loncom/loncron - view

File: [LON-CAPA] / loncom / loncron
Revision 1.85: download - view: text, annotated - select for diffs
Tue Jul 20 02:42:27 2010 UTC (13 years, 9 months ago) by raeburn
Branches: MAIN
CVS tags: HEAD

- Control hosting of user sessions.
  - To accommodate multi-domain library servers, use internet domain names (e.g., msu.edu)
    instead of LON-CAPA domains in interface DC uses to set options for session hosting.

  - Provide mechanism to retrieve lonHostID for a hostname, so we can tell which
    domain determines session hosting config for external users on a multi-domain
    machine.

  - New routines in lonnet.pm:
     &get_server_homeID()  -- gets lonHostID for a hostname
     &internet_dom() -- gets internet domain name for a LON-CAPA Host ID
     &get_internet_names() -- gets internet domain names for all domains
                              for a server given one of the LC host IDs for the server.

   - New file in /home/httpd/lonTabs:
     - serverhomeIDs.tab (populated by loncron). : paired hostname:lonHostID

   - New subroutine in loncron: &write_serverhomeIDs() -- populates serverhomeIDs.tab
   - New subroutine in lond: &server_homeID_handler() -- returns LON-CAPA Host ID of server
   - New scalar in lond: $clienthomedom -- set to domain of LON-CAPA Host ID of client
   - New subroutine in domainprefs.pm: &build_location_hashes()
       -- used to set up options for session hosting as internet domain names
           - where two names (e.g., msu.edu, loncapa.org) are controlled by the same
             institution these will receive a single checkbox.

1: #!/usr/bin/perl 2: 3: # Housekeeping program, started by cron, loncontrol and loncron.pl 4: # 5: # $Id: loncron,v 1.85 2010/07/20 02:42:27 raeburn Exp $ 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/ 28: # 29: 30: $|=1; 31: use strict; 32: 33: use lib '/home/httpd/lib/perl/'; 34: use LONCAPA::Configuration; 35: use Apache::lonnet; 36: use Apache::loncommon; 37: 38: use IO::File; 39: use IO::Socket; 40: use HTML::Entities; 41: use Getopt::Long; 42: #globals 43: use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount); 44: 45: my $statusdir="/home/httpd/html/lon-status"; 46: 47: 48: # --------------------------------------------------------- Output error status 49: 50: sub log { 51: my $fh=shift; 52: if ($fh) { print $fh @_ } 53: } 54: 55: sub errout { 56: my $fh=shift; 57: &log($fh,(<<ENDERROUT)); 58: <table border="2" bgcolor="#CCCCCC"> 59: <tr><td>Notices</td><td>$notices</td></tr> 60: <tr><td>Warnings</td><td>$warnings</td></tr> 61: <tr><td>Errors</td><td>$errors</td></tr> 62: </table><p><a href="#top">Top</a></p> 63: ENDERROUT 64: } 65: 66: sub rotate_logfile { 67: my ($file,$fh,$description) = @_; 68: my $size=(stat($file))[7]; 69: if ($size>40000) { 70: &log($fh,"<p>Rotating $description ...</p>"); 71: rename("$file.2","$file.3"); 72: rename("$file.1","$file.2"); 73: rename("$file","$file.1"); 74: } 75: } 76: 77: sub start_daemon { 78: my ($fh,$daemon,$pidfile,$args) = @_; 79: my $progname=$daemon; 80: if ($daemon eq 'lonc') { 81: $progname='loncnew'; 82: } 83: my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors"; 84: &rotate_logfile($error_fname,$fh,'error logs'); 85: if ($daemon eq 'lonc') { 86: &clean_sockets($fh); 87: } 88: system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors"); 89: sleep 1; 90: if (-e $pidfile) { 91: &log($fh,"<p>Seems like it started ...</p>"); 92: my $lfh=IO::File->new("$pidfile"); 93: my $daemonpid=<$lfh>; 94: chomp($daemonpid); 95: if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) { 96: return 1; 97: } else { 98: return 0; 99: } 100: } 101: &log($fh,"<p>Seems like that did not work!</p>"); 102: $errors++; 103: return 0; 104: } 105: 106: sub checkon_daemon { 107: my ($fh,$daemon,$maxsize,$send,$args)=@_; 108: 109: my $result; 110: &log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>'); 111: printf("%-15s ",$daemon); 112: if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){ 113: open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|"); 114: while (my $line=<DFH>) { 115: &log($fh,"$line"); 116: if ($line=~/INFO/) { $notices++; } 117: if ($line=~/WARNING/) { $notices++; } 118: if ($line=~/CRITICAL/) { $warnings++; } 119: }; 120: close (DFH); 121: } 122: &log($fh,"</tt></p>"); 123: 124: my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid"; 125: 126: my $restartflag=1; 127: my $daemonpid; 128: if (-e $pidfile) { 129: my $lfh=IO::File->new("$pidfile"); 130: $daemonpid=<$lfh>; 131: chomp($daemonpid); 132: if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) { 133: &log($fh,"<h3>$daemon at pid $daemonpid responding"); 134: if ($send) { &log($fh,", sending $send"); } 135: &log($fh,"</h3>"); 136: if ($send eq 'USR1') { kill USR1 => $daemonpid; } 137: if ($send eq 'USR2') { kill USR2 => $daemonpid; } 138: $restartflag=0; 139: if ($send eq 'USR2') { 140: $result = 'reloaded'; 141: print "reloaded\n"; 142: } else { 143: $result = 'running'; 144: print "running\n"; 145: } 146: } else { 147: $errors++; 148: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>"); 149: $restartflag=1; 150: &log($fh,"<h3>Decided to clean up stale .pid file and restart $daemon</h3>"); 151: } 152: } 153: if ($restartflag==1) { 154: $simplestatus{$daemon}='off'; 155: $errors++; 156: my $kadaemon=$daemon; 157: if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; } 158: &log($fh,'<br><font color="red">Killall '.$daemon.': '. 159: `killall $kadaemon 2>&1`.' - '); 160: sleep 1; 161: &log($fh,unlink($pidfile).' - '. 162: `killall -9 $kadaemon 2>&1`. 163: '</font><br>'); 164: &log($fh,"<h3>$daemon not running, trying to start</h3>"); 165: 166: if (&start_daemon($fh,$daemon,$pidfile,$args)) { 167: &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>"); 168: $simplestatus{$daemon}='restarted'; 169: $result = 'started'; 170: print "started\n"; 171: } else { 172: $errors++; 173: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>"); 174: &log($fh,"<p>Give it one more try ...</p>"); 175: print " "; 176: if (&start_daemon($fh,$daemon,$pidfile,$args)) { 177: &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>"); 178: $simplestatus{$daemon}='restarted'; 179: $result = 'started'; 180: print "started\n"; 181: } else { 182: $result = 'failed'; 183: print " failed\n"; 184: $simplestatus{$daemon}='failed'; 185: $errors++; $errors++; 186: &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>"); 187: &log($fh,"<p>Unable to start $daemon</p>"); 188: } 189: } 190: 191: if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){ 192: &log($fh,"<p><pre>"); 193: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|"); 194: while (my $line=<DFH>) { 195: &log($fh,"$line"); 196: if ($line=~/WARNING/) { $notices++; } 197: if ($line=~/CRITICAL/) { $notices++; } 198: }; 199: close (DFH); 200: &log($fh,"</pre></p>"); 201: } 202: } 203: 204: my $fname="$perlvar{'lonDaemons'}/logs/$daemon.log"; 205: &rotate_logfile($fname,$fh,'logs'); 206: 207: &errout($fh); 208: return $result; 209: } 210: 211: # --------------------------------------------------------------------- Machine 212: sub log_machine_info { 213: my ($fh)=@_; 214: &log($fh,'<hr /><a name="machine" /><h2>Machine Information</h2>'); 215: &log($fh,"<h3>loadavg</h3>"); 216: 217: open (LOADAVGH,"/proc/loadavg"); 218: my $loadavg=<LOADAVGH>; 219: close (LOADAVGH); 220: 221: &log($fh,"<tt>$loadavg</tt>"); 222: 223: my @parts=split(/\s+/,$loadavg); 224: if ($parts[1]>4.0) { 225: $errors++; 226: } elsif ($parts[1]>2.0) { 227: $warnings++; 228: } elsif ($parts[1]>1.0) { 229: $notices++; 230: } 231: 232: &log($fh,"<h3>df</h3>"); 233: &log($fh,"<pre>"); 234: 235: open (DFH,"df|"); 236: while (my $line=<DFH>) { 237: &log($fh,&encode_entities($line,'<>&"')); 238: @parts=split(/\s+/,$line); 239: my $usage=$parts[4]; 240: $usage=~s/\W//g; 241: if ($usage>90) { 242: $warnings++; 243: $notices++; 244: } elsif ($usage>80) { 245: $warnings++; 246: } elsif ($usage>60) { 247: $notices++; 248: } 249: if ($usage>95) { $warnings++; $warnings++; $simplestatus{'diskfull'}++; } 250: } 251: close (DFH); 252: &log($fh,"</pre>"); 253: 254: 255: &log($fh,"<h3>ps</h3>"); 256: &log($fh,"<pre>"); 257: my $psproc=0; 258: 259: open (PSH,"ps aux --cols 140 |"); 260: while (my $line=<PSH>) { 261: &log($fh,&encode_entities($line,'<>&"')); 262: $psproc++; 263: } 264: close (PSH); 265: &log($fh,"</pre>"); 266: 267: if ($psproc>200) { $notices++; } 268: if ($psproc>250) { $notices++; } 269: 270: &log($fh,"<h3>distprobe</h3>"); 271: &log($fh,"<pre>"); 272: open(DSH,"$perlvar{'lonDaemons'}/distprobe |"); 273: while (my $line=<DSH>) { 274: &log($fh,&encode_entities($line,'<>&"')); 275: $psproc++; 276: } 277: close(DSH); 278: &log($fh,"</pre>"); 279: 280: &errout($fh); 281: } 282: 283: sub start_logging { 284: my $fh=IO::File->new(">$statusdir/newstatus.html"); 285: my %simplestatus=(); 286: my $now=time; 287: my $date=localtime($now); 288: 289: 290: &log($fh,(<<ENDHEADERS)); 291: <html> 292: <head> 293: <title>LON Status Report $perlvar{'lonHostID'}</title> 294: </head> 295: <body bgcolor="#AAAAAA"> 296: <a name="top" /> 297: <h1>LON Status Report $perlvar{'lonHostID'}</h1> 298: <h2>$date ($now)</h2> 299: <ol> 300: <li><a href="#configuration">Configuration</a></li> 301: <li><a href="#machine">Machine Information</a></li> 302: <li><a href="#tmp">Temporary Files</a></li> 303: <li><a href="#tokens">Session Tokens</a></li> 304: <li><a href="#httpd">httpd</a></li> 305: <li><a href="#lonsql">lonsql</a></li> 306: <li><a href="#lond">lond</a></li> 307: <li><a href="#lonc">lonc</a></li> 308: <li><a href="#lonnet">lonnet</a></li> 309: <li><a href="#connections">Connections</a></li> 310: <li><a href="#delayed">Delayed Messages</a></li> 311: <li><a href="#errcount">Error Count</a></li> 312: </ol> 313: <hr /> 314: <a name="configuration" /> 315: <h2>Configuration</h2> 316: <h3>PerlVars</h3> 317: <table border="2"> 318: ENDHEADERS 319: 320: foreach my $varname (sort(keys(%perlvar))) { 321: &log($fh,"<tr><td>$varname</td><td>". 322: &encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n"); 323: } 324: &log($fh,"</table><h3>Hosts</h3><table border='2'>"); 325: my %hostname = &Apache::lonnet::all_hostnames(); 326: foreach my $id (sort(keys(%hostname))) { 327: my $role = (&Apache::lonnet::is_library($id) ? 'library' 328: : 'access'); 329: &log($fh, 330: "<tr><td>$id</td><td>".&Apache::lonnet::host_domain($id). 331: "</td><td>".$role. 332: "</td><td>".&Apache::lonnet::hostname($id)."</td></tr>\n"); 333: } 334: &log($fh,"</table><h3>Spare Hosts</h3><ul>"); 335: foreach my $type (sort(keys(%Apache::lonnet::spareid))) { 336: &log($fh,"<li>$type\n<ol>"); 337: foreach my $id (@{ $Apache::lonnet::spareid{$type} }) { 338: &log($fh,"<li>$id</li>\n"); 339: } 340: &log($fh,"</ol>\n</li>\n"); 341: } 342: &log($fh,"</ul>\n"); 343: return $fh; 344: } 345: 346: # --------------------------------------------------------------- clean out tmp 347: sub clean_tmp { 348: my ($fh)=@_; 349: &log($fh,'<hr /><a name="tmp" /><h2>Temporary Files</h2>'); 350: my ($cleaned,$old,$removed) = (0,0,0); 351: my %errors = ( 352: dir => [], 353: file => [], 354: failopen => [], 355: ); 356: my %error_titles = ( 357: dir => 'failed to remove empty directory:', 358: file => 'failed to unlike stale file', 359: failopen => 'failed to open file or directory' 360: ); 361: ($cleaned,$old,$removed) = &recursive_clean_tmp('',$cleaned,$old,$removed,\%errors); 362: &log($fh,"Cleaned up: ".$cleaned." files; removed: $removed empty directories; (found: $old old checkout tokens)"); 363: foreach my $key (sort(keys(%errors))) { 364: if (ref($errors{$key}) eq 'ARRAY') { 365: if (@{$errors{$key}} > 0) { 366: &log($fh,"Error during cleanup ($error_titles{$key}):<ul><li>". 367: join('</li><li><tt>',@{$errors{$key}}).'</tt></li></ul><br />'); 368: } 369: } 370: } 371: } 372: 373: sub recursive_clean_tmp { 374: my ($subdir,$cleaned,$old,$removed,$errors) = @_; 375: my $base = "$perlvar{'lonDaemons'}/tmp"; 376: my $path = $base; 377: next if ($subdir =~ m{\.\./}); 378: next unless (ref($errors) eq 'HASH'); 379: unless ($subdir eq '') { 380: $path .= '/'.$subdir; 381: } 382: if (opendir(my $dh,"$path")) { 383: while (my $file = readdir($dh)) { 384: next if ($file =~ /^\.\.?$/); 385: my $fname = "$path/$file"; 386: if (-d $fname) { 387: my $innerdir; 388: if ($subdir eq '') { 389: $innerdir = $file; 390: } else { 391: $innerdir = $subdir.'/'.$file; 392: } 393: ($cleaned,$old,$removed) = 394: &recursive_clean_tmp($innerdir,$cleaned,$old,$removed,$errors); 395: my @doms = &Apache::lonnet::current_machine_domains(); 396: 397: if (open(my $dirhandle,$fname)) { 398: unless (($innerdir eq 'helprequests') || 399: (($innerdir =~ /^addcourse/) && ($innerdir !~ m{/\d+$}))) { 400: my @contents = grep {!/^\.\.?$/} readdir($dirhandle); 401: join('&&',@contents)."\n"; 402: if (scalar(grep {!/^\.\.?$/} readdir($dirhandle)) == 0) { 403: closedir($dirhandle); 404: if ($fname =~ m{^\Q$perlvar{'lonDaemons'}\E/tmp/}) { 405: if (rmdir($fname)) { 406: $removed ++; 407: } elsif (ref($errors->{dir}) eq 'ARRAY') { 408: push(@{$errors->{dir}},$fname); 409: } 410: } 411: } 412: } else { 413: closedir($dirhandle); 414: } 415: } 416: } else { 417: my ($dev,$ino,$mode,$nlink, 418: $uid,$gid,$rdev,$size, 419: $atime,$mtime,$ctime, 420: $blksize,$blocks)=stat($fname); 421: my $now=time; 422: my $since=$now-$mtime; 423: if ($since>$perlvar{'lonExpire'}) { 424: if ($subdir eq '') { 425: my $line=''; 426: if ($fname =~ /\.db$/) { 427: if (unlink($fname)) { 428: $cleaned++; 429: } elsif (ref($errors->{file}) eq 'ARRAY') { 430: push(@{$errors->{file}},$fname); 431: } 432: } elsif (open(PROBE,$fname)) { 433: my $line=''; 434: $line=<PROBE>; 435: close(PROBE); 436: if ($line=~/^CHECKOUTTOKEN\&/) { 437: if ($since>365*$perlvar{'lonExpire'}) { 438: if (unlink($fname)) { 439: $cleaned++; 440: } elsif (ref($errors->{file}) eq 'ARRAY') { 441: push(@{$errors->{file}},$fname); 442: } 443: } else { 444: $old++; 445: } 446: } else { 447: if (unlink($fname)) { 448: $cleaned++; 449: } elsif (ref($errors->{file}) eq 'ARRAY') { 450: push(@{$errors->{file}},$fname); 451: } 452: } 453: } elsif (ref($errors->{failopen}) eq 'ARRAY') { 454: push(@{$errors->{failopen}},$fname); 455: } 456: } else { 457: if (unlink($fname)) { 458: $cleaned++; 459: } elsif (ref($errors->{file}) eq 'ARRAY') { 460: push(@{$errors->{file}},$fname); 461: } 462: } 463: } 464: } 465: } 466: closedir($dh); 467: } elsif (ref($errors->{failopen}) eq 'ARRAY') { 468: push(@{$errors->{failopen}},$path); 469: } 470: return ($cleaned,$old,$removed); 471: } 472: 473: # ------------------------------------------------------------ clean out lonIDs 474: sub clean_lonIDs { 475: my ($fh)=@_; 476: &log($fh,'<hr /><a name="tokens" /><h2>Session Tokens</h2>'); 477: my $cleaned=0; 478: my $active=0; 479: while (my $fname=<$perlvar{'lonIDsDir'}/*>) { 480: my ($dev,$ino,$mode,$nlink, 481: $uid,$gid,$rdev,$size, 482: $atime,$mtime,$ctime, 483: $blksize,$blocks)=stat($fname); 484: my $now=time; 485: my $since=$now-$mtime; 486: if ($since>$perlvar{'lonExpire'}) { 487: $cleaned++; 488: &log($fh,"Unlinking $fname<br>"); 489: unlink("$fname"); 490: } else { 491: $active++; 492: } 493: } 494: &log($fh,"<p>Cleaned up ".$cleaned." stale session token(s).</p>"); 495: &log($fh,"<h3>$active open session(s)</h3>"); 496: } 497: 498: # ----------------------------------------------------------- clean out sockets 499: sub clean_sockets { 500: my ($fh)=@_; 501: my $cleaned=0; 502: opendir(SOCKETS,$perlvar{'lonSockDir'}); 503: while (my $fname=readdir(SOCKETS)) { 504: next if (-d $fname 505: || $fname=~/(mysqlsock|maximasock|rsock|\Q$perlvar{'lonSockDir'}\E)/); 506: $cleaned++; 507: &log($fh,"Unlinking $fname<br />"); 508: unlink("/home/httpd/sockets/$fname"); 509: } 510: &log($fh,"<p>Cleaned up ".$cleaned." stale sockets.</p>"); 511: } 512: 513: 514: # ----------------------------------------------------------------------- httpd 515: sub check_httpd_logs { 516: my ($fh)=@_; 517: &log($fh,'<hr /><a name="httpd" /><h2>httpd</h2><h3>Access Log</h3><pre>'); 518: 519: open (DFH,"tail -n25 /etc/httpd/logs/access_log|"); 520: while (my $line=<DFH>) { &log($fh,&encode_entities($line,'<>&"')) }; 521: close (DFH); 522: 523: &log($fh,"</pre><h3>Error Log</h3><pre>"); 524: 525: open (DFH,"tail -n25 /etc/httpd/logs/error_log|"); 526: while (my $line=<DFH>) { 527: &log($fh,"$line"); 528: if ($line=~/\[error\]/) { $notices++; } 529: } 530: close (DFH); 531: &log($fh,"</pre>"); 532: &errout($fh); 533: } 534: 535: # ---------------------------------------------------------------------- lonnet 536: 537: sub rotate_lonnet_logs { 538: my ($fh)=@_; 539: &log($fh,'<hr /><a name="lonnet" /><h2>lonnet</h2><h3>Temp Log</h3><pre>'); 540: print "checking logs\n"; 541: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){ 542: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|"); 543: while (my $line=<DFH>) { 544: &log($fh,&encode_entities($line,'<>&"')); 545: } 546: close (DFH); 547: } 548: &log($fh,"</pre><h3>Perm Log</h3><pre>"); 549: 550: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") { 551: open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|"); 552: while (my $line=<DFH>) { 553: &log($fh,&encode_entities($line,'<>&"')); 554: } 555: close (DFH); 556: } else { &log($fh,"No perm log\n") } 557: 558: my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log"; 559: &rotate_logfile($fname,$fh,'lonnet log'); 560: 561: &log($fh,"</pre>"); 562: &errout($fh); 563: } 564: 565: sub rotate_other_logs { 566: my ($fh) = @_; 567: my %logs = ( 568: autoenroll => 'Auto Enroll log', 569: autocreate => 'Create Course log', 570: searchcat => 'Search Cataloguing log', 571: autoupdate => 'Auto Update log', 572: refreshcourseids_db => 'Refresh CourseIDs db log', 573: ); 574: foreach my $item (keys(%logs)) { 575: my $fname=$perlvar{'lonDaemons'}.'/logs/'.$item.'.log'; 576: &rotate_logfile($fname,$fh,$logs{$item}); 577: } 578: } 579: 580: # ----------------------------------------------------------------- Connections 581: sub test_connections { 582: my ($fh)=@_; 583: &log($fh,'<hr /><a name="connections" /><h2>Connections</h2>'); 584: print "testing connections\n"; 585: &log($fh,"<table border='2'>"); 586: my ($good,$bad)=(0,0); 587: my %hostname = &Apache::lonnet::all_hostnames(); 588: foreach my $tryserver (sort(keys(%hostname))) { 589: print("."); 590: my $result; 591: my $answer=&Apache::lonnet::reply("ping",$tryserver); 592: if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { 593: $result="<b>ok</b>"; 594: $good++; 595: } else { 596: $result=$answer; 597: $warnings++; 598: if ($answer eq 'con_lost') { 599: $bad++; 600: $warnings++; 601: } else { 602: $good++; #self connection 603: } 604: } 605: if ($answer =~ /con_lost/) { print(" $tryserver down\n"); } 606: &log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n"); 607: } 608: &log($fh,"</table>"); 609: print "\n$good good, $bad bad connections\n"; 610: &errout($fh); 611: } 612: 613: 614: # ------------------------------------------------------------ Delayed messages 615: sub check_delayed_msg { 616: my ($fh)=@_; 617: &log($fh,'<hr /><a name="delayed" /><h2>Delayed Messages</h2>'); 618: print "checking buffers\n"; 619: 620: &log($fh,'<h3>Scanning Permanent Log</h3>'); 621: 622: my $unsend=0; 623: 624: my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log"); 625: while (my $line=<$dfh>) { 626: my ($time,$sdf,$dserv,$dcmd)=split(/:/,$line); 627: if ($sdf eq 'F') { 628: my $local=localtime($time); 629: &log($fh,"<b>Failed: $time, $dserv, $dcmd</b><br>"); 630: $warnings++; 631: } 632: if ($sdf eq 'S') { $unsend--; } 633: if ($sdf eq 'D') { $unsend++; } 634: } 635: 636: &log($fh,"<p>Total unsend messages: <b>$unsend</b></p>\n"); 637: $warnings=$warnings+5*$unsend; 638: 639: if ($unsend) { $simplestatus{'unsend'}=$unsend; } 640: &log($fh,"<h3>Outgoing Buffer</h3>\n<pre>"); 641: # list directory with delayed messages and remember offline servers 642: my %servers=(); 643: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|"); 644: while (my $line=<DFH>) { 645: my ($server)=($line=~/\.(\w+)$/); 646: if ($server) { $servers{$server}=1; } 647: &log($fh,&encode_entities($line,'<>&"')); 648: } 649: &log($fh,"</pre>\n"); 650: close (DFH); 651: # pong to all servers that have delayed messages 652: # this will trigger a reverse connection, which should flush the buffers 653: foreach my $tryserver (keys %servers) { 654: my $answer=&Apache::lonnet::reply("pong",$tryserver); 655: &log($fh,"Pong to $tryserver: $answer<br />"); 656: } 657: } 658: 659: sub finish_logging { 660: my ($fh)=@_; 661: &log($fh,"<a name='errcount' />\n"); 662: $totalcount=$notices+4*$warnings+100*$errors; 663: &errout($fh); 664: &log($fh,"<h1>Total Error Count: $totalcount</h1>"); 665: my $now=time; 666: my $date=localtime($now); 667: &log($fh,"<hr />$date ($now)</body></html>\n"); 668: print "lon-status webpage updated\n"; 669: $fh->close(); 670: 671: if ($errors) { $simplestatus{'errors'}=$errors; } 672: if ($warnings) { $simplestatus{'warnings'}=$warnings; } 673: if ($notices) { $simplestatus{'notices'}=$notices; } 674: $simplestatus{'time'}=time; 675: } 676: 677: sub log_simplestatus { 678: rename("$statusdir/newstatus.html","$statusdir/index.html"); 679: 680: my $sfh=IO::File->new(">$statusdir/loncron_simple.txt"); 681: foreach (keys %simplestatus) { 682: print $sfh $_.'='.$simplestatus{$_}.'&'; 683: } 684: print $sfh "\n"; 685: $sfh->close(); 686: } 687: 688: sub write_loncaparevs { 689: if (open(my $fh,">$perlvar{'lonTabDir'}/loncaparevs.tab")) { 690: my %hostname = &Apache::lonnet::all_hostnames(); 691: foreach my $id (sort(keys(%hostname))) { 692: if ($id ne '') { 693: my $loncaparev = &Apache::lonnet::get_server_loncaparev('',$id,1,'loncron'); 694: if ($loncaparev =~ /^[\d.\-]+$/) { 695: print $fh $id.':'.$loncaparev."\n"; 696: } 697: } 698: } 699: close($fh); 700: } 701: return; 702: } 703: 704: sub write_serverhomeIDs { 705: if (open(my $fh,">$perlvar{'lonTabDir'}/serverhomeIDs.tab")) { 706: my %name_to_host = &Apache::lonnet::all_names(); 707: foreach my $name (sort(keys(%name_to_host))) { 708: if ($name ne '') { 709: if (ref($name_to_host{$name}) eq 'ARRAY') { 710: my $serverhomeID = &Apache::lonnet::get_server_homeID($name,1,'loncron'); 711: if ($serverhomeID ne '') { 712: print $fh $name.':'.$serverhomeID."\n"; 713: } else { 714: print $fh $name.':'.$name_to_host{$name}->[0]."\n"; 715: } 716: } 717: } 718: } 719: close($fh); 720: } 721: return; 722: } 723: 724: sub send_mail { 725: print "sending mail\n"; 726: my $defdom = $perlvar{'lonDefDomain'}; 727: my $origmail = $perlvar{'lonAdmEMail'}; 728: my $emailto = &Apache::loncommon::build_recipient_list(undef, 729: 'lonstatusmail',$defdom,$origmail); 730: if ($totalcount>2500) { 731: $emailto.=",$perlvar{'lonSysEMail'}"; 732: } 733: my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; 734: 735: my $result=system("metasend -b -S 4000000 -t $emailto -s '$subj' -f $statusdir/index.html -m text/html >& /dev/null"); 736: if ($result != 0) { 737: $result=system("mail -s '$subj' $emailto < $statusdir/index.html"); 738: } 739: } 740: 741: sub usage { 742: print(<<USAGE); 743: loncron - housekeeping program that checks up on various parts of Lon-CAPA 744: 745: Options: 746: --help Display 747: --noemail Do not send the status email 748: --justcheckconnections Only check the current status of the lonc/d 749: connections, do not send emails do not 750: check if the daemons are running, do not 751: generate lon-status 752: --justcheckdaemons Only check that all of the Lon-CAPA daemons are 753: running, do not send emails do not 754: check the lonc/d connections, do not 755: generate lon-status 756: --justreload Only tell the daemons to reload the config files, 757: do not send emails do not 758: check if the daemons are running, do not 759: generate lon-status 760: 761: USAGE 762: } 763: 764: # ================================================================ Main Program 765: sub main () { 766: my ($help,$justcheckdaemons,$noemail,$justcheckconnections, 767: $justreload); 768: &GetOptions("help" => \$help, 769: "justcheckdaemons" => \$justcheckdaemons, 770: "noemail" => \$noemail, 771: "justcheckconnections" => \$justcheckconnections, 772: "justreload" => \$justreload 773: ); 774: if ($help) { &usage(); return; } 775: # --------------------------------- Read loncapa_apache.conf and loncapa.conf 776: my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf'); 777: %perlvar=%{$perlvarref}; 778: undef $perlvarref; 779: delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed 780: delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed 781: chdir($perlvar{'lonDaemons'}); 782: # --------------------------------------- Make sure that LON-CAPA is configured 783: # I only test for one thing here (lonHostID). This is just a safeguard. 784: if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) { 785: print("Unconfigured machine.\n"); 786: my $emailto=$perlvar{'lonSysEMail'}; 787: my $hostname=`/bin/hostname`; 788: chop $hostname; 789: $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell 790: my $subj="LON: Unconfigured machine $hostname"; 791: system("echo 'Unconfigured machine $hostname.' |\ 792: mailto $emailto -s '$subj' > /dev/null"); 793: exit 1; 794: } 795: 796: # ----------------------------- Make sure this process is running from user=www 797: my $wwwid=getpwnam('www'); 798: if ($wwwid!=$<) { 799: print("User ID mismatch. This program must be run as user 'www'\n"); 800: my $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; 801: my $subj="LON: $perlvar{'lonHostID'} User ID mismatch"; 802: system("echo 'User ID mismatch. loncron must be run as user www.' |\ 803: mailto $emailto -s '$subj' > /dev/null"); 804: exit 1; 805: } 806: 807: # -------------------------------------------- Force reload of host information 808: &Apache::lonnet::load_hosts_tab(1); 809: &Apache::lonnet::load_domain_tab(1); 810: &Apache::lonnet::get_iphost(1); 811: 812: # ----------------------------------------- Force firewall update for lond port 813: 814: if ((!$justcheckdaemons) && (!$justreload)) { 815: my $now = time; 816: my $tmpfile = $perlvar{'lonDaemons'}.'/tmp/lciptables_iphost_'. 817: $now.$$.int(rand(10000)); 818: if (open(my $fh,">$tmpfile")) { 819: my %iphosts = &Apache::lonnet::get_iphost(); 820: foreach my $key (keys(%iphosts)) { 821: print $fh "$key\n"; 822: } 823: close($fh); 824: my $execpath = $perlvar{'lonDaemons'}.'/lciptables'; 825: system("$execpath $tmpfile"); 826: unlink($fh); 827: } 828: } 829: 830: # ---------------------------------------------------------------- Start report 831: 832: $errors=0; 833: $warnings=0; 834: $notices=0; 835: 836: 837: my $fh; 838: if (!$justcheckdaemons && !$justcheckconnections && !$justreload) { 839: $fh=&start_logging(); 840: 841: &log_machine_info($fh); 842: &clean_tmp($fh); 843: &clean_lonIDs($fh); 844: &check_httpd_logs($fh); 845: &rotate_lonnet_logs($fh); 846: &rotate_other_logs($fh); 847: } 848: if (!$justcheckconnections && !$justreload) { 849: &checkon_daemon($fh,'lonmemcached',40000); 850: &checkon_daemon($fh,'lonsql',200000); 851: if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') { 852: &checkon_daemon($fh,'lond',40000,'USR2'); 853: } 854: &checkon_daemon($fh,'lonc',40000,'USR1'); 855: &checkon_daemon($fh,'lonmaxima',40000); 856: &checkon_daemon($fh,'lonr',40000); 857: } 858: if ($justreload) { 859: &checkon_daemon($fh,'lond',40000,'USR2'); 860: &checkon_daemon($fh,'lonc',40000,'USR2'); 861: } 862: if ($justcheckconnections) { 863: &test_connections($fh); 864: } 865: if (!$justcheckdaemons && !$justcheckconnections && !$justreload) { 866: &check_delayed_msg($fh); 867: &finish_logging($fh); 868: &log_simplestatus(); 869: 870: if ($totalcount>200 && !$noemail) { &send_mail(); } 871: } 872: if (!$justcheckconnections && !$justreload) { 873: &write_loncaparevs(); 874: &write_serverhomeIDs(); 875: } 876: } 877: 878: &main(); 879: 1; 880: 881: 882: 883: 884: 885: 886: 887: