--- loncom/loncron 2001/03/15 13:20:46 1.18 +++ loncom/loncron 2003/03/10 17:39:21 1.37 @@ -9,9 +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 -# 02/12/2001 Scott Harrison +# 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; @@ -46,17 +51,12 @@ 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 $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed # --------------------------------------- Make sure that LON-CAPA is configured # I only test for one thing here (lonHostID). This is just a safeguard. @@ -88,14 +88,21 @@ if ($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-"); + } + } } } @@ -143,6 +150,7 @@ print $fh (<lonsql
  • lond
  • lonc +
  • lonhttpd
  • lonnet
  • Connections
  • Delayed Messages @@ -155,17 +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. $id\n"; } @@ -201,7 +209,8 @@ while ($line=) { $usage=$parts[4]; $usage=~s/\W//g; if ($usage>90) { - $warnings++; + $warnings++; + $notices++; } elsif ($usage>80) { $warnings++; } elsif ($usage>60) { @@ -211,11 +220,29 @@ while ($line=) { } 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, @@ -224,12 +251,24 @@ 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

      '; @@ -258,13 +297,13 @@ 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++; } 
@@ -275,13 +314,12 @@ print $fh "
      "; # ---------------------------------------------------------------------- lonsql -# -# Do not run for now -# + +my $restartflag=1; if ($perlvar{'lonRole'} eq "library") { 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=) { 
@@ -295,29 +333,42 @@ if ($perlvar{'lonRole'} eq "library") {
     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

      "; } - } else { + } + 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 120; + sleep 2; if (-e $lonsqlfile) { print $fh "Seems like it started ...

      "; my $lfh=IO::File->new("$lonsqlfile"); my $lonsqlpid=<$lfh>; chomp($lonsqlpid); - sleep 30; + sleep 2; if (kill 0 => $lonsqlpid) { print $fh "

      lonsql at pid $lonsqlpid responding

      "; } else { @@ -326,7 +377,7 @@ if ($perlvar{'lonRole'} eq "library") { print $fh "Give it one more try ...

      "; system( "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors"); - sleep 120; + sleep 2; } } else { print $fh "Seems like that did not work!

      "; @@ -364,9 +415,10 @@ if ($perlvar{'lonRole'} eq "library") { # ------------------------------------------------------------------------ lond print $fh '

      lond

      Log

      ';
+print "lond\n";
 
 if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
-open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
+open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lond.log|");
 while ($line=) { 
    print $fh "$line";
    if ($line=~/INFO/) { $notices++; }
@@ -379,48 +431,40 @@ print $fh "
      "; my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid"; -my $restartflag=1; +$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++; 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; - } + $restartflag=1; print $fh - "

      Deciding to clean up stale .pid file and restart lond

      "; + "

      Decided to clean up stale .pid file and restart lond

      "; } } 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( "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); - sleep 120; + 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 { @@ -429,7 +473,7 @@ if ($restartflag==1) { print $fh "Give it one more try ...

      "; system( "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); - sleep 120; + sleep 2; } } else { print $fh "Seems like that did not work!

      "; @@ -466,9 +510,10 @@ if ($size>40000) { # ------------------------------------------------------------------------ lonc print $fh '

      lonc

      Log

      ';
+print "lonc\n";
 
 if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
-open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
+open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonc.log|");
 while ($line=) { 
    print $fh "$line";
    if ($line=~/INFO/) { $notices++; }
@@ -493,37 +538,29 @@ if (-e $loncfile) {
    } 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 $fh 
-	  "
      Deciding to clean up stale .pid file and restart lonc
      ";
+	  "
      Decided to clean up stale .pid file and restart lonc
      ";
    }
 } 
 if ($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 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
-   sleep 120;
+   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 $loncpid responding
      ";
        } else {
@@ -532,7 +569,7 @@ if ($restartflag==1) {
           print $fh "Give it one more try ...
      ";
  	  system(
  "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
-          sleep 120;
+          sleep 2;
        }
    } else {
        print $fh "Seems like that did not work!
      ";
@@ -567,9 +604,106 @@ if ($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=) { 
@@ -608,7 +742,7 @@ 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'}") {
@@ -627,6 +761,7 @@ print $fh "
      ";
 # ------------------------------------------------------------ Delayed messages
 
 print $fh '
      Delayed Messages
      ';
+print "buffers\n";
 
 print $fh '
      Scanning Permanent Log
      ';
 
@@ -663,13 +798,17 @@ 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>600) {
+      $emailto.=",$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");