Annotation of loncom/loncron, revision 1.58
1.1 albertel 1: #!/usr/bin/perl
2:
1.47 albertel 3: # Housekeeping program, started by cron, loncontrol and loncron.pl
4: #
1.58 ! albertel 5: # $Id: loncron,v 1.57 2005/02/14 00:29:48 albertel Exp $
1.47 albertel 6: #
7: # Copyright Michigan State University Board of Trustees
8: #
9: # This file is part of the LearningOnline Network with CAPA (LON-CAPA).
10: #
11: # LON-CAPA is free software; you can redistribute it and/or modify
12: # it under the terms of the GNU General Public License as published by
13: # the Free Software Foundation; either version 2 of the License, or
14: # (at your option) any later version.
15: #
16: # LON-CAPA is distributed in the hope that it will be useful,
17: # but WITHOUT ANY WARRANTY; without even the implied warranty of
18: # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19: # GNU General Public License for more details.
20: #
21: # You should have received a copy of the GNU General Public License
22: # along with LON-CAPA; if not, write to the Free Software
23: # Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
24: #
25: # /home/httpd/html/adm/gpl.txt
26: #
27: # http://www.lon-capa.org/
1.1 albertel 28: #
1.24 www 29:
30: $|=1;
1.48 albertel 31: use strict;
1.1 albertel 32:
1.26 harris41 33: use lib '/home/httpd/lib/perl/';
34: use LONCAPA::Configuration;
35:
1.1 albertel 36: use IO::File;
37: use IO::Socket;
1.48 albertel 38: use HTML::Entities;
1.49 albertel 39: use Getopt::Long;
1.46 albertel 40: #globals
41: use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount);
42:
43: my $statusdir="/home/httpd/html/lon-status";
44:
1.1 albertel 45:
46: # -------------------------------------------------- Non-critical communication
47: sub reply {
48: my ($cmd,$server)=@_;
49: my $peerfile="$perlvar{'lonSockDir'}/$server";
50: my $client=IO::Socket::UNIX->new(Peer =>"$peerfile",
51: Type => SOCK_STREAM,
52: Timeout => 10)
53: or return "con_lost";
54: print $client "$cmd\n";
55: my $answer=<$client>;
56: chomp($answer);
57: if (!$answer) { $answer="con_lost"; }
58: return $answer;
59: }
60:
61: # --------------------------------------------------------- Output error status
62:
1.46 albertel 63: sub log {
64: my $fh=shift;
65: if ($fh) { print $fh @_ }
66: }
67:
1.1 albertel 68: sub errout {
69: my $fh=shift;
1.46 albertel 70: &log($fh,(<<ENDERROUT));
1.48 albertel 71: <table border="2" bgcolor="#CCCCCC">
1.1 albertel 72: <tr><td>Notices</td><td>$notices</td></tr>
73: <tr><td>Warnings</td><td>$warnings</td></tr>
74: <tr><td>Errors</td><td>$errors</td></tr>
1.48 albertel 75: </table><p><a href="#top">Top</a></p>
1.1 albertel 76: ENDERROUT
77: }
78:
1.42 albertel 79: sub start_daemon {
1.50 albertel 80: my ($fh,$daemon,$pidfile,$args) = @_;
1.44 albertel 81: my $progname=$daemon;
1.50 albertel 82: if ($daemon eq 'lonc' && $args eq 'new') {
1.44 albertel 83: $progname='loncnew';
84: print "new ";
85: }
1.51 albertel 86: my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors";
87: my $size=(stat($error_fname))[7];
88: if ($size>40000) {
89: &log($fh,"<p>Rotating error logs ...</p>");
90: rename("$error_fname.2","$error_fname.3");
91: rename("$error_fname.1","$error_fname.2");
92: rename("$error_fname","$error_fname.1");
93: }
94: system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors");
1.42 albertel 95: sleep 2;
96: if (-e $pidfile) {
1.48 albertel 97: &log($fh,"<p>Seems like it started ...</p>");
1.42 albertel 98: my $lfh=IO::File->new("$pidfile");
99: my $daemonpid=<$lfh>;
100: chomp($daemonpid);
101: sleep 2;
102: if (kill 0 => $daemonpid) {
103: return 1;
104: } else {
105: return 0;
106: }
107: }
1.48 albertel 108: &log($fh,"<p>Seems like that did not work!</p>");
1.42 albertel 109: $errors++;
110: return 0;
111: }
112:
113: sub checkon_daemon {
1.50 albertel 114: my ($fh,$daemon,$maxsize,$sendusr1,$args)=@_;
1.42 albertel 115:
1.48 albertel 116: &log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>');
1.57 albertel 117: printf("%-15s ",$daemon);
1.42 albertel 118: if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
119: open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
1.46 albertel 120: while (my $line=<DFH>) {
121: &log($fh,"$line");
1.42 albertel 122: if ($line=~/INFO/) { $notices++; }
123: if ($line=~/WARNING/) { $notices++; }
124: if ($line=~/CRITICAL/) { $warnings++; }
125: };
126: close (DFH);
127: }
1.48 albertel 128: &log($fh,"</tt></p>");
1.42 albertel 129:
130: my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid";
131:
132: my $restartflag=1;
1.46 albertel 133: my $daemonpid;
1.42 albertel 134: if (-e $pidfile) {
135: my $lfh=IO::File->new("$pidfile");
1.46 albertel 136: $daemonpid=<$lfh>;
1.42 albertel 137: chomp($daemonpid);
138: if (kill 0 => $daemonpid) {
1.46 albertel 139: &log($fh,"<h3>$daemon at pid $daemonpid responding");
140: if ($sendusr1) { &log($fh,", sending USR1"); }
141: &log($fh,"</h3>");
1.42 albertel 142: if ($sendusr1) { kill USR1 => $daemonpid; }
143: $restartflag=0;
144: print "running\n";
145: } else {
146: $errors++;
1.46 albertel 147: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.42 albertel 148: $restartflag=1;
1.46 albertel 149: &log($fh,"<h3>Decided to clean up stale .pid file and restart $daemon</h3>");
1.42 albertel 150: }
151: }
152: if ($restartflag==1) {
153: $simplestatus{$daemon}='off';
154: $errors++;
1.57 albertel 155: my $kadaemon=$daemon;
156: if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; }
1.46 albertel 157: &log($fh,'<br><font color="red">Killall '.$daemon.': '.
1.57 albertel 158: `killall $kadaemon 2>&1`.' - ');
1.42 albertel 159: sleep 2;
1.46 albertel 160: &log($fh,unlink($pidfile).' - '.
1.57 albertel 161: `killall -9 $kadaemon 2>&1`.
1.46 albertel 162: '</font><br>');
163: &log($fh,"<h3>$daemon not running, trying to start</h3>");
1.42 albertel 164:
1.50 albertel 165: if (&start_daemon($fh,$daemon,$pidfile,$args)) {
1.46 albertel 166: &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
1.42 albertel 167: $simplestatus{$daemon}='restarted';
168: print "started\n";
169: } else {
170: $errors++;
1.46 albertel 171: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.48 albertel 172: &log($fh,"<p>Give it one more try ...</p>");
1.42 albertel 173: print " ";
1.50 albertel 174: if (&start_daemon($fh,$daemon,$pidfile,$args)) {
1.46 albertel 175: &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
1.42 albertel 176: $simplestatus{$daemon}='restarted';
177: print "started\n";
178: } else {
179: print " failed\n";
180: $simplestatus{$daemon}='failed';
181: $errors++; $errors++;
1.46 albertel 182: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.48 albertel 183: &log($fh,"<p>Unable to start $daemon</p>");
1.42 albertel 184: }
185: }
186:
187: if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
1.46 albertel 188: &log($fh,"<p><pre>");
1.42 albertel 189: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|");
1.46 albertel 190: while (my $line=<DFH>) {
191: &log($fh,"$line");
1.42 albertel 192: if ($line=~/WARNING/) { $notices++; }
193: if ($line=~/CRITICAL/) { $notices++; }
194: };
195: close (DFH);
1.48 albertel 196: &log($fh,"</pre></p>");
1.42 albertel 197: }
198: }
199:
1.46 albertel 200: my $fname="$perlvar{'lonDaemons'}/logs/$daemon.log";
1.42 albertel 201:
202: my ($dev,$ino,$mode,$nlink,
203: $uid,$gid,$rdev,$size,
204: $atime,$mtime,$ctime,
205: $blksize,$blocks)=stat($fname);
206:
207: if ($size>$maxsize) {
1.48 albertel 208: &log($fh,"<p>Rotating logs ...</p>");
1.42 albertel 209: rename("$fname.2","$fname.3");
210: rename("$fname.1","$fname.2");
211: rename("$fname","$fname.1");
212: }
213:
214: &errout($fh);
215: }
1.1 albertel 216:
1.46 albertel 217: # --------------------------------------------------------------------- Machine
218: sub log_machine_info {
219: my ($fh)=@_;
1.48 albertel 220: &log($fh,'<hr /><a name="machine" /><h2>Machine Information</h2>');
1.46 albertel 221: &log($fh,"<h3>loadavg</h3>");
222:
223: open (LOADAVGH,"/proc/loadavg");
224: my $loadavg=<LOADAVGH>;
225: close (LOADAVGH);
226:
227: &log($fh,"<tt>$loadavg</tt>");
228:
229: my @parts=split(/\s+/,$loadavg);
230: if ($parts[1]>4.0) {
231: $errors++;
232: } elsif ($parts[1]>2.0) {
233: $warnings++;
234: } elsif ($parts[1]>1.0) {
235: $notices++;
236: }
1.13 harris41 237:
1.46 albertel 238: &log($fh,"<h3>df</h3>");
239: &log($fh,"<pre>");
1.14 harris41 240:
1.46 albertel 241: open (DFH,"df|");
242: while (my $line=<DFH>) {
1.48 albertel 243: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 244: @parts=split(/\s+/,$line);
245: my $usage=$parts[4];
246: $usage=~s/\W//g;
247: if ($usage>90) {
248: $warnings++;
249: $notices++;
250: } elsif ($usage>80) {
251: $warnings++;
252: } elsif ($usage>60) {
253: $notices++;
1.31 albertel 254: }
1.46 albertel 255: if ($usage>95) { $warnings++; $warnings++; $simplestatus{'diskfull'}++; }
1.1 albertel 256: }
1.46 albertel 257: close (DFH);
258: &log($fh,"</pre>");
1.1 albertel 259:
260:
1.46 albertel 261: &log($fh,"<h3>ps</h3>");
262: &log($fh,"<pre>");
263: my $psproc=0;
1.1 albertel 264:
1.53 albertel 265: open (PSH,"ps aux --cols 140 |");
1.46 albertel 266: while (my $line=<PSH>) {
1.48 albertel 267: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 268: $psproc++;
269: }
270: close (PSH);
271: &log($fh,"</pre>");
1.1 albertel 272:
1.46 albertel 273: if ($psproc>200) { $notices++; }
274: if ($psproc>250) { $notices++; }
1.1 albertel 275:
1.46 albertel 276: &errout($fh);
277: }
1.1 albertel 278:
1.46 albertel 279: sub start_logging {
280: my ($hostdom,$hostrole,$hostname,$spareid)=@_;
1.43 albertel 281: my $fh=IO::File->new(">$statusdir/newstatus.html");
282: my %simplestatus=();
1.46 albertel 283: my $now=time;
284: my $date=localtime($now);
1.43 albertel 285:
1.46 albertel 286:
287: &log($fh,(<<ENDHEADERS));
1.1 albertel 288: <html>
289: <head>
290: <title>LON Status Report $perlvar{'lonHostID'}</title>
291: </head>
1.3 www 292: <body bgcolor="#AAAAAA">
1.48 albertel 293: <a name="top" />
1.1 albertel 294: <h1>LON Status Report $perlvar{'lonHostID'}</h1>
295: <h2>$date ($now)</h2>
296: <ol>
1.48 albertel 297: <li><a href="#configuration">Configuration</a></li>
298: <li><a href="#machine">Machine Information</a></li>
299: <li><a href="#tmp">Temporary Files</a></li>
300: <li><a href="#tokens">Session Tokens</a></li>
301: <li><a href="#httpd">httpd</a></li>
302: <li><a href="#lonsql">lonsql</a></li>
303: <li><a href="#lond">lond</a></li>
304: <li><a href="#lonc">lonc</a></li>
305: <li><a href="#lonhttpd">lonhttpd</a></li>
306: <li><a href="#lonnet">lonnet</a></li>
307: <li><a href="#connections">Connections</a></li>
308: <li><a href="#delayed">Delayed Messages</a></li>
309: <li><a href="#errcount">Error Count</a></li>
1.1 albertel 310: </ol>
1.48 albertel 311: <hr />
312: <a name="configuration" />
1.1 albertel 313: <h2>Configuration</h2>
314: <h3>PerlVars</h3>
1.48 albertel 315: <table border="2">
1.1 albertel 316: ENDHEADERS
317:
1.46 albertel 318: foreach my $varname (sort(keys(%perlvar))) {
1.48 albertel 319: &log($fh,"<tr><td>$varname</td><td>".
320: &encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n");
1.43 albertel 321: }
1.48 albertel 322: &log($fh,"</table><h3>Hosts</h3><table border='2'>");
1.46 albertel 323: foreach my $id (sort(keys(%{$hostname}))) {
324: &log($fh,
325: "<tr><td>$id</td><td>".$hostdom->{$id}.
326: "</td><td>".$hostrole->{$id}.
327: "</td><td>".$hostname->{$id}."</td></tr>\n");
328: }
329: &log($fh,"</table><h3>Spare Hosts</h3><ol>");
330: foreach my $id (sort(keys(%{$spareid}))) {
1.48 albertel 331: &log($fh,"<li>$id\n</li>");
1.43 albertel 332: }
1.46 albertel 333: &log($fh,"</ol>\n");
334: return $fh;
335: }
1.11 www 336:
337: # --------------------------------------------------------------- clean out tmp
1.46 albertel 338: sub clean_tmp {
339: my ($fh)=@_;
1.48 albertel 340: &log($fh,'<hr /><a name="tmp" /><h2>Temporary Files</h2>');
1.46 albertel 341: my $cleaned=0;
342: my $old=0;
343: while (my $fname=<$perlvar{'lonDaemons'}/tmp/*>) {
1.43 albertel 344: my ($dev,$ino,$mode,$nlink,
345: $uid,$gid,$rdev,$size,
346: $atime,$mtime,$ctime,
347: $blksize,$blocks)=stat($fname);
1.46 albertel 348: my $now=time;
349: my $since=$now-$mtime;
1.43 albertel 350: if ($since>$perlvar{'lonExpire'}) {
1.46 albertel 351: my $line='';
1.43 albertel 352: if (open(PROBE,$fname)) {
353: $line=<PROBE>;
354: close(PROBE);
355: }
356: unless ($line=~/^CHECKOUTTOKEN\&/) {
357: $cleaned++;
358: unlink("$fname");
359: } else {
360: if ($since>365*$perlvar{'lonExpire'}) {
361: $cleaned++;
362: unlink("$fname");
363: } else { $old++; }
364: }
365: }
366: }
1.46 albertel 367: &log($fh,"Cleaned up ".$cleaned." files (".$old." old checkout tokens).");
368: }
1.11 www 369:
370: # ------------------------------------------------------------ clean out lonIDs
1.46 albertel 371: sub clean_lonIDs {
372: my ($fh)=@_;
1.48 albertel 373: &log($fh,'<hr /><a name="tokens" /><h2>Session Tokens</h2>');
1.46 albertel 374: my $cleaned=0;
375: my $active=0;
376: while (my $fname=<$perlvar{'lonIDsDir'}/*>) {
1.43 albertel 377: my ($dev,$ino,$mode,$nlink,
378: $uid,$gid,$rdev,$size,
379: $atime,$mtime,$ctime,
380: $blksize,$blocks)=stat($fname);
1.46 albertel 381: my $now=time;
382: my $since=$now-$mtime;
1.43 albertel 383: if ($since>$perlvar{'lonExpire'}) {
384: $cleaned++;
1.46 albertel 385: &log($fh,"Unlinking $fname<br>");
1.43 albertel 386: unlink("$fname");
387: } else {
388: $active++;
389: }
1.46 albertel 390: }
1.48 albertel 391: &log($fh,"<p>Cleaned up ".$cleaned." stale session token(s).</p>");
1.46 albertel 392: &log($fh,"<h3>$active open session(s)</h3>");
393: }
1.43 albertel 394:
1.11 www 395:
1.1 albertel 396: # ----------------------------------------------------------------------- httpd
1.46 albertel 397: sub check_httpd_logs {
398: my ($fh)=@_;
1.48 albertel 399: &log($fh,'<hr /><a name="httpd" /><h2>httpd</h2><h3>Access Log</h3><pre>');
1.43 albertel 400:
401: open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
1.48 albertel 402: while (my $line=<DFH>) { &log($fh,&encode_entities($line,'<>&"')) };
1.43 albertel 403: close (DFH);
1.46 albertel 404:
405: &log($fh,"</pre><h3>Error Log</h3><pre>");
406:
1.43 albertel 407: open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
1.46 albertel 408: while (my $line=<DFH>) {
409: &log($fh,"$line");
1.43 albertel 410: if ($line=~/\[error\]/) { $notices++; }
1.46 albertel 411: }
1.43 albertel 412: close (DFH);
1.46 albertel 413: &log($fh,"</pre>");
1.43 albertel 414: &errout($fh);
1.46 albertel 415: }
1.1 albertel 416:
417: # ---------------------------------------------------------------------- lonnet
418:
1.48 albertel 419: sub rotate_lonnet_logs {
1.46 albertel 420: my ($fh)=@_;
1.48 albertel 421: &log($fh,'<hr /><a name="lonnet" /><h2>lonnet</h2><h3>Temp Log</h3><pre>');
1.43 albertel 422: print "checking logs\n";
423: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
424: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
1.46 albertel 425: while (my $line=<DFH>) {
1.48 albertel 426: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 427: }
1.43 albertel 428: close (DFH);
429: }
1.46 albertel 430: &log($fh,"</pre><h3>Perm Log</h3><pre>");
1.43 albertel 431:
432: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
433: open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
1.46 albertel 434: while (my $line=<DFH>) {
1.48 albertel 435: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 436: }
1.43 albertel 437: close (DFH);
1.46 albertel 438: } else { &log($fh,"No perm log\n") }
1.43 albertel 439:
1.46 albertel 440: my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
1.43 albertel 441:
442: my ($dev,$ino,$mode,$nlink,
443: $uid,$gid,$rdev,$size,
444: $atime,$mtime,$ctime,
445: $blksize,$blocks)=stat($fname);
446:
447: if ($size>40000) {
1.48 albertel 448: &log($fh,"<p>Rotating logs ...</p>");
1.43 albertel 449: rename("$fname.2","$fname.3");
450: rename("$fname.1","$fname.2");
451: rename("$fname","$fname.1");
452: }
1.1 albertel 453:
1.46 albertel 454: &log($fh,"</pre>");
1.43 albertel 455: &errout($fh);
1.46 albertel 456: }
457:
1.43 albertel 458: # ----------------------------------------------------------------- Connections
1.46 albertel 459: sub test_connections {
460: my ($fh,$hostname)=@_;
1.48 albertel 461: &log($fh,'<hr /><a name="connections" /><h2>Connections</h2>');
1.43 albertel 462: print "testing connections\n";
1.48 albertel 463: &log($fh,"<table border='2'>");
1.49 albertel 464: my ($good,$bad)=(0,0);
1.46 albertel 465: foreach my $tryserver (sort(keys(%{$hostname}))) {
1.43 albertel 466: print(".");
1.46 albertel 467: my $result;
1.55 albertel 468: my $answer=reply("ping",$tryserver);
1.43 albertel 469: if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
470: $result="<b>ok</b>";
1.49 albertel 471: $good++;
1.43 albertel 472: } else {
473: $result=$answer;
474: $warnings++;
1.49 albertel 475: if ($answer eq 'con_lost') {
476: $bad++;
477: $warnings++;
1.50 albertel 478: } else {
479: $good++; #self connection
1.49 albertel 480: }
1.43 albertel 481: }
482: if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }
1.46 albertel 483: &log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n");
1.1 albertel 484: }
1.46 albertel 485: &log($fh,"</table>");
1.49 albertel 486: print "\n$good good, $bad bad connections\n";
1.43 albertel 487: &errout($fh);
1.46 albertel 488: }
489:
490:
1.1 albertel 491: # ------------------------------------------------------------ Delayed messages
1.46 albertel 492: sub check_delayed_msg {
493: my ($fh)=@_;
1.48 albertel 494: &log($fh,'<hr /><a name="delayed" /><h2>Delayed Messages</h2>');
1.43 albertel 495: print "checking buffers\n";
1.46 albertel 496:
497: &log($fh,'<h3>Scanning Permanent Log</h3>');
1.1 albertel 498:
1.46 albertel 499: my $unsend=0;
1.1 albertel 500:
1.46 albertel 501: my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
502: while (my $line=<$dfh>) {
503: my ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
504: if ($sdf eq 'F') {
505: my $local=localtime($time);
506: &log($fh,"<b>Failed: $time, $dserv, $dcmd</b><br>");
507: $warnings++;
1.43 albertel 508: }
1.46 albertel 509: if ($sdf eq 'S') { $unsend--; }
510: if ($sdf eq 'D') { $unsend++; }
1.1 albertel 511: }
1.46 albertel 512:
1.48 albertel 513: &log($fh,"<p>Total unsend messages: <b>$unsend</b></p>\n");
1.43 albertel 514: $warnings=$warnings+5*$unsend;
1.1 albertel 515:
1.43 albertel 516: if ($unsend) { $simplestatus{'unsend'}=$unsend; }
1.48 albertel 517: &log($fh,"<h3>Outgoing Buffer</h3>\n<pre>");
1.1 albertel 518:
1.43 albertel 519: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
1.46 albertel 520: while (my $line=<DFH>) {
1.48 albertel 521: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 522: }
1.48 albertel 523: &log($fh,"</pre>\n");
1.43 albertel 524: close (DFH);
1.46 albertel 525: }
1.1 albertel 526:
1.46 albertel 527: sub finish_logging {
528: my ($fh)=@_;
1.48 albertel 529: &log($fh,"<a name='errcount' />\n");
1.43 albertel 530: $totalcount=$notices+4*$warnings+100*$errors;
531: &errout($fh);
1.46 albertel 532: &log($fh,"<h1>Total Error Count: $totalcount</h1>");
533: my $now=time;
534: my $date=localtime($now);
1.48 albertel 535: &log($fh,"<hr />$date ($now)</body></html>\n");
1.43 albertel 536: print "lon-status webpage updated\n";
537: $fh->close();
1.46 albertel 538:
539: if ($errors) { $simplestatus{'errors'}=$errors; }
540: if ($warnings) { $simplestatus{'warnings'}=$warnings; }
541: if ($notices) { $simplestatus{'notices'}=$notices; }
542: $simplestatus{'time'}=time;
1.1 albertel 543: }
544:
1.46 albertel 545: sub log_simplestatus {
546: rename ("$statusdir/newstatus.html","$statusdir/index.html");
547:
1.43 albertel 548: my $sfh=IO::File->new(">$statusdir/loncron_simple.txt");
549: foreach (keys %simplestatus) {
550: print $sfh $_.'='.$simplestatus{$_}.'&';
551: }
552: print $sfh "\n";
553: $sfh->close();
1.41 www 554: }
1.46 albertel 555:
556: sub send_mail {
1.43 albertel 557: print "sending mail\n";
1.46 albertel 558: my $emailto="$perlvar{'lonAdmEMail'}";
1.54 www 559: if ($totalcount>2500) {
1.43 albertel 560: $emailto.=",$perlvar{'lonSysEMail'}";
561: }
1.46 albertel 562: my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";
1.52 albertel 563:
1.58 ! albertel 564: my $result=system("metasend -b -S 4000000 -t $emailto -s '$subj' -f $statusdir/index.html -m text/html >& /dev/null");
1.52 albertel 565: if ($result != 0) {
566: $result=system("mail -s '$subj' $emailto < $statusdir/index.html");
567: }
1.1 albertel 568: }
1.46 albertel 569:
1.49 albertel 570: sub usage {
571: print(<<USAGE);
572: loncron - housekeeping program that checks up on various parts of Lon-CAPA
573:
574: Options:
575: --help Display help
576: --oldlonc When starting the lonc daemon use 'lonc' not 'loncnew'
577: --noemail Do not send the status email
578: --justcheckconnections Only check the current status of the lonc/d
579: connections, do not send emails do not
580: check if the daemons are running, do not
581: generate lon-status
582: --justcheckdaemons Only check that all of the Lon-CAPA daemons are
583: running, do not send emails do not
584: check the lonc/d connections, do not
585: generate lon-status
586:
587: USAGE
588: }
589:
1.46 albertel 590: # ================================================================ Main Program
591: sub main () {
1.49 albertel 592: my ($oldlonc,$help,$justcheckdaemons,$noemail,$justcheckconnections);
593: &GetOptions("help" => \$help,
594: "oldlonc" => \$oldlonc,
595: "justcheckdaemons" => \$justcheckdaemons,
596: "noemail" => \$noemail,
597: "justcheckconnections" => \$justcheckconnections
598: );
599: if ($help) { &usage(); return; }
1.46 albertel 600: # --------------------------------- Read loncapa_apache.conf and loncapa.conf
601: my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
602: %perlvar=%{$perlvarref};
603: undef $perlvarref;
604: delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
605: delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
606:
607: # --------------------------------------- Make sure that LON-CAPA is configured
608: # I only test for one thing here (lonHostID). This is just a safeguard.
609: if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) {
610: print("Unconfigured machine.\n");
611: my $emailto=$perlvar{'lonSysEMail'};
612: my $hostname=`/bin/hostname`;
613: chop $hostname;
614: $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
615: my $subj="LON: Unconfigured machine $hostname";
616: system("echo 'Unconfigured machine $hostname.' |\
617: mailto $emailto -s '$subj' > /dev/null");
618: exit 1;
619: }
620:
621: # ----------------------------- Make sure this process is running from user=www
622: my $wwwid=getpwnam('www');
623: if ($wwwid!=$<) {
624: print("User ID mismatch. This program must be run as user 'www'\n");
625: my $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
626: my $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
627: system("echo 'User ID mismatch. loncron must be run as user www.' |\
628: mailto $emailto -s '$subj' > /dev/null");
629: exit 1;
630: }
631:
632: # ------------------------------------------------------------- Read hosts file
633: my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
634:
635: my (%hostname,%hostdom,%hostrole,%spareid);
636: while (my $configline=<$config>) {
1.49 albertel 637: next if ($configline =~ /^(\#|\s*\$)/);
1.56 albertel 638: my ($id,$domain,$role,$name)=split(/:/,$configline);
639: if ($id && $domain && $role && $name) {
640: $name=~s/\s//g;
1.46 albertel 641: $hostname{$id}=$name;
642: $hostdom{$id}=$domain;
643: $hostrole{$id}=$role;
644: }
645: }
646: undef $config;
647:
648: # ------------------------------------------------------ Read spare server file
649: $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
650:
651: while (my $configline=<$config>) {
652: chomp($configline);
653: if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
654: $spareid{$configline}=1;
655: }
656: }
657: undef $config;
658:
659: # ---------------------------------------------------------------- Start report
660:
661: $errors=0;
662: $warnings=0;
663: $notices=0;
664:
665:
1.49 albertel 666: my $fh;
667: if (!$justcheckdaemons && !$justcheckconnections) {
668: $fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);
669:
670: &log_machine_info($fh);
671: &clean_tmp($fh);
672: &clean_lonIDs($fh);
673: &check_httpd_logs($fh);
674: &rotate_lonnet_logs($fh);
675: }
676: if (!$justcheckconnections) {
677: &checkon_daemon($fh,'lonsql',200000);
678: &checkon_daemon($fh,'lond',40000,1);
1.50 albertel 679: my $args='new';
680: if ($oldlonc) { $args = ''; }
681: &checkon_daemon($fh,'lonc',40000,1,$args);
1.49 albertel 682: &checkon_daemon($fh,'lonhttpd',40000);
1.57 albertel 683: &checkon_daemon($fh,'lonmemcached',40000);
1.49 albertel 684: }
685: if (!$justcheckdaemons) {
686: &test_connections($fh,\%hostname);
687: }
688: if (!$justcheckdaemons && !$justcheckconnections) {
689: &check_delayed_msg($fh);
690: &finish_logging($fh);
691: &log_simplestatus();
692:
693: if ($totalcount>200 && !$noemail) { &send_mail(); }
694: }
1.46 albertel 695: }
696:
697: &main();
1.1 albertel 698: 1;
699:
700:
701:
702:
703:
704:
705:
706:
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>
![[BACK]](/icons/back.gif){kind=icon}
Return to loncron CVS log ![[TXT]](/icons/text.gif){kind=icon}
![[DIR]](/icons/dir.gif){kind=icon}
Up to [LON-CAPA] / loncom