Annotation of loncom/loncron, revision 1.72
1.1 albertel 1: #!/usr/bin/perl
2:
1.47 albertel 3: # Housekeeping program, started by cron, loncontrol and loncron.pl
4: #
1.72 ! albertel 5: # $Id: loncron,v 1.71 2007/04/03 00:47:26 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;
1.72 ! albertel 35: use Apache::lonnet;
1.26 harris41 36:
1.1 albertel 37: use IO::File;
38: use IO::Socket;
1.48 albertel 39: use HTML::Entities;
1.49 albertel 40: use Getopt::Long;
1.46 albertel 41: #globals
42: use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount);
43:
44: my $statusdir="/home/httpd/html/lon-status";
45:
1.1 albertel 46:
47: # --------------------------------------------------------- Output error status
48:
1.46 albertel 49: sub log {
50: my $fh=shift;
51: if ($fh) { print $fh @_ }
52: }
53:
1.1 albertel 54: sub errout {
55: my $fh=shift;
1.46 albertel 56: &log($fh,(<<ENDERROUT));
1.48 albertel 57: <table border="2" bgcolor="#CCCCCC">
1.1 albertel 58: <tr><td>Notices</td><td>$notices</td></tr>
59: <tr><td>Warnings</td><td>$warnings</td></tr>
60: <tr><td>Errors</td><td>$errors</td></tr>
1.48 albertel 61: </table><p><a href="#top">Top</a></p>
1.1 albertel 62: ENDERROUT
63: }
64:
1.42 albertel 65: sub start_daemon {
1.50 albertel 66: my ($fh,$daemon,$pidfile,$args) = @_;
1.44 albertel 67: my $progname=$daemon;
1.71 albertel 68: if ($daemon eq 'lonc') {
1.44 albertel 69: $progname='loncnew';
70: }
1.51 albertel 71: my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors";
72: my $size=(stat($error_fname))[7];
73: if ($size>40000) {
74: &log($fh,"<p>Rotating error logs ...</p>");
75: rename("$error_fname.2","$error_fname.3");
76: rename("$error_fname.1","$error_fname.2");
77: rename("$error_fname","$error_fname.1");
78: }
79: system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors");
1.60 albertel 80: sleep 1;
1.42 albertel 81: if (-e $pidfile) {
1.48 albertel 82: &log($fh,"<p>Seems like it started ...</p>");
1.42 albertel 83: my $lfh=IO::File->new("$pidfile");
84: my $daemonpid=<$lfh>;
85: chomp($daemonpid);
1.62 albertel 86: if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
1.42 albertel 87: return 1;
88: } else {
89: return 0;
90: }
91: }
1.48 albertel 92: &log($fh,"<p>Seems like that did not work!</p>");
1.42 albertel 93: $errors++;
94: return 0;
95: }
96:
97: sub checkon_daemon {
1.59 albertel 98: my ($fh,$daemon,$maxsize,$send,$args)=@_;
1.42 albertel 99:
1.63 albertel 100: my $result;
1.48 albertel 101: &log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>');
1.57 albertel 102: printf("%-15s ",$daemon);
1.42 albertel 103: if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
104: open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
1.46 albertel 105: while (my $line=<DFH>) {
106: &log($fh,"$line");
1.42 albertel 107: if ($line=~/INFO/) { $notices++; }
108: if ($line=~/WARNING/) { $notices++; }
109: if ($line=~/CRITICAL/) { $warnings++; }
110: };
111: close (DFH);
112: }
1.48 albertel 113: &log($fh,"</tt></p>");
1.42 albertel 114:
115: my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid";
116:
117: my $restartflag=1;
1.46 albertel 118: my $daemonpid;
1.42 albertel 119: if (-e $pidfile) {
120: my $lfh=IO::File->new("$pidfile");
1.46 albertel 121: $daemonpid=<$lfh>;
1.42 albertel 122: chomp($daemonpid);
1.62 albertel 123: if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
1.46 albertel 124: &log($fh,"<h3>$daemon at pid $daemonpid responding");
1.59 albertel 125: if ($send) { &log($fh,", sending $send"); }
1.46 albertel 126: &log($fh,"</h3>");
1.59 albertel 127: if ($send eq 'USR1') { kill USR1 => $daemonpid; }
128: if ($send eq 'USR2') { kill USR2 => $daemonpid; }
1.42 albertel 129: $restartflag=0;
1.59 albertel 130: if ($send eq 'USR2') {
1.63 albertel 131: $result = 'reloaded';
1.59 albertel 132: print "reloaded\n";
133: } else {
1.63 albertel 134: $result = 'running';
1.59 albertel 135: print "running\n";
136: }
1.42 albertel 137: } else {
138: $errors++;
1.46 albertel 139: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.42 albertel 140: $restartflag=1;
1.46 albertel 141: &log($fh,"<h3>Decided to clean up stale .pid file and restart $daemon</h3>");
1.42 albertel 142: }
143: }
144: if ($restartflag==1) {
145: $simplestatus{$daemon}='off';
146: $errors++;
1.57 albertel 147: my $kadaemon=$daemon;
148: if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; }
1.46 albertel 149: &log($fh,'<br><font color="red">Killall '.$daemon.': '.
1.57 albertel 150: `killall $kadaemon 2>&1`.' - ');
1.60 albertel 151: sleep 1;
1.46 albertel 152: &log($fh,unlink($pidfile).' - '.
1.57 albertel 153: `killall -9 $kadaemon 2>&1`.
1.46 albertel 154: '</font><br>');
155: &log($fh,"<h3>$daemon not running, trying to start</h3>");
1.42 albertel 156:
1.50 albertel 157: if (&start_daemon($fh,$daemon,$pidfile,$args)) {
1.46 albertel 158: &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
1.42 albertel 159: $simplestatus{$daemon}='restarted';
1.63 albertel 160: $result = 'started';
1.42 albertel 161: print "started\n";
162: } else {
163: $errors++;
1.46 albertel 164: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.48 albertel 165: &log($fh,"<p>Give it one more try ...</p>");
1.42 albertel 166: print " ";
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 {
1.63 albertel 173: $result = 'failed';
1.42 albertel 174: print " failed\n";
175: $simplestatus{$daemon}='failed';
176: $errors++; $errors++;
1.46 albertel 177: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
1.48 albertel 178: &log($fh,"<p>Unable to start $daemon</p>");
1.42 albertel 179: }
180: }
181:
182: if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
1.46 albertel 183: &log($fh,"<p><pre>");
1.42 albertel 184: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|");
1.46 albertel 185: while (my $line=<DFH>) {
186: &log($fh,"$line");
1.42 albertel 187: if ($line=~/WARNING/) { $notices++; }
188: if ($line=~/CRITICAL/) { $notices++; }
189: };
190: close (DFH);
1.48 albertel 191: &log($fh,"</pre></p>");
1.42 albertel 192: }
193: }
194:
1.46 albertel 195: my $fname="$perlvar{'lonDaemons'}/logs/$daemon.log";
1.42 albertel 196:
197: my ($dev,$ino,$mode,$nlink,
198: $uid,$gid,$rdev,$size,
199: $atime,$mtime,$ctime,
200: $blksize,$blocks)=stat($fname);
201:
202: if ($size>$maxsize) {
1.48 albertel 203: &log($fh,"<p>Rotating logs ...</p>");
1.42 albertel 204: rename("$fname.2","$fname.3");
205: rename("$fname.1","$fname.2");
206: rename("$fname","$fname.1");
207: }
208:
209: &errout($fh);
1.63 albertel 210: return $result;
1.42 albertel 211: }
1.1 albertel 212:
1.46 albertel 213: # --------------------------------------------------------------------- Machine
214: sub log_machine_info {
215: my ($fh)=@_;
1.48 albertel 216: &log($fh,'<hr /><a name="machine" /><h2>Machine Information</h2>');
1.46 albertel 217: &log($fh,"<h3>loadavg</h3>");
218:
219: open (LOADAVGH,"/proc/loadavg");
220: my $loadavg=<LOADAVGH>;
221: close (LOADAVGH);
222:
223: &log($fh,"<tt>$loadavg</tt>");
224:
225: my @parts=split(/\s+/,$loadavg);
226: if ($parts[1]>4.0) {
227: $errors++;
228: } elsif ($parts[1]>2.0) {
229: $warnings++;
230: } elsif ($parts[1]>1.0) {
231: $notices++;
232: }
1.13 harris41 233:
1.46 albertel 234: &log($fh,"<h3>df</h3>");
235: &log($fh,"<pre>");
1.14 harris41 236:
1.46 albertel 237: open (DFH,"df|");
238: while (my $line=<DFH>) {
1.48 albertel 239: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 240: @parts=split(/\s+/,$line);
241: my $usage=$parts[4];
242: $usage=~s/\W//g;
243: if ($usage>90) {
244: $warnings++;
245: $notices++;
246: } elsif ($usage>80) {
247: $warnings++;
248: } elsif ($usage>60) {
249: $notices++;
1.31 albertel 250: }
1.46 albertel 251: if ($usage>95) { $warnings++; $warnings++; $simplestatus{'diskfull'}++; }
1.1 albertel 252: }
1.46 albertel 253: close (DFH);
254: &log($fh,"</pre>");
1.1 albertel 255:
256:
1.46 albertel 257: &log($fh,"<h3>ps</h3>");
258: &log($fh,"<pre>");
259: my $psproc=0;
1.1 albertel 260:
1.53 albertel 261: open (PSH,"ps aux --cols 140 |");
1.46 albertel 262: while (my $line=<PSH>) {
1.48 albertel 263: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 264: $psproc++;
265: }
266: close (PSH);
267: &log($fh,"</pre>");
1.1 albertel 268:
1.46 albertel 269: if ($psproc>200) { $notices++; }
270: if ($psproc>250) { $notices++; }
1.1 albertel 271:
1.61 albertel 272: &log($fh,"<h3>distprobe</h3>");
273: &log($fh,"<pre>");
274: open(DSH,"$perlvar{'lonDaemons'}/distprobe |");
275: while (my $line=<DSH>) {
276: &log($fh,&encode_entities($line,'<>&"'));
277: $psproc++;
278: }
279: close(DSH);
280: &log($fh,"</pre>");
281:
1.46 albertel 282: &errout($fh);
283: }
1.1 albertel 284:
1.46 albertel 285: sub start_logging {
1.43 albertel 286: my $fh=IO::File->new(">$statusdir/newstatus.html");
287: my %simplestatus=();
1.46 albertel 288: my $now=time;
289: my $date=localtime($now);
1.43 albertel 290:
1.46 albertel 291:
292: &log($fh,(<<ENDHEADERS));
1.1 albertel 293: <html>
294: <head>
295: <title>LON Status Report $perlvar{'lonHostID'}</title>
296: </head>
1.3 www 297: <body bgcolor="#AAAAAA">
1.48 albertel 298: <a name="top" />
1.1 albertel 299: <h1>LON Status Report $perlvar{'lonHostID'}</h1>
300: <h2>$date ($now)</h2>
301: <ol>
1.48 albertel 302: <li><a href="#configuration">Configuration</a></li>
303: <li><a href="#machine">Machine Information</a></li>
304: <li><a href="#tmp">Temporary Files</a></li>
305: <li><a href="#tokens">Session Tokens</a></li>
306: <li><a href="#httpd">httpd</a></li>
307: <li><a href="#lonsql">lonsql</a></li>
308: <li><a href="#lond">lond</a></li>
309: <li><a href="#lonc">lonc</a></li>
310: <li><a href="#lonhttpd">lonhttpd</a></li>
311: <li><a href="#lonnet">lonnet</a></li>
312: <li><a href="#connections">Connections</a></li>
313: <li><a href="#delayed">Delayed Messages</a></li>
314: <li><a href="#errcount">Error Count</a></li>
1.1 albertel 315: </ol>
1.48 albertel 316: <hr />
317: <a name="configuration" />
1.1 albertel 318: <h2>Configuration</h2>
319: <h3>PerlVars</h3>
1.48 albertel 320: <table border="2">
1.1 albertel 321: ENDHEADERS
322:
1.46 albertel 323: foreach my $varname (sort(keys(%perlvar))) {
1.48 albertel 324: &log($fh,"<tr><td>$varname</td><td>".
325: &encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n");
1.43 albertel 326: }
1.48 albertel 327: &log($fh,"</table><h3>Hosts</h3><table border='2'>");
1.72 ! albertel 328: my %hostname = &Apache::lonnet::all_hostnames();
! 329: foreach my $id (sort(keys(%hostname))) {
! 330: my $role = (&Apache::lonnet::is_library($id) ? 'library'
! 331: : 'access');
1.46 albertel 332: &log($fh,
1.72 ! albertel 333: "<tr><td>$id</td><td>".&Apache::lonnet::host_domain($id).
! 334: "</td><td>".$role.
! 335: "</td><td>".&Apache::lonnet::hostname($id)."</td></tr>\n");
! 336: }
! 337: &log($fh,"</table><h3>Spare Hosts</h3><ul>");
! 338: foreach my $type (sort(keys(%Apache::lonnet::spareid))) {
! 339: &log($fh,"<li>$type\n<ol>");
! 340: foreach my $id (@{ $Apache::lonnet::spareid{$type} }) {
! 341: &log($fh,"<li>$id</li>\n");
! 342: }
! 343: &log($fh,"</ol>\n</li>\n");
1.43 albertel 344: }
1.72 ! albertel 345: &log($fh,"</ul>\n");
1.46 albertel 346: return $fh;
347: }
1.11 www 348:
349: # --------------------------------------------------------------- clean out tmp
1.46 albertel 350: sub clean_tmp {
351: my ($fh)=@_;
1.48 albertel 352: &log($fh,'<hr /><a name="tmp" /><h2>Temporary Files</h2>');
1.46 albertel 353: my $cleaned=0;
354: my $old=0;
355: while (my $fname=<$perlvar{'lonDaemons'}/tmp/*>) {
1.43 albertel 356: my ($dev,$ino,$mode,$nlink,
357: $uid,$gid,$rdev,$size,
358: $atime,$mtime,$ctime,
359: $blksize,$blocks)=stat($fname);
1.46 albertel 360: my $now=time;
361: my $since=$now-$mtime;
1.43 albertel 362: if ($since>$perlvar{'lonExpire'}) {
1.46 albertel 363: my $line='';
1.43 albertel 364: if (open(PROBE,$fname)) {
365: $line=<PROBE>;
366: close(PROBE);
367: }
368: unless ($line=~/^CHECKOUTTOKEN\&/) {
369: $cleaned++;
370: unlink("$fname");
371: } else {
372: if ($since>365*$perlvar{'lonExpire'}) {
373: $cleaned++;
374: unlink("$fname");
375: } else { $old++; }
376: }
377: }
378: }
1.46 albertel 379: &log($fh,"Cleaned up ".$cleaned." files (".$old." old checkout tokens).");
380: }
1.11 www 381:
382: # ------------------------------------------------------------ clean out lonIDs
1.46 albertel 383: sub clean_lonIDs {
384: my ($fh)=@_;
1.48 albertel 385: &log($fh,'<hr /><a name="tokens" /><h2>Session Tokens</h2>');
1.46 albertel 386: my $cleaned=0;
387: my $active=0;
388: while (my $fname=<$perlvar{'lonIDsDir'}/*>) {
1.43 albertel 389: my ($dev,$ino,$mode,$nlink,
390: $uid,$gid,$rdev,$size,
391: $atime,$mtime,$ctime,
392: $blksize,$blocks)=stat($fname);
1.46 albertel 393: my $now=time;
394: my $since=$now-$mtime;
1.43 albertel 395: if ($since>$perlvar{'lonExpire'}) {
396: $cleaned++;
1.46 albertel 397: &log($fh,"Unlinking $fname<br>");
1.43 albertel 398: unlink("$fname");
399: } else {
400: $active++;
401: }
1.46 albertel 402: }
1.48 albertel 403: &log($fh,"<p>Cleaned up ".$cleaned." stale session token(s).</p>");
1.46 albertel 404: &log($fh,"<h3>$active open session(s)</h3>");
405: }
1.43 albertel 406:
1.11 www 407:
1.1 albertel 408: # ----------------------------------------------------------------------- httpd
1.46 albertel 409: sub check_httpd_logs {
410: my ($fh)=@_;
1.48 albertel 411: &log($fh,'<hr /><a name="httpd" /><h2>httpd</h2><h3>Access Log</h3><pre>');
1.43 albertel 412:
413: open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
1.48 albertel 414: while (my $line=<DFH>) { &log($fh,&encode_entities($line,'<>&"')) };
1.43 albertel 415: close (DFH);
1.46 albertel 416:
417: &log($fh,"</pre><h3>Error Log</h3><pre>");
418:
1.43 albertel 419: open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
1.46 albertel 420: while (my $line=<DFH>) {
421: &log($fh,"$line");
1.43 albertel 422: if ($line=~/\[error\]/) { $notices++; }
1.46 albertel 423: }
1.43 albertel 424: close (DFH);
1.46 albertel 425: &log($fh,"</pre>");
1.43 albertel 426: &errout($fh);
1.46 albertel 427: }
1.1 albertel 428:
429: # ---------------------------------------------------------------------- lonnet
430:
1.48 albertel 431: sub rotate_lonnet_logs {
1.46 albertel 432: my ($fh)=@_;
1.48 albertel 433: &log($fh,'<hr /><a name="lonnet" /><h2>lonnet</h2><h3>Temp Log</h3><pre>');
1.43 albertel 434: print "checking logs\n";
435: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
436: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
1.46 albertel 437: while (my $line=<DFH>) {
1.48 albertel 438: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 439: }
1.43 albertel 440: close (DFH);
441: }
1.46 albertel 442: &log($fh,"</pre><h3>Perm Log</h3><pre>");
1.43 albertel 443:
444: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
445: open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
1.46 albertel 446: while (my $line=<DFH>) {
1.48 albertel 447: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 448: }
1.43 albertel 449: close (DFH);
1.46 albertel 450: } else { &log($fh,"No perm log\n") }
1.43 albertel 451:
1.46 albertel 452: my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
1.43 albertel 453:
454: my ($dev,$ino,$mode,$nlink,
455: $uid,$gid,$rdev,$size,
456: $atime,$mtime,$ctime,
457: $blksize,$blocks)=stat($fname);
458:
459: if ($size>40000) {
1.48 albertel 460: &log($fh,"<p>Rotating logs ...</p>");
1.43 albertel 461: rename("$fname.2","$fname.3");
462: rename("$fname.1","$fname.2");
463: rename("$fname","$fname.1");
464: }
1.1 albertel 465:
1.46 albertel 466: &log($fh,"</pre>");
1.43 albertel 467: &errout($fh);
1.46 albertel 468: }
469:
1.43 albertel 470: # ----------------------------------------------------------------- Connections
1.46 albertel 471: sub test_connections {
1.72 ! albertel 472: my ($fh)=@_;
1.48 albertel 473: &log($fh,'<hr /><a name="connections" /><h2>Connections</h2>');
1.43 albertel 474: print "testing connections\n";
1.48 albertel 475: &log($fh,"<table border='2'>");
1.49 albertel 476: my ($good,$bad)=(0,0);
1.72 ! albertel 477: my %hostname = &Apache::lonnet::all_hostnames();
! 478: foreach my $tryserver (sort(keys(%hostname))) {
1.43 albertel 479: print(".");
1.46 albertel 480: my $result;
1.72 ! albertel 481: my $answer=&Apache::lonnet::reply("ping",$tryserver);
1.43 albertel 482: if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
483: $result="<b>ok</b>";
1.49 albertel 484: $good++;
1.43 albertel 485: } else {
486: $result=$answer;
487: $warnings++;
1.49 albertel 488: if ($answer eq 'con_lost') {
489: $bad++;
490: $warnings++;
1.50 albertel 491: } else {
492: $good++; #self connection
1.49 albertel 493: }
1.43 albertel 494: }
495: if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }
1.46 albertel 496: &log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n");
1.1 albertel 497: }
1.46 albertel 498: &log($fh,"</table>");
1.49 albertel 499: print "\n$good good, $bad bad connections\n";
1.43 albertel 500: &errout($fh);
1.46 albertel 501: }
502:
503:
1.1 albertel 504: # ------------------------------------------------------------ Delayed messages
1.46 albertel 505: sub check_delayed_msg {
1.72 ! albertel 506: my ($fh)=@_;
1.48 albertel 507: &log($fh,'<hr /><a name="delayed" /><h2>Delayed Messages</h2>');
1.43 albertel 508: print "checking buffers\n";
1.46 albertel 509:
510: &log($fh,'<h3>Scanning Permanent Log</h3>');
1.1 albertel 511:
1.46 albertel 512: my $unsend=0;
1.1 albertel 513:
1.46 albertel 514: my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
515: while (my $line=<$dfh>) {
516: my ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
517: if ($sdf eq 'F') {
518: my $local=localtime($time);
519: &log($fh,"<b>Failed: $time, $dserv, $dcmd</b><br>");
520: $warnings++;
1.43 albertel 521: }
1.46 albertel 522: if ($sdf eq 'S') { $unsend--; }
523: if ($sdf eq 'D') { $unsend++; }
1.1 albertel 524: }
1.46 albertel 525:
1.48 albertel 526: &log($fh,"<p>Total unsend messages: <b>$unsend</b></p>\n");
1.43 albertel 527: $warnings=$warnings+5*$unsend;
1.1 albertel 528:
1.43 albertel 529: if ($unsend) { $simplestatus{'unsend'}=$unsend; }
1.48 albertel 530: &log($fh,"<h3>Outgoing Buffer</h3>\n<pre>");
1.68 www 531: # list directory with delayed messages and remember offline servers
532: my %servers=();
1.43 albertel 533: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
1.68 www 534: while (my $line=<DFH>) {
535: my ($server)=($line=~/\.(\w+)$/);
536: if ($server) { $servers{$server}=1; }
1.48 albertel 537: &log($fh,&encode_entities($line,'<>&"'));
1.46 albertel 538: }
1.48 albertel 539: &log($fh,"</pre>\n");
1.43 albertel 540: close (DFH);
1.68 www 541: # pong to all servers that have delayed messages
542: # this will trigger a reverse connection, which should flush the buffers
543: foreach my $tryserver (keys %servers) {
1.72 ! albertel 544: my $answer=&Apache::lonnet::reply("pong",$tryserver);
1.69 www 545: &log($fh,"Pong to $tryserver: $answer<br />");
1.68 www 546: }
1.46 albertel 547: }
1.1 albertel 548:
1.46 albertel 549: sub finish_logging {
550: my ($fh)=@_;
1.48 albertel 551: &log($fh,"<a name='errcount' />\n");
1.43 albertel 552: $totalcount=$notices+4*$warnings+100*$errors;
553: &errout($fh);
1.46 albertel 554: &log($fh,"<h1>Total Error Count: $totalcount</h1>");
555: my $now=time;
556: my $date=localtime($now);
1.48 albertel 557: &log($fh,"<hr />$date ($now)</body></html>\n");
1.43 albertel 558: print "lon-status webpage updated\n";
559: $fh->close();
1.46 albertel 560:
561: if ($errors) { $simplestatus{'errors'}=$errors; }
562: if ($warnings) { $simplestatus{'warnings'}=$warnings; }
563: if ($notices) { $simplestatus{'notices'}=$notices; }
564: $simplestatus{'time'}=time;
1.1 albertel 565: }
566:
1.46 albertel 567: sub log_simplestatus {
568: rename ("$statusdir/newstatus.html","$statusdir/index.html");
569:
1.43 albertel 570: my $sfh=IO::File->new(">$statusdir/loncron_simple.txt");
571: foreach (keys %simplestatus) {
572: print $sfh $_.'='.$simplestatus{$_}.'&';
573: }
574: print $sfh "\n";
575: $sfh->close();
1.41 www 576: }
1.46 albertel 577:
578: sub send_mail {
1.43 albertel 579: print "sending mail\n";
1.46 albertel 580: my $emailto="$perlvar{'lonAdmEMail'}";
1.54 www 581: if ($totalcount>2500) {
1.43 albertel 582: $emailto.=",$perlvar{'lonSysEMail'}";
583: }
1.46 albertel 584: my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";
1.52 albertel 585:
1.58 albertel 586: my $result=system("metasend -b -S 4000000 -t $emailto -s '$subj' -f $statusdir/index.html -m text/html >& /dev/null");
1.52 albertel 587: if ($result != 0) {
588: $result=system("mail -s '$subj' $emailto < $statusdir/index.html");
589: }
1.1 albertel 590: }
1.46 albertel 591:
1.49 albertel 592: sub usage {
593: print(<<USAGE);
594: loncron - housekeeping program that checks up on various parts of Lon-CAPA
595:
596: Options:
1.71 albertel 597: --help Display
1.49 albertel 598: --noemail Do not send the status email
599: --justcheckconnections Only check the current status of the lonc/d
600: connections, do not send emails do not
601: check if the daemons are running, do not
602: generate lon-status
603: --justcheckdaemons Only check that all of the Lon-CAPA daemons are
604: running, do not send emails do not
605: check the lonc/d connections, do not
606: generate lon-status
1.59 albertel 607: --justreload Only tell the daemons to reload the config files,
608: do not send emails do not
609: check if the daemons are running, do not
610: generate lon-status
1.49 albertel 611:
612: USAGE
613: }
614:
1.46 albertel 615: # ================================================================ Main Program
616: sub main () {
1.71 albertel 617: my ($help,$justcheckdaemons,$noemail,$justcheckconnections,
1.59 albertel 618: $justreload);
1.49 albertel 619: &GetOptions("help" => \$help,
620: "justcheckdaemons" => \$justcheckdaemons,
621: "noemail" => \$noemail,
1.59 albertel 622: "justcheckconnections" => \$justcheckconnections,
623: "justreload" => \$justreload
1.49 albertel 624: );
625: if ($help) { &usage(); return; }
1.46 albertel 626: # --------------------------------- Read loncapa_apache.conf and loncapa.conf
627: my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
628: %perlvar=%{$perlvarref};
629: undef $perlvarref;
630: delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
631: delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
632:
633: # --------------------------------------- Make sure that LON-CAPA is configured
634: # I only test for one thing here (lonHostID). This is just a safeguard.
635: if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) {
636: print("Unconfigured machine.\n");
637: my $emailto=$perlvar{'lonSysEMail'};
638: my $hostname=`/bin/hostname`;
639: chop $hostname;
640: $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
641: my $subj="LON: Unconfigured machine $hostname";
642: system("echo 'Unconfigured machine $hostname.' |\
643: mailto $emailto -s '$subj' > /dev/null");
644: exit 1;
645: }
646:
647: # ----------------------------- Make sure this process is running from user=www
648: my $wwwid=getpwnam('www');
649: if ($wwwid!=$<) {
650: print("User ID mismatch. This program must be run as user 'www'\n");
651: my $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
652: my $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
653: system("echo 'User ID mismatch. loncron must be run as user www.' |\
654: mailto $emailto -s '$subj' > /dev/null");
655: exit 1;
656: }
657:
1.72 ! albertel 658: # -------------------------------------------- Force reload of host information
! 659: &Apache::lonnet::load_hosts_tab(1);
! 660: &Apache::lonnet::load_domain_tab(1);
! 661: &Apache::lonnet::get_iphost(1);
1.46 albertel 662:
663: # ---------------------------------------------------------------- Start report
664:
665: $errors=0;
666: $warnings=0;
667: $notices=0;
668:
669:
1.49 albertel 670: my $fh;
1.59 albertel 671: if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
1.72 ! albertel 672: $fh=&start_logging();
1.49 albertel 673:
674: &log_machine_info($fh);
675: &clean_tmp($fh);
676: &clean_lonIDs($fh);
677: &check_httpd_logs($fh);
678: &rotate_lonnet_logs($fh);
679: }
1.59 albertel 680: if (!$justcheckconnections && !$justreload) {
1.49 albertel 681: &checkon_daemon($fh,'lonsql',200000);
1.63 albertel 682: if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') {
683: &checkon_daemon($fh,'lond',40000,'USR2');
684: }
1.71 albertel 685: &checkon_daemon($fh,'lonc',40000,'USR1');
1.49 albertel 686: &checkon_daemon($fh,'lonhttpd',40000);
1.57 albertel 687: &checkon_daemon($fh,'lonmemcached',40000);
1.70 raeburn 688: &checkon_daemon($fh,'lonmaxima',40000);
1.49 albertel 689: }
1.59 albertel 690: if ($justreload) {
691: &checkon_daemon($fh,'lond',40000,'USR2');
1.71 albertel 692: &checkon_daemon($fh,'lonc',40000,'USR2');
1.59 albertel 693: }
1.63 albertel 694: if ($justcheckconnections) {
1.72 ! albertel 695: &test_connections($fh);
1.49 albertel 696: }
1.59 albertel 697: if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
1.72 ! albertel 698: &check_delayed_msg($fh);
1.49 albertel 699: &finish_logging($fh);
700: &log_simplestatus();
701:
702: if ($totalcount>200 && !$noemail) { &send_mail(); }
703: }
1.46 albertel 704: }
705:
706: &main();
1.1 albertel 707: 1;
708:
709:
710:
711:
712:
713:
714:
715:
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