File:  [LON-CAPA] / loncom / loncron
Revision 1.6: download - view: text, annotated - select for diffs
Thu Oct 26 22:21:02 2000 UTC (23 years, 6 months ago) by harris41
Branches: MAIN
CVS tags: HEAD
disabling lonsql for now so that newly installed library machines do not crash

    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: 
   13: use IO::File;
   14: use IO::Socket;
   15: 
   16: # -------------------------------------------------- Non-critical communication
   17: sub reply {
   18:     my ($cmd,$server)=@_;
   19:     my $peerfile="$perlvar{'lonSockDir'}/$server";
   20:     my $client=IO::Socket::UNIX->new(Peer    =>"$peerfile",
   21:                                      Type    => SOCK_STREAM,
   22:                                      Timeout => 10)
   23:        or return "con_lost";
   24:     print $client "$cmd\n";
   25:     my $answer=<$client>;
   26:     chomp($answer);
   27:     if (!$answer) { $answer="con_lost"; }
   28:     return $answer;
   29: }
   30: 
   31: # --------------------------------------------------------- Output error status
   32: 
   33: sub errout {
   34:    my $fh=shift;
   35:    print $fh (<<ENDERROUT);
   36:      <p><table border=2 bgcolor="#CCCCCC">
   37:      <tr><td>Notices</td><td>$notices</td></tr>
   38:      <tr><td>Warnings</td><td>$warnings</td></tr>
   39:      <tr><td>Errors</td><td>$errors</td></tr>
   40:      </table><p><a href="#top">Top</a><p>
   41: ENDERROUT
   42: }
   43: 
   44: # ================================================================ Main Program
   45: 
   46: 
   47: # ------------------------------------------------------------ Read access.conf
   48: {
   49:     my $config=IO::File->new("/etc/httpd/conf/access.conf");
   50: 
   51:     while (my $configline=<$config>) {
   52:         if ($configline =~ /PerlSetVar/) {
   53: 	   my ($dummy,$varname,$varvalue)=split(/\s+/,$configline);
   54:            $perlvar{$varname}=$varvalue;
   55:         }
   56:     }
   57: }
   58: 
   59: # ------------------------------------------------------------- Read hosts file
   60: {
   61:     my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
   62: 
   63:     while (my $configline=<$config>) {
   64:        my ($id,$domain,$role,$name,$ip)=split(/:/,$configline);
   65:        $hostname{$id}=$name;
   66:        $hostdom{$id}=$domain;
   67:        $hostrole{$id}=$role;
   68:        $hostip{$id}=$ip;
   69:        if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) {
   70: 	   $libserv{$id}=$name;
   71:        }
   72:     }
   73: }
   74: 
   75: # ------------------------------------------------------ Read spare server file
   76: {
   77:     my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
   78: 
   79:     while (my $configline=<$config>) {
   80:        chomp($configline);
   81:        if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
   82:           $spareid{$configline}=1;
   83:        }
   84:     }
   85: }
   86: 
   87: # ---------------------------------------------------------------- Start report
   88: 
   89: $statusdir="/home/httpd/html/lon-status";
   90: 
   91: $errors=0;
   92: $warnings=0;
   93: $notices=0;
   94: 
   95: $now=time;
   96: $date=localtime($now);
   97: 
   98: {
   99: my $fh=IO::File->new(">$statusdir/newstatus.html");
  100: 
  101: print $fh (<<ENDHEADERS);
  102: <html>
  103: <head>
  104: <title>LON Status Report $perlvar{'lonHostID'}</title>
  105: </head>
  106: <body bgcolor="#AAAAAA">
  107: <a name="top">
  108: <h1>LON Status Report $perlvar{'lonHostID'}</h1>
  109: <h2>$date ($now)</h2>
  110: <ol>
  111: <li><a href="#configuration">Configuration</a>
  112: <li><a href="#machine">Machine Information</a>
  113: <li><a href="#httpd">httpd</a>
  114: <li><a href="#lond">lond</a>
  115: <li><a href="#lonc">lonc</a>
  116: <li><a href="#lonnet">lonnet</a>
  117: <li><a href="#connections">Connections</a>
  118: <li><a href="#delayed">Delayed Messages</a>
  119: <li><a href="#errcount">Error Count</a>
  120: </ol>
  121: <hr>
  122: <a name="configuration">
  123: <h2>Configuration</h2>
  124: <h3>PerlVars</h3>
  125: <table border=2>
  126: ENDHEADERS
  127: 
  128: foreach $varname (keys %perlvar) {
  129:     print $fh "<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n";
  130: }
  131: print $fh "</table><h3>Hosts</h3><table border=2>";
  132: foreach $id (keys %hostname) {
  133: print $fh 
  134:     "<tr><td>$id</td><td>$hostdom{$id}</td><td>$hostrole{$id}</td>";
  135: print $fh "<td>$hostname{$id}</td><td>$hostip{$id}</td></tr>\n";
  136: }
  137: print $fh "</table><h3>Spare Hosts</h3><ol>";
  138: foreach $id (keys %spareid) {
  139:     print $fh "<li>$id\n";
  140: }
  141: 
  142: print $fh "</ol>\n";
  143: 
  144: # --------------------------------------------------------------------- Machine
  145: 
  146: print $fh '<hr><a name="machine"><h2>Machine Information</h2>';
  147: print $fh "<h3>loadavg</h3>";
  148: 
  149: open (LOADAVGH,"/proc/loadavg");
  150: $loadavg=<LOADAVGH>;
  151: close (LOADAVGH);
  152: 
  153: print $fh "<tt>$loadavg</tt>";
  154: 
  155: @parts=split(/\s+/,$loadavg);
  156: if ($parts[1]>4.0) {
  157:     $errors++;
  158: } elsif ($parts[1]>2.0) {
  159:     $warnings++;
  160: } elsif ($parts[1]>1.0) {
  161:     $notices++;
  162: }
  163: 
  164: print $fh "<h3>df</h3>";
  165: print $fh "<pre>";
  166: 
  167: open (DFH,"df|");
  168: while ($line=<DFH>) { 
  169:    print $fh "$line"; 
  170:    @parts=split(/\s+/,$line);
  171:    $usage=$parts[4];
  172:    $usage=~s/\W//g;
  173:    if ($usage>90) { 
  174:       $warnings++; 
  175:    } elsif ($usage>80) {
  176:       $warnings++;
  177:    } elsif ($usage>60) {
  178:       $notices++;
  179:    }
  180:    if ($usage>95) { $warnings++; $warnings++ }
  181: }
  182: close (DFH);
  183: print $fh "</pre>";
  184: &errout($fh);
  185: # ----------------------------------------------------------------------- httpd
  186: 
  187: print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>';
  188: 
  189: open (DFH,"tail -n40 /etc/httpd/logs/access_log|");
  190: while ($line=<DFH>) { print $fh "$line" };
  191: close (DFH);
  192: 
  193: print $fh "</pre><h3>Error Log</h3><pre>";
  194: 
  195: open (DFH,"tail -n50 /etc/httpd/logs/error_log|");
  196: while ($line=<DFH>) { 
  197:    print $fh "$line";
  198:    if ($line=~/\[error\]/) { $notices++; } 
  199: };
  200: close (DFH);
  201: print $fh "</pre>";
  202: &errout($fh);
  203: 
  204: 
  205: # ------------------------------------------------------------------------ lonsql
  206: 
  207: if ($perlvar{'lonRole'} eq "library" && 1==0) {
  208: 
  209:     print $fh '<hr><a name="lond"><h2>lonsql</h2><h3>Log</h3><pre>';
  210:     
  211:     if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
  212: 	open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
  213: 	while ($line=<DFH>) { 
  214: 	    print $fh "$line";
  215: 	    if ($line=~/INFO/) { $notices++; }
  216: 	    if ($line=~/WARNING/) { $notices++; }
  217: 	    if ($line=~/CRITICAL/) { $warnings++; }
  218: 	};
  219: 	close (DFH);
  220:     }
  221:     print $fh "</pre>";
  222:     
  223:     my $lonsqlfile="$perlvar{'lonDaemons'}/logs/lonsql.pid";
  224:     
  225:     if (-e $lonsqlfile) {
  226: 	my $lfh=IO::File->new("$lonsqlfile");
  227: 	my $lonsqlpid=<$lfh>;
  228: 	chomp($lonsqlpid);
  229: 	if (kill 0 => $lonsqlpid) {
  230: 	    print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";
  231: 	} else {
  232: 	    $errors++; $errors++;
  233: 	    print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
  234: 	}
  235:     } else {
  236: 	$errors++;
  237: 	print $fh "<h3>lonsql not running, trying to start</h3>";
  238: 	system("$perlvar{'lonDaemons'}/lonsql");
  239: 	sleep 120;
  240: 	if (-e $lonsqlfile) {
  241: 	    print $fh "Seems like it started ...<p>";
  242: 	    my $lfh=IO::File->new("$lonsqlfile");
  243: 	    my $lonsqlpid=<$lfh>;
  244: 	    chomp($lonsqlpid);
  245: 	    sleep 30;
  246: 	    if (kill 0 => $lonsqlpid) {
  247: 		print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";
  248: 	    } else {
  249: 		$errors++; $errors++;
  250: 		print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
  251: 		print $fh "Give it one more try ...<p>";
  252: 		system("$perlvar{'lonDaemons'}/lonsql");
  253: 		sleep 120;
  254: 	    }
  255: 	} else {
  256: 	    print $fh "Seems like that did not work!<p>";
  257: 	    $errors++;
  258: 	}
  259: 	if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
  260: 	    print $fh "<p><pre>";
  261: 	    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
  262: 	    while ($line=<DFH>) { 
  263: 		print $fh "$line";
  264: 		if ($line=~/WARNING/) { $notices++; }
  265: 		if ($line=~/CRITICAL/) { $notices++; }
  266: 	    };
  267: 	    close (DFH);
  268: 	    print $fh "</pre>";
  269: 	}
  270:     }
  271: 
  272:     $fname="$perlvar{'lonDaemons'}/logs/lonsql.log";
  273: 
  274:     my ($dev,$ino,$mode,$nlink,
  275: 	$uid,$gid,$rdev,$size,
  276: 	$atime,$mtime,$ctime,
  277: 	$blksize,$blocks)=stat($fname);
  278: 
  279:     if ($size>40000) {
  280: 	print $fh "Rotating logs ...<p>";
  281: 	rename("$fname.2","$fname.3");
  282: 	rename("$fname.1","$fname.2");
  283: 	rename("$fname","$fname.1");
  284:     }
  285: 
  286:     &errout($fh);
  287: }
  288: # ------------------------------------------------------------------------ lond
  289: 
  290: print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>';
  291: 
  292: if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
  293: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
  294: while ($line=<DFH>) { 
  295:    print $fh "$line";
  296:    if ($line=~/INFO/) { $notices++; }
  297:    if ($line=~/WARNING/) { $notices++; }
  298:    if ($line=~/CRITICAL/) { $warnings++; }
  299: };
  300: close (DFH);
  301: }
  302: print $fh "</pre>";
  303: 
  304: my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
  305: 
  306: if (-e $londfile) {
  307:    my $lfh=IO::File->new("$londfile");
  308:    my $londpid=<$lfh>;
  309:    chomp($londpid);
  310:    if (kill 0 => $londpid) {
  311:       print $fh "<h3>lond at pid $londpid responding</h3>";
  312:    } else {
  313:       $errors++; $errors++;
  314:       print $fh "<h3>lond at pid $londpid not responding</h3>";
  315:    }
  316: } else {
  317:    $errors++;
  318:    print $fh "<h3>lond not running, trying to start</h3>";
  319:    system("$perlvar{'lonDaemons'}/lond");
  320:    sleep 120;
  321:    if (-e $londfile) {
  322:        print $fh "Seems like it started ...<p>";
  323:        my $lfh=IO::File->new("$londfile");
  324:        my $londpid=<$lfh>;
  325:        chomp($londpid);
  326:        sleep 30;
  327:        if (kill 0 => $londpid) {
  328:           print $fh "<h3>lond at pid $londpid responding</h3>";
  329:        } else {
  330:           $errors++; $errors++;
  331:           print $fh "<h3>lond at pid $londpid not responding</h3>";
  332:           print $fh "Give it one more try ...<p>";
  333:           system("$perlvar{'lonDaemons'}/lond");
  334:           sleep 120;
  335:        }
  336:    } else {
  337:        print $fh "Seems like that did not work!<p>";
  338:        $errors++;
  339:    }
  340:    if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
  341:     print $fh "<p><pre>";
  342:     open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
  343:     while ($line=<DFH>) { 
  344:       print $fh "$line";
  345:       if ($line=~/WARNING/) { $notices++; }
  346:       if ($line=~/CRITICAL/) { $notices++; }
  347:     };
  348:     close (DFH);
  349:     print $fh "</pre>";
  350:    }
  351: }
  352: 
  353: $fname="$perlvar{'lonDaemons'}/logs/lond.log";
  354: 
  355:                           my ($dev,$ino,$mode,$nlink,
  356:                               $uid,$gid,$rdev,$size,
  357:                               $atime,$mtime,$ctime,
  358:                               $blksize,$blocks)=stat($fname);
  359: 
  360: if ($size>40000) {
  361:     print $fh "Rotating logs ...<p>";
  362:     rename("$fname.2","$fname.3");
  363:     rename("$fname.1","$fname.2");
  364:     rename("$fname","$fname.1");
  365: }
  366: 
  367: &errout($fh);
  368: # ------------------------------------------------------------------------ lonc
  369: 
  370: print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>';
  371: 
  372: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
  373: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
  374: while ($line=<DFH>) { 
  375:    print $fh "$line";
  376:    if ($line=~/INFO/) { $notices++; }
  377:    if ($line=~/WARNING/) { $notices++; }
  378:    if ($line=~/CRITICAL/) { $warnings++; }
  379: };
  380: close (DFH);
  381: }
  382: print $fh "</pre>";
  383: 
  384: my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
  385: 
  386: if (-e $loncfile) {
  387:    my $lfh=IO::File->new("$loncfile");
  388:    my $loncpid=<$lfh>;
  389:    chomp($loncpid);
  390:    if (kill 0 => $loncpid) {
  391:       print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";
  392:       kill USR1 => $loncpid;
  393:    } else {
  394:       $errors++; $errors++;
  395:       print $fh "<h3>lonc at pid $loncpid not responding</h3>";
  396:    }
  397: } else {
  398:    $errors++;
  399:    print $fh "<h3>lonc not running, trying to start</h3>";
  400:    system("$perlvar{'lonDaemons'}/lonc");
  401:    sleep 120;
  402:    if (-e $loncfile) {
  403:        print $fh "Seems like it started ...<p>";
  404:        my $lfh=IO::File->new("$loncfile");
  405:        my $loncpid=<$lfh>;
  406:        chomp($loncpid);
  407:        sleep 30;
  408:        if (kill 0 => $loncpid) {
  409:           print $fh "<h3>lonc at pid $loncpid responding</h3>";
  410:        } else {
  411:           $errors++; $errors++;
  412:           print $fh "<h3>lonc at pid $loncpid not responding</h3>";
  413:           print $fh "Give it one more try ...<p>";
  414:           system("$perlvar{'lonDaemons'}/lonc");
  415:           sleep 120;
  416:        }
  417:    } else {
  418:        print $fh "Seems like that did not work!<p>";
  419:        $errors++;
  420:    }
  421:    if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") {
  422:     print $fh "<p><pre>";
  423:     open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
  424:     while ($line=<DFH>) { 
  425:       print $fh "$line";
  426:       if ($line=~/WARNING/) { $notices++; }
  427:       if ($line=~/CRITICAL/) { $notices++; }
  428:     };
  429:     close (DFH);
  430:     print $fh "</pre>";
  431:    }
  432: }
  433: 
  434: $fname="$perlvar{'lonDaemons'}/logs/lonc.log";
  435: 
  436:                           my ($dev,$ino,$mode,$nlink,
  437:                               $uid,$gid,$rdev,$size,
  438:                               $atime,$mtime,$ctime,
  439:                               $blksize,$blocks)=stat($fname);
  440: 
  441: if ($size>40000) {
  442:     print $fh "Rotating logs ...<p>";
  443:     rename("$fname.2","$fname.3");
  444:     rename("$fname.1","$fname.2");
  445:     rename("$fname","$fname.1");
  446: }
  447: 
  448:    
  449: &errout($fh);
  450: # ---------------------------------------------------------------------- lonnet
  451: 
  452: print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>';
  453: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
  454: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
  455: while ($line=<DFH>) { 
  456:     print $fh "$line";
  457: };
  458: close (DFH);
  459: }
  460: print $fh "</pre><h3>Perm Log</h3>";
  461: 
  462: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
  463:     open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
  464: while ($line=<DFH>) { 
  465:    print $fh "$line";
  466: };
  467: close (DFH);
  468: } else { print $fh "No perm log\n" }
  469: 
  470: $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
  471: 
  472:                           my ($dev,$ino,$mode,$nlink,
  473:                               $uid,$gid,$rdev,$size,
  474:                               $atime,$mtime,$ctime,
  475:                               $blksize,$blocks)=stat($fname);
  476: 
  477: if ($size>40000) {
  478:     print $fh "Rotating logs ...<p>";
  479:     rename("$fname.2","$fname.3");
  480:     rename("$fname.1","$fname.2");
  481:     rename("$fname","$fname.1");
  482: }
  483: 
  484: print $fh "</pre>";
  485: &errout($fh);
  486: # ----------------------------------------------------------------- Connections
  487: 
  488: print $fh '<hr><a name="connections"><h2>Connections</h2>';
  489: 
  490: print $fh "<table border=2>";
  491: foreach $tryserver (keys %hostname) {
  492: 
  493:     $answer=reply("pong",$tryserver);
  494:     if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
  495: 	$result="<b>ok</b>";
  496:     } else {
  497:         $result=$answer;
  498:         $warnings++;
  499:         if ($answer eq 'con_lost') { $warnings++; }
  500:     }
  501:     print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n";
  502: 
  503: }
  504: print $fh "</table>";
  505: 
  506: &errout($fh);
  507: # ------------------------------------------------------------ Delayed messages
  508: 
  509: print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
  510: 
  511: print $fh '<h3>Scanning Permanent Log</h3>';
  512: 
  513: $unsend=0;
  514: {
  515:     my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
  516:     while ($line=<$dfh>) {
  517: 	($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
  518:         if ($sdf eq 'F') { 
  519: 	    $local=localtime($time);
  520:             print "<b>Failed: $time, $dserv, $dcmd</b><br>";
  521:             $warnings++;
  522:         }
  523:         if ($sdf eq 'S') { $unsend--; }
  524:         if ($sdf eq 'D') { $unsend++; }
  525:     }
  526: }
  527: print $fh "Total unsend messages: <b>$unsend</b><p>\n";
  528: $warnings=$warnings+5*$unsend;
  529: 
  530: print $fh "<h3>Outgoing Buffer</h3>";
  531: 
  532: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
  533: while ($line=<DFH>) { 
  534:     print $fh "$line<br>";
  535: };
  536: close (DFH);
  537: 
  538: # ------------------------------------------------------------------------- End
  539: print $fh "<a name=errcount>\n";
  540: $totalcount=$notices+4*$warnings+100*$errors;
  541: &errout($fh);
  542: print $fh "<h1>Total Error Count: $totalcount</h1>";
  543: $now=time;
  544: $date=localtime($now);
  545: print $fh "<hr>$date ($now)</body></html>\n";
  546: 
  547: }
  548: 
  549: rename ("$statusdir/newstatus.html","$statusdir/index.html");
  550: 
  551: if ($totalcount>200) {
  552:    $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
  553:    $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; 
  554:    system(
  555:  "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");
  556: }
  557: 1;
  558: 
  559: 
  560: 
  561: 
  562: 
  563: 
  564: 
  565: 

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