File:  [LON-CAPA] / loncom / loncron
Revision 1.39: download - view: text, annotated - select for diffs
Tue Jun 10 22:07:26 2003 UTC (20 years, 10 months ago) by albertel
Branches: MAIN
CVS tags: version_0_99_3, version_0_99_2, HEAD
- always start lonsql

    1: #!/usr/bin/perl
    2: 
    3: # The LearningOnline Network
    4: # Housekeeping program, started by cron
    5: #
    6: # (TCP networking package
    7: # 6/1/99,6/2,6/10,6/11,6/12,6/14,6/26,6/28,6/29,6/30,
    8: # 7/1,7/2,7/9,7/10,7/12 Gerd Kortemeyer)
    9: #
   10: # 7/14,7/15,7/19,7/21,7/22,11/18,
   11: # 2/8 Gerd Kortemeyer
   12: # 12/23 Gerd Kortemeyer
   13: # YEAR=2001
   14: # 09/04,09/06,11/26 Gerd Kortemeyer
   15: 
   16: $|=1;
   17: 
   18: use lib '/home/httpd/lib/perl/';
   19: use LONCAPA::Configuration;
   20: 
   21: use IO::File;
   22: use IO::Socket;
   23: 
   24: # -------------------------------------------------- Non-critical communication
   25: sub reply {
   26:     my ($cmd,$server)=@_;
   27:     my $peerfile="$perlvar{'lonSockDir'}/$server";
   28:     my $client=IO::Socket::UNIX->new(Peer    =>"$peerfile",
   29:                                      Type    => SOCK_STREAM,
   30:                                      Timeout => 10)
   31:        or return "con_lost";
   32:     print $client "$cmd\n";
   33:     my $answer=<$client>;
   34:     chomp($answer);
   35:     if (!$answer) { $answer="con_lost"; }
   36:     return $answer;
   37: }
   38: 
   39: # --------------------------------------------------------- Output error status
   40: 
   41: sub errout {
   42:    my $fh=shift;
   43:    print $fh (<<ENDERROUT);
   44:      <p><table border=2 bgcolor="#CCCCCC">
   45:      <tr><td>Notices</td><td>$notices</td></tr>
   46:      <tr><td>Warnings</td><td>$warnings</td></tr>
   47:      <tr><td>Errors</td><td>$errors</td></tr>
   48:      </table><p><a href="#top">Top</a><p>
   49: ENDERROUT
   50: }
   51: 
   52: # ================================================================ Main Program
   53: 
   54: # --------------------------------- Read loncapa_apache.conf and loncapa.conf
   55: my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
   56: %perlvar=%{$perlvarref};
   57: undef $perlvarref;
   58: delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
   59: delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
   60: 
   61: # --------------------------------------- Make sure that LON-CAPA is configured
   62: # I only test for one thing here (lonHostID).  This is just a safeguard.
   63: if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) {
   64:    print("Unconfigured machine.\n");
   65:    $emailto=$perlvar{'lonSysEMail'};
   66:    $hostname=`/bin/hostname`;
   67:    chop $hostname;
   68:    $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
   69:    $subj="LON: Unconfigured machine $hostname";
   70:    system("echo 'Unconfigured machine $hostname.' |\
   71:  mailto $emailto -s '$subj' > /dev/null");
   72:     exit 1;
   73: }
   74: 
   75: # ----------------------------- Make sure this process is running from user=www
   76: my $wwwid=getpwnam('www');
   77: if ($wwwid!=$<) {
   78:    print("User ID mismatch.  This program must be run as user 'www'\n");
   79:    $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
   80:    $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
   81:    system("echo 'User ID mismatch.  loncron must be run as user www.' |\
   82:  mailto $emailto -s '$subj' > /dev/null");
   83:    exit 1;
   84: }
   85: 
   86: # ------------------------------------------------------------- Read hosts file
   87: {
   88:     my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
   89: 
   90:     while (my $configline=<$config>) {
   91: 	my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline);
   92: 	if ($id && $domain && $role && $name && $ip) {
   93: 	    $hostname{$id}=$name;
   94: 	    $hostdom{$id}=$domain;
   95: 	    $hostip{$id}=$ip;
   96: 	    $hostrole{$id}=$role;
   97: 	    if ($domdescr) { $domaindescription{$domain}=$domdescr; }
   98: 	    if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) {
   99: 		$libserv{$id}=$name;
  100: 	    }
  101: 	} else {
  102: 	    if ($configline) {
  103: #		&logthis("Skipping hosts.tab line -$configline-");
  104: 	    }
  105: 	}
  106:     }
  107: }
  108: 
  109: # ------------------------------------------------------ Read spare server file
  110: {
  111:     my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
  112: 
  113:     while (my $configline=<$config>) {
  114:        chomp($configline);
  115:        if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
  116:           $spareid{$configline}=1;
  117:        }
  118:     }
  119: }
  120: 
  121: # ---------------------------------------------------------------- Start report
  122: 
  123: $statusdir="/home/httpd/html/lon-status";
  124: 
  125: $errors=0;
  126: $warnings=0;
  127: $notices=0;
  128: 
  129: $now=time;
  130: $date=localtime($now);
  131: 
  132: {
  133: my $fh=IO::File->new(">$statusdir/newstatus.html");
  134: 
  135: print $fh (<<ENDHEADERS);
  136: <html>
  137: <head>
  138: <title>LON Status Report $perlvar{'lonHostID'}</title>
  139: </head>
  140: <body bgcolor="#AAAAAA">
  141: <a name="top">
  142: <h1>LON Status Report $perlvar{'lonHostID'}</h1>
  143: <h2>$date ($now)</h2>
  144: <ol>
  145: <li><a href="#configuration">Configuration</a>
  146: <li><a href="#machine">Machine Information</a>
  147: <li><a href="#tmp">Temporary Files</a>
  148: <li><a href="#tokens">Session Tokens</a>
  149: <li><a href="#httpd">httpd</a>
  150: <li><a href="#lonsql">lonsql</a>
  151: <li><a href="#lond">lond</a>
  152: <li><a href="#lonc">lonc</a>
  153: <li><a href="#lonhttpd">lonhttpd</a>
  154: <li><a href="#lonnet">lonnet</a>
  155: <li><a href="#connections">Connections</a>
  156: <li><a href="#delayed">Delayed Messages</a>
  157: <li><a href="#errcount">Error Count</a>
  158: </ol>
  159: <hr>
  160: <a name="configuration">
  161: <h2>Configuration</h2>
  162: <h3>PerlVars</h3>
  163: <table border=2>
  164: ENDHEADERS
  165: 
  166: foreach $varname (sort(keys(%perlvar))) {
  167:     print $fh "<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n";
  168: }
  169: print $fh "</table><h3>Hosts</h3><table border=2>";
  170: foreach $id (sort(keys(%hostname))) {
  171:     print $fh 
  172: 	"<tr><td>$id</td><td>$hostdom{$id}</td><td>$hostrole{$id}</td>";
  173:     print $fh "<td>$hostname{$id}</td><td>$hostip{$id}</td></tr>\n";
  174: }
  175: print $fh "</table><h3>Spare Hosts</h3><ol>";
  176: foreach $id (sort(keys(%spareid))) {
  177:     print $fh "<li>$id\n";
  178: }
  179: 
  180: print $fh "</ol>\n";
  181: 
  182: # --------------------------------------------------------------------- Machine
  183: 
  184: print $fh '<hr><a name="machine"><h2>Machine Information</h2>';
  185: print $fh "<h3>loadavg</h3>";
  186: 
  187: open (LOADAVGH,"/proc/loadavg");
  188: $loadavg=<LOADAVGH>;
  189: close (LOADAVGH);
  190: 
  191: print $fh "<tt>$loadavg</tt>";
  192: 
  193: @parts=split(/\s+/,$loadavg);
  194: if ($parts[1]>4.0) {
  195:     $errors++;
  196: } elsif ($parts[1]>2.0) {
  197:     $warnings++;
  198: } elsif ($parts[1]>1.0) {
  199:     $notices++;
  200: }
  201: 
  202: print $fh "<h3>df</h3>";
  203: print $fh "<pre>";
  204: 
  205: open (DFH,"df|");
  206: while ($line=<DFH>) { 
  207:    print $fh "$line"; 
  208:    @parts=split(/\s+/,$line);
  209:    $usage=$parts[4];
  210:    $usage=~s/\W//g;
  211:    if ($usage>90) { 
  212:       $warnings++;
  213:       $notices++; 
  214:    } elsif ($usage>80) {
  215:       $warnings++;
  216:    } elsif ($usage>60) {
  217:       $notices++;
  218:    }
  219:    if ($usage>95) { $warnings++; $warnings++ }
  220: }
  221: close (DFH);
  222: print $fh "</pre>";
  223: 
  224: 
  225: print $fh "<h3>ps</h3>";
  226: print $fh "<pre>";
  227: $psproc=0;
  228: 
  229: open (PSH,"ps -aux|");
  230: while ($line=<PSH>) { 
  231:    print $fh "$line"; 
  232:    $psproc++;
  233: }
  234: close (PSH);
  235: print $fh "</pre>";
  236: 
  237: if ($psproc>200) { $notices++; }
  238: if ($psproc>250) { $notices++; }
  239: 
  240: &errout($fh);
  241: 
  242: # --------------------------------------------------------------- clean out tmp
  243: print $fh '<hr><a name="tmp"><h2>Temporary Files</h2>';
  244: $cleaned=0;
  245: $old=0;
  246: while ($fname=<$perlvar{'lonDaemons'}/tmp/*>) {
  247:                           my ($dev,$ino,$mode,$nlink,
  248:                               $uid,$gid,$rdev,$size,
  249:                               $atime,$mtime,$ctime,
  250:                               $blksize,$blocks)=stat($fname);
  251:                           $now=time;
  252:                           $since=$now-$mtime;
  253:                           if ($since>$perlvar{'lonExpire'}) {
  254:                               $line='';
  255:                               if (open(PROBE,$fname)) {
  256: 				  $line=<PROBE>;
  257:                                   close(PROBE);
  258: 			      }
  259: 			      unless ($line=~/^CHECKOUTTOKEN\&/) {
  260:                                  $cleaned++;
  261:                                  unlink("$fname");
  262: 			      } else {
  263: 				  if ($since>365*$perlvar{'lonExpire'}) {
  264:                                      $cleaned++;
  265:                                      unlink("$fname");
  266: 				 } else { $old++; }
  267:                               }
  268:                           }
  269:     
  270: }
  271: print $fh "Cleaned up ".$cleaned." files (".$old." old checkout tokens).";
  272: 
  273: # ------------------------------------------------------------ clean out lonIDs
  274: print $fh '<hr><a name="tokens"><h2>Session Tokens</h2>';
  275: $cleaned=0;
  276: $active=0;
  277: while ($fname=<$perlvar{'lonIDsDir'}/*>) {
  278:                           my ($dev,$ino,$mode,$nlink,
  279:                               $uid,$gid,$rdev,$size,
  280:                               $atime,$mtime,$ctime,
  281:                               $blksize,$blocks)=stat($fname);
  282:                           $now=time;
  283:                           $since=$now-$mtime;
  284:                           if ($since>$perlvar{'lonExpire'}) {
  285:                               $cleaned++;
  286:                               print $fh "Unlinking $fname<br>";
  287:                               unlink("$fname");
  288:                           } else {
  289:                               $active++;
  290:                           }
  291:     
  292: }
  293: print $fh "<p>Cleaned up ".$cleaned." stale session token(s).";
  294: print $fh "<h3>$active open session(s)</h3>";
  295: 
  296: # ----------------------------------------------------------------------- httpd
  297: 
  298: print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>';
  299: 
  300: open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
  301: while ($line=<DFH>) { print $fh "$line" };
  302: close (DFH);
  303: 
  304: print $fh "</pre><h3>Error Log</h3><pre>";
  305: 
  306: open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
  307: while ($line=<DFH>) { 
  308:    print $fh "$line";
  309:    if ($line=~/\[error\]/) { $notices++; } 
  310: };
  311: close (DFH);
  312: print $fh "</pre>";
  313: &errout($fh);
  314: 
  315: 
  316: # ---------------------------------------------------------------------- lonsql
  317: 
  318: my $restartflag=1;
  319:     print $fh '<hr><a name="lonsql"><h2>lonsql</h2><h3>Log</h3><pre>';
  320:     print "lonsql\n";
  321:     if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
  322: 	open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
  323: 	while ($line=<DFH>) { 
  324: 	    print $fh "$line";
  325: 	    if ($line=~/INFO/) { $notices++; }
  326: 	    if ($line=~/WARNING/) { $notices++; }
  327: 	    if ($line=~/CRITICAL/) { $warnings++; }
  328: 	};
  329: 	close (DFH);
  330:     }
  331:     print $fh "</pre>";
  332:     
  333:     my $lonsqlfile="$perlvar{'lonDaemons'}/logs/lonsql.pid";
  334:  
  335:     $restartflag=1;
  336:    
  337:     if (-e $lonsqlfile) {
  338: 	my $lfh=IO::File->new("$lonsqlfile");
  339: 	my $lonsqlpid=<$lfh>;
  340: 	chomp($lonsqlpid);
  341: 	if (kill 0 => $lonsqlpid) {
  342: 	    print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";
  343: 	    $restartflag=0;
  344: 	} else {
  345: 	    $errors++; $errors++;
  346: 	    print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
  347: 		$restartflag=1;
  348: 	print $fh 
  349: 	    "<h3>Decided to clean up stale .pid file and restart lonsql</h3>";
  350: 	}
  351:     }
  352:     if ($restartflag==1) {
  353: 	$errors++;
  354: 	         print $fh '<br><font color="red">Killall lonsql: '.
  355:                     system('killall lonsql').' - ';
  356:                     sleep 2;
  357:                     print $fh unlink($lonsqlfile).' - '.
  358:                               system('killall -9 lonsql').
  359:                     '</font><br>';
  360: 	print $fh "<h3>lonsql not running, trying to start</h3>";
  361: 	system(
  362:  "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors");
  363: 	sleep 2;
  364: 	if (-e $lonsqlfile) {
  365: 	    print $fh "Seems like it started ...<p>";
  366: 	    my $lfh=IO::File->new("$lonsqlfile");
  367: 	    my $lonsqlpid=<$lfh>;
  368: 	    chomp($lonsqlpid);
  369: 	    sleep 2;
  370: 	    if (kill 0 => $lonsqlpid) {
  371: 		print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";
  372: 	    } else {
  373: 		$errors++; $errors++;
  374: 		print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
  375: 		print $fh "Give it one more try ...<p>";
  376: 		system(
  377:  "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors");
  378: 		sleep 2;
  379: 	    }
  380: 	} else {
  381: 	    print $fh "Seems like that did not work!<p>";
  382: 	    $errors++;
  383: 	}
  384: 	if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
  385: 	    print $fh "<p><pre>";
  386: 	    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
  387: 	    while ($line=<DFH>) { 
  388: 		print $fh "$line";
  389: 		if ($line=~/WARNING/) { $notices++; }
  390: 		if ($line=~/CRITICAL/) { $notices++; }
  391: 	    };
  392: 	    close (DFH);
  393: 	    print $fh "</pre>";
  394: 	}
  395:     }
  396: 
  397:     $fname="$perlvar{'lonDaemons'}/logs/lonsql.log";
  398: 
  399:     my ($dev,$ino,$mode,$nlink,
  400: 	$uid,$gid,$rdev,$size,
  401: 	$atime,$mtime,$ctime,
  402: 	$blksize,$blocks)=stat($fname);
  403: 
  404:     if ($size>200000) {
  405: 	print $fh "Rotating logs ...<p>";
  406: 	rename("$fname.2","$fname.3");
  407: 	rename("$fname.1","$fname.2");
  408: 	rename("$fname","$fname.1");
  409:     }
  410: 
  411:     &errout($fh);
  412: # ------------------------------------------------------------------------ lond
  413: 
  414: print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>';
  415: print "lond\n";
  416: 
  417: if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
  418: open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lond.log|");
  419: while ($line=<DFH>) { 
  420:    print $fh "$line";
  421:    if ($line=~/INFO/) { $notices++; }
  422:    if ($line=~/WARNING/) { $notices++; }
  423:    if ($line=~/CRITICAL/) { $warnings++; }
  424: };
  425: close (DFH);
  426: }
  427: print $fh "</pre>";
  428: 
  429: my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
  430: 
  431: $restartflag=1;
  432: if (-e $londfile) {    
  433:    my $lfh=IO::File->new("$londfile");
  434:    my $londpid=<$lfh>;
  435:    chomp($londpid);
  436:    if (kill 0 => $londpid) {
  437:       print $fh "<h3>lond at pid $londpid responding, sending USR1</h3>";
  438:       kill USR1 => $londpid;
  439:       $restartflag=0;
  440:    } else {
  441:       $errors++;
  442:       print $fh "<h3>lond at pid $londpid not responding</h3>";
  443:       $restartflag=1;
  444:       print $fh 
  445: 	  "<h3>Decided to clean up stale .pid file and restart lond</h3>";
  446:    }
  447: } 
  448: if ($restartflag==1) {
  449:    $errors++;
  450: 	  print $fh '<br><font color="red">Killall lond: '.
  451:                     system('killall lond').' - ';
  452:           sleep 2;
  453:           print $fh unlink($londfile).' - '.system('killall -9 lond').
  454:                     '</font><br>';
  455:    print $fh "<h3>lond not running, trying to start</h3>";
  456:    system(
  457:      "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
  458:    sleep 2;
  459:    if (-e $londfile) {
  460:        print $fh "Seems like it started ...<p>";
  461:        my $lfh=IO::File->new("$londfile");
  462:        my $londpid=<$lfh>;
  463:        chomp($londpid);
  464:        sleep 2;
  465:        if (kill 0 => $londpid) {
  466:           print $fh "<h3>lond at pid $londpid responding</h3>";
  467:        } else {
  468:           $errors++; $errors++;
  469:           print $fh "<h3>lond at pid $londpid not responding</h3>";
  470:           print $fh "Give it one more try ...<p>";
  471: 	  system(
  472:  "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
  473:           sleep 2;
  474:        }
  475:    } else {
  476:        print $fh "Seems like that did not work!<p>";
  477:        $errors++;
  478:    }
  479:    if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
  480:     print $fh "<p><pre>";
  481:     open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
  482:     while ($line=<DFH>) { 
  483:       print $fh "$line";
  484:       if ($line=~/WARNING/) { $notices++; }
  485:       if ($line=~/CRITICAL/) { $notices++; }
  486:     };
  487:     close (DFH);
  488:     print $fh "</pre>";
  489:    }
  490: }
  491: 
  492: $fname="$perlvar{'lonDaemons'}/logs/lond.log";
  493: 
  494:                           my ($dev,$ino,$mode,$nlink,
  495:                               $uid,$gid,$rdev,$size,
  496:                               $atime,$mtime,$ctime,
  497:                               $blksize,$blocks)=stat($fname);
  498: 
  499: if ($size>40000) {
  500:     print $fh "Rotating logs ...<p>";
  501:     rename("$fname.2","$fname.3");
  502:     rename("$fname.1","$fname.2");
  503:     rename("$fname","$fname.1");
  504: }
  505: 
  506: &errout($fh);
  507: # ------------------------------------------------------------------------ lonc
  508: 
  509: print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>';
  510: print "lonc\n";
  511: 
  512: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
  513: open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonc.log|");
  514: while ($line=<DFH>) { 
  515:    print $fh "$line";
  516:    if ($line=~/INFO/) { $notices++; }
  517:    if ($line=~/WARNING/) { $notices++; }
  518:    if ($line=~/CRITICAL/) { $warnings++; }
  519: };
  520: close (DFH);
  521: }
  522: print $fh "</pre>";
  523: 
  524: my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
  525: 
  526: $restartflag=1;
  527: if (-e $loncfile) {
  528:    my $lfh=IO::File->new("$loncfile");
  529:    my $loncpid=<$lfh>;
  530:    chomp($loncpid);
  531:    if (kill 0 => $loncpid) {
  532:       print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";
  533:       kill USR1 => $loncpid;
  534:       $restartflag=0;
  535:    } else {
  536:       $errors++;
  537:       print $fh "<h3>lonc at pid $loncpid not responding</h3>";
  538:       # Solution: kill parent and children processes, remove .pid and restart
  539: 	  $restartflag=1;
  540:       print $fh 
  541: 	  "<h3>Decided to clean up stale .pid file and restart lonc</h3>";
  542:    }
  543: } 
  544: if ($restartflag==1) {
  545:    $errors++;
  546: 	  print $fh '<br><font color="red">Killall lonc: '.
  547: 	            system('killall lonc').' - ';
  548:           sleep 2;
  549:           print $fh unlink($loncfile).' - '.system('killall -9 lonc').
  550:                     '</font><br>';
  551:    print $fh "<h3>lonc not running, trying to start</h3>";
  552: 	system(
  553:  "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
  554:    sleep 2;
  555:    if (-e $loncfile) {
  556:        print $fh "Seems like it started ...<p>";
  557:        my $lfh=IO::File->new("$loncfile");
  558:        my $loncpid=<$lfh>;
  559:        chomp($loncpid);
  560:        sleep 2;
  561:        if (kill 0 => $loncpid) {
  562:           print $fh "<h3>lonc at pid $loncpid responding</h3>";
  563:        } else {
  564:           $errors++; $errors++;
  565:           print $fh "<h3>lonc at pid $loncpid not responding</h3>";
  566:           print $fh "Give it one more try ...<p>";
  567:  	  system(
  568:  "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
  569:           sleep 2;
  570:        }
  571:    } else {
  572:        print $fh "Seems like that did not work!<p>";
  573:        $errors++;
  574:    }
  575:    if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") {
  576:     print $fh "<p><pre>";
  577:     open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
  578:     while ($line=<DFH>) { 
  579:       print $fh "$line";
  580:       if ($line=~/WARNING/) { $notices++; }
  581:       if ($line=~/CRITICAL/) { $notices++; }
  582:     };
  583:     close (DFH);
  584:     print $fh "</pre>";
  585:    }
  586: }
  587: 
  588: $fname="$perlvar{'lonDaemons'}/logs/lonc.log";
  589: 
  590:                           my ($dev,$ino,$mode,$nlink,
  591:                               $uid,$gid,$rdev,$size,
  592:                               $atime,$mtime,$ctime,
  593:                               $blksize,$blocks)=stat($fname);
  594: 
  595: if ($size>40000) {
  596:     print $fh "Rotating logs ...<p>";
  597:     rename("$fname.2","$fname.3");
  598:     rename("$fname.1","$fname.2");
  599:     rename("$fname","$fname.1");
  600: }
  601: 
  602:    
  603: &errout($fh);
  604: # -------------------------------------------------------------------- lonhttpd
  605: 
  606: print $fh '<hr><a name="lonhttpd"><h2>lonhttpd</h2><h3>Log</h3><pre>';
  607: print "lonhttpd\n";
  608: 
  609: if (-e "$perlvar{'lonDaemons'}/logs/lonhttpd.log"){
  610: open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonhttpd.log|");
  611: while ($line=<DFH>) { 
  612:    print $fh "$line";
  613:    if ($line=~/INFO/) { $notices++; }
  614:    if ($line=~/WARNING/) { $notices++; }
  615:    if ($line=~/CRITICAL/) { $warnings++; }
  616: };
  617: close (DFH);
  618: }
  619: print $fh "</pre>";
  620: 
  621: my $lonhttpdfile="$perlvar{'lonDaemons'}/logs/lonhttpd.pid";
  622: 
  623: $restartflag=1;
  624: if (-e $lonhttpdfile) {
  625:    my $lfh=IO::File->new("$lonhttpdfile");
  626:    my $lonhttpdpid=<$lfh>;
  627:    chomp($lonhttpdpid);
  628:    if (kill 0 => $lonhttpdpid) {
  629:       print $fh "<h3>lonhttpd at pid $lonhttpdpid responding</h3>";
  630:       $restartflag=0;
  631:    } else {
  632:       $errors++;
  633:       print $fh "<h3>lonhttpd at pid $lonhttpdpid not responding</h3>";
  634:       # Solution: kill parent and children processes, remove .pid and restart
  635: 	  $restartflag=1;
  636:       print $fh 
  637: 	  "<h3>Decided to clean up stale .pid file and restart lonhttpd</h3>";
  638:    }
  639: } 
  640: if ($restartflag==1) {
  641:    $errors++;
  642: 	  print $fh '<br><font color="red">Killall lonhttpd: '.
  643: 	            system('killall lonhttpd').' - ';
  644:           sleep 2;
  645:           print $fh unlink($lonhttpdfile).' - '.system('killall -9 lonhttpd').
  646:                     '</font><br>';
  647:    print $fh "<h3>lonhttpd not running, trying to start</h3>";
  648: 	system(
  649:  "$perlvar{'lonDaemons'}/lonhttpd 2>>$perlvar{'lonDaemons'}/logs/lonhttpd_errors");
  650:    sleep 2;
  651:    if (-e $lonhttpdfile) {
  652:        print $fh "Seems like it started ...<p>";
  653:        my $lfh=IO::File->new("$lonhttpdfile");
  654:        my $lonhttpdpid=<$lfh>;
  655:        chomp($lonhttpdpid);
  656:        sleep 2;
  657:        if (kill 0 => $lonhttpdpid) {
  658:           print $fh "<h3>lonhttpd at pid $lonhttpdpid responding</h3>";
  659:        } else {
  660:           $errors++; $errors++;
  661:           print $fh "<h3>lonhttpd at pid $lonhttpdpid not responding</h3>";
  662:           print $fh "Give it one more try ...<p>";
  663:  	  system(
  664:  "$perlvar{'lonDaemons'}/lonhttpd 2>>$perlvar{'lonDaemons'}/logs/lonhttpd_errors");
  665:           sleep 2;
  666:        }
  667:    } else {
  668:        print $fh "Seems like that did not work!<p>";
  669:        $errors++;
  670:    }
  671:    if (-e "$perlvar{'lonDaemons'}/logs/lonhttpd.log") {
  672:     print $fh "<p><pre>";
  673:     open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonhttpd.log|");
  674:     while ($line=<DFH>) { 
  675:       print $fh "$line";
  676:       if ($line=~/WARNING/) { $notices++; }
  677:       if ($line=~/CRITICAL/) { $notices++; }
  678:     };
  679:     close (DFH);
  680:     print $fh "</pre>";
  681:    }
  682: }
  683: 
  684: $fname="$perlvar{'lonDaemons'}/logs/lonhttpd.log";
  685: 
  686:                           my ($dev,$ino,$mode,$nlink,
  687:                               $uid,$gid,$rdev,$size,
  688:                               $atime,$mtime,$ctime,
  689:                               $blksize,$blocks)=stat($fname);
  690: 
  691: if ($size>40000) {
  692:     print $fh "Rotating logs ...<p>";
  693:     rename("$fname.2","$fname.3");
  694:     rename("$fname.1","$fname.2");
  695:     rename("$fname","$fname.1");
  696: }
  697: 
  698:    
  699: &errout($fh);
  700: # ---------------------------------------------------------------------- lonnet
  701: 
  702: print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>';
  703: print "lonnet\n";
  704: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
  705: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
  706: while ($line=<DFH>) { 
  707:     print $fh "$line";
  708: };
  709: close (DFH);
  710: }
  711: print $fh "</pre><h3>Perm Log</h3><pre>";
  712: 
  713: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
  714:     open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
  715: while ($line=<DFH>) { 
  716:    print $fh "$line";
  717: };
  718: close (DFH);
  719: } else { print $fh "No perm log\n" }
  720: 
  721: $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
  722: 
  723:                           my ($dev,$ino,$mode,$nlink,
  724:                               $uid,$gid,$rdev,$size,
  725:                               $atime,$mtime,$ctime,
  726:                               $blksize,$blocks)=stat($fname);
  727: 
  728: if ($size>40000) {
  729:     print $fh "Rotating logs ...<p>";
  730:     rename("$fname.2","$fname.3");
  731:     rename("$fname.1","$fname.2");
  732:     rename("$fname","$fname.1");
  733: }
  734: 
  735: print $fh "</pre>";
  736: &errout($fh);
  737: # ----------------------------------------------------------------- Connections
  738: 
  739: print $fh '<hr><a name="connections"><h2>Connections</h2>';
  740: 
  741: print $fh "<table border=2>";
  742: foreach $tryserver (sort(keys(%hostname))) {
  743: 
  744:     $answer=reply("pong",$tryserver);
  745:     if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
  746: 	$result="<b>ok</b>";
  747:     } else {
  748:         $result=$answer;
  749:         $warnings++;
  750:         if ($answer eq 'con_lost') { $warnings++; }
  751:     }
  752:     print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n";
  753: 
  754: }
  755: print $fh "</table>";
  756: 
  757: &errout($fh);
  758: # ------------------------------------------------------------ Delayed messages
  759: 
  760: print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
  761: print "buffers\n";
  762: 
  763: print $fh '<h3>Scanning Permanent Log</h3>';
  764: 
  765: $unsend=0;
  766: {
  767:     my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
  768:     while ($line=<$dfh>) {
  769: 	($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
  770:         if ($sdf eq 'F') { 
  771: 	    $local=localtime($time);
  772:             print "<b>Failed: $time, $dserv, $dcmd</b><br>";
  773:             $warnings++;
  774:         }
  775:         if ($sdf eq 'S') { $unsend--; }
  776:         if ($sdf eq 'D') { $unsend++; }
  777:     }
  778: }
  779: print $fh "Total unsend messages: <b>$unsend</b><p>\n";
  780: $warnings=$warnings+5*$unsend;
  781: 
  782: print $fh "<h3>Outgoing Buffer</h3>";
  783: 
  784: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
  785: while ($line=<DFH>) { 
  786:     print $fh "$line<br>";
  787: };
  788: close (DFH);
  789: 
  790: # ------------------------------------------------------------------------- End
  791: print $fh "<a name=errcount>\n";
  792: $totalcount=$notices+4*$warnings+100*$errors;
  793: &errout($fh);
  794: print $fh "<h1>Total Error Count: $totalcount</h1>";
  795: $now=time;
  796: $date=localtime($now);
  797: print $fh "<hr>$date ($now)</body></html>\n";
  798: print "writing done\n";
  799: }
  800: 
  801: rename ("$statusdir/newstatus.html","$statusdir/index.html");
  802: 
  803: if ($totalcount>200) {
  804:    print "mailing\n";
  805:    $emailto="$perlvar{'lonAdmEMail'}";
  806:    if ($totalcount>600) {
  807:       $emailto.=",$perlvar{'lonSysEMail'}";
  808:    }
  809:    $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; 
  810:    system(
  811:  "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");
  812: }
  813: 1;
  814: 
  815: 
  816: 
  817: 
  818: 
  819: 
  820: 
  821: 

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