Annotation of loncom/loncron, revision 1.95
1.1 albertel 1: #!/usr/bin/perl
2:
1.47 albertel 3: # Housekeeping program, started by cron, loncontrol and loncron.pl
4: #
1.95 ! raeburn 5: # $Id: loncron,v 1.94 2011/11/08 22:25:25 raeburn 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;
1.89 raeburn 35: use LONCAPA;
1.72 albertel 36: use Apache::lonnet;
1.79 raeburn 37: use Apache::loncommon;
1.26 harris41 38:
1.1 albertel 39: use IO::File;
40: use IO::Socket;
1.48 albertel 41: use HTML::Entities;
1.49 albertel 42: use Getopt::Long;
1.46 albertel 43: #globals
44: use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount);
45:
46: my $statusdir="/home/httpd/html/lon-status";
47:
1.1 albertel 48:
49: # --------------------------------------------------------- Output error status
50:
1.46 albertel 51: sub log {
52: my $fh=shift;
53: if ($fh) { print $fh @_ }
54: }
55:
1.1 albertel 56: sub errout {
57: my $fh=shift;
1.46 albertel 58: &log($fh,(<<ENDERROUT));
1.48 albertel 59: <table border="2" bgcolor="#CCCCCC">
1.1 albertel 60: <tr><td>Notices</td><td>$notices</td></tr>
61: <tr><td>Warnings</td><td>$warnings</td></tr>
62: <tr><td>Errors</td><td>$errors</td></tr>
1.48 albertel 63: </table><p><a href="#top">Top</a></p>
1.1 albertel 64: ENDERROUT
65: }
66:
1.73 albertel 67: sub rotate_logfile {
68: my ($file,$fh,$description) = @_;
69: my $size=(stat($file))[7];
70: if ($size>40000) {
71: &log($fh,"<p>Rotating $description ...</p>");
72: rename("$file.2","$file.3");
73: rename("$file.1","$file.2");
74: rename("$file","$file.1");
75: }
76: }
77:
1.42 albertel 78: sub start_daemon {
1.50 albertel 79: my ($fh,$daemon,$pidfile,$args) = @_;
1.44 albertel 80: my $progname=$daemon;
1.71 albertel 81: if ($daemon eq 'lonc') {
1.44 albertel 82: $progname='loncnew';
83: }
1.51 albertel 84: my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors";
1.73 albertel 85: &rotate_logfile($error_fname,$fh,'error logs');
1.74 albertel 86: if ($daemon eq 'lonc') {
87: &clean_sockets($fh);
88: }
1.51 albertel 89: system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors");
1.60 albertel 90: sleep 1;
1.42 albertel 91: if (-e $pidfile) {
1.48 albertel 92: &log($fh,"<p>Seems like it started ...</p>");
1.42 albertel 93: my $lfh=IO::File->new("$pidfile");
94: my $daemonpid=<$lfh>;
95: chomp($daemonpid);
1.62 albertel 96: if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
1.42 albertel 97: return 1;
98: } else {
99: return 0;
100: }
101: }
1.48 albertel 102: &log($fh,"<p>Seems like that did not work!</p>");
1.42 albertel 103: $errors++;
104: return 0;
105: }
106:
107: sub checkon_daemon {
1.59 albertel 108: my ($fh,$daemon,$maxsize,$send,$args)=@_;
1.42 albertel 109:
1.63 albertel 110: my $result;
1.48 albertel 111: &log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>');
1.57 albertel 112: printf("%-15s ",$daemon);
1.42 albertel 113: if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
114: open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
1.46 albertel 115: while (my $line=<DFH>) {
116: &log($fh,"$line");
1.42 albertel 117: if ($line=~/INFO/) { $notices++; }
118: if ($line=~/WARNING/) { $notices++; }
119: if ($line=~/CRITICAL/) { $warnings++; }
120: };
121: close (DFH);
122: }
1.48 albertel 123: &log($fh,"</tt></p>");
1.42 albertel 124:
125: my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid";
126:
127: my $restartflag=1;
1.46 albertel 128: my $daemonpid;
1.42 albertel 129: if (-e $pidfile) {
130: my $lfh=IO::File->new("$pidfile");
1.46 albertel 131: $daemonpid=<$lfh>;
1.42 albertel 132: chomp($daemonpid);
1.62 albertel 133: if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
1.46 albertel 134: &log($fh,"<h3>$daemon at pid $daemonpid responding");
1.59 albertel 135: if ($send) { &log($fh,", sending $send"); }
1.46 albertel 136: &log($fh,"</h3>");
1.59 albertel 137: if ($send eq 'USR1') { kill USR1 => $daemonpid; }
138: if ($send eq 'USR2') { kill USR2 => $daemonpid; }
1.42 albertel 139: $restartflag=0;
1.59 albertel 140: if ($send eq 'USR2') {
1.63 albertel 141: $result = 'reloaded';
1.59 albertel 142: print "reloaded\n";
143: } else {
1.63 albertel 144: $result = 'running';
1.59 albertel 145: print "running\n";
146: }
1.42 albertel 147: } else {
148: $errors++;
1.46 albertel 149: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.42 albertel 150: $restartflag=1;
1.46 albertel 151: &log($fh,"<h3>Decided to clean up stale .pid file and restart $daemon</h3>");
1.42 albertel 152: }
153: }
154: if ($restartflag==1) {
155: $simplestatus{$daemon}='off';
156: $errors++;
1.57 albertel 157: my $kadaemon=$daemon;
158: if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; }
1.46 albertel 159: &log($fh,'<br><font color="red">Killall '.$daemon.': '.
1.57 albertel 160: `killall $kadaemon 2>&1`.' - ');
1.60 albertel 161: sleep 1;
1.46 albertel 162: &log($fh,unlink($pidfile).' - '.
1.57 albertel 163: `killall -9 $kadaemon 2>&1`.
1.46 albertel 164: '</font><br>');
165: &log($fh,"<h3>$daemon not running, trying to start</h3>");
1.42 albertel 166:
1.50 albertel 167: if (&start_daemon($fh,$daemon,$pidfile,$args)) {
1.46 albertel 168: &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
1.42 albertel 169: $simplestatus{$daemon}='restarted';
1.63 albertel 170: $result = 'started';
1.42 albertel 171: print "started\n";
172: } else {
173: $errors++;
1.46 albertel 174: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.48 albertel 175: &log($fh,"<p>Give it one more try ...</p>");
1.42 albertel 176: print " ";
1.50 albertel 177: if (&start_daemon($fh,$daemon,$pidfile,$args)) {
1.46 albertel 178: &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
1.42 albertel 179: $simplestatus{$daemon}='restarted';
1.63 albertel 180: $result = 'started';
1.42 albertel 181: print "started\n";
182: } else {
1.63 albertel 183: $result = 'failed';
1.42 albertel 184: print " failed\n";
185: $simplestatus{$daemon}='failed';
186: $errors++; $errors++;
1.46 albertel 187: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.48 albertel 188: &log($fh,"<p>Unable to start $daemon</p>");
1.42 albertel 189: }
190: }
191:
192: if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
1.46 albertel 193: &log($fh,"<p><pre>");
1.42 albertel 194: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|");
1.46 albertel 195: while (my $line=<DFH>) {
196: &log($fh,"$line");
1.42 albertel 197: if ($line=~/WARNING/) { $notices++; }
198: if ($line=~/CRITICAL/) { $notices++; }
199: };
200: close (DFH);
1.48 albertel 201: &log($fh,"</pre></p>");
1.42 albertel 202: }
203: }
204:
1.46 albertel 205: my $fname="$perlvar{'lonDaemons'}/logs/$daemon.log";
1.73 albertel 206: &rotate_logfile($fname,$fh,'logs');
1.42 albertel 207:
208: &errout($fh);
1.63 albertel 209: return $result;
1.42 albertel 210: }
1.1 albertel 211:
1.46 albertel 212: # --------------------------------------------------------------------- Machine
213: sub log_machine_info {
214: my ($fh)=@_;
1.48 albertel 215: &log($fh,'<hr /><a name="machine" /><h2>Machine Information</h2>');
1.46 albertel 216: &log($fh,"<h3>loadavg</h3>");
217:
218: open (LOADAVGH,"/proc/loadavg");
219: my $loadavg=<LOADAVGH>;
220: close (LOADAVGH);
221:
222: &log($fh,"<tt>$loadavg</tt>");
223:
224: my @parts=split(/\s+/,$loadavg);
225: if ($parts[1]>4.0) {
226: $errors++;
227: } elsif ($parts[1]>2.0) {
228: $warnings++;
229: } elsif ($parts[1]>1.0) {
230: $notices++;
231: }
1.13 harris41 232:
1.46 albertel 233: &log($fh,"<h3>df</h3>");
234: &log($fh,"<pre>");
1.14 harris41 235:
1.46 albertel 236: open (DFH,"df|");
237: while (my $line=<DFH>) {
1.48 albertel 238: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 239: @parts=split(/\s+/,$line);
240: my $usage=$parts[4];
241: $usage=~s/\W//g;
242: if ($usage>90) {
243: $warnings++;
244: $notices++;
245: } elsif ($usage>80) {
246: $warnings++;
247: } elsif ($usage>60) {
248: $notices++;
1.31 albertel 249: }
1.46 albertel 250: if ($usage>95) { $warnings++; $warnings++; $simplestatus{'diskfull'}++; }
1.1 albertel 251: }
1.46 albertel 252: close (DFH);
253: &log($fh,"</pre>");
1.1 albertel 254:
255:
1.46 albertel 256: &log($fh,"<h3>ps</h3>");
257: &log($fh,"<pre>");
258: my $psproc=0;
1.1 albertel 259:
1.53 albertel 260: open (PSH,"ps aux --cols 140 |");
1.46 albertel 261: while (my $line=<PSH>) {
1.48 albertel 262: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 263: $psproc++;
264: }
265: close (PSH);
266: &log($fh,"</pre>");
1.1 albertel 267:
1.46 albertel 268: if ($psproc>200) { $notices++; }
269: if ($psproc>250) { $notices++; }
1.1 albertel 270:
1.61 albertel 271: &log($fh,"<h3>distprobe</h3>");
272: &log($fh,"<pre>");
273: open(DSH,"$perlvar{'lonDaemons'}/distprobe |");
274: while (my $line=<DSH>) {
275: &log($fh,&encode_entities($line,'<>&"'));
276: $psproc++;
277: }
278: close(DSH);
279: &log($fh,"</pre>");
280:
1.46 albertel 281: &errout($fh);
282: }
1.1 albertel 283:
1.46 albertel 284: sub start_logging {
1.43 albertel 285: my $fh=IO::File->new(">$statusdir/newstatus.html");
286: my %simplestatus=();
1.46 albertel 287: my $now=time;
288: my $date=localtime($now);
1.43 albertel 289:
1.46 albertel 290:
291: &log($fh,(<<ENDHEADERS));
1.1 albertel 292: <html>
293: <head>
294: <title>LON Status Report $perlvar{'lonHostID'}</title>
295: </head>
1.3 www 296: <body bgcolor="#AAAAAA">
1.48 albertel 297: <a name="top" />
1.1 albertel 298: <h1>LON Status Report $perlvar{'lonHostID'}</h1>
299: <h2>$date ($now)</h2>
300: <ol>
1.48 albertel 301: <li><a href="#configuration">Configuration</a></li>
302: <li><a href="#machine">Machine Information</a></li>
303: <li><a href="#tmp">Temporary Files</a></li>
304: <li><a href="#tokens">Session Tokens</a></li>
305: <li><a href="#httpd">httpd</a></li>
306: <li><a href="#lonsql">lonsql</a></li>
307: <li><a href="#lond">lond</a></li>
308: <li><a href="#lonc">lonc</a></li>
309: <li><a href="#lonnet">lonnet</a></li>
310: <li><a href="#connections">Connections</a></li>
311: <li><a href="#delayed">Delayed Messages</a></li>
312: <li><a href="#errcount">Error Count</a></li>
1.1 albertel 313: </ol>
1.48 albertel 314: <hr />
315: <a name="configuration" />
1.1 albertel 316: <h2>Configuration</h2>
317: <h3>PerlVars</h3>
1.48 albertel 318: <table border="2">
1.1 albertel 319: ENDHEADERS
320:
1.46 albertel 321: foreach my $varname (sort(keys(%perlvar))) {
1.48 albertel 322: &log($fh,"<tr><td>$varname</td><td>".
323: &encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n");
1.43 albertel 324: }
1.48 albertel 325: &log($fh,"</table><h3>Hosts</h3><table border='2'>");
1.72 albertel 326: my %hostname = &Apache::lonnet::all_hostnames();
327: foreach my $id (sort(keys(%hostname))) {
328: my $role = (&Apache::lonnet::is_library($id) ? 'library'
329: : 'access');
1.46 albertel 330: &log($fh,
1.72 albertel 331: "<tr><td>$id</td><td>".&Apache::lonnet::host_domain($id).
332: "</td><td>".$role.
333: "</td><td>".&Apache::lonnet::hostname($id)."</td></tr>\n");
334: }
335: &log($fh,"</table><h3>Spare Hosts</h3><ul>");
336: foreach my $type (sort(keys(%Apache::lonnet::spareid))) {
337: &log($fh,"<li>$type\n<ol>");
338: foreach my $id (@{ $Apache::lonnet::spareid{$type} }) {
339: &log($fh,"<li>$id</li>\n");
340: }
341: &log($fh,"</ol>\n</li>\n");
1.43 albertel 342: }
1.72 albertel 343: &log($fh,"</ul>\n");
1.46 albertel 344: return $fh;
345: }
1.11 www 346:
347: # --------------------------------------------------------------- clean out tmp
1.46 albertel 348: sub clean_tmp {
349: my ($fh)=@_;
1.48 albertel 350: &log($fh,'<hr /><a name="tmp" /><h2>Temporary Files</h2>');
1.82 raeburn 351: my ($cleaned,$old,$removed) = (0,0,0);
352: my %errors = (
353: dir => [],
354: file => [],
355: failopen => [],
356: );
357: my %error_titles = (
358: dir => 'failed to remove empty directory:',
359: file => 'failed to unlike stale file',
360: failopen => 'failed to open file or directory'
361: );
362: ($cleaned,$old,$removed) = &recursive_clean_tmp('',$cleaned,$old,$removed,\%errors);
363: &log($fh,"Cleaned up: ".$cleaned." files; removed: $removed empty directories; (found: $old old checkout tokens)");
364: foreach my $key (sort(keys(%errors))) {
365: if (ref($errors{$key}) eq 'ARRAY') {
366: if (@{$errors{$key}} > 0) {
367: &log($fh,"Error during cleanup ($error_titles{$key}):<ul><li>".
368: join('</li><li><tt>',@{$errors{$key}}).'</tt></li></ul><br />');
369: }
370: }
371: }
372: }
373:
374: sub recursive_clean_tmp {
375: my ($subdir,$cleaned,$old,$removed,$errors) = @_;
376: my $base = "$perlvar{'lonDaemons'}/tmp";
377: my $path = $base;
378: next if ($subdir =~ m{\.\./});
379: next unless (ref($errors) eq 'HASH');
380: unless ($subdir eq '') {
381: $path .= '/'.$subdir;
382: }
383: if (opendir(my $dh,"$path")) {
384: while (my $file = readdir($dh)) {
385: next if ($file =~ /^\.\.?$/);
386: my $fname = "$path/$file";
387: if (-d $fname) {
388: my $innerdir;
389: if ($subdir eq '') {
390: $innerdir = $file;
391: } else {
392: $innerdir = $subdir.'/'.$file;
393: }
394: ($cleaned,$old,$removed) =
395: &recursive_clean_tmp($innerdir,$cleaned,$old,$removed,$errors);
396: my @doms = &Apache::lonnet::current_machine_domains();
397:
398: if (open(my $dirhandle,$fname)) {
399: unless (($innerdir eq 'helprequests') ||
400: (($innerdir =~ /^addcourse/) && ($innerdir !~ m{/\d+$}))) {
401: my @contents = grep {!/^\.\.?$/} readdir($dirhandle);
402: join('&&',@contents)."\n";
403: if (scalar(grep {!/^\.\.?$/} readdir($dirhandle)) == 0) {
404: closedir($dirhandle);
405: if ($fname =~ m{^\Q$perlvar{'lonDaemons'}\E/tmp/}) {
406: if (rmdir($fname)) {
407: $removed ++;
408: } elsif (ref($errors->{dir}) eq 'ARRAY') {
409: push(@{$errors->{dir}},$fname);
410: }
411: }
412: }
413: } else {
414: closedir($dirhandle);
415: }
416: }
417: } else {
418: my ($dev,$ino,$mode,$nlink,
419: $uid,$gid,$rdev,$size,
420: $atime,$mtime,$ctime,
421: $blksize,$blocks)=stat($fname);
422: my $now=time;
423: my $since=$now-$mtime;
424: if ($since>$perlvar{'lonExpire'}) {
425: if ($subdir eq '') {
426: my $line='';
427: if ($fname =~ /\.db$/) {
428: if (unlink($fname)) {
429: $cleaned++;
430: } elsif (ref($errors->{file}) eq 'ARRAY') {
431: push(@{$errors->{file}},$fname);
432: }
433: } elsif (open(PROBE,$fname)) {
434: my $line='';
435: $line=<PROBE>;
436: close(PROBE);
437: if ($line=~/^CHECKOUTTOKEN\&/) {
438: if ($since>365*$perlvar{'lonExpire'}) {
439: if (unlink($fname)) {
440: $cleaned++;
441: } elsif (ref($errors->{file}) eq 'ARRAY') {
442: push(@{$errors->{file}},$fname);
443: }
444: } else {
445: $old++;
446: }
447: } else {
448: if (unlink($fname)) {
449: $cleaned++;
450: } elsif (ref($errors->{file}) eq 'ARRAY') {
451: push(@{$errors->{file}},$fname);
452: }
453: }
454: } elsif (ref($errors->{failopen}) eq 'ARRAY') {
455: push(@{$errors->{failopen}},$fname);
456: }
457: } else {
458: if (unlink($fname)) {
459: $cleaned++;
460: } elsif (ref($errors->{file}) eq 'ARRAY') {
461: push(@{$errors->{file}},$fname);
462: }
463: }
464: }
465: }
466: }
467: closedir($dh);
468: } elsif (ref($errors->{failopen}) eq 'ARRAY') {
469: push(@{$errors->{failopen}},$path);
1.43 albertel 470: }
1.82 raeburn 471: return ($cleaned,$old,$removed);
1.46 albertel 472: }
1.11 www 473:
474: # ------------------------------------------------------------ clean out lonIDs
1.46 albertel 475: sub clean_lonIDs {
476: my ($fh)=@_;
1.48 albertel 477: &log($fh,'<hr /><a name="tokens" /><h2>Session Tokens</h2>');
1.46 albertel 478: my $cleaned=0;
479: my $active=0;
480: while (my $fname=<$perlvar{'lonIDsDir'}/*>) {
1.43 albertel 481: my ($dev,$ino,$mode,$nlink,
482: $uid,$gid,$rdev,$size,
483: $atime,$mtime,$ctime,
484: $blksize,$blocks)=stat($fname);
1.46 albertel 485: my $now=time;
486: my $since=$now-$mtime;
1.43 albertel 487: if ($since>$perlvar{'lonExpire'}) {
488: $cleaned++;
1.46 albertel 489: &log($fh,"Unlinking $fname<br>");
1.43 albertel 490: unlink("$fname");
491: } else {
492: $active++;
493: }
1.46 albertel 494: }
1.48 albertel 495: &log($fh,"<p>Cleaned up ".$cleaned." stale session token(s).</p>");
1.46 albertel 496: &log($fh,"<h3>$active open session(s)</h3>");
497: }
1.43 albertel 498:
1.74 albertel 499: # ----------------------------------------------------------- clean out sockets
500: sub clean_sockets {
501: my ($fh)=@_;
502: my $cleaned=0;
503: opendir(SOCKETS,$perlvar{'lonSockDir'});
504: while (my $fname=readdir(SOCKETS)) {
505: next if (-d $fname
1.80 www 506: || $fname=~/(mysqlsock|maximasock|rsock|\Q$perlvar{'lonSockDir'}\E)/);
1.74 albertel 507: $cleaned++;
508: &log($fh,"Unlinking $fname<br />");
509: unlink("/home/httpd/sockets/$fname");
510: }
511: &log($fh,"<p>Cleaned up ".$cleaned." stale sockets.</p>");
512: }
513:
1.11 www 514:
1.1 albertel 515: # ----------------------------------------------------------------------- httpd
1.46 albertel 516: sub check_httpd_logs {
517: my ($fh)=@_;
1.94 raeburn 518: if (open(PIPE,"./lchttpdlogs|")) {
1.93 raeburn 519: while (my $line=<PIPE>) {
520: &log($fh,$line);
521: if ($line=~/\[error\]/) { $notices++; }
522: }
523: close(PIPE);
1.46 albertel 524: }
1.43 albertel 525: &errout($fh);
1.46 albertel 526: }
1.1 albertel 527:
528: # ---------------------------------------------------------------------- lonnet
529:
1.48 albertel 530: sub rotate_lonnet_logs {
1.46 albertel 531: my ($fh)=@_;
1.48 albertel 532: &log($fh,'<hr /><a name="lonnet" /><h2>lonnet</h2><h3>Temp Log</h3><pre>');
1.43 albertel 533: print "checking logs\n";
534: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
535: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
1.46 albertel 536: while (my $line=<DFH>) {
1.48 albertel 537: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 538: }
1.43 albertel 539: close (DFH);
540: }
1.46 albertel 541: &log($fh,"</pre><h3>Perm Log</h3><pre>");
1.43 albertel 542:
543: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
544: open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
1.46 albertel 545: while (my $line=<DFH>) {
1.48 albertel 546: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 547: }
1.43 albertel 548: close (DFH);
1.46 albertel 549: } else { &log($fh,"No perm log\n") }
1.43 albertel 550:
1.46 albertel 551: my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
1.73 albertel 552: &rotate_logfile($fname,$fh,'lonnet log');
1.1 albertel 553:
1.46 albertel 554: &log($fh,"</pre>");
1.43 albertel 555: &errout($fh);
1.46 albertel 556: }
557:
1.73 albertel 558: sub rotate_other_logs {
559: my ($fh) = @_;
1.83 raeburn 560: my %logs = (
561: autoenroll => 'Auto Enroll log',
562: autocreate => 'Create Course log',
563: searchcat => 'Search Cataloguing log',
564: autoupdate => 'Auto Update log',
565: refreshcourseids_db => 'Refresh CourseIDs db log',
566: );
567: foreach my $item (keys(%logs)) {
568: my $fname=$perlvar{'lonDaemons'}.'/logs/'.$item.'.log';
569: &rotate_logfile($fname,$fh,$logs{$item});
570: }
1.73 albertel 571: }
572:
1.43 albertel 573: # ----------------------------------------------------------------- Connections
1.46 albertel 574: sub test_connections {
1.72 albertel 575: my ($fh)=@_;
1.48 albertel 576: &log($fh,'<hr /><a name="connections" /><h2>Connections</h2>');
1.43 albertel 577: print "testing connections\n";
1.48 albertel 578: &log($fh,"<table border='2'>");
1.49 albertel 579: my ($good,$bad)=(0,0);
1.72 albertel 580: my %hostname = &Apache::lonnet::all_hostnames();
581: foreach my $tryserver (sort(keys(%hostname))) {
1.43 albertel 582: print(".");
1.46 albertel 583: my $result;
1.72 albertel 584: my $answer=&Apache::lonnet::reply("ping",$tryserver);
1.43 albertel 585: if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
586: $result="<b>ok</b>";
1.49 albertel 587: $good++;
1.43 albertel 588: } else {
589: $result=$answer;
590: $warnings++;
1.49 albertel 591: if ($answer eq 'con_lost') {
592: $bad++;
593: $warnings++;
1.50 albertel 594: } else {
595: $good++; #self connection
1.49 albertel 596: }
1.43 albertel 597: }
598: if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }
1.46 albertel 599: &log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n");
1.1 albertel 600: }
1.46 albertel 601: &log($fh,"</table>");
1.49 albertel 602: print "\n$good good, $bad bad connections\n";
1.43 albertel 603: &errout($fh);
1.46 albertel 604: }
605:
606:
1.1 albertel 607: # ------------------------------------------------------------ Delayed messages
1.46 albertel 608: sub check_delayed_msg {
1.72 albertel 609: my ($fh)=@_;
1.48 albertel 610: &log($fh,'<hr /><a name="delayed" /><h2>Delayed Messages</h2>');
1.43 albertel 611: print "checking buffers\n";
1.46 albertel 612:
613: &log($fh,'<h3>Scanning Permanent Log</h3>');
1.1 albertel 614:
1.46 albertel 615: my $unsend=0;
1.1 albertel 616:
1.46 albertel 617: my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
618: while (my $line=<$dfh>) {
619: my ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
620: if ($sdf eq 'F') {
621: my $local=localtime($time);
622: &log($fh,"<b>Failed: $time, $dserv, $dcmd</b><br>");
623: $warnings++;
1.43 albertel 624: }
1.46 albertel 625: if ($sdf eq 'S') { $unsend--; }
626: if ($sdf eq 'D') { $unsend++; }
1.1 albertel 627: }
1.46 albertel 628:
1.48 albertel 629: &log($fh,"<p>Total unsend messages: <b>$unsend</b></p>\n");
1.95 ! raeburn 630: if ($unsend > 0) {
! 631: $warnings=$warnings+5*$unsend;
! 632: }
1.1 albertel 633:
1.43 albertel 634: if ($unsend) { $simplestatus{'unsend'}=$unsend; }
1.48 albertel 635: &log($fh,"<h3>Outgoing Buffer</h3>\n<pre>");
1.68 www 636: # list directory with delayed messages and remember offline servers
637: my %servers=();
1.43 albertel 638: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
1.68 www 639: while (my $line=<DFH>) {
640: my ($server)=($line=~/\.(\w+)$/);
641: if ($server) { $servers{$server}=1; }
1.48 albertel 642: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 643: }
1.48 albertel 644: &log($fh,"</pre>\n");
1.43 albertel 645: close (DFH);
1.95 ! raeburn 646: my %hostname = &Apache::lonnet::all_hostnames();
! 647: my $numhosts = scalar(keys(%hostname));
1.68 www 648: # pong to all servers that have delayed messages
649: # this will trigger a reverse connection, which should flush the buffers
1.95 ! raeburn 650: foreach my $tryserver (sort(keys(%servers))) {
! 651: if ($hostname{$tryserver} || !$numhosts) {
! 652: my $answer;
! 653: eval {
! 654: local $SIG{ ALRM } = sub { die "TIMEOUT" };
! 655: alarm(20);
! 656: $answer = &Apache::lonnet::reply("pong",$tryserver);
! 657: alarm(0);
! 658: };
! 659: if ($@ && $@ =~ m/TIMEOUT/) {
! 660: &log($fh,"Attempted pong to $tryserver timed out<br />");
! 661: print "time out while contacting: $tryserver for pong\n";
! 662: } else {
! 663: &log($fh,"Pong to $tryserver: $answer<br />");
! 664: }
1.91 raeburn 665: } else {
1.95 ! raeburn 666: &log($fh,"$tryserver has delayed messages, but is not part of the cluster -- skipping 'Pong'.<br />");
1.91 raeburn 667: }
1.68 www 668: }
1.46 albertel 669: }
1.1 albertel 670:
1.46 albertel 671: sub finish_logging {
672: my ($fh)=@_;
1.48 albertel 673: &log($fh,"<a name='errcount' />\n");
1.43 albertel 674: $totalcount=$notices+4*$warnings+100*$errors;
675: &errout($fh);
1.46 albertel 676: &log($fh,"<h1>Total Error Count: $totalcount</h1>");
677: my $now=time;
678: my $date=localtime($now);
1.48 albertel 679: &log($fh,"<hr />$date ($now)</body></html>\n");
1.43 albertel 680: print "lon-status webpage updated\n";
681: $fh->close();
1.46 albertel 682:
683: if ($errors) { $simplestatus{'errors'}=$errors; }
684: if ($warnings) { $simplestatus{'warnings'}=$warnings; }
685: if ($notices) { $simplestatus{'notices'}=$notices; }
686: $simplestatus{'time'}=time;
1.1 albertel 687: }
688:
1.46 albertel 689: sub log_simplestatus {
1.73 albertel 690: rename("$statusdir/newstatus.html","$statusdir/index.html");
1.46 albertel 691:
1.43 albertel 692: my $sfh=IO::File->new(">$statusdir/loncron_simple.txt");
693: foreach (keys %simplestatus) {
694: print $sfh $_.'='.$simplestatus{$_}.'&';
695: }
696: print $sfh "\n";
697: $sfh->close();
1.41 www 698: }
1.46 albertel 699:
1.84 raeburn 700: sub write_loncaparevs {
1.90 raeburn 701: print "Retrieving LON-CAPA version information\n";
1.84 raeburn 702: if (open(my $fh,">$perlvar{'lonTabDir'}/loncaparevs.tab")) {
703: my %hostname = &Apache::lonnet::all_hostnames();
704: foreach my $id (sort(keys(%hostname))) {
705: if ($id ne '') {
1.90 raeburn 706: my $loncaparev;
707: eval {
708: local $SIG{ ALRM } = sub { die "TIMEOUT" };
709: alarm(10);
710: $loncaparev =
711: &Apache::lonnet::get_server_loncaparev('',$id,1,'loncron');
712: alarm(0);
713: };
714: if ($@ && $@ =~ m/TIMEOUT/) {
715: print "time out while contacting lonHost: $id for version\n";
716: }
1.86 raeburn 717: if ($loncaparev =~ /^[\w.\-]+$/) {
1.84 raeburn 718: print $fh $id.':'.$loncaparev."\n";
719: }
720: }
721: }
722: close($fh);
723: }
724: return;
725: }
726:
1.85 raeburn 727: sub write_serverhomeIDs {
1.90 raeburn 728: print "Retrieving LON-CAPA lonHostID information\n";
1.85 raeburn 729: if (open(my $fh,">$perlvar{'lonTabDir'}/serverhomeIDs.tab")) {
730: my %name_to_host = &Apache::lonnet::all_names();
731: foreach my $name (sort(keys(%name_to_host))) {
732: if ($name ne '') {
733: if (ref($name_to_host{$name}) eq 'ARRAY') {
1.90 raeburn 734: my $serverhomeID;
735: eval {
736: local $SIG{ ALRM } = sub { die "TIMEOUT" };
737: alarm(10);
738: $serverhomeID =
739: &Apache::lonnet::get_server_homeID($name,1,'loncron');
740: alarm(0);
741: };
742: if ($@ && $@ =~ m/TIMEOUT/) {
743: print "Time out while contacting server: $name\n";
744: }
1.85 raeburn 745: if ($serverhomeID ne '') {
746: print $fh $name.':'.$serverhomeID."\n";
747: } else {
748: print $fh $name.':'.$name_to_host{$name}->[0]."\n";
749: }
750: }
751: }
752: }
753: close($fh);
754: }
755: return;
756: }
757:
1.46 albertel 758: sub send_mail {
1.43 albertel 759: print "sending mail\n";
1.79 raeburn 760: my $defdom = $perlvar{'lonDefDomain'};
761: my $origmail = $perlvar{'lonAdmEMail'};
1.78 raeburn 762: my $emailto = &Apache::loncommon::build_recipient_list(undef,
763: 'lonstatusmail',$defdom,$origmail);
1.54 www 764: if ($totalcount>2500) {
1.43 albertel 765: $emailto.=",$perlvar{'lonSysEMail'}";
766: }
1.46 albertel 767: my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";
1.52 albertel 768:
1.58 albertel 769: my $result=system("metasend -b -S 4000000 -t $emailto -s '$subj' -f $statusdir/index.html -m text/html >& /dev/null");
1.52 albertel 770: if ($result != 0) {
771: $result=system("mail -s '$subj' $emailto < $statusdir/index.html");
772: }
1.1 albertel 773: }
1.46 albertel 774:
1.49 albertel 775: sub usage {
776: print(<<USAGE);
777: loncron - housekeeping program that checks up on various parts of Lon-CAPA
778:
779: Options:
1.71 albertel 780: --help Display
1.49 albertel 781: --noemail Do not send the status email
782: --justcheckconnections Only check the current status of the lonc/d
783: connections, do not send emails do not
784: check if the daemons are running, do not
785: generate lon-status
786: --justcheckdaemons Only check that all of the Lon-CAPA daemons are
787: running, do not send emails do not
788: check the lonc/d connections, do not
789: generate lon-status
1.59 albertel 790: --justreload Only tell the daemons to reload the config files,
791: do not send emails do not
792: check if the daemons are running, do not
793: generate lon-status
1.49 albertel 794:
795: USAGE
796: }
797:
1.46 albertel 798: # ================================================================ Main Program
799: sub main () {
1.71 albertel 800: my ($help,$justcheckdaemons,$noemail,$justcheckconnections,
1.59 albertel 801: $justreload);
1.49 albertel 802: &GetOptions("help" => \$help,
803: "justcheckdaemons" => \$justcheckdaemons,
804: "noemail" => \$noemail,
1.59 albertel 805: "justcheckconnections" => \$justcheckconnections,
806: "justreload" => \$justreload
1.49 albertel 807: );
808: if ($help) { &usage(); return; }
1.46 albertel 809: # --------------------------------- Read loncapa_apache.conf and loncapa.conf
810: my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
811: %perlvar=%{$perlvarref};
812: undef $perlvarref;
813: delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
814: delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
1.75 albertel 815: chdir($perlvar{'lonDaemons'});
1.46 albertel 816: # --------------------------------------- Make sure that LON-CAPA is configured
817: # I only test for one thing here (lonHostID). This is just a safeguard.
818: if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) {
819: print("Unconfigured machine.\n");
820: my $emailto=$perlvar{'lonSysEMail'};
821: my $hostname=`/bin/hostname`;
822: chop $hostname;
823: $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
824: my $subj="LON: Unconfigured machine $hostname";
825: system("echo 'Unconfigured machine $hostname.' |\
826: mailto $emailto -s '$subj' > /dev/null");
827: exit 1;
828: }
829:
830: # ----------------------------- Make sure this process is running from user=www
831: my $wwwid=getpwnam('www');
832: if ($wwwid!=$<) {
833: print("User ID mismatch. This program must be run as user 'www'\n");
834: my $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
835: my $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
836: system("echo 'User ID mismatch. loncron must be run as user www.' |\
837: mailto $emailto -s '$subj' > /dev/null");
838: exit 1;
839: }
840:
1.72 albertel 841: # -------------------------------------------- Force reload of host information
842: &Apache::lonnet::load_hosts_tab(1);
843: &Apache::lonnet::load_domain_tab(1);
844: &Apache::lonnet::get_iphost(1);
1.46 albertel 845:
1.81 raeburn 846: # ----------------------------------------- Force firewall update for lond port
847:
848: if ((!$justcheckdaemons) && (!$justreload)) {
849: my $now = time;
850: my $tmpfile = $perlvar{'lonDaemons'}.'/tmp/lciptables_iphost_'.
851: $now.$$.int(rand(10000));
852: if (open(my $fh,">$tmpfile")) {
853: my %iphosts = &Apache::lonnet::get_iphost();
854: foreach my $key (keys(%iphosts)) {
855: print $fh "$key\n";
856: }
857: close($fh);
1.89 raeburn 858: if (&LONCAPA::try_to_lock('/tmp/lock_lciptables')) {
859: my $execpath = $perlvar{'lonDaemons'}.'/lciptables';
860: system("$execpath $tmpfile");
861: unlink('/tmp/lock_lciptables'); # Remove the lock file.
862: }
1.88 raeburn 863: unlink($tmpfile);
1.81 raeburn 864: }
865: }
866:
1.46 albertel 867: # ---------------------------------------------------------------- Start report
868:
869: $errors=0;
870: $warnings=0;
871: $notices=0;
872:
873:
1.49 albertel 874: my $fh;
1.59 albertel 875: if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
1.72 albertel 876: $fh=&start_logging();
1.49 albertel 877:
878: &log_machine_info($fh);
879: &clean_tmp($fh);
880: &clean_lonIDs($fh);
881: &check_httpd_logs($fh);
882: &rotate_lonnet_logs($fh);
1.73 albertel 883: &rotate_other_logs($fh);
1.49 albertel 884: }
1.59 albertel 885: if (!$justcheckconnections && !$justreload) {
1.76 albertel 886: &checkon_daemon($fh,'lonmemcached',40000);
1.49 albertel 887: &checkon_daemon($fh,'lonsql',200000);
1.63 albertel 888: if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') {
889: &checkon_daemon($fh,'lond',40000,'USR2');
890: }
1.71 albertel 891: &checkon_daemon($fh,'lonc',40000,'USR1');
1.70 raeburn 892: &checkon_daemon($fh,'lonmaxima',40000);
1.80 www 893: &checkon_daemon($fh,'lonr',40000);
1.49 albertel 894: }
1.59 albertel 895: if ($justreload) {
896: &checkon_daemon($fh,'lond',40000,'USR2');
1.71 albertel 897: &checkon_daemon($fh,'lonc',40000,'USR2');
1.59 albertel 898: }
1.63 albertel 899: if ($justcheckconnections) {
1.72 albertel 900: &test_connections($fh);
1.49 albertel 901: }
1.59 albertel 902: if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
1.72 albertel 903: &check_delayed_msg($fh);
1.49 albertel 904: &finish_logging($fh);
905: &log_simplestatus();
1.87 raeburn 906: &write_loncaparevs();
907: &write_serverhomeIDs();
1.49 albertel 908:
909: if ($totalcount>200 && !$noemail) { &send_mail(); }
910: }
1.46 albertel 911: }
912:
913: &main();
1.1 albertel 914: 1;
915:
916:
917:
918:
919:
920:
921:
922:
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