display help
                    590:    --oldlonc  When starting the lonc daemon use 'lonc' not 'loncnew'
                    591:    --noemail  Do not send the status email
                    592:    --justcheckconnections  Only check the current status of the lonc/d
                    593:                                 connections, do not send emails do not
                    594:                                 check if the daemons are running, do not
                    595:                                 generate lon-status
                    596:    --justcheckdaemons      Only check that all of the Lon-CAPA daemons are
                    597:                                 running, do not send emails do not
                    598:                                 check the lonc/d connections, do not
                    599:                                 generate lon-status
1.59      albertel  600:    --justreload            Only tell the daemons to reload the config files,
                    601: 				do not send emails do not
                    602:                                 check if the daemons are running, do not
                    603:                                 generate lon-status
1.49      albertel  604:                            
                    605: USAGE
                    606: }
                    607: 
1.46      albertel  608: # ================================================================ Main Program
                    609: sub main () {
1.59      albertel  610:     my ($oldlonc,$help,$justcheckdaemons,$noemail,$justcheckconnections,
                    611: 	$justreload);
1.49      albertel  612:     &GetOptions("help"                 => \$help,
                    613: 		"oldlonc"              => \$oldlonc,
                    614: 		"justcheckdaemons"     => \$justcheckdaemons,
                    615: 		"noemail"              => \$noemail,
1.59      albertel  616: 		"justcheckconnections" => \$justcheckconnections,
                    617: 		"justreload"           => \$justreload
1.49      albertel  618: 		);
                    619:     if ($help) { &usage(); return; }
1.46      albertel  620: # --------------------------------- Read loncapa_apache.conf and loncapa.conf
                    621:     my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
                    622:     %perlvar=%{$perlvarref};
                    623:     undef $perlvarref;
                    624:     delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
                    625:     delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
                    626: 
                    627: # --------------------------------------- Make sure that LON-CAPA is configured
                    628: # I only test for one thing here (lonHostID).  This is just a safeguard.
                    629:     if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) {
                    630: 	print("Unconfigured machine.\n");
                    631: 	my $emailto=$perlvar{'lonSysEMail'};
                    632: 	my $hostname=`/bin/hostname`;
                    633: 	chop $hostname;
                    634: 	$hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
                    635: 	my $subj="LON: Unconfigured machine $hostname";
                    636: 	system("echo 'Unconfigured machine $hostname.' |\
                    637:  mailto $emailto -s '$subj' > /dev/null");
                    638: 	exit 1;
                    639:     }
                    640: 
                    641: # ----------------------------- Make sure this process is running from user=www
                    642:     my $wwwid=getpwnam('www');
                    643:     if ($wwwid!=$<) {
                    644: 	print("User ID mismatch.  This program must be run as user 'www'\n");
                    645: 	my $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
                    646: 	my $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
                    647: 	system("echo 'User ID mismatch.  loncron must be run as user www.' |\
                    648:  mailto $emailto -s '$subj' > /dev/null");
                    649: 	exit 1;
                    650:     }
                    651: 
                    652: # ------------------------------------------------------------- Read hosts file
                    653:     my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
                    654:     
                    655:     my (%hostname,%hostdom,%hostrole,%spareid);
                    656:     while (my $configline=<$config>) {
1.49      albertel  657: 	next if ($configline =~ /^(\#|\s*\$)/);
1.56      albertel  658: 	my ($id,$domain,$role,$name)=split(/:/,$configline);
                    659: 	if ($id && $domain && $role && $name) {
                    660: 	    $name=~s/\s//g;
1.46      albertel  661: 	    $hostname{$id}=$name;
                    662: 	    $hostdom{$id}=$domain;
                    663: 	    $hostrole{$id}=$role;
                    664: 	}
                    665:     }
                    666:     undef $config;
                    667: 
                    668: # ------------------------------------------------------ Read spare server file
                    669:     $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
                    670:     
                    671:     while (my $configline=<$config>) {
                    672: 	chomp($configline);
                    673: 	if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
                    674: 	    $spareid{$configline}=1;
                    675: 	}
                    676:     }
                    677:     undef $config;
                    678: 
                    679: # ---------------------------------------------------------------- Start report
                    680: 
                    681:     $errors=0;
                    682:     $warnings=0;
                    683:     $notices=0;
                    684: 
                    685: 	
1.49      albertel  686:     my $fh;
1.59      albertel  687:     if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
1.49      albertel  688: 	$fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);
                    689: 
                    690: 	&log_machine_info($fh);
                    691: 	&clean_tmp($fh);
                    692: 	&clean_lonIDs($fh);
                    693: 	&check_httpd_logs($fh);
                    694: 	&rotate_lonnet_logs($fh);
                    695:     }
1.59      albertel  696:     if (!$justcheckconnections && !$justreload) {
1.49      albertel  697: 	&checkon_daemon($fh,'lonsql',200000);
1.59      albertel  698: 	&checkon_daemon($fh,'lond',40000,'USR1');
1.50      albertel  699: 	my $args='new';
                    700: 	if ($oldlonc) { $args = ''; }
1.59      albertel  701: 	&checkon_daemon($fh,'lonc',40000,'USR1',$args);
1.49      albertel  702: 	&checkon_daemon($fh,'lonhttpd',40000);
1.57      albertel  703: 	&checkon_daemon($fh,'lonmemcached',40000);
1.49      albertel  704:     }
1.59      albertel  705:     if ($justreload) {
                    706: 	&checkon_daemon($fh,'lond',40000,'USR2');
                    707: 	my $args='new';
                    708: 	if ($oldlonc) { $args = ''; }
                    709: 	&checkon_daemon($fh,'lonc',40000,'USR2',$args);
                    710:     }
                    711:     if (!$justcheckdaemons && !$justreload) {
1.49      albertel  712: 	&test_connections($fh,\%hostname);
                    713:     }
1.59      albertel  714:     if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
1.49      albertel  715: 	&check_delayed_msg($fh);
                    716: 	&finish_logging($fh);
                    717: 	&log_simplestatus();
                    718: 	
                    719: 	if ($totalcount>200 && !$noemail) { &send_mail(); }
                    720:     }
1.46      albertel  721: }
                    722: 
                    723: &main();
1.1       albertel  724: 1;
                    725: 
                    726: 
                    727: 
                    728: 
                    729: 
                    730: 
                    731: 
                    732: 

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