--- loncom/loncron 2003/06/10 22:07:26 1.39 +++ loncom/loncron 2003/10/14 15:36:21 1.44 @@ -49,6 +49,133 @@ sub errout { ENDERROUT } +sub start_daemon { + my ($fh,$daemon,$pidfile) = @_; + my $progname=$daemon; + if ($daemon eq 'lonc' && $ARGV[0] eq 'new') { + $progname='loncnew'; + print "new "; + } + system("$perlvar{'lonDaemons'}/$progname 2>>$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 # --------------------------------- Read loncapa_apache.conf and loncapa.conf @@ -61,13 +188,13 @@ delete $perlvar{'lonSqlAccess'}; # remov # --------------------------------------- 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.' |\ + 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; } @@ -75,18 +202,18 @@ if ('{[[[[lonHostID]]]]}' eq $perlvar{'l # ----------------------------- 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.' |\ + 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; + exit 1; } # ------------------------------------------------------------- Read hosts file { my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab"); - + while (my $configline=<$config>) { my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline); if ($id && $domain && $role && $name && $ip) { @@ -109,12 +236,12 @@ if ($wwwid!=$<) { # ------------------------------------------------------ Read spare server file { my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab"); - + while (my $configline=<$config>) { - chomp($configline); - if (($configline) && ($configline ne $perlvar{'lonHostID'})) { - $spareid{$configline}=1; - } + chomp($configline); + if (($configline) && ($configline ne $perlvar{'lonHostID'})) { + $spareid{$configline}=1; + } } } @@ -130,9 +257,10 @@ $now=time; $date=localtime($now); { -my $fh=IO::File->new(">$statusdir/newstatus.html"); - -print $fh (<new(">$statusdir/newstatus.html"); + my %simplestatus=(); + + print $fh (< LON Status Report $perlvar{'lonHostID'} @@ -163,652 +291,297 @@ print $fh (< ENDHEADERS -foreach $varname (sort(keys(%perlvar))) { - print $fh "$varname$perlvar{$varname}\n"; -} -print $fh "

Hosts

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

Spare Hosts

    "; -foreach $id (sort(keys(%spareid))) { - print $fh "
  1. $id\n"; -} - -print $fh "
\n"; + foreach $varname (sort(keys(%perlvar))) { + print $fh "$varname$perlvar{$varname}\n"; + } + print $fh "

Hosts

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

Spare Hosts

    "; + foreach $id (sort(keys(%spareid))) { + print $fh "
  1. $id\n"; + } + + print $fh "
\n"; # --------------------------------------------------------------------- Machine - -print $fh '

Machine Information

'; -print $fh "

loadavg

"; - -open (LOADAVGH,"/proc/loadavg"); -$loadavg=; -close (LOADAVGH); - -print $fh "$loadavg"; - -@parts=split(/\s+/,$loadavg); -if ($parts[1]>4.0) { - $errors++; -} elsif ($parts[1]>2.0) { - $warnings++; -} elsif ($parts[1]>1.0) { - $notices++; -} - -print $fh "

df

"; -print $fh "
";
-
-open (DFH,"df|");
-while ($line=) { 
-   print $fh "$line"; 
-   @parts=split(/\s+/,$line);
-   $usage=$parts[4];
-   $usage=~s/\W//g;
-   if ($usage>90) { 
-      $warnings++;
-      $notices++; 
-   } elsif ($usage>80) {
-      $warnings++;
-   } elsif ($usage>60) {
-      $notices++;
-   }
-   if ($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 '

Machine Information

'; + print $fh "

loadavg

"; -} -print $fh "

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

$active open session(s)

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

httpd

Access Log

';
+    open (LOADAVGH,"/proc/loadavg");
+    $loadavg=;
+    close (LOADAVGH);
+    
+    print $fh "$loadavg";
+    
+    @parts=split(/\s+/,$loadavg);
+    if ($parts[1]>4.0) {
+	$errors++;
+    } elsif ($parts[1]>2.0) {
+	$warnings++;
+    } elsif ($parts[1]>1.0) {
+	$notices++;
+    }
 
-open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
-while ($line=) { print $fh "$line" };
-close (DFH);
-
-print $fh "

Error Log

";
-
-open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
-while ($line=) { 
-   print $fh "$line";
-   if ($line=~/\[error\]/) { $notices++; } 
-};
-close (DFH);
-print $fh "
"; -&errout($fh); + print $fh "

df

"; + print $fh "
";
 
+    open (DFH,"df|");
+    while ($line=) { 
+	print $fh "$line"; 
+	@parts=split(/\s+/,$line);
+	$usage=$parts[4];
+	$usage=~s/\W//g;
+	if ($usage>90) { 
+	    $warnings++;
+	    $notices++; 
+	} elsif ($usage>80) {
+	    $warnings++;
+	} elsif ($usage>60) {
+	    $notices++;
+	}
+	if ($usage>95) { $warnings++; $warnings++; $simplestatus{'diskfull'}++; }
+    }
+    close (DFH);
+    print $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 "

ps

"; + print $fh "
";
+    $psproc=0;
+
+    open (PSH,"ps -aux|");
+    while ($line=) { 
+	print $fh "$line"; 
+	$psproc++;
     }
+    close (PSH);
     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

"; + + 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 { - $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; + if ($since>365*$perlvar{'lonExpire'}) { + $cleaned++; + unlink("$fname"); + } else { $old++; } } - } 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 "
"; } + } + print $fh "Cleaned up ".$cleaned." files (".$old." old checkout tokens)."; - $fname="$perlvar{'lonDaemons'}/logs/lonsql.log"; - - my ($dev,$ino,$mode,$nlink, - $uid,$gid,$rdev,$size, - $atime,$mtime,$ctime, - $blksize,$blocks)=stat($fname); +# ------------------------------------------------------------ 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 { + $active++; + } - if ($size>200000) { - print $fh "Rotating logs ...

"; - rename("$fname.2","$fname.3"); - rename("$fname.1","$fname.2"); - rename("$fname","$fname.1"); } + print $fh "

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

$active open session(s)

"; - &errout($fh); -# ------------------------------------------------------------------------ lond - -print $fh '

lond

Log

';
-print "lond\n";
+# ----------------------------------------------------------------------- httpd
 
-if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
-open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lond.log|");
-while ($line=) { 
-   print $fh "$line";
-   if ($line=~/INFO/) { $notices++; }
-   if ($line=~/WARNING/) { $notices++; }
-   if ($line=~/CRITICAL/) { $warnings++; }
-};
-close (DFH);
-}
-print $fh "
"; + print $fh '

httpd

Access Log

';
+    
+    open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
+    while ($line=) { print $fh "$line" };
+    close (DFH);
 
-my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
+    print $fh "

Error Log

";
 
-$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, sending USR1

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

lond at pid $londpid not responding

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

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 2; - if (-e $londfile) { - print $fh "Seems like it started ...

"; - my $lfh=IO::File->new("$londfile"); - my $londpid=<$lfh>; - chomp($londpid); - 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( - "$perlvar{'lonDaemons'}/lond 2>>$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 $perlvar{'lonDaemons'}/logs/lond.log|");
+    open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
     while ($line=) { 
-      print $fh "$line";
-      if ($line=~/WARNING/) { $notices++; }
-      if ($line=~/CRITICAL/) { $notices++; }
+	print $fh "$line";
+	if ($line=~/\[error\]/) { $notices++; } 
     };
     close (DFH);
     print $fh "
"; - } -} - -$fname="$perlvar{'lonDaemons'}/logs/lond.log"; + &errout($fh); - 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); -# ------------------------------------------------------------------------ lonc +# ---------------------------------------------------------------------- lonsql -print $fh '


lonc

Log

';
-print "lonc\n";
+    &checkon_daemon($fh,'lonsql',200000);
 
-if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
-open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonc.log|");
-while ($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"; +# ------------------------------------------------------------------------ lond -$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

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

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 2; - if (-e $loncfile) { - print $fh "Seems like it started ...

"; - my $lfh=IO::File->new("$loncfile"); - my $loncpid=<$lfh>; - chomp($loncpid); - 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( - "$perlvar{'lonDaemons'}/lonc 2>>$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 $fh "
"; - } -} + &checkon_daemon($fh,'lond',40000,1); -$fname="$perlvar{'lonDaemons'}/logs/lonc.log"; +# ------------------------------------------------------------------------ lonc - 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($fh,'lonc',40000,1); - -&errout($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"; + &checkon_daemon($fh,'lonhttpd',40000); -$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="$perlvar{'lonDaemons'}/logs/lonhttpd.log"; +# ---------------------------------------------------------------------- lonnet - my ($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"); -} + 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=) { 
+	    print $fh "$line";
+	};
+	close (DFH);
+    }
+    print $fh "

Perm Log

";
+    
+    if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
+	open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
+	while ($line=) { 
+	    print $fh "$line";
+	};
+	close (DFH);
+    } else { print $fh "No perm log\n" }
 
-   
-&errout($fh);
-# ---------------------------------------------------------------------- lonnet
+    $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
 
-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";
-};
-close (DFH);
-}
-print $fh "

Perm Log

";
+    my ($dev,$ino,$mode,$nlink,
+	$uid,$gid,$rdev,$size,
+	$atime,$mtime,$ctime,
+	$blksize,$blocks)=stat($fname);
 
-if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
-    open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
-while ($line=) { 
-   print $fh "$line";
-};
-close (DFH);
-} else { print $fh "No perm log\n" }
-
-$fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
-
-                          my ($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"); -} + if ($size>40000) { + print $fh "Rotating logs ...

"; + rename("$fname.2","$fname.3"); + rename("$fname.1","$fname.2"); + rename("$fname","$fname.1"); + } -print $fh "

"; -&errout($fh); + print $fh "
"; + &errout($fh); # ----------------------------------------------------------------- Connections -print $fh '

Connections

'; - -print $fh ""; -foreach $tryserver (sort(keys(%hostname))) { + print $fh '

Connections

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

Delayed Messages

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

Delayed Messages

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

Scanning Permanent Log

'; + print $fh '

Scanning Permanent Log

'; -$unsend=0; -{ - my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log"); - while ($line=<$dfh>) { - ($time,$sdf,$dserv,$dcmd)=split(/:/,$line); - if ($sdf eq 'F') { - $local=localtime($time); - print "Failed: $time, $dserv, $dcmd
"; - $warnings++; - } - if ($sdf eq 'S') { $unsend--; } - if ($sdf eq 'D') { $unsend++; } + $unsend=0; + { + my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log"); + while ($line=<$dfh>) { + ($time,$sdf,$dserv,$dcmd)=split(/:/,$line); + if ($sdf eq 'F') { + $local=localtime($time); + print $fh "Failed: $time, $dserv, $dcmd
"; + $warnings++; + } + if ($sdf eq 'S') { $unsend--; } + if ($sdf eq 'D') { $unsend++; } + } } -} -print $fh "Total unsend messages: $unsend

\n"; -$warnings=$warnings+5*$unsend; + print $fh "Total unsend messages: $unsend

\n"; + $warnings=$warnings+5*$unsend; -print $fh "

Outgoing Buffer

"; + if ($unsend) { $simplestatus{'unsend'}=$unsend; } + print $fh "

Outgoing Buffer

"; -open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|"); -while ($line=) { - print $fh "$line
"; -}; -close (DFH); + open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|"); + while ($line=) { + print $fh "$line
"; + }; + close (DFH); # ------------------------------------------------------------------------- End -print $fh "
\n"; -$totalcount=$notices+4*$warnings+100*$errors; -&errout($fh); -print $fh "

Total Error Count: $totalcount

"; -$now=time; -$date=localtime($now); -print $fh "
$date ($now)\n"; -print "writing done\n"; -} + print $fh "
\n"; + $totalcount=$notices+4*$warnings+100*$errors; + &errout($fh); + print $fh "

Total Error Count: $totalcount

"; + $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) { - 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"); + 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"); } 1;