[BACK]Return to loncron CVS log [TXT] [DIR] Up to [LON-CAPA] / loncom

Diff for /loncom/loncron between versions 1.48 and 1.72

version 1.48, 2004/05/11 20:19:46 version 1.72, 2007/04/11 23:08:43
Line 32  use strict; Line 32  use strict;
   
 use lib '/home/httpd/lib/perl/';  use lib '/home/httpd/lib/perl/';
 use LONCAPA::Configuration;  use LONCAPA::Configuration;
   use Apache::lonnet;
   
 use IO::File;  use IO::File;
 use IO::Socket;  use IO::Socket;
 use HTML::Entities;  use HTML::Entities;
   use Getopt::Long;
 #globals  #globals
 use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount);  use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount);
   
 my $statusdir="/home/httpd/html/lon-status";  my $statusdir="/home/httpd/html/lon-status";
   
   
 # -------------------------------------------------- Non-critical communication  
 sub reply {  
     my ($cmd,$server)=@_;  
     my $peerfile="$perlvar{'lonSockDir'}/$server";  
     my $client=IO::Socket::UNIX->new(Peer    =>"$peerfile",  
                                      Type    => SOCK_STREAM,  
                                      Timeout => 10)  
        or return "con_lost";  
     print $client "$cmd\n";  
     my $answer=<$client>;  
     chomp($answer);  
     if (!$answer) { $answer="con_lost"; }  
     return $answer;  
 }  
   
 # --------------------------------------------------------- Output error status  # --------------------------------------------------------- Output error status
   
 sub log {  sub log {
Line 76  ENDERROUT Line 63  ENDERROUT
 }  }
   
 sub start_daemon {  sub start_daemon {
     my ($fh,$daemon,$pidfile) = @_;      my ($fh,$daemon,$pidfile,$args) = @_;
     my $progname=$daemon;      my $progname=$daemon;
     if ($daemon eq 'lonc' && $ARGV[0] eq 'new') {      if ($daemon eq 'lonc') {
  $progname='loncnew';    $progname='loncnew'; 
  print "new ";  
     }      }
     system("$perlvar{'lonDaemons'}/$progname 2>>$perlvar{'lonDaemons'}/logs/${daemon}_errors");      my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors";
     sleep 2;      my $size=(stat($error_fname))[7];
       if ($size>40000) {
    &log($fh,"<p>Rotating error logs ...</p>");
    rename("$error_fname.2","$error_fname.3");
    rename("$error_fname.1","$error_fname.2");
    rename("$error_fname","$error_fname.1");
       }
       system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors");
       sleep 1;
     if (-e $pidfile) {      if (-e $pidfile) {
  &log($fh,"<p>Seems like it started ...</p>");   &log($fh,"<p>Seems like it started ...</p>");
  my $lfh=IO::File->new("$pidfile");   my $lfh=IO::File->new("$pidfile");
  my $daemonpid=<$lfh>;   my $daemonpid=<$lfh>;
  chomp($daemonpid);   chomp($daemonpid);
  sleep 2;   if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
  if (kill 0 => $daemonpid) {  
     return 1;      return 1;
  } else {   } else {
     return 0;      return 0;
Line 102  sub start_daemon { Line 95  sub start_daemon {
 }  }
   
 sub checkon_daemon {  sub checkon_daemon {
     my ($fh,$daemon,$maxsize,$sendusr1)=@_;      my ($fh,$daemon,$maxsize,$send,$args)=@_;
   
       my $result;
     &log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>');      &log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>');
     printf("%-10s ",$daemon);      printf("%-15s ",$daemon);
     if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){      if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
  open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");   open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
  while (my $line=<DFH>) {    while (my $line=<DFH>) { 
Line 126  sub checkon_daemon { Line 120  sub checkon_daemon {
  my $lfh=IO::File->new("$pidfile");   my $lfh=IO::File->new("$pidfile");
  $daemonpid=<$lfh>;   $daemonpid=<$lfh>;
  chomp($daemonpid);   chomp($daemonpid);
  if (kill 0 => $daemonpid) {   if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
     &log($fh,"<h3>$daemon at pid $daemonpid responding");      &log($fh,"<h3>$daemon at pid $daemonpid responding");
     if ($sendusr1) { &log($fh,", sending USR1"); }      if ($send) { &log($fh,", sending $send"); }
     &log($fh,"</h3>");      &log($fh,"</h3>");
     if ($sendusr1) { kill USR1 => $daemonpid; }      if ($send eq 'USR1') { kill USR1 => $daemonpid; }
       if ($send eq 'USR2') { kill USR2 => $daemonpid; }
     $restartflag=0;      $restartflag=0;
     print "running\n";      if ($send eq 'USR2') {
    $result = 'reloaded';
    print "reloaded\n";
       } else {
    $result = 'running';
    print "running\n";
       }
  } else {   } else {
     $errors++;      $errors++;
     &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");      &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
Line 143  sub checkon_daemon { Line 144  sub checkon_daemon {
     if ($restartflag==1) {      if ($restartflag==1) {
  $simplestatus{$daemon}='off';   $simplestatus{$daemon}='off';
  $errors++;   $errors++;
    my $kadaemon=$daemon;
    if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; }
  &log($fh,'<br><font color="red">Killall '.$daemon.': '.   &log($fh,'<br><font color="red">Killall '.$daemon.': '.
     `killall $daemon 2>&1`.' - ');      `killall $kadaemon 2>&1`.' - ');
  sleep 2;   sleep 1;
  &log($fh,unlink($pidfile).' - '.   &log($fh,unlink($pidfile).' - '.
     `killall -9 $daemon 2>&1`.      `killall -9 $kadaemon 2>&1`.
     '</font><br>');      '</font><br>');
  &log($fh,"<h3>$daemon not running, trying to start</h3>");   &log($fh,"<h3>$daemon not running, trying to start</h3>");
   
  if (&start_daemon($fh,$daemon,$pidfile)) {   if (&start_daemon($fh,$daemon,$pidfile,$args)) {
     &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");      &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
     $simplestatus{$daemon}='restarted';      $simplestatus{$daemon}='restarted';
       $result = 'started';
     print "started\n";      print "started\n";
  } else {   } else {
     $errors++;      $errors++;
     &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");      &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
     &log($fh,"<p>Give it one more try ...</p>");      &log($fh,"<p>Give it one more try ...</p>");
     print " ";      print " ";
     if (&start_daemon($fh,$daemon,$pidfile)) {      if (&start_daemon($fh,$daemon,$pidfile,$args)) {
  &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");   &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
  $simplestatus{$daemon}='restarted';   $simplestatus{$daemon}='restarted';
    $result = 'started';
  print "started\n";   print "started\n";
     } else {      } else {
    $result = 'failed';
  print " failed\n";   print " failed\n";
  $simplestatus{$daemon}='failed';   $simplestatus{$daemon}='failed';
  $errors++; $errors++;   $errors++; $errors++;
Line 201  sub checkon_daemon { Line 207  sub checkon_daemon {
     }      }
   
     &errout($fh);      &errout($fh);
       return $result;
 }  }
   
 # --------------------------------------------------------------------- Machine  # --------------------------------------------------------------------- Machine
Line 251  sub log_machine_info { Line 258  sub log_machine_info {
     &log($fh,"<pre>");      &log($fh,"<pre>");
     my $psproc=0;      my $psproc=0;
   
     open (PSH,"ps -aux --cols 140 |");      open (PSH,"ps aux --cols 140 |");
     while (my $line=<PSH>) {       while (my $line=<PSH>) { 
  &log($fh,&encode_entities($line,'<>&"'));    &log($fh,&encode_entities($line,'<>&"')); 
  $psproc++;   $psproc++;
Line 262  sub log_machine_info { Line 269  sub log_machine_info {
     if ($psproc>200) { $notices++; }      if ($psproc>200) { $notices++; }
     if ($psproc>250) { $notices++; }      if ($psproc>250) { $notices++; }
   
       &log($fh,"<h3>distprobe</h3>");
       &log($fh,"<pre>");
       open(DSH,"$perlvar{'lonDaemons'}/distprobe |");
       while (my $line=<DSH>) { 
    &log($fh,&encode_entities($line,'<>&"')); 
    $psproc++;
       }
       close(DSH);
       &log($fh,"</pre>");
   
     &errout($fh);      &errout($fh);
 }  }
   
 sub start_logging {  sub start_logging {
     my ($hostdom,$hostrole,$hostname,$spareid)=@_;  
     my $fh=IO::File->new(">$statusdir/newstatus.html");      my $fh=IO::File->new(">$statusdir/newstatus.html");
     my %simplestatus=();      my %simplestatus=();
     my $now=time;      my $now=time;
Line 309  ENDHEADERS Line 325  ENDHEADERS
      &encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n");       &encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n");
     }      }
     &log($fh,"</table><h3>Hosts</h3><table border='2'>");      &log($fh,"</table><h3>Hosts</h3><table border='2'>");
     foreach my $id (sort(keys(%{$hostname}))) {      my %hostname = &Apache::lonnet::all_hostnames();
       foreach my $id (sort(keys(%hostname))) {
    my $role = (&Apache::lonnet::is_library($id) ? 'library'
                                        : 'access');
  &log($fh,   &log($fh,
     "<tr><td>$id</td><td>".$hostdom->{$id}.      "<tr><td>$id</td><td>".&Apache::lonnet::host_domain($id).
     "</td><td>".$hostrole->{$id}.      "</td><td>".$role.
     "</td><td>".$hostname->{$id}."</td></tr>\n");      "</td><td>".&Apache::lonnet::hostname($id)."</td></tr>\n");
     }      }
     &log($fh,"</table><h3>Spare Hosts</h3><ol>");      &log($fh,"</table><h3>Spare Hosts</h3><ul>");
     foreach my $id (sort(keys(%{$spareid}))) {      foreach my $type (sort(keys(%Apache::lonnet::spareid))) {
  &log($fh,"<li>$id\n</li>");   &log($fh,"<li>$type\n<ol>");
    foreach my $id (@{ $Apache::lonnet::spareid{$type} }) {
       &log($fh,"<li>$id</li>\n");
    }
    &log($fh,"</ol>\n</li>\n");
     }      }
     &log($fh,"</ol>\n");      &log($fh,"</ul>\n");
     return $fh;      return $fh;
 }  }
   
Line 446  sub rotate_lonnet_logs { Line 469  sub rotate_lonnet_logs {
   
 # ----------------------------------------------------------------- Connections  # ----------------------------------------------------------------- Connections
 sub test_connections {  sub test_connections {
     my ($fh,$hostname)=@_;      my ($fh)=@_;
     &log($fh,'<hr /><a name="connections" /><h2>Connections</h2>');      &log($fh,'<hr /><a name="connections" /><h2>Connections</h2>');
     print "testing connections\n";      print "testing connections\n";
     &log($fh,"<table border='2'>");      &log($fh,"<table border='2'>");
     foreach my $tryserver (sort(keys(%{$hostname}))) {      my ($good,$bad)=(0,0);
       my %hostname = &Apache::lonnet::all_hostnames();
       foreach my $tryserver (sort(keys(%hostname))) {
  print(".");   print(".");
  my $result;   my $result;
  my $answer=reply("pong",$tryserver);   my $answer=&Apache::lonnet::reply("ping",$tryserver);
  if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {   if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
     $result="<b>ok</b>";      $result="<b>ok</b>";
       $good++;
  } else {   } else {
     $result=$answer;      $result=$answer;
     $warnings++;      $warnings++;
     if ($answer eq 'con_lost') { $warnings++; }      if ($answer eq 'con_lost') {
    $bad++;
    $warnings++;
       } else {
    $good++; #self connection
       }
  }   }
  if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }   if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }
  &log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n");   &log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n");
     }      }
     &log($fh,"</table>");      &log($fh,"</table>");
       print "\n$good good, $bad bad connections\n";
     &errout($fh);      &errout($fh);
 }  }
   
Line 497  sub check_delayed_msg { Line 528  sub check_delayed_msg {
   
     if ($unsend) { $simplestatus{'unsend'}=$unsend; }      if ($unsend) { $simplestatus{'unsend'}=$unsend; }
     &log($fh,"<h3>Outgoing Buffer</h3>\n<pre>");      &log($fh,"<h3>Outgoing Buffer</h3>\n<pre>");
   # list directory with delayed messages and remember offline servers
       my %servers=();
     open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");      open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
     while (my $line=<DFH>) {       while (my $line=<DFH>) {
           my ($server)=($line=~/\.(\w+)$/);
           if ($server) { $servers{$server}=1; }
  &log($fh,&encode_entities($line,'<>&"'));   &log($fh,&encode_entities($line,'<>&"'));
     }      }
     &log($fh,"</pre>\n");      &log($fh,"</pre>\n");
     close (DFH);      close (DFH);
   # pong to all servers that have delayed messages
   # this will trigger a reverse connection, which should flush the buffers
       foreach my $tryserver (keys %servers) {
    my $answer=&Apache::lonnet::reply("pong",$tryserver);
    &log($fh,"Pong to $tryserver: $answer<br />");
       }
 }  }
   
 sub finish_logging {  sub finish_logging {
Line 538  sub log_simplestatus { Line 578  sub log_simplestatus {
 sub send_mail {  sub send_mail {
     print "sending mail\n";      print "sending mail\n";
     my $emailto="$perlvar{'lonAdmEMail'}";      my $emailto="$perlvar{'lonAdmEMail'}";
     if ($totalcount>1000) {      if ($totalcount>2500) {
  $emailto.=",$perlvar{'lonSysEMail'}";   $emailto.=",$perlvar{'lonSysEMail'}";
     }      }
     my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";       my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; 
     system("metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");  
       my $result=system("metasend -b -S 4000000 -t $emailto -s '$subj' -f $statusdir/index.html -m text/html >& /dev/null");
       if ($result != 0) {
    $result=system("mail -s '$subj' $emailto < $statusdir/index.html");
       }
   }
   
   sub usage {
       print(<<USAGE);
   loncron - housekeeping program that checks up on various parts of Lon-CAPA
   
   Options:
      --help     Display 
      --noemail  Do not send the status email
      --justcheckconnections  Only check the current status of the lonc/d
                                   connections, do not send emails do not
                                   check if the daemons are running, do not
                                   generate lon-status
      --justcheckdaemons      Only check that all of the Lon-CAPA daemons are
                                   running, do not send emails do not
                                   check the lonc/d connections, do not
                                   generate lon-status
      --justreload            Only tell the daemons to reload the config files,
    do not send emails do not
                                   check if the daemons are running, do not
                                   generate lon-status
                              
   USAGE
 }  }
   
 # ================================================================ Main Program  # ================================================================ Main Program
 sub main () {  sub main () {
       my ($help,$justcheckdaemons,$noemail,$justcheckconnections,
    $justreload);
       &GetOptions("help"                 => \$help,
    "justcheckdaemons"     => \$justcheckdaemons,
    "noemail"              => \$noemail,
    "justcheckconnections" => \$justcheckconnections,
    "justreload"           => \$justreload
    );
       if ($help) { &usage(); return; }
 # --------------------------------- Read loncapa_apache.conf and loncapa.conf  # --------------------------------- Read loncapa_apache.conf and loncapa.conf
     my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');      my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
     %perlvar=%{$perlvarref};      %perlvar=%{$perlvarref};
Line 579  sub main () { Line 655  sub main () {
  exit 1;   exit 1;
     }      }
   
 # ------------------------------------------------------------- Read hosts file  # -------------------------------------------- Force reload of host information
     my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");      &Apache::lonnet::load_hosts_tab(1);
           &Apache::lonnet::load_domain_tab(1);
     my (%hostname,%hostdom,%hostrole,%spareid);      &Apache::lonnet::get_iphost(1);
     while (my $configline=<$config>) {  
  next if ($configline =~ /^(\#|\s*$)/);  
  my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline);  
  if ($id && $domain && $role && $name && $ip) {  
     $hostname{$id}=$name;  
     $hostdom{$id}=$domain;  
     $hostrole{$id}=$role;  
  }  
     }  
     undef $config;  
   
 # ------------------------------------------------------ Read spare server file  
     $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");  
       
     while (my $configline=<$config>) {  
  chomp($configline);  
  if (($configline) && ($configline ne $perlvar{'lonHostID'})) {  
     $spareid{$configline}=1;  
  }  
     }  
     undef $config;  
   
 # ---------------------------------------------------------------- Start report  # ---------------------------------------------------------------- Start report
   
Line 611  sub main () { Line 666  sub main () {
     $warnings=0;      $warnings=0;
     $notices=0;      $notices=0;
   
     my $fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);  
   
     &log_machine_info($fh);  
     &clean_tmp($fh);  
     &clean_lonIDs($fh);  
     &check_httpd_logs($fh);  
     &rotate_lonnet_logs($fh);  
     &checkon_daemon($fh,'lonsql',200000);  
     &checkon_daemon($fh,'lond',40000,1);  
     &checkon_daemon($fh,'lonc',40000,1);  
     &checkon_daemon($fh,'lonhttpd',40000);  
       
     &test_connections($fh,\%hostname);  
     &check_delayed_msg($fh);  
   
     &finish_logging($fh);  
     &log_simplestatus();  
   
     if ($totalcount>200) { &send_mail(); }      my $fh;
       if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
    $fh=&start_logging();
   
    &log_machine_info($fh);
    &clean_tmp($fh);
    &clean_lonIDs($fh);
    &check_httpd_logs($fh);
    &rotate_lonnet_logs($fh);
       }
       if (!$justcheckconnections && !$justreload) {
    &checkon_daemon($fh,'lonsql',200000);
    if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') {
       &checkon_daemon($fh,'lond',40000,'USR2');
    }
    &checkon_daemon($fh,'lonc',40000,'USR1');
    &checkon_daemon($fh,'lonhttpd',40000);
    &checkon_daemon($fh,'lonmemcached',40000);
           &checkon_daemon($fh,'lonmaxima',40000);
       }
       if ($justreload) {
    &checkon_daemon($fh,'lond',40000,'USR2');
    &checkon_daemon($fh,'lonc',40000,'USR2');
       }
       if ($justcheckconnections) {
    &test_connections($fh);
       }
       if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
    &check_delayed_msg($fh);
    &finish_logging($fh);
    &log_simplestatus();
   
    if ($totalcount>200 && !$noemail) { &send_mail(); }
       }
 }  }
   
 &main();  &main();
Removed from v.1.48  
changed lines
  Added in v.1.72

FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>