--- loncom/loncron 1999/11/18 19:52:46 1.2 +++ loncom/loncron 2003/07/29 21:00:21 1.40 @@ -7,7 +7,16 @@ # 6/1/99,6/2,6/10,6/11,6/12,6/14,6/26,6/28,6/29,6/30, # 7/1,7/2,7/9,7/10,7/12 Gerd Kortemeyer) # -# 7/14,7/15,7/19,7/21,7/22,11/18 Gerd Kortemeyer +# 7/14,7/15,7/19,7/21,7/22,11/18, +# 2/8 Gerd Kortemeyer +# 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; @@ -42,17 +51,36 @@ ENDERROUT # ================================================================ Main Program - -# ------------------------------------------------------------ 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; - } - } +# --------------------------------- 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; +} + +# ----------------------------- 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 @@ -60,14 +88,21 @@ ENDERROUT 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-"); + } + } } } @@ -102,16 +137,20 @@ print $fh (< LON Status Report $perlvar{'lonHostID'} - +

LON Status Report $perlvar{'lonHostID'}

$date ($now)

  1. Configuration
  2. Machine Information +
  3. Temporary Files +
  4. Session Tokens
  5. httpd +
  6. lonsql
  7. lond
  8. lonc +
  9. lonhttpd
  10. lonnet
  11. Connections
  12. Delayed Messages @@ -124,17 +163,17 @@ 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"; } @@ -152,7 +191,7 @@ close (LOADAVGH); print $fh "$loadavg"; @parts=split(/\s+/,$loadavg); -if ($parts[1]>3.0) { +if ($parts[1]>4.0) { $errors++; } elsif ($parts[1]>2.0) { $warnings++; @@ -170,28 +209,101 @@ while ($line=) { $usage=$parts[4]; $usage=~s/\W//g; if ($usage>90) { - $errors++; + $warnings++; + $notices++; } elsif ($usage>80) { $warnings++; } elsif ($usage>60) { $notices++; } - if ($usage>95) { $errors++; } + if ($usage>95) { $warnings++; $warnings++ } } 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, + $atime,$mtime,$ctime, + $blksize,$blocks)=stat($fname); + $now=time; + $since=$now-$mtime; + if ($since>$perlvar{'lonExpire'}) { + $line=''; + if (open(PROBE,$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 (".$old." old checkout tokens)."; + +# ------------------------------------------------------------ clean out lonIDs +print $fh '

      Session Tokens

      '; +$cleaned=0; +$active=0; +while ($fname=<$perlvar{'lonIDsDir'}/*>) { + my ($dev,$ino,$mode,$nlink, + $uid,$gid,$rdev,$size, + $atime,$mtime,$ctime, + $blksize,$blocks)=stat($fname); + $now=time; + $since=$now-$mtime; + if ($since>$perlvar{'lonExpire'}) { + $cleaned++; + print $fh "Unlinking $fname
      "; + unlink("$fname"); + } else { + $active++; + } + +} +print $fh "

      Cleaned up ".$cleaned." stale session token(s)."; +print $fh "

      $active open session(s)

      "; + # ----------------------------------------------------------------------- httpd print $fh '

      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=~/\[error\]/) { $notices++; } 
      @@ -199,15 +311,116 @@ while ($line=) {
       close (DFH);
       print $fh "
      "; &errout($fh); + + +# ---------------------------------------------------------------------- lonsql + +my $restartflag=1; + print $fh '

      lonsql

      Log

      ';
      +    print "lonsql\n";
      +    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"; + + $restartflag=1; + + 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

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

      lonsql at pid $lonsqlpid not responding

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

      Decided to clean up stale .pid file and restart lonsql

      "; + } + } + if ($restartflag==1) { + $errors++; + print $fh '
      Killall lonsql: '. + system('killall lonsql').' - '; + sleep 2; + print $fh unlink($lonsqlfile).' - '. + system('killall -9 lonsql'). + '
      '; + print $fh "

      lonsql not running, trying to start

      "; + system( + "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors"); + sleep 2; + if (-e $lonsqlfile) { + print $fh "Seems like it started ...

      "; + my $lfh=IO::File->new("$lonsqlfile"); + my $lonsqlpid=<$lfh>; + chomp($lonsqlpid); + sleep 2; + 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( + "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors"); + sleep 2; + } + } 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>200000) { + print $fh "Rotating logs ...

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


      lond

      Log

      ';
      +print "lond\n";
       
       if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
      -open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lond.log|");
      +open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lond.log|");
       while ($line=) { 
          print $fh "$line";
      -   if ($line=~/giving up/) { $notices++; }
      +   if ($line=~/INFO/) { $notices++; }
      +   if ($line=~/WARNING/) { $notices++; }
      +   if ($line=~/CRITICAL/) { $warnings++; }
       };
       close (DFH);
       }
      @@ -215,40 +428,65 @@ print $fh "
      "; my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid"; -if (-e $londfile) { +$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

      "; + print $fh "

      lond at pid $londpid responding, sending USR1

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

      lond at pid $londpid not responding

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

      Decided to clean up stale .pid file and restart lond

      "; } -} else { +} +if ($restartflag==1) { $errors++; + print $fh '
      Killall lond: '. + system('killall lond').' - '; + sleep 2; + print $fh unlink($londfile).' - '.system('killall -9 lond'). + '
      '; print $fh "

      lond not running, trying to start

      "; - system("$perlvar{'lonDaemons'}/lond"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); + sleep 2; if (-e $londfile) { print $fh "Seems like it started ...

      "; my $lfh=IO::File->new("$londfile"); my $londpid=<$lfh>; chomp($londpid); - sleep 30; + sleep 2; 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("$perlvar{'lonDaemons'}/lond"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); + sleep 2; } } 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"; @@ -269,12 +507,15 @@ if ($size>40000) { # ------------------------------------------------------------------------ lonc print $fh '

      lonc

      Log

      ';
      +print "lonc\n";
       
       if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
      -open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonc.log|");
      +open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonc.log|");
       while ($line=) { 
          print $fh "$line";
      -   if ($line=~/died/) { $notices++; }
      +   if ($line=~/INFO/) { $notices++; }
      +   if ($line=~/WARNING/) { $notices++; }
      +   if ($line=~/CRITICAL/) { $warnings++; }
       };
       close (DFH);
       }
      @@ -282,6 +523,7 @@ print $fh "
      "; my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid"; +$restartflag=1; if (-e $loncfile) { my $lfh=IO::File->new("$loncfile"); my $loncpid=<$lfh>; @@ -289,34 +531,58 @@ if (-e $loncfile) { if (kill 0 => $loncpid) { print $fh "

      lonc at pid $loncpid responding, sending USR1

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

      lonc at pid $loncpid not responding

      "; + # Solution: kill parent and children processes, remove .pid and restart + $restartflag=1; + print $fh + "

      Decided to clean up stale .pid file and restart lonc

      "; } -} else { +} +if ($restartflag==1) { $errors++; + print $fh '
      Killall lonc: '. + system('killall lonc').' - '; + sleep 2; + print $fh unlink($loncfile).' - '.system('killall -9 lonc'). + '
      '; print $fh "

      lonc not running, trying to start

      "; - system("$perlvar{'lonDaemons'}/lonc"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors"); + sleep 2; if (-e $loncfile) { print $fh "Seems like it started ...

      "; my $lfh=IO::File->new("$loncfile"); my $loncpid=<$lfh>; chomp($loncpid); - sleep 30; + sleep 2; 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("$perlvar{'lonDaemons'}/lonc"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors"); + sleep 2; } } 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"; @@ -335,20 +601,114 @@ if ($size>40000) { &errout($fh); +# -------------------------------------------------------------------- lonhttpd + +print $fh '

      lonhttpd

      Log

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

      lonhttpd at pid $lonhttpdpid responding

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

      lonhttpd at pid $lonhttpdpid not responding

      "; + # Solution: kill parent and children processes, remove .pid and restart + $restartflag=1; + print $fh + "

      Decided to clean up stale .pid file and restart lonhttpd

      "; + } +} +if ($restartflag==1) { + $errors++; + print $fh '
      Killall lonhttpd: '. + system('killall lonhttpd').' - '; + sleep 2; + print $fh unlink($lonhttpdfile).' - '.system('killall -9 lonhttpd'). + '
      '; + print $fh "

      lonhttpd not running, trying to start

      "; + system( + "$perlvar{'lonDaemons'}/lonhttpd 2>>$perlvar{'lonDaemons'}/logs/lonhttpd_errors"); + sleep 2; + if (-e $lonhttpdfile) { + print $fh "Seems like it started ...

      "; + my $lfh=IO::File->new("$lonhttpdfile"); + my $lonhttpdpid=<$lfh>; + chomp($lonhttpdpid); + sleep 2; + if (kill 0 => $lonhttpdpid) { + print $fh "

      lonhttpd at pid $lonhttpdpid responding

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

      lonhttpd at pid $lonhttpdpid not responding

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

      "; + system( + "$perlvar{'lonDaemons'}/lonhttpd 2>>$perlvar{'lonDaemons'}/logs/lonhttpd_errors"); + sleep 2; + } + } else { + print $fh "Seems like that did not work!

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

      ";
      +    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonhttpd.log|");
      +    while ($line=) { 
      +      print $fh "$line";
      +      if ($line=~/WARNING/) { $notices++; }
      +      if ($line=~/CRITICAL/) { $notices++; }
      +    };
      +    close (DFH);
      +    print $fh "
      "; + } +} + +$fname="$perlvar{'lonDaemons'}/logs/lonhttpd.log"; + + my ($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"); +} + + +&errout($fh); # ---------------------------------------------------------------------- lonnet print $fh '


      lonnet

      Temp Log

      ';
      +print "lonnet\n";
       if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
       open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
       while ($line=) { 
           print $fh "$line";
      -    if ($line=~/Delayed/) { $warnings++; }
      -    if ($line=~/giving up/) { $warnings++; }
      -    if ($line=~/FAILED/) { $errors++; }
       };
       close (DFH);
       }
      -print $fh "

      Perm Log

      "; +print $fh "

      Perm Log

      ";
       
       if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
           open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
      @@ -379,7 +739,7 @@ print $fh "
      "; print $fh '

      Connections

      '; print $fh ""; -foreach $tryserver (keys %hostname) { +foreach $tryserver (sort(keys(%hostname))) { $answer=reply("pong",$tryserver); if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { @@ -398,6 +758,7 @@ print $fh "
      "; # ------------------------------------------------------------ Delayed messages print $fh '

      Delayed Messages

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

      Scanning Permanent Log

      '; @@ -434,13 +795,17 @@ print $fh "

      Total Error Count: $total $now=time; $date=localtime($now); print $fh "
      $date ($now)\n"; - +print "writing done\n"; } rename ("$statusdir/newstatus.html","$statusdir/index.html"); if ($totalcount>200) { - $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; + print "mailing\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");