version 1.2, 1999/11/18 19:52:46
|
version 1.7, 2000/12/06 21:11:35
|
Line 7
|
Line 7
|
# 6/1/99,6/2,6/10,6/11,6/12,6/14,6/26,6/28,6/29,6/30, |
# 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) |
# 7/1,7/2,7/9,7/10,7/12 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::File; |
use IO::Socket; |
use IO::Socket; |
Line 102 print $fh (<<ENDHEADERS);
|
Line 103 print $fh (<<ENDHEADERS);
|
<head> |
<head> |
<title>LON Status Report $perlvar{'lonHostID'}</title> |
<title>LON Status Report $perlvar{'lonHostID'}</title> |
</head> |
</head> |
<body bgcolor="#FFFFFF"> |
<body bgcolor="#AAAAAA"> |
<a name="top"> |
<a name="top"> |
<h1>LON Status Report $perlvar{'lonHostID'}</h1> |
<h1>LON Status Report $perlvar{'lonHostID'}</h1> |
<h2>$date ($now)</h2> |
<h2>$date ($now)</h2> |
Line 152 close (LOADAVGH);
|
Line 153 close (LOADAVGH);
|
print $fh "<tt>$loadavg</tt>"; |
print $fh "<tt>$loadavg</tt>"; |
|
|
@parts=split(/\s+/,$loadavg); |
@parts=split(/\s+/,$loadavg); |
if ($parts[1]>3.0) { |
if ($parts[1]>4.0) { |
$errors++; |
$errors++; |
} elsif ($parts[1]>2.0) { |
} elsif ($parts[1]>2.0) { |
$warnings++; |
$warnings++; |
Line 170 while ($line=<DFH>) {
|
Line 171 while ($line=<DFH>) {
|
$usage=$parts[4]; |
$usage=$parts[4]; |
$usage=~s/\W//g; |
$usage=~s/\W//g; |
if ($usage>90) { |
if ($usage>90) { |
$errors++; |
$warnings++; |
} elsif ($usage>80) { |
} elsif ($usage>80) { |
$warnings++; |
$warnings++; |
} elsif ($usage>60) { |
} elsif ($usage>60) { |
$notices++; |
$notices++; |
} |
} |
if ($usage>95) { $errors++; } |
if ($usage>95) { $warnings++; $warnings++ } |
} |
} |
close (DFH); |
close (DFH); |
print $fh "</pre>"; |
print $fh "</pre>"; |
Line 199 while ($line=<DFH>) {
|
Line 200 while ($line=<DFH>) {
|
close (DFH); |
close (DFH); |
print $fh "</pre>"; |
print $fh "</pre>"; |
&errout($fh); |
&errout($fh); |
|
|
|
|
|
# ------------------------------------------------------------------------ lonsql |
|
|
|
if ($perlvar{'lonRole'} eq "library" && 1==0) { |
|
|
|
print $fh '<hr><a name="lond"><h2>lonsql</h2><h3>Log</h3><pre>'; |
|
|
|
if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){ |
|
open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|"); |
|
while ($line=<DFH>) { |
|
print $fh "$line"; |
|
if ($line=~/INFO/) { $notices++; } |
|
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $warnings++; } |
|
}; |
|
close (DFH); |
|
} |
|
print $fh "</pre>"; |
|
|
|
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 "<h3>lonsql at pid $lonsqlpid responding</h3>"; |
|
} else { |
|
$errors++; $errors++; |
|
print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>"; |
|
} |
|
} else { |
|
$errors++; |
|
print $fh "<h3>lonsql not running, trying to start</h3>"; |
|
system("$perlvar{'lonDaemons'}/lonsql"); |
|
sleep 120; |
|
if (-e $lonsqlfile) { |
|
print $fh "Seems like it started ...<p>"; |
|
my $lfh=IO::File->new("$lonsqlfile"); |
|
my $lonsqlpid=<$lfh>; |
|
chomp($lonsqlpid); |
|
sleep 30; |
|
if (kill 0 => $lonsqlpid) { |
|
print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>"; |
|
} else { |
|
$errors++; $errors++; |
|
print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>"; |
|
print $fh "Give it one more try ...<p>"; |
|
system("$perlvar{'lonDaemons'}/lonsql"); |
|
sleep 120; |
|
} |
|
} else { |
|
print $fh "Seems like that did not work!<p>"; |
|
$errors++; |
|
} |
|
if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){ |
|
print $fh "<p><pre>"; |
|
open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|"); |
|
while ($line=<DFH>) { |
|
print $fh "$line"; |
|
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $notices++; } |
|
}; |
|
close (DFH); |
|
print $fh "</pre>"; |
|
} |
|
} |
|
|
|
$fname="$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 $fh "Rotating logs ...<p>"; |
|
rename("$fname.2","$fname.3"); |
|
rename("$fname.1","$fname.2"); |
|
rename("$fname","$fname.1"); |
|
} |
|
|
|
&errout($fh); |
|
} |
# ------------------------------------------------------------------------ lond |
# ------------------------------------------------------------------------ lond |
|
|
print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>'; |
print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>'; |
|
|
if (-e "$perlvar{'lonDaemons'}/logs/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=<DFH>) { |
while ($line=<DFH>) { |
print $fh "$line"; |
print $fh "$line"; |
if ($line=~/giving up/) { $notices++; } |
if ($line=~/INFO/) { $notices++; } |
|
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $warnings++; } |
}; |
}; |
close (DFH); |
close (DFH); |
} |
} |
Line 215 print $fh "</pre>";
|
Line 303 print $fh "</pre>";
|
|
|
my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid"; |
my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid"; |
|
|
if (-e $londfile) { |
my $restartflag=1; |
|
if (-e $londfile) { |
my $lfh=IO::File->new("$londfile"); |
my $lfh=IO::File->new("$londfile"); |
my $londpid=<$lfh>; |
my $londpid=<$lfh>; |
chomp($londpid); |
chomp($londpid); |
if (kill 0 => $londpid) { |
if (kill 0 => $londpid) { |
print $fh "<h3>lond at pid $londpid responding</h3>"; |
print $fh "<h3>lond at pid $londpid responding</h3>"; |
|
$restartflag=0; |
} else { |
} else { |
$errors++; $errors++; |
$errors++; $errors++; |
print $fh "<h3>lond at pid $londpid not responding</h3>"; |
print $fh "<h3>lond at pid $londpid not responding</h3>"; |
|
# Intelligently handle this. |
|
# Possibility #1: there is no process |
|
# Solution: remove .pid file and restart |
|
if (getpgrp($londpid)==-1) { |
|
unlink($londfile); |
|
$restartflag=1; |
|
} |
|
# 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 |
|
kill 2 => $londpid; |
|
unlink($londfile); |
|
$restartflag=1; |
|
print $fh "<h3>Deciding to clean up stale .pid file and restart lond</h3>"; |
} |
} |
} else { |
} |
|
if ($restartflag==1) { |
$errors++; |
$errors++; |
print $fh "<h3>lond not running, trying to start</h3>"; |
print $fh "<h3>lond not running, trying to start</h3>"; |
system("$perlvar{'lonDaemons'}/lond"); |
system("$perlvar{'lonDaemons'}/lond"); |
Line 249 if (-e $londfile) {
|
Line 355 if (-e $londfile) {
|
print $fh "Seems like that did not work!<p>"; |
print $fh "Seems like that did not work!<p>"; |
$errors++; |
$errors++; |
} |
} |
|
if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){ |
|
print $fh "<p><pre>"; |
|
open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|"); |
|
while ($line=<DFH>) { |
|
print $fh "$line"; |
|
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $notices++; } |
|
}; |
|
close (DFH); |
|
print $fh "</pre>"; |
|
} |
} |
} |
|
|
$fname="$perlvar{'lonDaemons'}/logs/lond.log"; |
$fname="$perlvar{'lonDaemons'}/logs/lond.log"; |
Line 271 if ($size>40000) {
|
Line 388 if ($size>40000) {
|
print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>'; |
print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>'; |
|
|
if (-e "$perlvar{'lonDaemons'}/logs/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=<DFH>) { |
while ($line=<DFH>) { |
print $fh "$line"; |
print $fh "$line"; |
if ($line=~/died/) { $notices++; } |
if ($line=~/INFO/) { $notices++; } |
|
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $warnings++; } |
}; |
}; |
close (DFH); |
close (DFH); |
} |
} |
Line 282 print $fh "</pre>";
|
Line 401 print $fh "</pre>";
|
|
|
my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid"; |
my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid"; |
|
|
|
$restartflag=1; |
if (-e $loncfile) { |
if (-e $loncfile) { |
my $lfh=IO::File->new("$loncfile"); |
my $lfh=IO::File->new("$loncfile"); |
my $loncpid=<$lfh>; |
my $loncpid=<$lfh>; |
Line 289 if (-e $loncfile) {
|
Line 409 if (-e $loncfile) {
|
if (kill 0 => $loncpid) { |
if (kill 0 => $loncpid) { |
print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>"; |
print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>"; |
kill USR1 => $loncpid; |
kill USR1 => $loncpid; |
|
$restartflag=0; |
} else { |
} else { |
$errors++; $errors++; |
$errors++; $errors++; |
print $fh "<h3>lonc at pid $loncpid not responding</h3>"; |
print $fh "<h3>lonc at pid $loncpid not responding</h3>"; |
|
# Intelligently handle this. |
|
# Possibility #1: there is no process |
|
# Solution: remove .pid file and restart |
|
if (getpgrp($loncpid)==-1) { |
|
unlink($loncfile); |
|
$restartflag=1; |
|
} |
|
# 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 |
|
kill 2 => $loncpid; |
|
unlink($loncfile); |
|
$restartflag=1; |
|
print $fh "<h3>Deciding to clean up stale .pid file and restart lonc</h3>"; |
} |
} |
} else { |
} |
|
if ($restartflag==1) { |
$errors++; |
$errors++; |
print $fh "<h3>lonc not running, trying to start</h3>"; |
print $fh "<h3>lonc not running, trying to start</h3>"; |
system("$perlvar{'lonDaemons'}/lonc"); |
system("$perlvar{'lonDaemons'}/lonc"); |
Line 317 if (-e $loncfile) {
|
Line 454 if (-e $loncfile) {
|
print $fh "Seems like that did not work!<p>"; |
print $fh "Seems like that did not work!<p>"; |
$errors++; |
$errors++; |
} |
} |
|
if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") { |
|
print $fh "<p><pre>"; |
|
open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|"); |
|
while ($line=<DFH>) { |
|
print $fh "$line"; |
|
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $notices++; } |
|
}; |
|
close (DFH); |
|
print $fh "</pre>"; |
|
} |
} |
} |
|
|
$fname="$perlvar{'lonDaemons'}/logs/lonc.log"; |
$fname="$perlvar{'lonDaemons'}/logs/lonc.log"; |
Line 342 if (-e "$perlvar{'lonDaemons'}/logs/lonn
|
Line 490 if (-e "$perlvar{'lonDaemons'}/logs/lonn
|
open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|"); |
open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|"); |
while ($line=<DFH>) { |
while ($line=<DFH>) { |
print $fh "$line"; |
print $fh "$line"; |
if ($line=~/Delayed/) { $warnings++; } |
|
if ($line=~/giving up/) { $warnings++; } |
|
if ($line=~/FAILED/) { $errors++; } |
|
}; |
}; |
close (DFH); |
close (DFH); |
} |
} |