--- loncom/loncron 1999/11/18 19:52:46 1.2 +++ loncom/loncron 2000/12/08 23:33:36 1.9 @@ -7,7 +7,8 @@ # 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 use IO::File; use IO::Socket; @@ -102,7 +103,7 @@ print $fh (< LON Status Report $perlvar{'lonHostID'} - +

LON Status Report $perlvar{'lonHostID'}

$date ($now)

@@ -152,7 +153,7 @@ 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) { $warnings++; @@ -170,13 +171,13 @@ while ($line=) { $usage=$parts[4]; $usage=~s/\W//g; if ($usage>90) { - $errors++; + $warnings++; } elsif ($usage>80) { $warnings++; } elsif ($usage>60) { $notices++; } - if ($usage>95) { $errors++; } + if ($usage>95) { $warnings++; $warnings++ } } close (DFH); print $fh ""; @@ -199,15 +200,102 @@ while ($line=) { close (DFH); print $fh ""; &errout($fh); + + +# ------------------------------------------------------------------------ lonsql + +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"); + } + + &errout($fh); +} # ------------------------------------------------------------------------ lond print $fh '

lond

Log

';
 
 if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
-open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lond.log|");
+open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
 while ($line=) { 
    print $fh "$line";
-   if ($line=~/giving up/) { $notices++; }
+   if ($line=~/INFO/) { $notices++; }
+   if ($line=~/WARNING/) { $notices++; }
+   if ($line=~/CRITICAL/) { $warnings++; }
 };
 close (DFH);
 }
@@ -215,17 +303,38 @@ print $fh "
"; my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid"; -if (-e $londfile) { +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++; $errors++; + $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); + $restartflag=1; + } + else { + # Possibility #2: there is a zombie process + # Possibility #3: there is a live process that is not responding + # for an unknown reason + # Solution: kill process, remove .pid and restart + `killall -9 lond`; + unlink($londfile); + $restartflag=1; + } + print $fh + "

Deciding to clean up stale .pid file and restart lond

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

lond not running, trying to start

"; system("$perlvar{'lonDaemons'}/lond"); @@ -249,6 +358,17 @@ if (-e $londfile) { 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"; @@ -271,10 +391,12 @@ if ($size>40000) { print $fh '

lonc

Log

';
 
 if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
-open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonc.log|");
+open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
 while ($line=) { 
    print $fh "$line";
-   if ($line=~/died/) { $notices++; }
+   if ($line=~/INFO/) { $notices++; }
+   if ($line=~/WARNING/) { $notices++; }
+   if ($line=~/CRITICAL/) { $warnings++; }
 };
 close (DFH);
 }
@@ -282,6 +404,7 @@ print $fh "
"; my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid"; +$restartflag=1; if (-e $loncfile) { my $lfh=IO::File->new("$loncfile"); my $loncpid=<$lfh>; @@ -289,11 +412,31 @@ 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

"; + # Intelligently handle this. + # Possibility #1: there is no process + # Solution: remove .pid file and restart + if (getpgrp($loncpid)==-1) { + unlink($loncfile); + $restartflag=1; + } + else { + # Possibility #2: there is a zombie process + # Possibility #3: there is a live process that is not responding + # for an unknown reason + # Solution: kill process, remove .pid and restart + `killall -9 lonc`; + unlink($loncfile); + $restartflag=1; + } + print $fh + "

Deciding to clean up stale .pid file and restart lonc

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

lonc not running, trying to start

"; system("$perlvar{'lonDaemons'}/lonc"); @@ -317,6 +460,17 @@ if (-e $loncfile) { 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"; @@ -342,9 +496,6 @@ if (-e "$perlvar{'lonDaemons'}/logs/lonn 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); }