Annotation of loncom/loncron, revision 1.10
1.1 albertel 1: #!/usr/bin/perl
2:
3: # The LearningOnline Network
4: # Housekeeping program, started by cron
5: #
6: # (TCP networking package
7: # 6/1/99,6/2,6/10,6/11,6/12,6/14,6/26,6/28,6/29,6/30,
8: # 7/1,7/2,7/9,7/10,7/12 Gerd Kortemeyer)
9: #
1.3 www 10: # 7/14,7/15,7/19,7/21,7/22,11/18,
11: # 2/8 Gerd Kortemeyer
1.1 albertel 12:
13: use IO::File;
14: use IO::Socket;
15:
16: # -------------------------------------------------- Non-critical communication
17: sub reply {
18: my ($cmd,$server)=@_;
19: my $peerfile="$perlvar{'lonSockDir'}/$server";
20: my $client=IO::Socket::UNIX->new(Peer =>"$peerfile",
21: Type => SOCK_STREAM,
22: Timeout => 10)
23: or return "con_lost";
24: print $client "$cmd\n";
25: my $answer=<$client>;
26: chomp($answer);
27: if (!$answer) { $answer="con_lost"; }
28: return $answer;
29: }
30:
31: # --------------------------------------------------------- Output error status
32:
33: sub errout {
34: my $fh=shift;
35: print $fh (<<ENDERROUT);
36: <p><table border=2 bgcolor="#CCCCCC">
37: <tr><td>Notices</td><td>$notices</td></tr>
38: <tr><td>Warnings</td><td>$warnings</td></tr>
39: <tr><td>Errors</td><td>$errors</td></tr>
40: </table><p><a href="#top">Top</a><p>
41: ENDERROUT
42: }
43:
44: # ================================================================ Main Program
45:
46:
47: # ------------------------------------------------------------ Read access.conf
48: {
49: my $config=IO::File->new("/etc/httpd/conf/access.conf");
50:
51: while (my $configline=<$config>) {
52: if ($configline =~ /PerlSetVar/) {
53: my ($dummy,$varname,$varvalue)=split(/\s+/,$configline);
54: $perlvar{$varname}=$varvalue;
55: }
56: }
57: }
58:
59: # ------------------------------------------------------------- Read hosts file
60: {
61: my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
62:
63: while (my $configline=<$config>) {
64: my ($id,$domain,$role,$name,$ip)=split(/:/,$configline);
65: $hostname{$id}=$name;
66: $hostdom{$id}=$domain;
67: $hostrole{$id}=$role;
68: $hostip{$id}=$ip;
69: if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) {
70: $libserv{$id}=$name;
71: }
72: }
73: }
74:
75: # ------------------------------------------------------ Read spare server file
76: {
77: my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
78:
79: while (my $configline=<$config>) {
80: chomp($configline);
81: if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
82: $spareid{$configline}=1;
83: }
84: }
85: }
86:
87: # ---------------------------------------------------------------- Start report
88:
89: $statusdir="/home/httpd/html/lon-status";
90:
91: $errors=0;
92: $warnings=0;
93: $notices=0;
94:
95: $now=time;
96: $date=localtime($now);
97:
98: {
99: my $fh=IO::File->new(">$statusdir/newstatus.html");
100:
101: print $fh (<<ENDHEADERS);
102: <html>
103: <head>
104: <title>LON Status Report $perlvar{'lonHostID'}</title>
105: </head>
1.3 www 106: <body bgcolor="#AAAAAA">
1.1 albertel 107: <a name="top">
108: <h1>LON Status Report $perlvar{'lonHostID'}</h1>
109: <h2>$date ($now)</h2>
110: <ol>
111: <li><a href="#configuration">Configuration</a>
112: <li><a href="#machine">Machine Information</a>
113: <li><a href="#httpd">httpd</a>
114: <li><a href="#lond">lond</a>
115: <li><a href="#lonc">lonc</a>
116: <li><a href="#lonnet">lonnet</a>
117: <li><a href="#connections">Connections</a>
118: <li><a href="#delayed">Delayed Messages</a>
119: <li><a href="#errcount">Error Count</a>
120: </ol>
121: <hr>
122: <a name="configuration">
123: <h2>Configuration</h2>
124: <h3>PerlVars</h3>
125: <table border=2>
126: ENDHEADERS
127:
128: foreach $varname (keys %perlvar) {
129: print $fh "<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n";
130: }
131: print $fh "</table><h3>Hosts</h3><table border=2>";
132: foreach $id (keys %hostname) {
133: print $fh
134: "<tr><td>$id</td><td>$hostdom{$id}</td><td>$hostrole{$id}</td>";
135: print $fh "<td>$hostname{$id}</td><td>$hostip{$id}</td></tr>\n";
136: }
137: print $fh "</table><h3>Spare Hosts</h3><ol>";
138: foreach $id (keys %spareid) {
139: print $fh "<li>$id\n";
140: }
141:
142: print $fh "</ol>\n";
143:
144: # --------------------------------------------------------------------- Machine
145:
146: print $fh '<hr><a name="machine"><h2>Machine Information</h2>';
147: print $fh "<h3>loadavg</h3>";
148:
149: open (LOADAVGH,"/proc/loadavg");
150: $loadavg=<LOADAVGH>;
151: close (LOADAVGH);
152:
153: print $fh "<tt>$loadavg</tt>";
154:
155: @parts=split(/\s+/,$loadavg);
1.4 www 156: if ($parts[1]>4.0) {
1.1 albertel 157: $errors++;
158: } elsif ($parts[1]>2.0) {
159: $warnings++;
160: } elsif ($parts[1]>1.0) {
161: $notices++;
162: }
163:
164: print $fh "<h3>df</h3>";
165: print $fh "<pre>";
166:
167: open (DFH,"df|");
168: while ($line=<DFH>) {
169: print $fh "$line";
170: @parts=split(/\s+/,$line);
171: $usage=$parts[4];
172: $usage=~s/\W//g;
173: if ($usage>90) {
1.4 www 174: $warnings++;
1.1 albertel 175: } elsif ($usage>80) {
176: $warnings++;
177: } elsif ($usage>60) {
178: $notices++;
179: }
1.4 www 180: if ($usage>95) { $warnings++; $warnings++ }
1.1 albertel 181: }
182: close (DFH);
183: print $fh "</pre>";
184: &errout($fh);
185: # ----------------------------------------------------------------------- httpd
186:
187: print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>';
188:
189: open (DFH,"tail -n40 /etc/httpd/logs/access_log|");
190: while ($line=<DFH>) { print $fh "$line" };
191: close (DFH);
192:
193: print $fh "</pre><h3>Error Log</h3><pre>";
194:
195: open (DFH,"tail -n50 /etc/httpd/logs/error_log|");
196: while ($line=<DFH>) {
197: print $fh "$line";
198: if ($line=~/\[error\]/) { $notices++; }
199: };
200: close (DFH);
201: print $fh "</pre>";
202: &errout($fh);
1.5 harris41 203:
204:
205: # ------------------------------------------------------------------------ lonsql
206:
1.6 harris41 207: if ($perlvar{'lonRole'} eq "library" && 1==0) {
1.5 harris41 208:
209: print $fh '<hr><a name="lond"><h2>lonsql</h2><h3>Log</h3><pre>';
210:
211: if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
212: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
213: while ($line=<DFH>) {
214: print $fh "$line";
215: if ($line=~/INFO/) { $notices++; }
216: if ($line=~/WARNING/) { $notices++; }
217: if ($line=~/CRITICAL/) { $warnings++; }
218: };
219: close (DFH);
220: }
221: print $fh "</pre>";
222:
223: my $lonsqlfile="$perlvar{'lonDaemons'}/logs/lonsql.pid";
224:
225: if (-e $lonsqlfile) {
226: my $lfh=IO::File->new("$lonsqlfile");
227: my $lonsqlpid=<$lfh>;
228: chomp($lonsqlpid);
229: if (kill 0 => $lonsqlpid) {
230: print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";
231: } else {
232: $errors++; $errors++;
233: print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
234: }
235: } else {
236: $errors++;
237: print $fh "<h3>lonsql not running, trying to start</h3>";
238: system("$perlvar{'lonDaemons'}/lonsql");
239: sleep 120;
240: if (-e $lonsqlfile) {
241: print $fh "Seems like it started ...<p>";
242: my $lfh=IO::File->new("$lonsqlfile");
243: my $lonsqlpid=<$lfh>;
244: chomp($lonsqlpid);
245: sleep 30;
246: if (kill 0 => $lonsqlpid) {
247: print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";
248: } else {
249: $errors++; $errors++;
250: print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
251: print $fh "Give it one more try ...<p>";
252: system("$perlvar{'lonDaemons'}/lonsql");
253: sleep 120;
254: }
255: } else {
256: print $fh "Seems like that did not work!<p>";
257: $errors++;
258: }
259: if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
260: print $fh "<p><pre>";
261: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
262: while ($line=<DFH>) {
263: print $fh "$line";
264: if ($line=~/WARNING/) { $notices++; }
265: if ($line=~/CRITICAL/) { $notices++; }
266: };
267: close (DFH);
268: print $fh "</pre>";
269: }
270: }
271:
272: $fname="$perlvar{'lonDaemons'}/logs/lonsql.log";
273:
274: my ($dev,$ino,$mode,$nlink,
275: $uid,$gid,$rdev,$size,
276: $atime,$mtime,$ctime,
277: $blksize,$blocks)=stat($fname);
278:
279: if ($size>40000) {
280: print $fh "Rotating logs ...<p>";
281: rename("$fname.2","$fname.3");
282: rename("$fname.1","$fname.2");
283: rename("$fname","$fname.1");
284: }
285:
286: &errout($fh);
287: }
1.1 albertel 288: # ------------------------------------------------------------------------ lond
289:
290: print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>';
291:
292: if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
1.3 www 293: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
1.1 albertel 294: while ($line=<DFH>) {
295: print $fh "$line";
1.3 www 296: if ($line=~/INFO/) { $notices++; }
1.4 www 297: if ($line=~/WARNING/) { $notices++; }
298: if ($line=~/CRITICAL/) { $warnings++; }
1.1 albertel 299: };
300: close (DFH);
301: }
302: print $fh "</pre>";
303:
304: my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
305:
1.7 harris41 306: my $restartflag=1;
307: if (-e $londfile) {
1.1 albertel 308: my $lfh=IO::File->new("$londfile");
309: my $londpid=<$lfh>;
310: chomp($londpid);
311: if (kill 0 => $londpid) {
312: print $fh "<h3>lond at pid $londpid responding</h3>";
1.7 harris41 313: $restartflag=0;
1.1 albertel 314: } else {
1.8 harris41 315: $errors++;
1.1 albertel 316: print $fh "<h3>lond at pid $londpid not responding</h3>";
1.7 harris41 317: # Intelligently handle this.
318: # Possibility #1: there is no process
319: # Solution: remove .pid file and restart
320: if (getpgrp($londpid)==-1) {
321: unlink($londfile);
322: $restartflag=1;
323: }
1.8 harris41 324: else {
1.10 ! harris41 325: # Possibility #2: there is a live process that is not responding
1.7 harris41 326: # for an unknown reason
1.10 ! harris41 327: # Solution: kill parent and children processes, remove .pid and restart
1.9 harris41 328: `killall -9 lond`;
1.8 harris41 329: unlink($londfile);
330: $restartflag=1;
331: }
332: print $fh
333: "<h3>Deciding to clean up stale .pid file and restart lond</h3>";
1.1 albertel 334: }
1.7 harris41 335: }
336: if ($restartflag==1) {
1.1 albertel 337: $errors++;
338: print $fh "<h3>lond not running, trying to start</h3>";
339: system("$perlvar{'lonDaemons'}/lond");
340: sleep 120;
341: if (-e $londfile) {
342: print $fh "Seems like it started ...<p>";
343: my $lfh=IO::File->new("$londfile");
344: my $londpid=<$lfh>;
345: chomp($londpid);
346: sleep 30;
347: if (kill 0 => $londpid) {
348: print $fh "<h3>lond at pid $londpid responding</h3>";
349: } else {
350: $errors++; $errors++;
351: print $fh "<h3>lond at pid $londpid not responding</h3>";
352: print $fh "Give it one more try ...<p>";
353: system("$perlvar{'lonDaemons'}/lond");
354: sleep 120;
355: }
356: } else {
357: print $fh "Seems like that did not work!<p>";
358: $errors++;
359: }
1.3 www 360: if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
361: print $fh "<p><pre>";
362: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
363: while ($line=<DFH>) {
364: print $fh "$line";
1.4 www 365: if ($line=~/WARNING/) { $notices++; }
366: if ($line=~/CRITICAL/) { $notices++; }
1.3 www 367: };
368: close (DFH);
369: print $fh "</pre>";
370: }
1.1 albertel 371: }
372:
373: $fname="$perlvar{'lonDaemons'}/logs/lond.log";
374:
375: my ($dev,$ino,$mode,$nlink,
376: $uid,$gid,$rdev,$size,
377: $atime,$mtime,$ctime,
378: $blksize,$blocks)=stat($fname);
379:
380: if ($size>40000) {
381: print $fh "Rotating logs ...<p>";
382: rename("$fname.2","$fname.3");
383: rename("$fname.1","$fname.2");
384: rename("$fname","$fname.1");
385: }
386:
387: &errout($fh);
388: # ------------------------------------------------------------------------ lonc
389:
390: print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>';
391:
392: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
1.3 www 393: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
1.1 albertel 394: while ($line=<DFH>) {
395: print $fh "$line";
1.3 www 396: if ($line=~/INFO/) { $notices++; }
1.4 www 397: if ($line=~/WARNING/) { $notices++; }
398: if ($line=~/CRITICAL/) { $warnings++; }
1.1 albertel 399: };
400: close (DFH);
401: }
402: print $fh "</pre>";
403:
404: my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
405:
1.7 harris41 406: $restartflag=1;
1.1 albertel 407: if (-e $loncfile) {
408: my $lfh=IO::File->new("$loncfile");
409: my $loncpid=<$lfh>;
410: chomp($loncpid);
411: if (kill 0 => $loncpid) {
412: print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";
413: kill USR1 => $loncpid;
1.7 harris41 414: $restartflag=0;
1.1 albertel 415: } else {
1.8 harris41 416: $errors++;
1.1 albertel 417: print $fh "<h3>lonc at pid $loncpid not responding</h3>";
1.7 harris41 418: # Intelligently handle this.
419: # Possibility #1: there is no process
420: # Solution: remove .pid file and restart
421: if (getpgrp($loncpid)==-1) {
422: unlink($loncfile);
423: $restartflag=1;
424: }
1.8 harris41 425: else {
1.10 ! harris41 426: # Possibility #2: there is a live process that is not responding
1.7 harris41 427: # for an unknown reason
1.10 ! harris41 428: # Solution: kill parent and children processes, remove .pid and restart
1.9 harris41 429: `killall -9 lonc`;
1.8 harris41 430: unlink($loncfile);
431: $restartflag=1;
432: }
433: print $fh
434: "<h3>Deciding to clean up stale .pid file and restart lonc</h3>";
1.1 albertel 435: }
1.7 harris41 436: }
437: if ($restartflag==1) {
1.1 albertel 438: $errors++;
439: print $fh "<h3>lonc not running, trying to start</h3>";
440: system("$perlvar{'lonDaemons'}/lonc");
441: sleep 120;
442: if (-e $loncfile) {
443: print $fh "Seems like it started ...<p>";
444: my $lfh=IO::File->new("$loncfile");
445: my $loncpid=<$lfh>;
446: chomp($loncpid);
447: sleep 30;
448: if (kill 0 => $loncpid) {
449: print $fh "<h3>lonc at pid $loncpid responding</h3>";
450: } else {
451: $errors++; $errors++;
452: print $fh "<h3>lonc at pid $loncpid not responding</h3>";
453: print $fh "Give it one more try ...<p>";
454: system("$perlvar{'lonDaemons'}/lonc");
455: sleep 120;
456: }
457: } else {
458: print $fh "Seems like that did not work!<p>";
459: $errors++;
460: }
1.3 www 461: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") {
462: print $fh "<p><pre>";
463: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
464: while ($line=<DFH>) {
465: print $fh "$line";
1.4 www 466: if ($line=~/WARNING/) { $notices++; }
467: if ($line=~/CRITICAL/) { $notices++; }
1.3 www 468: };
469: close (DFH);
470: print $fh "</pre>";
471: }
1.1 albertel 472: }
473:
474: $fname="$perlvar{'lonDaemons'}/logs/lonc.log";
475:
476: my ($dev,$ino,$mode,$nlink,
477: $uid,$gid,$rdev,$size,
478: $atime,$mtime,$ctime,
479: $blksize,$blocks)=stat($fname);
480:
481: if ($size>40000) {
482: print $fh "Rotating logs ...<p>";
483: rename("$fname.2","$fname.3");
484: rename("$fname.1","$fname.2");
485: rename("$fname","$fname.1");
486: }
487:
488:
489: &errout($fh);
490: # ---------------------------------------------------------------------- lonnet
491:
492: print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>';
493: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
494: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
495: while ($line=<DFH>) {
496: print $fh "$line";
497: };
498: close (DFH);
499: }
500: print $fh "</pre><h3>Perm Log</h3>";
501:
502: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
503: open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
504: while ($line=<DFH>) {
505: print $fh "$line";
506: };
507: close (DFH);
508: } else { print $fh "No perm log\n" }
509:
510: $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
511:
512: my ($dev,$ino,$mode,$nlink,
513: $uid,$gid,$rdev,$size,
514: $atime,$mtime,$ctime,
515: $blksize,$blocks)=stat($fname);
516:
517: if ($size>40000) {
518: print $fh "Rotating logs ...<p>";
519: rename("$fname.2","$fname.3");
520: rename("$fname.1","$fname.2");
521: rename("$fname","$fname.1");
522: }
523:
524: print $fh "</pre>";
525: &errout($fh);
526: # ----------------------------------------------------------------- Connections
527:
528: print $fh '<hr><a name="connections"><h2>Connections</h2>';
529:
530: print $fh "<table border=2>";
531: foreach $tryserver (keys %hostname) {
532:
533: $answer=reply("pong",$tryserver);
534: if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
535: $result="<b>ok</b>";
536: } else {
537: $result=$answer;
538: $warnings++;
539: if ($answer eq 'con_lost') { $warnings++; }
540: }
541: print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n";
542:
543: }
544: print $fh "</table>";
545:
546: &errout($fh);
547: # ------------------------------------------------------------ Delayed messages
548:
549: print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
550:
551: print $fh '<h3>Scanning Permanent Log</h3>';
552:
553: $unsend=0;
554: {
555: my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
556: while ($line=<$dfh>) {
557: ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
558: if ($sdf eq 'F') {
559: $local=localtime($time);
560: print "<b>Failed: $time, $dserv, $dcmd</b><br>";
561: $warnings++;
562: }
563: if ($sdf eq 'S') { $unsend--; }
564: if ($sdf eq 'D') { $unsend++; }
565: }
566: }
567: print $fh "Total unsend messages: <b>$unsend</b><p>\n";
568: $warnings=$warnings+5*$unsend;
569:
570: print $fh "<h3>Outgoing Buffer</h3>";
571:
572: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
573: while ($line=<DFH>) {
574: print $fh "$line<br>";
575: };
576: close (DFH);
577:
578: # ------------------------------------------------------------------------- End
579: print $fh "<a name=errcount>\n";
580: $totalcount=$notices+4*$warnings+100*$errors;
581: &errout($fh);
582: print $fh "<h1>Total Error Count: $totalcount</h1>";
583: $now=time;
584: $date=localtime($now);
585: print $fh "<hr>$date ($now)</body></html>\n";
586:
587: }
588:
589: rename ("$statusdir/newstatus.html","$statusdir/index.html");
590:
591: if ($totalcount>200) {
592: $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
593: $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";
594: system(
595: "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");
596: }
597: 1;
598:
599:
600:
601:
602:
603:
604:
605:
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