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: #
10: # 7/14,7/15,7/19,7/21,7/22,11/18,
11: # 2/8 Gerd Kortemeyer
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>
106: <body bgcolor="#AAAAAA">
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+/,$loadavg);
156: if ($parts[1]>4.0) {
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) {
174: $warnings++;
175: } elsif ($usage>80) {
176: $warnings++;
177: } elsif ($usage>60) {
178: $notices++;
179: }
180: if ($usage>95) { $warnings++; $warnings++ }
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);
203:
204:
205: # ------------------------------------------------------------------------ lonsql
206:
207: if ($perlvar{'lonRole'} eq "library" && 1==0) {
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: }
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"){
293: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
294: while ($line=<DFH>) {
295: print $fh "$line";
296: if ($line=~/INFO/) { $notices++; }
297: if ($line=~/WARNING/) { $notices++; }
298: if ($line=~/CRITICAL/) { $warnings++; }
299: };
300: close (DFH);
301: }
302: print $fh "</pre>";
303:
304: my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
305:
306: my $restartflag=1;
307: if (-e $londfile) {
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>";
313: $restartflag=0;
314: } else {
315: $errors++; $errors++;
316: print $fh "<h3>lond at pid $londpid not responding</h3>";
317: # Intelligently handle this.
318: # Possibility #1: there is no process
319: # Solution: remove .pid file and restart
320: if (getpgrp($londpid)==-1) {
321: unlink($londfile);
322: $restartflag=1;
323: }
324: # Possibility #2: there is a zombie process
325: # Possibility #3: there is a live process that is not responding
326: # for an unknown reason
327: # Solution: kill process, remove .pid and restart
328: kill 2 => $londpid;
329: unlink($londfile);
330: $restartflag=1;
331: print $fh "<h3>Deciding to clean up stale .pid file and restart lond</h3>";
332: }
333: }
334: if ($restartflag==1) {
335: $errors++;
336: print $fh "<h3>lond not running, trying to start</h3>";
337: system("$perlvar{'lonDaemons'}/lond");
338: sleep 120;
339: if (-e $londfile) {
340: print $fh "Seems like it started ...<p>";
341: my $lfh=IO::File->new("$londfile");
342: my $londpid=<$lfh>;
343: chomp($londpid);
344: sleep 30;
345: if (kill 0 => $londpid) {
346: print $fh "<h3>lond at pid $londpid responding</h3>";
347: } else {
348: $errors++; $errors++;
349: print $fh "<h3>lond at pid $londpid not responding</h3>";
350: print $fh "Give it one more try ...<p>";
351: system("$perlvar{'lonDaemons'}/lond");
352: sleep 120;
353: }
354: } else {
355: print $fh "Seems like that did not work!<p>";
356: $errors++;
357: }
358: if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
359: print $fh "<p><pre>";
360: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
361: while ($line=<DFH>) {
362: print $fh "$line";
363: if ($line=~/WARNING/) { $notices++; }
364: if ($line=~/CRITICAL/) { $notices++; }
365: };
366: close (DFH);
367: print $fh "</pre>";
368: }
369: }
370:
371: $fname="$perlvar{'lonDaemons'}/logs/lond.log";
372:
373: my ($dev,$ino,$mode,$nlink,
374: $uid,$gid,$rdev,$size,
375: $atime,$mtime,$ctime,
376: $blksize,$blocks)=stat($fname);
377:
378: if ($size>40000) {
379: print $fh "Rotating logs ...<p>";
380: rename("$fname.2","$fname.3");
381: rename("$fname.1","$fname.2");
382: rename("$fname","$fname.1");
383: }
384:
385: &errout($fh);
386: # ------------------------------------------------------------------------ lonc
387:
388: print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>';
389:
390: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
391: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
392: while ($line=<DFH>) {
393: print $fh "$line";
394: if ($line=~/INFO/) { $notices++; }
395: if ($line=~/WARNING/) { $notices++; }
396: if ($line=~/CRITICAL/) { $warnings++; }
397: };
398: close (DFH);
399: }
400: print $fh "</pre>";
401:
402: my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
403:
404: $restartflag=1;
405: if (-e $loncfile) {
406: my $lfh=IO::File->new("$loncfile");
407: my $loncpid=<$lfh>;
408: chomp($loncpid);
409: if (kill 0 => $loncpid) {
410: print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";
411: kill USR1 => $loncpid;
412: $restartflag=0;
413: } else {
414: $errors++; $errors++;
415: print $fh "<h3>lonc at pid $loncpid not responding</h3>";
416: # Intelligently handle this.
417: # Possibility #1: there is no process
418: # Solution: remove .pid file and restart
419: if (getpgrp($loncpid)==-1) {
420: unlink($loncfile);
421: $restartflag=1;
422: }
423: # Possibility #2: there is a zombie process
424: # Possibility #3: there is a live process that is not responding
425: # for an unknown reason
426: # Solution: kill process, remove .pid and restart
427: kill 2 => $loncpid;
428: unlink($loncfile);
429: $restartflag=1;
430: print $fh "<h3>Deciding to clean up stale .pid file and restart lonc</h3>";
431: }
432: }
433: if ($restartflag==1) {
434: $errors++;
435: print $fh "<h3>lonc not running, trying to start</h3>";
436: system("$perlvar{'lonDaemons'}/lonc");
437: sleep 120;
438: if (-e $loncfile) {
439: print $fh "Seems like it started ...<p>";
440: my $lfh=IO::File->new("$loncfile");
441: my $loncpid=<$lfh>;
442: chomp($loncpid);
443: sleep 30;
444: if (kill 0 => $loncpid) {
445: print $fh "<h3>lonc at pid $loncpid responding</h3>";
446: } else {
447: $errors++; $errors++;
448: print $fh "<h3>lonc at pid $loncpid not responding</h3>";
449: print $fh "Give it one more try ...<p>";
450: system("$perlvar{'lonDaemons'}/lonc");
451: sleep 120;
452: }
453: } else {
454: print $fh "Seems like that did not work!<p>";
455: $errors++;
456: }
457: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") {
458: print $fh "<p><pre>";
459: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
460: while ($line=<DFH>) {
461: print $fh "$line";
462: if ($line=~/WARNING/) { $notices++; }
463: if ($line=~/CRITICAL/) { $notices++; }
464: };
465: close (DFH);
466: print $fh "</pre>";
467: }
468: }
469:
470: $fname="$perlvar{'lonDaemons'}/logs/lonc.log";
471:
472: my ($dev,$ino,$mode,$nlink,
473: $uid,$gid,$rdev,$size,
474: $atime,$mtime,$ctime,
475: $blksize,$blocks)=stat($fname);
476:
477: if ($size>40000) {
478: print $fh "Rotating logs ...<p>";
479: rename("$fname.2","$fname.3");
480: rename("$fname.1","$fname.2");
481: rename("$fname","$fname.1");
482: }
483:
484:
485: &errout($fh);
486: # ---------------------------------------------------------------------- lonnet
487:
488: print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>';
489: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
490: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
491: while ($line=<DFH>) {
492: print $fh "$line";
493: };
494: close (DFH);
495: }
496: print $fh "</pre><h3>Perm Log</h3>";
497:
498: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
499: open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
500: while ($line=<DFH>) {
501: print $fh "$line";
502: };
503: close (DFH);
504: } else { print $fh "No perm log\n" }
505:
506: $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
507:
508: my ($dev,$ino,$mode,$nlink,
509: $uid,$gid,$rdev,$size,
510: $atime,$mtime,$ctime,
511: $blksize,$blocks)=stat($fname);
512:
513: if ($size>40000) {
514: print $fh "Rotating logs ...<p>";
515: rename("$fname.2","$fname.3");
516: rename("$fname.1","$fname.2");
517: rename("$fname","$fname.1");
518: }
519:
520: print $fh "</pre>";
521: &errout($fh);
522: # ----------------------------------------------------------------- Connections
523:
524: print $fh '<hr><a name="connections"><h2>Connections</h2>';
525:
526: print $fh "<table border=2>";
527: foreach $tryserver (keys %hostname) {
528:
529: $answer=reply("pong",$tryserver);
530: if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
531: $result="<b>ok</b>";
532: } else {
533: $result=$answer;
534: $warnings++;
535: if ($answer eq 'con_lost') { $warnings++; }
536: }
537: print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n";
538:
539: }
540: print $fh "</table>";
541:
542: &errout($fh);
543: # ------------------------------------------------------------ Delayed messages
544:
545: print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
546:
547: print $fh '<h3>Scanning Permanent Log</h3>';
548:
549: $unsend=0;
550: {
551: my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
552: while ($line=<$dfh>) {
553: ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
554: if ($sdf eq 'F') {
555: $local=localtime($time);
556: print "<b>Failed: $time, $dserv, $dcmd</b><br>";
557: $warnings++;
558: }
559: if ($sdf eq 'S') { $unsend--; }
560: if ($sdf eq 'D') { $unsend++; }
561: }
562: }
563: print $fh "Total unsend messages: <b>$unsend</b><p>\n";
564: $warnings=$warnings+5*$unsend;
565:
566: print $fh "<h3>Outgoing Buffer</h3>";
567:
568: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
569: while ($line=<DFH>) {
570: print $fh "$line<br>";
571: };
572: close (DFH);
573:
574: # ------------------------------------------------------------------------- End
575: print $fh "<a name=errcount>\n";
576: $totalcount=$notices+4*$warnings+100*$errors;
577: &errout($fh);
578: print $fh "<h1>Total Error Count: $totalcount</h1>";
579: $now=time;
580: $date=localtime($now);
581: print $fh "<hr>$date ($now)</body></html>\n";
582:
583: }
584:
585: rename ("$statusdir/newstatus.html","$statusdir/index.html");
586:
587: if ($totalcount>200) {
588: $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
589: $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";
590: system(
591: "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");
592: }
593: 1;
594:
595:
596:
597:
598:
599:
600:
601:
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