Annotation of loncom/loncron, revision 1.26

1.1       albertel    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: #
1.3       www        10: # 7/14,7/15,7/19,7/21,7/22,11/18,
                     11: # 2/8 Gerd Kortemeyer
1.20      harris41   12: # 12/6/2000,12/8 Scott Harrison
1.11      www        13: # 12/23 Gerd Kortemeyer
1.22      harris41   14: # YEAR=2001
                     15: # 1/10/2001, 2/12/, 2/26, 3/15, 04/11, 04/21,8/27 Scott Harrison
1.25      www        16: # 09/04,09/06,11/26 Gerd Kortemeyer
1.26    ! harris41   17: # YEAR=2002
        !            18: # 5/11/2002 Scott Harrison
1.24      www        19: 
                     20: $|=1;
1.1       albertel   21: 
1.26    ! harris41   22: use lib '/home/httpd/lib/perl/';
        !            23: use LONCAPA::Configuration;
        !            24: 
1.1       albertel   25: use IO::File;
                     26: use IO::Socket;
                     27: 
                     28: # -------------------------------------------------- Non-critical communication
                     29: sub reply {
                     30:     my ($cmd,$server)=@_;
                     31:     my $peerfile="$perlvar{'lonSockDir'}/$server";
                     32:     my $client=IO::Socket::UNIX->new(Peer    =>"$peerfile",
                     33:                                      Type    => SOCK_STREAM,
                     34:                                      Timeout => 10)
                     35:        or return "con_lost";
                     36:     print $client "$cmd\n";
                     37:     my $answer=<$client>;
                     38:     chomp($answer);
                     39:     if (!$answer) { $answer="con_lost"; }
                     40:     return $answer;
                     41: }
                     42: 
                     43: # --------------------------------------------------------- Output error status
                     44: 
                     45: sub errout {
                     46:    my $fh=shift;
                     47:    print $fh (<<ENDERROUT);
                     48:      <p><table border=2 bgcolor="#CCCCCC">
                     49:      <tr><td>Notices</td><td>$notices</td></tr>
                     50:      <tr><td>Warnings</td><td>$warnings</td></tr>
                     51:      <tr><td>Errors</td><td>$errors</td></tr>
                     52:      </table><p><a href="#top">Top</a><p>
                     53: ENDERROUT
                     54: }
                     55: 
                     56: # ================================================================ Main Program
                     57: 
1.26    ! harris41   58: # ------------------------------------------- Read access.conf and loncapa.conf
        !            59: my $perlvarref=LONCAPA::Configuration::read_conf('access.conf','loncapa.conf');
        !            60: my %perlvar=%{$perlvarref};
        !            61: undef $perlvarref;
        !            62: delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
        !            63: delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
1.13      harris41   64: 
1.14      harris41   65: # --------------------------------------- Make sure that LON-CAPA is configured
                     66: # I only test for one thing here (lonHostID).  This is just a safeguard.
                     67: if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) {
1.15      harris41   68:    print("Unconfigured machine.\n");
1.14      harris41   69:    $emailto=$perlvar{'lonSysEMail'};
                     70:    $hostname=`/bin/hostname`;
                     71:    chop $hostname;
                     72:    $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
                     73:    $subj="LON: Unconfigured machine $hostname";
                     74:    system("echo 'Unconfigured machine $hostname.' |\
                     75:  mailto $emailto -s '$subj' > /dev/null");
                     76:     exit 1;
                     77: }
                     78: 
1.13      harris41   79: # ----------------------------- Make sure this process is running from user=www
                     80: my $wwwid=getpwnam('www');
                     81: if ($wwwid!=$<) {
1.14      harris41   82:    print("User ID mismatch.  This program must be run as user 'www'\n");
1.13      harris41   83:    $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
                     84:    $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
                     85:    system("echo 'User ID mismatch.  loncron must be run as user www.' |\
                     86:  mailto $emailto -s '$subj' > /dev/null");
1.14      harris41   87:    exit 1;
1.1       albertel   88: }
                     89: 
                     90: # ------------------------------------------------------------- Read hosts file
                     91: {
                     92:     my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
                     93: 
                     94:     while (my $configline=<$config>) {
                     95:        my ($id,$domain,$role,$name,$ip)=split(/:/,$configline);
                     96:        $hostname{$id}=$name;
                     97:        $hostdom{$id}=$domain;
                     98:        $hostrole{$id}=$role;
                     99:        $hostip{$id}=$ip;
                    100:        if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) {
                    101: 	   $libserv{$id}=$name;
                    102:        }
                    103:     }
                    104: }
                    105: 
                    106: # ------------------------------------------------------ Read spare server file
                    107: {
                    108:     my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
                    109: 
                    110:     while (my $configline=<$config>) {
                    111:        chomp($configline);
                    112:        if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
                    113:           $spareid{$configline}=1;
                    114:        }
                    115:     }
                    116: }
                    117: 
                    118: # ---------------------------------------------------------------- Start report
                    119: 
                    120: $statusdir="/home/httpd/html/lon-status";
                    121: 
                    122: $errors=0;
                    123: $warnings=0;
                    124: $notices=0;
                    125: 
                    126: $now=time;
                    127: $date=localtime($now);
                    128: 
                    129: {
                    130: my $fh=IO::File->new(">$statusdir/newstatus.html");
                    131: 
                    132: print $fh (<<ENDHEADERS);
                    133: <html>
                    134: <head>
                    135: <title>LON Status Report $perlvar{'lonHostID'}</title>
                    136: </head>
1.3       www       137: <body bgcolor="#AAAAAA">
1.1       albertel  138: <a name="top">
                    139: <h1>LON Status Report $perlvar{'lonHostID'}</h1>
                    140: <h2>$date ($now)</h2>
                    141: <ol>
                    142: <li><a href="#configuration">Configuration</a>
                    143: <li><a href="#machine">Machine Information</a>
1.11      www       144: <li><a href="#tmp">Temporary Files</a>
                    145: <li><a href="#tokens">Session Tokens</a>
1.1       albertel  146: <li><a href="#httpd">httpd</a>
1.11      www       147: <li><a href="#lonsql">lonsql</a>
1.1       albertel  148: <li><a href="#lond">lond</a>
                    149: <li><a href="#lonc">lonc</a>
                    150: <li><a href="#lonnet">lonnet</a>
                    151: <li><a href="#connections">Connections</a>
                    152: <li><a href="#delayed">Delayed Messages</a>
                    153: <li><a href="#errcount">Error Count</a>
                    154: </ol>
                    155: <hr>
                    156: <a name="configuration">
                    157: <h2>Configuration</h2>
                    158: <h3>PerlVars</h3>
                    159: <table border=2>
                    160: ENDHEADERS
                    161: 
                    162: foreach $varname (keys %perlvar) {
                    163:     print $fh "<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n";
                    164: }
                    165: print $fh "</table><h3>Hosts</h3><table border=2>";
                    166: foreach $id (keys %hostname) {
                    167: print $fh 
                    168:     "<tr><td>$id</td><td>$hostdom{$id}</td><td>$hostrole{$id}</td>";
                    169: print $fh "<td>$hostname{$id}</td><td>$hostip{$id}</td></tr>\n";
                    170: }
                    171: print $fh "</table><h3>Spare Hosts</h3><ol>";
                    172: foreach $id (keys %spareid) {
                    173:     print $fh "<li>$id\n";
                    174: }
                    175: 
                    176: print $fh "</ol>\n";
                    177: 
                    178: # --------------------------------------------------------------------- Machine
                    179: 
                    180: print $fh '<hr><a name="machine"><h2>Machine Information</h2>';
                    181: print $fh "<h3>loadavg</h3>";
                    182: 
                    183: open (LOADAVGH,"/proc/loadavg");
                    184: $loadavg=<LOADAVGH>;
                    185: close (LOADAVGH);
                    186: 
                    187: print $fh "<tt>$loadavg</tt>";
                    188: 
                    189: @parts=split(/\s+/,$loadavg);
1.4       www       190: if ($parts[1]>4.0) {
1.1       albertel  191:     $errors++;
                    192: } elsif ($parts[1]>2.0) {
                    193:     $warnings++;
                    194: } elsif ($parts[1]>1.0) {
                    195:     $notices++;
                    196: }
                    197: 
                    198: print $fh "<h3>df</h3>";
                    199: print $fh "<pre>";
                    200: 
                    201: open (DFH,"df|");
                    202: while ($line=<DFH>) { 
                    203:    print $fh "$line"; 
                    204:    @parts=split(/\s+/,$line);
                    205:    $usage=$parts[4];
                    206:    $usage=~s/\W//g;
                    207:    if ($usage>90) { 
1.24      www       208:       $warnings++;
                    209:       $notices++; 
1.1       albertel  210:    } elsif ($usage>80) {
                    211:       $warnings++;
                    212:    } elsif ($usage>60) {
                    213:       $notices++;
                    214:    }
1.4       www       215:    if ($usage>95) { $warnings++; $warnings++ }
1.1       albertel  216: }
                    217: close (DFH);
                    218: print $fh "</pre>";
1.24      www       219: 
                    220: 
                    221: print $fh "<h3>ps</h3>";
                    222: print $fh "<pre>";
                    223: $psproc=0;
                    224: 
                    225: open (PSH,"ps -aux|");
                    226: while ($line=<PSH>) { 
                    227:    print $fh "$line"; 
                    228:    $psproc++;
                    229: }
                    230: close (PSH);
                    231: print $fh "</pre>";
                    232: 
                    233: if ($psproc>200) { $notices++; }
                    234: if ($psproc>250) { $notices++; }
                    235: 
1.1       albertel  236: &errout($fh);
1.11      www       237: 
                    238: # --------------------------------------------------------------- clean out tmp
                    239: print $fh '<hr><a name="tmp"><h2>Temporary Files</h2>';
                    240: $cleaned=0;
                    241: while ($fname=<$perlvar{'lonDaemons'}/tmp/*>) {
                    242:                           my ($dev,$ino,$mode,$nlink,
                    243:                               $uid,$gid,$rdev,$size,
                    244:                               $atime,$mtime,$ctime,
                    245:                               $blksize,$blocks)=stat($fname);
                    246:                           $now=time;
                    247:                           $since=$now-$mtime;
                    248:                           if ($since>$perlvar{'lonExpire'}) {
                    249:                               $cleaned++;
                    250:                               unlink("$fname");
                    251:                           }
                    252:     
                    253: }
                    254: print $fh "Cleaned up ".$cleaned." files.";
                    255: 
                    256: # ------------------------------------------------------------ clean out lonIDs
                    257: print $fh '<hr><a name="tokens"><h2>Session Tokens</h2>';
                    258: $cleaned=0;
                    259: $active=0;
                    260: while ($fname=<$perlvar{'lonIDsDir'}/*>) {
                    261:                           my ($dev,$ino,$mode,$nlink,
                    262:                               $uid,$gid,$rdev,$size,
                    263:                               $atime,$mtime,$ctime,
                    264:                               $blksize,$blocks)=stat($fname);
                    265:                           $now=time;
                    266:                           $since=$now-$mtime;
                    267:                           if ($since>$perlvar{'lonExpire'}) {
                    268:                               $cleaned++;
                    269:                               print $fh "Unlinking $fname<br>";
                    270:                               unlink("$fname");
                    271:                           } else {
                    272:                               $active++;
                    273:                           }
                    274:     
                    275: }
                    276: print $fh "<p>Cleaned up ".$cleaned." stale session token(s).";
                    277: print $fh "<h3>$active open session(s)</h3>";
                    278: 
1.1       albertel  279: # ----------------------------------------------------------------------- httpd
                    280: 
                    281: print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>';
                    282: 
1.23      www       283: open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
1.1       albertel  284: while ($line=<DFH>) { print $fh "$line" };
                    285: close (DFH);
                    286: 
                    287: print $fh "</pre><h3>Error Log</h3><pre>";
                    288: 
1.23      www       289: open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
1.1       albertel  290: while ($line=<DFH>) { 
                    291:    print $fh "$line";
                    292:    if ($line=~/\[error\]/) { $notices++; } 
                    293: };
                    294: close (DFH);
                    295: print $fh "</pre>";
                    296: &errout($fh);
1.5       harris41  297: 
                    298: 
1.11      www       299: # ---------------------------------------------------------------------- lonsql
1.22      harris41  300: 
                    301: my $restartflag=1;
1.18      harris41  302: if ($perlvar{'lonRole'} eq "library") {
1.5       harris41  303: 
1.11      www       304:     print $fh '<hr><a name="lonsql"><h2>lonsql</h2><h3>Log</h3><pre>';
1.23      www       305:     print "lonsql\n";
1.5       harris41  306:     if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
                    307: 	open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
                    308: 	while ($line=<DFH>) { 
                    309: 	    print $fh "$line";
                    310: 	    if ($line=~/INFO/) { $notices++; }
                    311: 	    if ($line=~/WARNING/) { $notices++; }
                    312: 	    if ($line=~/CRITICAL/) { $warnings++; }
                    313: 	};
                    314: 	close (DFH);
                    315:     }
                    316:     print $fh "</pre>";
                    317:     
                    318:     my $lonsqlfile="$perlvar{'lonDaemons'}/logs/lonsql.pid";
1.23      www       319:  
                    320:     $restartflag=1;
                    321:    
1.5       harris41  322:     if (-e $lonsqlfile) {
                    323: 	my $lfh=IO::File->new("$lonsqlfile");
                    324: 	my $lonsqlpid=<$lfh>;
                    325: 	chomp($lonsqlpid);
                    326: 	if (kill 0 => $lonsqlpid) {
                    327: 	    print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";
1.22      harris41  328: 	    $restartflag=0;
1.5       harris41  329: 	} else {
                    330: 	    $errors++; $errors++;
                    331: 	    print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
1.22      harris41  332: 		$restartflag=1;
1.23      www       333: 	print $fh 
                    334: 	    "<h3>Decided to clean up stale .pid file and restart lonsql</h3>";
1.5       harris41  335: 	}
1.22      harris41  336:     }
                    337:     if ($restartflag==1) {
1.5       harris41  338: 	$errors++;
1.23      www       339: 	         print $fh '<br><font color="red">Killall lonsql: '.
                    340:                     system('killall lonsql').' - ';
                    341:                     sleep 60;
                    342:                     print $fh unlink($lonsqlfile).' - '.
                    343:                               system('killall -9 lonsql').
                    344:                     '</font><br>';
1.5       harris41  345: 	print $fh "<h3>lonsql not running, trying to start</h3>";
1.16      harris41  346: 	system(
                    347:  "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors");
1.24      www       348: 	sleep 10;
1.5       harris41  349: 	if (-e $lonsqlfile) {
                    350: 	    print $fh "Seems like it started ...<p>";
                    351: 	    my $lfh=IO::File->new("$lonsqlfile");
                    352: 	    my $lonsqlpid=<$lfh>;
                    353: 	    chomp($lonsqlpid);
1.24      www       354: 	    sleep 30;
1.5       harris41  355: 	    if (kill 0 => $lonsqlpid) {
                    356: 		print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";
                    357: 	    } else {
                    358: 		$errors++; $errors++;
                    359: 		print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
                    360: 		print $fh "Give it one more try ...<p>";
1.16      harris41  361: 		system(
                    362:  "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors");
1.24      www       363: 		sleep 10;
1.5       harris41  364: 	    }
                    365: 	} else {
                    366: 	    print $fh "Seems like that did not work!<p>";
                    367: 	    $errors++;
                    368: 	}
                    369: 	if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
                    370: 	    print $fh "<p><pre>";
                    371: 	    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
                    372: 	    while ($line=<DFH>) { 
                    373: 		print $fh "$line";
                    374: 		if ($line=~/WARNING/) { $notices++; }
                    375: 		if ($line=~/CRITICAL/) { $notices++; }
                    376: 	    };
                    377: 	    close (DFH);
                    378: 	    print $fh "</pre>";
                    379: 	}
                    380:     }
                    381: 
                    382:     $fname="$perlvar{'lonDaemons'}/logs/lonsql.log";
                    383: 
                    384:     my ($dev,$ino,$mode,$nlink,
                    385: 	$uid,$gid,$rdev,$size,
                    386: 	$atime,$mtime,$ctime,
                    387: 	$blksize,$blocks)=stat($fname);
                    388: 
                    389:     if ($size>40000) {
                    390: 	print $fh "Rotating logs ...<p>";
                    391: 	rename("$fname.2","$fname.3");
                    392: 	rename("$fname.1","$fname.2");
                    393: 	rename("$fname","$fname.1");
                    394:     }
                    395: 
                    396:     &errout($fh);
                    397: }
1.1       albertel  398: # ------------------------------------------------------------------------ lond
                    399: 
                    400: print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>';
1.23      www       401: print "lond\n";
1.1       albertel  402: 
                    403: if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
1.23      www       404: open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lond.log|");
1.1       albertel  405: while ($line=<DFH>) { 
                    406:    print $fh "$line";
1.3       www       407:    if ($line=~/INFO/) { $notices++; }
1.4       www       408:    if ($line=~/WARNING/) { $notices++; }
                    409:    if ($line=~/CRITICAL/) { $warnings++; }
1.1       albertel  410: };
                    411: close (DFH);
                    412: }
                    413: print $fh "</pre>";
                    414: 
                    415: my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
                    416: 
1.22      harris41  417: $restartflag=1;
1.7       harris41  418: if (-e $londfile) {    
1.1       albertel  419:    my $lfh=IO::File->new("$londfile");
                    420:    my $londpid=<$lfh>;
                    421:    chomp($londpid);
                    422:    if (kill 0 => $londpid) {
1.25      www       423:       print $fh "<h3>lond at pid $londpid responding, sending USR1</h3>";
                    424:       kill USR1 => $londpid;
1.7       harris41  425:       $restartflag=0;
1.1       albertel  426:    } else {
1.8       harris41  427:       $errors++;
1.1       albertel  428:       print $fh "<h3>lond at pid $londpid not responding</h3>";
1.23      www       429:       $restartflag=1;
1.8       harris41  430:       print $fh 
1.23      www       431: 	  "<h3>Decided to clean up stale .pid file and restart lond</h3>";
1.1       albertel  432:    }
1.7       harris41  433: } 
                    434: if ($restartflag==1) {
1.1       albertel  435:    $errors++;
1.23      www       436: 	  print $fh '<br><font color="red">Killall lond: '.
                    437:                     system('killall lond').' - ';
                    438:           sleep 60;
                    439:           print $fh unlink($londfile).' - '.system('killall -9 lond').
                    440:                     '</font><br>';
1.1       albertel  441:    print $fh "<h3>lond not running, trying to start</h3>";
1.16      harris41  442:    system(
                    443:      "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
1.24      www       444:    sleep 10;
1.1       albertel  445:    if (-e $londfile) {
                    446:        print $fh "Seems like it started ...<p>";
                    447:        my $lfh=IO::File->new("$londfile");
                    448:        my $londpid=<$lfh>;
                    449:        chomp($londpid);
1.24      www       450:        sleep 30;
1.1       albertel  451:        if (kill 0 => $londpid) {
                    452:           print $fh "<h3>lond at pid $londpid responding</h3>";
                    453:        } else {
                    454:           $errors++; $errors++;
                    455:           print $fh "<h3>lond at pid $londpid not responding</h3>";
                    456:           print $fh "Give it one more try ...<p>";
1.16      harris41  457: 	  system(
                    458:  "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
1.24      www       459:           sleep 10;
1.1       albertel  460:        }
                    461:    } else {
                    462:        print $fh "Seems like that did not work!<p>";
                    463:        $errors++;
                    464:    }
1.3       www       465:    if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
                    466:     print $fh "<p><pre>";
                    467:     open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
                    468:     while ($line=<DFH>) { 
                    469:       print $fh "$line";
1.4       www       470:       if ($line=~/WARNING/) { $notices++; }
                    471:       if ($line=~/CRITICAL/) { $notices++; }
1.3       www       472:     };
                    473:     close (DFH);
                    474:     print $fh "</pre>";
                    475:    }
1.1       albertel  476: }
                    477: 
                    478: $fname="$perlvar{'lonDaemons'}/logs/lond.log";
                    479: 
                    480:                           my ($dev,$ino,$mode,$nlink,
                    481:                               $uid,$gid,$rdev,$size,
                    482:                               $atime,$mtime,$ctime,
                    483:                               $blksize,$blocks)=stat($fname);
                    484: 
                    485: if ($size>40000) {
                    486:     print $fh "Rotating logs ...<p>";
                    487:     rename("$fname.2","$fname.3");
                    488:     rename("$fname.1","$fname.2");
                    489:     rename("$fname","$fname.1");
                    490: }
                    491: 
                    492: &errout($fh);
                    493: # ------------------------------------------------------------------------ lonc
                    494: 
                    495: print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>';
1.23      www       496: print "lonc\n";
1.1       albertel  497: 
                    498: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
1.23      www       499: open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonc.log|");
1.1       albertel  500: while ($line=<DFH>) { 
                    501:    print $fh "$line";
1.3       www       502:    if ($line=~/INFO/) { $notices++; }
1.4       www       503:    if ($line=~/WARNING/) { $notices++; }
                    504:    if ($line=~/CRITICAL/) { $warnings++; }
1.1       albertel  505: };
                    506: close (DFH);
                    507: }
                    508: print $fh "</pre>";
                    509: 
                    510: my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
                    511: 
1.7       harris41  512: $restartflag=1;
1.1       albertel  513: if (-e $loncfile) {
                    514:    my $lfh=IO::File->new("$loncfile");
                    515:    my $loncpid=<$lfh>;
                    516:    chomp($loncpid);
                    517:    if (kill 0 => $loncpid) {
                    518:       print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";
                    519:       kill USR1 => $loncpid;
1.7       harris41  520:       $restartflag=0;
1.1       albertel  521:    } else {
1.8       harris41  522:       $errors++;
1.1       albertel  523:       print $fh "<h3>lonc at pid $loncpid not responding</h3>";
1.10      harris41  524:       # Solution: kill parent and children processes, remove .pid and restart
1.8       harris41  525: 	  $restartflag=1;
                    526:       print $fh 
1.23      www       527: 	  "<h3>Decided to clean up stale .pid file and restart lonc</h3>";
1.1       albertel  528:    }
1.7       harris41  529: } 
                    530: if ($restartflag==1) {
1.1       albertel  531:    $errors++;
1.23      www       532: 	  print $fh '<br><font color="red">Killall lonc: '.
                    533: 	            system('killall lonc').' - ';
                    534:           sleep 60;
                    535:           print $fh unlink($loncfile).' - '.system('killall -9 lonc').
                    536:                     '</font><br>';
1.1       albertel  537:    print $fh "<h3>lonc not running, trying to start</h3>";
1.16      harris41  538: 	system(
1.17      harris41  539:  "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
1.24      www       540:    sleep 10;
1.1       albertel  541:    if (-e $loncfile) {
                    542:        print $fh "Seems like it started ...<p>";
                    543:        my $lfh=IO::File->new("$loncfile");
                    544:        my $loncpid=<$lfh>;
                    545:        chomp($loncpid);
1.24      www       546:        sleep 30;
1.1       albertel  547:        if (kill 0 => $loncpid) {
                    548:           print $fh "<h3>lonc at pid $loncpid responding</h3>";
                    549:        } else {
                    550:           $errors++; $errors++;
                    551:           print $fh "<h3>lonc at pid $loncpid not responding</h3>";
                    552:           print $fh "Give it one more try ...<p>";
1.16      harris41  553:  	  system(
1.17      harris41  554:  "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
1.24      www       555:           sleep 10;
1.1       albertel  556:        }
                    557:    } else {
                    558:        print $fh "Seems like that did not work!<p>";
                    559:        $errors++;
                    560:    }
1.3       www       561:    if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") {
                    562:     print $fh "<p><pre>";
                    563:     open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
                    564:     while ($line=<DFH>) { 
                    565:       print $fh "$line";
1.4       www       566:       if ($line=~/WARNING/) { $notices++; }
                    567:       if ($line=~/CRITICAL/) { $notices++; }
1.3       www       568:     };
                    569:     close (DFH);
                    570:     print $fh "</pre>";
                    571:    }
1.1       albertel  572: }
                    573: 
                    574: $fname="$perlvar{'lonDaemons'}/logs/lonc.log";
                    575: 
                    576:                           my ($dev,$ino,$mode,$nlink,
                    577:                               $uid,$gid,$rdev,$size,
                    578:                               $atime,$mtime,$ctime,
                    579:                               $blksize,$blocks)=stat($fname);
                    580: 
                    581: if ($size>40000) {
                    582:     print $fh "Rotating logs ...<p>";
                    583:     rename("$fname.2","$fname.3");
                    584:     rename("$fname.1","$fname.2");
                    585:     rename("$fname","$fname.1");
                    586: }
                    587: 
                    588:    
                    589: &errout($fh);
                    590: # ---------------------------------------------------------------------- lonnet
                    591: 
                    592: print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>';
1.23      www       593: print "lonnet\n";
1.1       albertel  594: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
                    595: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
                    596: while ($line=<DFH>) { 
                    597:     print $fh "$line";
                    598: };
                    599: close (DFH);
                    600: }
1.11      www       601: print $fh "</pre><h3>Perm Log</h3><pre>";
1.1       albertel  602: 
                    603: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
                    604:     open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
                    605: while ($line=<DFH>) { 
                    606:    print $fh "$line";
                    607: };
                    608: close (DFH);
                    609: } else { print $fh "No perm log\n" }
                    610: 
                    611: $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
                    612: 
                    613:                           my ($dev,$ino,$mode,$nlink,
                    614:                               $uid,$gid,$rdev,$size,
                    615:                               $atime,$mtime,$ctime,
                    616:                               $blksize,$blocks)=stat($fname);
                    617: 
                    618: if ($size>40000) {
                    619:     print $fh "Rotating logs ...<p>";
                    620:     rename("$fname.2","$fname.3");
                    621:     rename("$fname.1","$fname.2");
                    622:     rename("$fname","$fname.1");
                    623: }
                    624: 
                    625: print $fh "</pre>";
                    626: &errout($fh);
                    627: # ----------------------------------------------------------------- Connections
                    628: 
                    629: print $fh '<hr><a name="connections"><h2>Connections</h2>';
                    630: 
                    631: print $fh "<table border=2>";
                    632: foreach $tryserver (keys %hostname) {
                    633: 
                    634:     $answer=reply("pong",$tryserver);
                    635:     if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
                    636: 	$result="<b>ok</b>";
                    637:     } else {
                    638:         $result=$answer;
                    639:         $warnings++;
                    640:         if ($answer eq 'con_lost') { $warnings++; }
                    641:     }
                    642:     print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n";
                    643: 
                    644: }
                    645: print $fh "</table>";
                    646: 
                    647: &errout($fh);
                    648: # ------------------------------------------------------------ Delayed messages
                    649: 
                    650: print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
1.23      www       651: print "buffers\n";
1.1       albertel  652: 
                    653: print $fh '<h3>Scanning Permanent Log</h3>';
                    654: 
                    655: $unsend=0;
                    656: {
                    657:     my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
                    658:     while ($line=<$dfh>) {
                    659: 	($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
                    660:         if ($sdf eq 'F') { 
                    661: 	    $local=localtime($time);
                    662:             print "<b>Failed: $time, $dserv, $dcmd</b><br>";
                    663:             $warnings++;
                    664:         }
                    665:         if ($sdf eq 'S') { $unsend--; }
                    666:         if ($sdf eq 'D') { $unsend++; }
                    667:     }
                    668: }
                    669: print $fh "Total unsend messages: <b>$unsend</b><p>\n";
                    670: $warnings=$warnings+5*$unsend;
                    671: 
                    672: print $fh "<h3>Outgoing Buffer</h3>";
                    673: 
                    674: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
                    675: while ($line=<DFH>) { 
                    676:     print $fh "$line<br>";
                    677: };
                    678: close (DFH);
                    679: 
                    680: # ------------------------------------------------------------------------- End
                    681: print $fh "<a name=errcount>\n";
                    682: $totalcount=$notices+4*$warnings+100*$errors;
                    683: &errout($fh);
                    684: print $fh "<h1>Total Error Count: $totalcount</h1>";
                    685: $now=time;
                    686: $date=localtime($now);
                    687: print $fh "<hr>$date ($now)</body></html>\n";
1.23      www       688: print "writing done\n";
1.1       albertel  689: }
                    690: 
                    691: rename ("$statusdir/newstatus.html","$statusdir/index.html");
                    692: 
                    693: if ($totalcount>200) {
1.23      www       694:    print "mailing\n";
1.1       albertel  695:    $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
                    696:    $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; 
                    697:    system(
1.24      www       698:  "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");
1.1       albertel  699: }
                    700: 1;
                    701: 
                    702: 
                    703: 
                    704: 
                    705: 
                    706: 
                    707: 
                    708: 

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