--- loncom/loncron 2001/01/10 17:20:44 1.12 +++ loncom/loncron 2003/09/10 19:13:09 1.42 @@ -9,8 +9,14 @@ # # 7/14,7/15,7/19,7/21,7/22,11/18, # 2/8 Gerd Kortemeyer -# Dec 00 Scott Harrison # 12/23 Gerd Kortemeyer +# YEAR=2001 +# 09/04,09/06,11/26 Gerd Kortemeyer + +$|=1; + +use lib '/home/httpd/lib/perl/'; +use LONCAPA::Configuration; use IO::File; use IO::Socket; @@ -43,25 +49,160 @@ sub errout { ENDERROUT } +sub start_daemon { + my ($fh,$daemon,$pidfile) = @_; + system("$perlvar{'lonDaemons'}/$daemon 2>>$perlvar{'lonDaemons'}/logs/${daemon}_errors"); + sleep 2; + if (-e $pidfile) { + print $fh "Seems like it started ...

"; + my $lfh=IO::File->new("$pidfile"); + my $daemonpid=<$lfh>; + chomp($daemonpid); + sleep 2; + if (kill 0 => $daemonpid) { + return 1; + } else { + return 0; + } + } + print $fh "Seems like that did not work!

"; + $errors++; + return 0; +} + +sub checkon_daemon { + my ($fh,$daemon,$maxsize,$sendusr1)=@_; + + print $fh '


'.$daemon.'

Log

';
+    printf("%-10s ",$daemon);
+    if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
+	open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
+	while ($line=) { 
+	    print $fh "$line";
+	    if ($line=~/INFO/) { $notices++; }
+	    if ($line=~/WARNING/) { $notices++; }
+	    if ($line=~/CRITICAL/) { $warnings++; }
+	};
+	close (DFH);
+    }
+    print $fh "
"; + + my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid"; + + my $restartflag=1; + + if (-e $pidfile) { + my $lfh=IO::File->new("$pidfile"); + my $daemonpid=<$lfh>; + chomp($daemonpid); + if (kill 0 => $daemonpid) { + print $fh "

$daemon at pid $daemonpid responding"; + if ($sendusr1) { print $fh ", sending USR1"; } + print $fh "

"; + if ($sendusr1) { kill USR1 => $daemonpid; } + $restartflag=0; + print "running\n"; + } else { + $errors++; + print $fh "

$daemon at pid $daemonpid not responding

"; + $restartflag=1; + print $fh "

Decided to clean up stale .pid file and restart $daemon

"; + } + } + if ($restartflag==1) { + $simplestatus{$daemon}='off'; + $errors++; + print $fh '
Killall '.$daemon.': '. + `killall $daemon 2>&1`.' - '; + sleep 2; + print $fh unlink($pidfile).' - '. + `killall -9 $daemon 2>&1`. + '
'; + print $fh "

$daemon not running, trying to start

"; + + if (&start_daemon($fh,$daemon,$pidfile)) { + print $fh "

$daemon at pid $daemonpid responding

"; + $simplestatus{$daemon}='restarted'; + print "started\n"; + } else { + $errors++; + print $fh "

$daemon at pid $daemonpid not responding

"; + print $fh "Give it one more try ...

"; + print " "; + if (&start_daemon($fh,$daemon,$pidfile)) { + print $fh "

$daemon at pid $daemonpid responding

"; + $simplestatus{$daemon}='restarted'; + print "started\n"; + } else { + print " failed\n"; + $simplestatus{$daemon}='failed'; + $errors++; $errors++; + print $fh "

$daemon at pid $daemonpid not responding

"; + print $fh "Unable to start $daemon

"; + } + } + + if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){ + print $fh "

";
+	    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|");
+	    while ($line=) { 
+		print $fh "$line";
+		if ($line=~/WARNING/) { $notices++; }
+		if ($line=~/CRITICAL/) { $notices++; }
+	    };
+	    close (DFH);
+	    print $fh "
"; + } + } + + $fname="$perlvar{'lonDaemons'}/logs/$daemon.log"; + + my ($dev,$ino,$mode,$nlink, + $uid,$gid,$rdev,$size, + $atime,$mtime,$ctime, + $blksize,$blocks)=stat($fname); + + if ($size>$maxsize) { + print $fh "Rotating logs ...

"; + rename("$fname.2","$fname.3"); + rename("$fname.1","$fname.2"); + rename("$fname","$fname.1"); + } + + &errout($fh); +} # ================================================================ Main Program -# ----------------------------- Make sure this process is running from user=www -my $wwwid=getpwnam('www'); -if ($wwwid!=$<) { - print("User ID mismatch. This program must be run as user 'www'\n") unless $noprint; +# --------------------------------- Read loncapa_apache.conf and loncapa.conf +my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf'); +%perlvar=%{$perlvarref}; +undef $perlvarref; +delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed +delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed + +# --------------------------------------- Make sure that LON-CAPA is configured +# I only test for one thing here (lonHostID). This is just a safeguard. +if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) { + print("Unconfigured machine.\n"); + $emailto=$perlvar{'lonSysEMail'}; + $hostname=`/bin/hostname`; + chop $hostname; + $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell + $subj="LON: Unconfigured machine $hostname"; + system("echo 'Unconfigured machine $hostname.' |\ + mailto $emailto -s '$subj' > /dev/null"); exit 1; } -# ------------------------------------------------------------ Read access.conf -{ - my $config=IO::File->new("/etc/httpd/conf/access.conf"); - - while (my $configline=<$config>) { - if ($configline =~ /PerlSetVar/) { - my ($dummy,$varname,$varvalue)=split(/\s+/,$configline); - $perlvar{$varname}=$varvalue; - } - } +# ----------------------------- Make sure this process is running from user=www +my $wwwid=getpwnam('www'); +if ($wwwid!=$<) { + print("User ID mismatch. This program must be run as user 'www'\n"); + $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; + $subj="LON: $perlvar{'lonHostID'} User ID mismatch"; + system("echo 'User ID mismatch. loncron must be run as user www.' |\ + mailto $emailto -s '$subj' > /dev/null"); + exit 1; } # ------------------------------------------------------------- Read hosts file @@ -69,14 +210,21 @@ if ($wwwid!=$<) { my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab"); while (my $configline=<$config>) { - my ($id,$domain,$role,$name,$ip)=split(/:/,$configline); - $hostname{$id}=$name; - $hostdom{$id}=$domain; - $hostrole{$id}=$role; - $hostip{$id}=$ip; - if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) { - $libserv{$id}=$name; - } + my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline); + if ($id && $domain && $role && $name && $ip) { + $hostname{$id}=$name; + $hostdom{$id}=$domain; + $hostip{$id}=$ip; + $hostrole{$id}=$role; + if ($domdescr) { $domaindescription{$domain}=$domdescr; } + if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) { + $libserv{$id}=$name; + } + } else { + if ($configline) { +# &logthis("Skipping hosts.tab line -$configline-"); + } + } } } @@ -105,6 +253,7 @@ $date=localtime($now); { my $fh=IO::File->new(">$statusdir/newstatus.html"); +my %simplestatus=(); print $fh (< @@ -124,6 +273,7 @@ print $fh (<lonsql

  • lond
  • lonc +
  • lonhttpd
  • lonnet
  • Connections
  • Delayed Messages @@ -136,17 +286,17 @@ print $fh (< ENDHEADERS -foreach $varname (keys %perlvar) { +foreach $varname (sort(keys(%perlvar))) { print $fh "$varname$perlvar{$varname}\n"; } print $fh "

    Hosts

    "; -foreach $id (keys %hostname) { -print $fh - ""; -print $fh "\n"; +foreach $id (sort(keys(%hostname))) { + print $fh + ""; + print $fh "\n"; } print $fh "
    $id$hostdom{$id}$hostrole{$id}$hostname{$id}$hostip{$id}
    $id$hostdom{$id}$hostrole{$id}$hostname{$id}$hostip{$id}

    Spare Hosts

      "; -foreach $id (keys %spareid) { +foreach $id (sort(keys(%spareid))) { print $fh "
    1. $id\n"; } @@ -182,21 +332,40 @@ while ($line=) { $usage=$parts[4]; $usage=~s/\W//g; if ($usage>90) { - $warnings++; + $warnings++; + $notices++; } elsif ($usage>80) { $warnings++; } elsif ($usage>60) { $notices++; } - if ($usage>95) { $warnings++; $warnings++ } + if ($usage>95) { $warnings++; $warnings++; $simplestatus{'diskfull'}++; } } close (DFH); print $fh ""; + + +print $fh "

      ps

      "; +print $fh "
      ";
      +$psproc=0;
      +
      +open (PSH,"ps -aux|");
      +while ($line=) { 
      +   print $fh "$line"; 
      +   $psproc++;
      +}
      +close (PSH);
      +print $fh "
      "; + +if ($psproc>200) { $notices++; } +if ($psproc>250) { $notices++; } + &errout($fh); # --------------------------------------------------------------- clean out tmp print $fh '

      Temporary Files

      '; $cleaned=0; +$old=0; while ($fname=<$perlvar{'lonDaemons'}/tmp/*>) { my ($dev,$ino,$mode,$nlink, $uid,$gid,$rdev,$size, @@ -205,12 +374,24 @@ while ($fname=<$perlvar{'lonDaemons'}/tm $now=time; $since=$now-$mtime; if ($since>$perlvar{'lonExpire'}) { - $cleaned++; - unlink("$fname"); + $line=''; + if (open(PROBE,$fname)) { + $line=; + close(PROBE); + } + unless ($line=~/^CHECKOUTTOKEN\&/) { + $cleaned++; + unlink("$fname"); + } else { + if ($since>365*$perlvar{'lonExpire'}) { + $cleaned++; + unlink("$fname"); + } else { $old++; } + } } } -print $fh "Cleaned up ".$cleaned." files."; +print $fh "Cleaned up ".$cleaned." files (".$old." old checkout tokens)."; # ------------------------------------------------------------ clean out lonIDs print $fh '

      Session Tokens

      '; @@ -239,13 +420,13 @@ print $fh "

      $active open session(s)

      httpd

      Access Log

      ';
       
      -open (DFH,"tail -n40 /etc/httpd/logs/access_log|");
      +open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
       while ($line=) { print $fh "$line" };
       close (DFH);
       
       print $fh "

      Error Log

      ";
       
      -open (DFH,"tail -n50 /etc/httpd/logs/error_log|");
      +open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
       while ($line=) { 
          print $fh "$line";
          if ($line=~/\[error\]/) { $notices++; } 
      @@ -256,295 +437,25 @@ print $fh "
      "; # ---------------------------------------------------------------------- lonsql -# -# Do not run for now -# -if ($perlvar{'lonRole'} eq "library" && 1==0) { - - print $fh '

      lonsql

      Log

      ';
      -    
      -    if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
      -	open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
      -	while ($line=) { 
      -	    print $fh "$line";
      -	    if ($line=~/INFO/) { $notices++; }
      -	    if ($line=~/WARNING/) { $notices++; }
      -	    if ($line=~/CRITICAL/) { $warnings++; }
      -	};
      -	close (DFH);
      -    }
      -    print $fh "
      "; - - my $lonsqlfile="$perlvar{'lonDaemons'}/logs/lonsql.pid"; - - if (-e $lonsqlfile) { - my $lfh=IO::File->new("$lonsqlfile"); - my $lonsqlpid=<$lfh>; - chomp($lonsqlpid); - if (kill 0 => $lonsqlpid) { - print $fh "

      lonsql at pid $lonsqlpid responding

      "; - } else { - $errors++; $errors++; - print $fh "

      lonsql at pid $lonsqlpid not responding

      "; - } - } else { - $errors++; - print $fh "

      lonsql not running, trying to start

      "; - system("$perlvar{'lonDaemons'}/lonsql"); - sleep 120; - if (-e $lonsqlfile) { - print $fh "Seems like it started ...

      "; - my $lfh=IO::File->new("$lonsqlfile"); - my $lonsqlpid=<$lfh>; - chomp($lonsqlpid); - sleep 30; - if (kill 0 => $lonsqlpid) { - print $fh "

      lonsql at pid $lonsqlpid responding

      "; - } else { - $errors++; $errors++; - print $fh "

      lonsql at pid $lonsqlpid not responding

      "; - print $fh "Give it one more try ...

      "; - system("$perlvar{'lonDaemons'}/lonsql"); - sleep 120; - } - } else { - print $fh "Seems like that did not work!

      "; - $errors++; - } - if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){ - print $fh "

      ";
      -	    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
      -	    while ($line=) { 
      -		print $fh "$line";
      -		if ($line=~/WARNING/) { $notices++; }
      -		if ($line=~/CRITICAL/) { $notices++; }
      -	    };
      -	    close (DFH);
      -	    print $fh "
      "; - } - } - - $fname="$perlvar{'lonDaemons'}/logs/lonsql.log"; - - my ($dev,$ino,$mode,$nlink, - $uid,$gid,$rdev,$size, - $atime,$mtime,$ctime, - $blksize,$blocks)=stat($fname); - if ($size>40000) { - print $fh "Rotating logs ...

      "; - rename("$fname.2","$fname.3"); - rename("$fname.1","$fname.2"); - rename("$fname","$fname.1"); - } +&checkon_daemon($fh,'lonsql',200000); - &errout($fh); -} # ------------------------------------------------------------------------ lond -print $fh '


      lond

      Log

      ';
      -
      -if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
      -open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
      -while ($line=) { 
      -   print $fh "$line";
      -   if ($line=~/INFO/) { $notices++; }
      -   if ($line=~/WARNING/) { $notices++; }
      -   if ($line=~/CRITICAL/) { $warnings++; }
      -};
      -close (DFH);
      -}
      -print $fh "
      "; - -my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid"; - -my $restartflag=1; -if (-e $londfile) { - my $lfh=IO::File->new("$londfile"); - my $londpid=<$lfh>; - chomp($londpid); - if (kill 0 => $londpid) { - print $fh "

      lond at pid $londpid responding

      "; - $restartflag=0; - } else { - $errors++; - print $fh "

      lond at pid $londpid not responding

      "; - # Intelligently handle this. - # Possibility #1: there is no process - # Solution: remove .pid file and restart - if (getpgrp($londpid)==-1) { - unlink($londfile); - $restartflag=1; - } - else { - # Possibility #2: there is a live process that is not responding - # for an unknown reason - # Solution: kill parent and children processes, remove .pid and restart - `killall -9 lond`; - unlink($londfile); - $restartflag=1; - } - print $fh - "

      Deciding to clean up stale .pid file and restart lond

      "; - } -} -if ($restartflag==1) { - $errors++; - print $fh "

      lond not running, trying to start

      "; - system("$perlvar{'lonDaemons'}/lond"); - sleep 120; - if (-e $londfile) { - print $fh "Seems like it started ...

      "; - my $lfh=IO::File->new("$londfile"); - my $londpid=<$lfh>; - chomp($londpid); - sleep 30; - if (kill 0 => $londpid) { - print $fh "

      lond at pid $londpid responding

      "; - } else { - $errors++; $errors++; - print $fh "

      lond at pid $londpid not responding

      "; - print $fh "Give it one more try ...

      "; - system("$perlvar{'lonDaemons'}/lond"); - sleep 120; - } - } else { - print $fh "Seems like that did not work!

      "; - $errors++; - } - if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){ - print $fh "

      ";
      -    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
      -    while ($line=) { 
      -      print $fh "$line";
      -      if ($line=~/WARNING/) { $notices++; }
      -      if ($line=~/CRITICAL/) { $notices++; }
      -    };
      -    close (DFH);
      -    print $fh "
      "; - } -} - -$fname="$perlvar{'lonDaemons'}/logs/lond.log"; - - my ($dev,$ino,$mode,$nlink, - $uid,$gid,$rdev,$size, - $atime,$mtime,$ctime, - $blksize,$blocks)=stat($fname); - -if ($size>40000) { - print $fh "Rotating logs ...

      "; - rename("$fname.2","$fname.3"); - rename("$fname.1","$fname.2"); - rename("$fname","$fname.1"); -} +&checkon_daemon($fh,'lond',40000,1); -&errout($fh); # ------------------------------------------------------------------------ lonc -print $fh '


      lonc

      Log

      ';
      -
      -if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
      -open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
      -while ($line=) { 
      -   print $fh "$line";
      -   if ($line=~/INFO/) { $notices++; }
      -   if ($line=~/WARNING/) { $notices++; }
      -   if ($line=~/CRITICAL/) { $warnings++; }
      -};
      -close (DFH);
      -}
      -print $fh "
      "; - -my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid"; - -$restartflag=1; -if (-e $loncfile) { - my $lfh=IO::File->new("$loncfile"); - my $loncpid=<$lfh>; - chomp($loncpid); - if (kill 0 => $loncpid) { - print $fh "

      lonc at pid $loncpid responding, sending USR1

      "; - kill USR1 => $loncpid; - $restartflag=0; - } else { - $errors++; - print $fh "

      lonc at pid $loncpid not responding

      "; - # Intelligently handle this. - # Possibility #1: there is no process - # Solution: remove .pid file and restart - if (getpgrp($loncpid)==-1) { - unlink($loncfile); - $restartflag=1; - } - else { - # Possibility #2: there is a live process that is not responding - # for an unknown reason - # Solution: kill parent and children processes, remove .pid and restart - `killall -9 lonc`; - unlink($loncfile); - $restartflag=1; - } - print $fh - "

      Deciding to clean up stale .pid file and restart lonc

      "; - } -} -if ($restartflag==1) { - $errors++; - print $fh "

      lonc not running, trying to start

      "; - system("$perlvar{'lonDaemons'}/lonc"); - sleep 120; - if (-e $loncfile) { - print $fh "Seems like it started ...

      "; - my $lfh=IO::File->new("$loncfile"); - my $loncpid=<$lfh>; - chomp($loncpid); - sleep 30; - if (kill 0 => $loncpid) { - print $fh "

      lonc at pid $loncpid responding

      "; - } else { - $errors++; $errors++; - print $fh "

      lonc at pid $loncpid not responding

      "; - print $fh "Give it one more try ...

      "; - system("$perlvar{'lonDaemons'}/lonc"); - sleep 120; - } - } else { - print $fh "Seems like that did not work!

      "; - $errors++; - } - if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") { - print $fh "

      ";
      -    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
      -    while ($line=) { 
      -      print $fh "$line";
      -      if ($line=~/WARNING/) { $notices++; }
      -      if ($line=~/CRITICAL/) { $notices++; }
      -    };
      -    close (DFH);
      -    print $fh "
      "; - } -} - -$fname="$perlvar{'lonDaemons'}/logs/lonc.log"; +&checkon_daemon($fh,'lonc',40000,1); - my ($dev,$ino,$mode,$nlink, - $uid,$gid,$rdev,$size, - $atime,$mtime,$ctime, - $blksize,$blocks)=stat($fname); +# -------------------------------------------------------------------- lonhttpd -if ($size>40000) { - print $fh "Rotating logs ...

      "; - rename("$fname.2","$fname.3"); - rename("$fname.1","$fname.2"); - rename("$fname","$fname.1"); -} +&checkon_daemon($fh,'lonhttpd',40000); - -&errout($fh); # ---------------------------------------------------------------------- lonnet print $fh '


      lonnet

      Temp Log

      ';
      +print "checking logs\n";
       if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
       open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
       while ($line=) { 
      @@ -581,10 +492,10 @@ print $fh "
      "; # ----------------------------------------------------------------- Connections print $fh '

      Connections

      '; - +print "testing connections\n"; print $fh ""; -foreach $tryserver (keys %hostname) { - +foreach $tryserver (sort(keys(%hostname))) { + print("."); $answer=reply("pong",$tryserver); if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { $result="ok"; @@ -593,6 +504,7 @@ foreach $tryserver (keys %hostname) { $warnings++; if ($answer eq 'con_lost') { $warnings++; } } + if ($answer =~ /con_lost/) { print(" $tryserver down\n"); } print $fh "\n"; } @@ -602,6 +514,7 @@ print $fh "
      $tryserver$result
      "; # ------------------------------------------------------------ Delayed messages print $fh '

      Delayed Messages

      '; +print "checking buffers\n"; print $fh '

      Scanning Permanent Log

      '; @@ -612,7 +525,7 @@ $unsend=0; ($time,$sdf,$dserv,$dcmd)=split(/:/,$line); if ($sdf eq 'F') { $local=localtime($time); - print "Failed: $time, $dserv, $dcmd
      "; + print $fh "Failed: $time, $dserv, $dcmd
      "; $warnings++; } if ($sdf eq 'S') { $unsend--; } @@ -622,6 +535,7 @@ $unsend=0; print $fh "Total unsend messages: $unsend

      \n"; $warnings=$warnings+5*$unsend; +if ($unsend) { $simplestatus{'unsend'}=$unsend; } print $fh "

      Outgoing Buffer

      "; open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|"); @@ -638,13 +552,29 @@ print $fh "

      Total Error Count: $total $now=time; $date=localtime($now); print $fh "
      $date ($now)\n"; - +print "lon-status webpage updated\n"; +$fh->close(); } +if ($errors) { $simplestatus{'errors'}=$errors; } +if ($warnings) { $simplestatus{'warnings'}=$warnings; } +if ($notices) { $simplestatus{'notices'}=$notices; } +$simplestatus{'time'}=time; rename ("$statusdir/newstatus.html","$statusdir/index.html"); - +{ +my $sfh=IO::File->new(">$statusdir/loncron_simple.txt"); +foreach (keys %simplestatus) { + print $sfh $_.'='.$simplestatus{$_}.'&'; +} +print $sfh "\n"; +$sfh->close(); +} if ($totalcount>200) { - $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; + print "sending mail\n"; + $emailto="$perlvar{'lonAdmEMail'}"; + if ($totalcount>1000) { + $emailto.=",$perlvar{'lonSysEMail'}"; + } $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; system( "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");