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 Gerd Kortemeyer
11:
12: use IO::File;
13: use IO::Socket;
14:
15: # -------------------------------------------------- Non-critical communication
16: sub reply {
17: my ($cmd,$server)=@_;
18: my $peerfile="$perlvar{'lonSockDir'}/$server";
19: my $client=IO::Socket::UNIX->new(Peer =>"$peerfile",
20: Type => SOCK_STREAM,
21: Timeout => 10)
22: or return "con_lost";
23: print $client "$cmd\n";
24: my $answer=<$client>;
25: chomp($answer);
26: if (!$answer) { $answer="con_lost"; }
27: return $answer;
28: }
29:
30: # --------------------------------------------------------- Output error status
31:
32: sub errout {
33: my $fh=shift;
34: print $fh (<<ENDERROUT);
35: <p><table border=2 bgcolor="#CCCCCC">
36: <tr><td>Notices</td><td>$notices</td></tr>
37: <tr><td>Warnings</td><td>$warnings</td></tr>
38: <tr><td>Errors</td><td>$errors</td></tr>
39: </table><p><a href="#top">Top</a><p>
40: ENDERROUT
41: }
42:
43: # ================================================================ Main Program
44:
45:
46: # ------------------------------------------------------------ Read access.conf
47: {
48: my $config=IO::File->new("/etc/httpd/conf/access.conf");
49:
50: while (my $configline=<$config>) {
51: if ($configline =~ /PerlSetVar/) {
52: my ($dummy,$varname,$varvalue)=split(/\s+/,$configline);
53: $perlvar{$varname}=$varvalue;
54: }
55: }
56: }
57:
58: # ------------------------------------------------------------- Read hosts file
59: {
60: my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
61:
62: while (my $configline=<$config>) {
63: my ($id,$domain,$role,$name,$ip)=split(/:/,$configline);
64: $hostname{$id}=$name;
65: $hostdom{$id}=$domain;
66: $hostrole{$id}=$role;
67: $hostip{$id}=$ip;
68: if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) {
69: $libserv{$id}=$name;
70: }
71: }
72: }
73:
74: # ------------------------------------------------------ Read spare server file
75: {
76: my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
77:
78: while (my $configline=<$config>) {
79: chomp($configline);
80: if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
81: $spareid{$configline}=1;
82: }
83: }
84: }
85:
86: # ---------------------------------------------------------------- Start report
87:
88: $statusdir="/home/httpd/html/lon-status";
89:
90: $errors=0;
91: $warnings=0;
92: $notices=0;
93:
94: $now=time;
95: $date=localtime($now);
96:
97: {
98: my $fh=IO::File->new(">$statusdir/newstatus.html");
99:
100: print $fh (<<ENDHEADERS);
101: <html>
102: <head>
103: <title>LON Status Report $perlvar{'lonHostID'}</title>
104: </head>
105: <body bgcolor="#FFFFFF">
106: <a name="top">
107: <h1>LON Status Report $perlvar{'lonHostID'}</h1>
108: <h2>$date ($now)</h2>
109: <ol>
110: <li><a href="#configuration">Configuration</a>
111: <li><a href="#machine">Machine Information</a>
112: <li><a href="#httpd">httpd</a>
113: <li><a href="#lond">lond</a>
114: <li><a href="#lonc">lonc</a>
115: <li><a href="#lonnet">lonnet</a>
116: <li><a href="#connections">Connections</a>
117: <li><a href="#delayed">Delayed Messages</a>
118: <li><a href="#errcount">Error Count</a>
119: </ol>
120: <hr>
121: <a name="configuration">
122: <h2>Configuration</h2>
123: <h3>PerlVars</h3>
124: <table border=2>
125: ENDHEADERS
126:
127: foreach $varname (keys %perlvar) {
128: print $fh "<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n";
129: }
130: print $fh "</table><h3>Hosts</h3><table border=2>";
131: foreach $id (keys %hostname) {
132: print $fh
133: "<tr><td>$id</td><td>$hostdom{$id}</td><td>$hostrole{$id}</td>";
134: print $fh "<td>$hostname{$id}</td><td>$hostip{$id}</td></tr>\n";
135: }
136: print $fh "</table><h3>Spare Hosts</h3><ol>";
137: foreach $id (keys %spareid) {
138: print $fh "<li>$id\n";
139: }
140:
141: print $fh "</ol>\n";
142:
143: # --------------------------------------------------------------------- Machine
144:
145: print $fh '<hr><a name="machine"><h2>Machine Information</h2>';
146: print $fh "<h3>loadavg</h3>";
147:
148: open (LOADAVGH,"/proc/loadavg");
149: $loadavg=<LOADAVGH>;
150: close (LOADAVGH);
151:
152: print $fh "<tt>$loadavg</tt>";
153:
154: @parts=split(/\s+/,$loadavg);
155: if ($parts[1]>3.0) {
156: $errors++;
157: } elsif ($parts[1]>2.0) {
158: $warnings++;
159: } elsif ($parts[1]>1.0) {
160: $notices++;
161: }
162:
163: print $fh "<h3>df</h3>";
164: print $fh "<pre>";
165:
166: open (DFH,"df|");
167: while ($line=<DFH>) {
168: print $fh "$line";
169: @parts=split(/\s+/,$line);
170: $usage=$parts[4];
171: $usage=~s/\W//g;
172: if ($usage>90) {
173: $errors++;
174: } elsif ($usage>80) {
175: $warnings++;
176: } elsif ($usage>60) {
177: $notices++;
178: }
179: if ($usage>95) { $errors++; }
180: }
181: close (DFH);
182: print $fh "</pre>";
183: &errout($fh);
184: # ----------------------------------------------------------------------- httpd
185:
186: print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>';
187:
188: open (DFH,"tail -n40 /etc/httpd/logs/access_log|");
189: while ($line=<DFH>) { print $fh "$line" };
190: close (DFH);
191:
192: print $fh "</pre><h3>Error Log</h3><pre>";
193:
194: open (DFH,"tail -n50 /etc/httpd/logs/error_log|");
195: while ($line=<DFH>) {
196: print $fh "$line";
197: if ($line=~/\[error\]/) { $notices++; }
198: };
199: close (DFH);
200: print $fh "</pre>";
201: &errout($fh);
202: # ------------------------------------------------------------------------ lond
203:
204: print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>';
205:
206: if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
207: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lond.log|");
208: while ($line=<DFH>) {
209: print $fh "$line";
210: if ($line=~/giving up/) { $notices++; }
211: };
212: close (DFH);
213: }
214: print $fh "</pre>";
215:
216: my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
217:
218: if (-e $londfile) {
219: my $lfh=IO::File->new("$londfile");
220: my $londpid=<$lfh>;
221: chomp($londpid);
222: if (kill 0 => $londpid) {
223: print $fh "<h3>lond at pid $londpid responding</h3>";
224: } else {
225: $errors++; $errors++;
226: print $fh "<h3>lond at pid $londpid not responding</h3>";
227: }
228: } else {
229: $errors++;
230: print $fh "<h3>lond not running, trying to start</h3>";
231: system("$perlvar{'lonDaemons'}/lond");
232: sleep 120;
233: if (-e $londfile) {
234: print $fh "Seems like it started ...<p>";
235: my $lfh=IO::File->new("$londfile");
236: my $londpid=<$lfh>;
237: chomp($londpid);
238: sleep 30;
239: if (kill 0 => $londpid) {
240: print $fh "<h3>lond at pid $londpid responding</h3>";
241: } else {
242: $errors++; $errors++;
243: print $fh "<h3>lond at pid $londpid not responding</h3>";
244: print $fh "Give it one more try ...<p>";
245: system("$perlvar{'lonDaemons'}/lond");
246: sleep 120;
247: }
248: } else {
249: print $fh "Seems like that did not work!<p>";
250: $errors++;
251: }
252: }
253:
254: $fname="$perlvar{'lonDaemons'}/logs/lond.log";
255:
256: my ($dev,$ino,$mode,$nlink,
257: $uid,$gid,$rdev,$size,
258: $atime,$mtime,$ctime,
259: $blksize,$blocks)=stat($fname);
260:
261: if ($size>40000) {
262: print $fh "Rotating logs ...<p>";
263: rename("$fname.2","$fname.3");
264: rename("$fname.1","$fname.2");
265: rename("$fname","$fname.1");
266: }
267:
268: &errout($fh);
269: # ------------------------------------------------------------------------ lonc
270:
271: print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>';
272:
273: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
274: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonc.log|");
275: while ($line=<DFH>) {
276: print $fh "$line";
277: if ($line=~/died/) { $notices++; }
278: };
279: close (DFH);
280: }
281: print $fh "</pre>";
282:
283: my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
284:
285: if (-e $loncfile) {
286: my $lfh=IO::File->new("$loncfile");
287: my $loncpid=<$lfh>;
288: chomp($loncpid);
289: if (kill 0 => $loncpid) {
290: print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";
291: kill USR1 => $loncpid;
292: } else {
293: $errors++; $errors++;
294: print $fh "<h3>lonc at pid $loncpid not responding</h3>";
295: }
296: } else {
297: $errors++;
298: print $fh "<h3>lonc not running, trying to start</h3>";
299: system("$perlvar{'lonDaemons'}/lonc");
300: sleep 120;
301: if (-e $loncfile) {
302: print $fh "Seems like it started ...<p>";
303: my $lfh=IO::File->new("$loncfile");
304: my $loncpid=<$lfh>;
305: chomp($loncpid);
306: sleep 30;
307: if (kill 0 => $loncpid) {
308: print $fh "<h3>lonc at pid $loncpid responding</h3>";
309: } else {
310: $errors++; $errors++;
311: print $fh "<h3>lonc at pid $loncpid not responding</h3>";
312: print $fh "Give it one more try ...<p>";
313: system("$perlvar{'lonDaemons'}/lonc");
314: sleep 120;
315: }
316: } else {
317: print $fh "Seems like that did not work!<p>";
318: $errors++;
319: }
320: }
321:
322: $fname="$perlvar{'lonDaemons'}/logs/lonc.log";
323:
324: my ($dev,$ino,$mode,$nlink,
325: $uid,$gid,$rdev,$size,
326: $atime,$mtime,$ctime,
327: $blksize,$blocks)=stat($fname);
328:
329: if ($size>40000) {
330: print $fh "Rotating logs ...<p>";
331: rename("$fname.2","$fname.3");
332: rename("$fname.1","$fname.2");
333: rename("$fname","$fname.1");
334: }
335:
336:
337: &errout($fh);
338: # ---------------------------------------------------------------------- lonnet
339:
340: print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>';
341: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
342: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
343: while ($line=<DFH>) {
344: print $fh "$line";
345: if ($line=~/Delayed/) { $warnings++; }
346: if ($line=~/giving up/) { $warnings++; }
347: if ($line=~/FAILED/) { $errors++; }
348: };
349: close (DFH);
350: }
351: print $fh "</pre><h3>Perm Log</h3>";
352:
353: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
354: open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
355: while ($line=<DFH>) {
356: print $fh "$line";
357: };
358: close (DFH);
359: } else { print $fh "No perm log\n" }
360:
361: $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
362:
363: my ($dev,$ino,$mode,$nlink,
364: $uid,$gid,$rdev,$size,
365: $atime,$mtime,$ctime,
366: $blksize,$blocks)=stat($fname);
367:
368: if ($size>40000) {
369: print $fh "Rotating logs ...<p>";
370: rename("$fname.2","$fname.3");
371: rename("$fname.1","$fname.2");
372: rename("$fname","$fname.1");
373: }
374:
375: print $fh "</pre>";
376: &errout($fh);
377: # ----------------------------------------------------------------- Connections
378:
379: print $fh '<hr><a name="connections"><h2>Connections</h2>';
380:
381: print $fh "<table border=2>";
382: foreach $tryserver (keys %hostname) {
383:
384: $answer=reply("pong",$tryserver);
385: if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
386: $result="<b>ok</b>";
387: } else {
388: $result=$answer;
389: $warnings++;
390: if ($answer eq 'con_lost') { $warnings++; }
391: }
392: print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n";
393:
394: }
395: print $fh "</table>";
396:
397: &errout($fh);
398: # ------------------------------------------------------------ Delayed messages
399:
400: print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
401:
402: print $fh '<h3>Scanning Permanent Log</h3>';
403:
404: $unsend=0;
405: {
406: my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
407: while ($line=<$dfh>) {
408: ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
409: if ($sdf eq 'F') {
410: $local=localtime($time);
411: print "<b>Failed: $time, $dserv, $dcmd</b><br>";
412: $warnings++;
413: }
414: if ($sdf eq 'S') { $unsend--; }
415: if ($sdf eq 'D') { $unsend++; }
416: }
417: }
418: print $fh "Total unsend messages: <b>$unsend</b><p>\n";
419: $warnings=$warnings+5*$unsend;
420:
421: print $fh "<h3>Outgoing Buffer</h3>";
422:
423: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
424: while ($line=<DFH>) {
425: print $fh "$line<br>";
426: };
427: close (DFH);
428:
429: # ------------------------------------------------------------------------- End
430: print $fh "<a name=errcount>\n";
431: $totalcount=$notices+4*$warnings+100*$errors;
432: &errout($fh);
433: print $fh "<h1>Total Error Count: $totalcount</h1>";
434: $now=time;
435: $date=localtime($now);
436: print $fh "<hr>$date ($now)</body></html>\n";
437:
438: }
439:
440: rename ("$statusdir/newstatus.html","$statusdir/index.html");
441:
442: if ($totalcount>200) {
443: $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
444: $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";
445: system(
446: "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");
447: }
448: 1;
449:
450:
451:
452:
453:
454:
455:
456:
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