Annotation of loncom/loncron, revision 1.4
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);
203: # ------------------------------------------------------------------------ lond
204:
205: print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>';
206:
207: if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
1.3 www 208: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
1.1 albertel 209: while ($line=<DFH>) {
210: print $fh "$line";
1.3 www 211: if ($line=~/INFO/) { $notices++; }
1.4 ! www 212: if ($line=~/WARNING/) { $notices++; }
! 213: if ($line=~/CRITICAL/) { $warnings++; }
1.1 albertel 214: };
215: close (DFH);
216: }
217: print $fh "</pre>";
218:
219: my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
220:
221: if (-e $londfile) {
222: my $lfh=IO::File->new("$londfile");
223: my $londpid=<$lfh>;
224: chomp($londpid);
225: if (kill 0 => $londpid) {
226: print $fh "<h3>lond at pid $londpid responding</h3>";
227: } else {
228: $errors++; $errors++;
229: print $fh "<h3>lond at pid $londpid not responding</h3>";
230: }
231: } else {
232: $errors++;
233: print $fh "<h3>lond not running, trying to start</h3>";
234: system("$perlvar{'lonDaemons'}/lond");
235: sleep 120;
236: if (-e $londfile) {
237: print $fh "Seems like it started ...<p>";
238: my $lfh=IO::File->new("$londfile");
239: my $londpid=<$lfh>;
240: chomp($londpid);
241: sleep 30;
242: if (kill 0 => $londpid) {
243: print $fh "<h3>lond at pid $londpid responding</h3>";
244: } else {
245: $errors++; $errors++;
246: print $fh "<h3>lond at pid $londpid not responding</h3>";
247: print $fh "Give it one more try ...<p>";
248: system("$perlvar{'lonDaemons'}/lond");
249: sleep 120;
250: }
251: } else {
252: print $fh "Seems like that did not work!<p>";
253: $errors++;
254: }
1.3 www 255: if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
256: print $fh "<p><pre>";
257: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
258: while ($line=<DFH>) {
259: print $fh "$line";
1.4 ! www 260: if ($line=~/WARNING/) { $notices++; }
! 261: if ($line=~/CRITICAL/) { $notices++; }
1.3 www 262: };
263: close (DFH);
264: print $fh "</pre>";
265: }
1.1 albertel 266: }
267:
268: $fname="$perlvar{'lonDaemons'}/logs/lond.log";
269:
270: my ($dev,$ino,$mode,$nlink,
271: $uid,$gid,$rdev,$size,
272: $atime,$mtime,$ctime,
273: $blksize,$blocks)=stat($fname);
274:
275: if ($size>40000) {
276: print $fh "Rotating logs ...<p>";
277: rename("$fname.2","$fname.3");
278: rename("$fname.1","$fname.2");
279: rename("$fname","$fname.1");
280: }
281:
282: &errout($fh);
283: # ------------------------------------------------------------------------ lonc
284:
285: print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>';
286:
287: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
1.3 www 288: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
1.1 albertel 289: while ($line=<DFH>) {
290: print $fh "$line";
1.3 www 291: if ($line=~/INFO/) { $notices++; }
1.4 ! www 292: if ($line=~/WARNING/) { $notices++; }
! 293: if ($line=~/CRITICAL/) { $warnings++; }
1.1 albertel 294: };
295: close (DFH);
296: }
297: print $fh "</pre>";
298:
299: my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
300:
301: if (-e $loncfile) {
302: my $lfh=IO::File->new("$loncfile");
303: my $loncpid=<$lfh>;
304: chomp($loncpid);
305: if (kill 0 => $loncpid) {
306: print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";
307: kill USR1 => $loncpid;
308: } else {
309: $errors++; $errors++;
310: print $fh "<h3>lonc at pid $loncpid not responding</h3>";
311: }
312: } else {
313: $errors++;
314: print $fh "<h3>lonc not running, trying to start</h3>";
315: system("$perlvar{'lonDaemons'}/lonc");
316: sleep 120;
317: if (-e $loncfile) {
318: print $fh "Seems like it started ...<p>";
319: my $lfh=IO::File->new("$loncfile");
320: my $loncpid=<$lfh>;
321: chomp($loncpid);
322: sleep 30;
323: if (kill 0 => $loncpid) {
324: print $fh "<h3>lonc at pid $loncpid responding</h3>";
325: } else {
326: $errors++; $errors++;
327: print $fh "<h3>lonc at pid $loncpid not responding</h3>";
328: print $fh "Give it one more try ...<p>";
329: system("$perlvar{'lonDaemons'}/lonc");
330: sleep 120;
331: }
332: } else {
333: print $fh "Seems like that did not work!<p>";
334: $errors++;
335: }
1.3 www 336: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") {
337: print $fh "<p><pre>";
338: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
339: while ($line=<DFH>) {
340: print $fh "$line";
1.4 ! www 341: if ($line=~/WARNING/) { $notices++; }
! 342: if ($line=~/CRITICAL/) { $notices++; }
1.3 www 343: };
344: close (DFH);
345: print $fh "</pre>";
346: }
1.1 albertel 347: }
348:
349: $fname="$perlvar{'lonDaemons'}/logs/lonc.log";
350:
351: my ($dev,$ino,$mode,$nlink,
352: $uid,$gid,$rdev,$size,
353: $atime,$mtime,$ctime,
354: $blksize,$blocks)=stat($fname);
355:
356: if ($size>40000) {
357: print $fh "Rotating logs ...<p>";
358: rename("$fname.2","$fname.3");
359: rename("$fname.1","$fname.2");
360: rename("$fname","$fname.1");
361: }
362:
363:
364: &errout($fh);
365: # ---------------------------------------------------------------------- lonnet
366:
367: print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>';
368: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
369: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
370: while ($line=<DFH>) {
371: print $fh "$line";
372: };
373: close (DFH);
374: }
375: print $fh "</pre><h3>Perm Log</h3>";
376:
377: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
378: open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
379: while ($line=<DFH>) {
380: print $fh "$line";
381: };
382: close (DFH);
383: } else { print $fh "No perm log\n" }
384:
385: $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
386:
387: my ($dev,$ino,$mode,$nlink,
388: $uid,$gid,$rdev,$size,
389: $atime,$mtime,$ctime,
390: $blksize,$blocks)=stat($fname);
391:
392: if ($size>40000) {
393: print $fh "Rotating logs ...<p>";
394: rename("$fname.2","$fname.3");
395: rename("$fname.1","$fname.2");
396: rename("$fname","$fname.1");
397: }
398:
399: print $fh "</pre>";
400: &errout($fh);
401: # ----------------------------------------------------------------- Connections
402:
403: print $fh '<hr><a name="connections"><h2>Connections</h2>';
404:
405: print $fh "<table border=2>";
406: foreach $tryserver (keys %hostname) {
407:
408: $answer=reply("pong",$tryserver);
409: if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
410: $result="<b>ok</b>";
411: } else {
412: $result=$answer;
413: $warnings++;
414: if ($answer eq 'con_lost') { $warnings++; }
415: }
416: print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n";
417:
418: }
419: print $fh "</table>";
420:
421: &errout($fh);
422: # ------------------------------------------------------------ Delayed messages
423:
424: print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
425:
426: print $fh '<h3>Scanning Permanent Log</h3>';
427:
428: $unsend=0;
429: {
430: my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
431: while ($line=<$dfh>) {
432: ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
433: if ($sdf eq 'F') {
434: $local=localtime($time);
435: print "<b>Failed: $time, $dserv, $dcmd</b><br>";
436: $warnings++;
437: }
438: if ($sdf eq 'S') { $unsend--; }
439: if ($sdf eq 'D') { $unsend++; }
440: }
441: }
442: print $fh "Total unsend messages: <b>$unsend</b><p>\n";
443: $warnings=$warnings+5*$unsend;
444:
445: print $fh "<h3>Outgoing Buffer</h3>";
446:
447: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
448: while ($line=<DFH>) {
449: print $fh "$line<br>";
450: };
451: close (DFH);
452:
453: # ------------------------------------------------------------------------- End
454: print $fh "<a name=errcount>\n";
455: $totalcount=$notices+4*$warnings+100*$errors;
456: &errout($fh);
457: print $fh "<h1>Total Error Count: $totalcount</h1>";
458: $now=time;
459: $date=localtime($now);
460: print $fh "<hr>$date ($now)</body></html>\n";
461:
462: }
463:
464: rename ("$statusdir/newstatus.html","$statusdir/index.html");
465:
466: if ($totalcount>200) {
467: $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
468: $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";
469: system(
470: "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");
471: }
472: 1;
473:
474:
475:
476:
477:
478:
479:
480:
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>