loncom/loncron - view

File: [LON-CAPA] / loncom / loncron
Revision 1.3: download - view: text, annotated - select for diffs
Tue Feb 8 17:34:24 2000 UTC (24 years, 4 months ago) by www
Branches: MAIN
CVS tags: HEAD

Put in additional sleep so that children do not die before reaper in place
(lonc)
Marked up logs
Put in report for after the lonc/lond start into loncron

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]>3.0) { 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: $errors++; 175: } elsif ($usage>80) { 176: $warnings++; 177: } elsif ($usage>60) { 178: $notices++; 179: } 180: if ($usage>95) { $errors++; } 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"){ 208: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|"); 209: while ($line=<DFH>) { 210: print $fh "$line"; 211: if ($line=~/INFO/) { $notices++; } 212: if ($line=~/WARNING/) { $warnings++; } 213: if ($line=~/CRITICAL/) { $errors++; } 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: $errors++; 254: } 255: if (-e "$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"; 260: if ($line=~/INFO/) { $notices++; } 261: if ($line=~/WARNING/) { $warnings++; } 262: if ($line=~/CRITICAL/) { $errors++; } 263: }; 264: close (DFH); 265: print $fh "</pre>"; 266: } 267: } 268: 269: $fname="$perlvar{'lonDaemons'}/logs/lond.log"; 270: 271: my ($dev,$ino,$mode,$nlink, 272: $uid,$gid,$rdev,$size, 273: $atime,$mtime,$ctime, 274: $blksize,$blocks)=stat($fname); 275: 276: if ($size>40000) { 277: print $fh "Rotating logs ...<p>"; 278: rename("$fname.2","$fname.3"); 279: rename("$fname.1","$fname.2"); 280: rename("$fname","$fname.1"); 281: } 282: 283: &errout($fh); 284: # ------------------------------------------------------------------------ lonc 285: 286: print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>'; 287: 288: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){ 289: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|"); 290: while ($line=<DFH>) { 291: print $fh "$line"; 292: if ($line=~/INFO/) { $notices++; } 293: if ($line=~/WARNING/) { $warnings++; } 294: if ($line=~/CRITICAL/) { $errors++; } 295: }; 296: close (DFH); 297: } 298: print $fh "</pre>"; 299: 300: my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid"; 301: 302: if (-e $loncfile) { 303: my $lfh=IO::File->new("$loncfile"); 304: my $loncpid=<$lfh>; 305: chomp($loncpid); 306: if (kill 0 => $loncpid) { 307: print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>"; 308: kill USR1 => $loncpid; 309: } else { 310: $errors++; $errors++; 311: print $fh "<h3>lonc at pid $loncpid not responding</h3>"; 312: } 313: } else { 314: $errors++; 315: print $fh "<h3>lonc not running, trying to start</h3>"; 316: system("$perlvar{'lonDaemons'}/lonc"); 317: sleep 120; 318: if (-e $loncfile) { 319: print $fh "Seems like it started ...<p>"; 320: my $lfh=IO::File->new("$loncfile"); 321: my $loncpid=<$lfh>; 322: chomp($loncpid); 323: sleep 30; 324: if (kill 0 => $loncpid) { 325: print $fh "<h3>lonc at pid $loncpid responding</h3>"; 326: } else { 327: $errors++; $errors++; 328: print $fh "<h3>lonc at pid $loncpid not responding</h3>"; 329: print $fh "Give it one more try ...<p>"; 330: system("$perlvar{'lonDaemons'}/lonc"); 331: sleep 120; 332: } 333: } else { 334: print $fh "Seems like that did not work!<p>"; 335: $errors++; 336: } 337: if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") { 338: print $fh "<p><pre>"; 339: open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|"); 340: while ($line=<DFH>) { 341: print $fh "$line"; 342: if ($line=~/INFO/) { $notices++; } 343: if ($line=~/WARNING/) { $warnings++; } 344: if ($line=~/CRITICAL/) { $errors++; } 345: }; 346: close (DFH); 347: print $fh "</pre>"; 348: } 349: } 350: 351: $fname="$perlvar{'lonDaemons'}/logs/lonc.log"; 352: 353: my ($dev,$ino,$mode,$nlink, 354: $uid,$gid,$rdev,$size, 355: $atime,$mtime,$ctime, 356: $blksize,$blocks)=stat($fname); 357: 358: if ($size>40000) { 359: print $fh "Rotating logs ...<p>"; 360: rename("$fname.2","$fname.3"); 361: rename("$fname.1","$fname.2"); 362: rename("$fname","$fname.1"); 363: } 364: 365: 366: &errout($fh); 367: # ---------------------------------------------------------------------- lonnet 368: 369: print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>'; 370: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){ 371: open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|"); 372: while ($line=<DFH>) { 373: print $fh "$line"; 374: }; 375: close (DFH); 376: } 377: print $fh "</pre><h3>Perm Log</h3>"; 378: 379: if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") { 380: open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|"); 381: while ($line=<DFH>) { 382: print $fh "$line"; 383: }; 384: close (DFH); 385: } else { print $fh "No perm log\n" } 386: 387: $fname="$perlvar{'lonDaemons'}/logs/lonnet.log"; 388: 389: my ($dev,$ino,$mode,$nlink, 390: $uid,$gid,$rdev,$size, 391: $atime,$mtime,$ctime, 392: $blksize,$blocks)=stat($fname); 393: 394: if ($size>40000) { 395: print $fh "Rotating logs ...<p>"; 396: rename("$fname.2","$fname.3"); 397: rename("$fname.1","$fname.2"); 398: rename("$fname","$fname.1"); 399: } 400: 401: print $fh "</pre>"; 402: &errout($fh); 403: # ----------------------------------------------------------------- Connections 404: 405: print $fh '<hr><a name="connections"><h2>Connections</h2>'; 406: 407: print $fh "<table border=2>"; 408: foreach $tryserver (keys %hostname) { 409: 410: $answer=reply("pong",$tryserver); 411: if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { 412: $result="<b>ok</b>"; 413: } else { 414: $result=$answer; 415: $warnings++; 416: if ($answer eq 'con_lost') { $warnings++; } 417: } 418: print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n"; 419: 420: } 421: print $fh "</table>"; 422: 423: &errout($fh); 424: # ------------------------------------------------------------ Delayed messages 425: 426: print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>'; 427: 428: print $fh '<h3>Scanning Permanent Log</h3>'; 429: 430: $unsend=0; 431: { 432: my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log"); 433: while ($line=<$dfh>) { 434: ($time,$sdf,$dserv,$dcmd)=split(/:/,$line); 435: if ($sdf eq 'F') { 436: $local=localtime($time); 437: print "<b>Failed: $time, $dserv, $dcmd</b><br>"; 438: $warnings++; 439: } 440: if ($sdf eq 'S') { $unsend--; } 441: if ($sdf eq 'D') { $unsend++; } 442: } 443: } 444: print $fh "Total unsend messages: <b>$unsend</b><p>\n"; 445: $warnings=$warnings+5*$unsend; 446: 447: print $fh "<h3>Outgoing Buffer</h3>"; 448: 449: open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|"); 450: while ($line=<DFH>) { 451: print $fh "$line<br>"; 452: }; 453: close (DFH); 454: 455: # ------------------------------------------------------------------------- End 456: print $fh "<a name=errcount>\n"; 457: $totalcount=$notices+4*$warnings+100*$errors; 458: &errout($fh); 459: print $fh "<h1>Total Error Count: $totalcount</h1>"; 460: $now=time; 461: $date=localtime($now); 462: print $fh "<hr>$date ($now)</body></html>\n"; 463: 464: } 465: 466: rename ("$statusdir/newstatus.html","$statusdir/index.html"); 467: 468: if ($totalcount>200) { 469: $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; 470: $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; 471: system( 472: "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html"); 473: } 474: 1; 475: 476: 477: 478: 479: 480: 481: 482:

Internal Server Error

The server encountered an internal error or misconfiguration and was unable to complete your request.

Please contact the server administrator at root@localhost to inform them of the time this error occurred, and the actions you performed just before this error.