version 1.3, 2000/02/08 17:34:24
|
version 1.21, 2001/04/24 15:55:13
|
Line 9
|
Line 9
|
# |
# |
# 7/14,7/15,7/19,7/21,7/22,11/18, |
# 7/14,7/15,7/19,7/21,7/22,11/18, |
# 2/8 Gerd Kortemeyer |
# 2/8 Gerd Kortemeyer |
|
# 12/6/2000,12/8 Scott Harrison |
|
# 12/23 Gerd Kortemeyer |
|
# 1/10/2001, 2/12/, 2/26, 3/15, 04/11, 04/21 Scott Harrison |
|
|
use IO::File; |
use IO::File; |
use IO::Socket; |
use IO::Socket; |
|
|
|
my $qflag=0; |
|
if (@ARGV) { |
|
my $arg=shift @ARGV; |
|
$qflag=1 if $arg eq 'quick'; |
|
} |
|
|
# -------------------------------------------------- Non-critical communication |
# -------------------------------------------------- Non-critical communication |
sub reply { |
sub reply { |
my ($cmd,$server)=@_; |
my ($cmd,$server)=@_; |
Line 43 ENDERROUT
|
Line 52 ENDERROUT
|
|
|
# ================================================================ Main Program |
# ================================================================ Main Program |
|
|
|
|
# ------------------------------------------------------------ Read access.conf |
# ------------------------------------------------------------ Read access.conf |
{ |
{ |
my $config=IO::File->new("/etc/httpd/conf/access.conf"); |
my $config=IO::File->new("/etc/httpd/conf/access.conf"); |
Line 54 ENDERROUT
|
Line 62 ENDERROUT
|
$perlvar{$varname}=$varvalue; |
$perlvar{$varname}=$varvalue; |
} |
} |
} |
} |
|
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 |
# ------------------------------------------------------------- Read hosts file |
Line 110 print $fh (<<ENDHEADERS);
|
Line 145 print $fh (<<ENDHEADERS);
|
<ol> |
<ol> |
<li><a href="#configuration">Configuration</a> |
<li><a href="#configuration">Configuration</a> |
<li><a href="#machine">Machine Information</a> |
<li><a href="#machine">Machine Information</a> |
|
<li><a href="#tmp">Temporary Files</a> |
|
<li><a href="#tokens">Session Tokens</a> |
<li><a href="#httpd">httpd</a> |
<li><a href="#httpd">httpd</a> |
|
<li><a href="#lonsql">lonsql</a> |
<li><a href="#lond">lond</a> |
<li><a href="#lond">lond</a> |
<li><a href="#lonc">lonc</a> |
<li><a href="#lonc">lonc</a> |
<li><a href="#lonnet">lonnet</a> |
<li><a href="#lonnet">lonnet</a> |
Line 153 close (LOADAVGH);
|
Line 191 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++; |
$warnings++; |
Line 171 while ($line=<DFH>) {
|
Line 209 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 $fh "</pre>"; |
&errout($fh); |
&errout($fh); |
|
|
|
# --------------------------------------------------------------- clean out tmp |
|
print $fh '<hr><a name="tmp"><h2>Temporary Files</h2>'; |
|
$cleaned=0; |
|
while ($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'}) { |
|
$cleaned++; |
|
unlink("$fname"); |
|
} |
|
|
|
} |
|
print $fh "Cleaned up ".$cleaned." files."; |
|
|
|
# ------------------------------------------------------------ clean out lonIDs |
|
print $fh '<hr><a name="tokens"><h2>Session Tokens</h2>'; |
|
$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 $fname<br>"; |
|
unlink("$fname"); |
|
} else { |
|
$active++; |
|
} |
|
|
|
} |
|
print $fh "<p>Cleaned up ".$cleaned." stale session token(s)."; |
|
print $fh "<h3>$active open session(s)</h3>"; |
|
|
# ----------------------------------------------------------------------- httpd |
# ----------------------------------------------------------------------- httpd |
|
|
print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>'; |
print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>'; |
Line 200 while ($line=<DFH>) {
|
Line 280 while ($line=<DFH>) {
|
close (DFH); |
close (DFH); |
print $fh "</pre>"; |
print $fh "</pre>"; |
&errout($fh); |
&errout($fh); |
|
|
|
|
|
# ---------------------------------------------------------------------- lonsql |
|
# |
|
# Do not run for now |
|
# |
|
if ($perlvar{'lonRole'} eq "library") { |
|
|
|
print $fh '<hr><a name="lonsql"><h2>lonsql</h2><h3>Log</h3><pre>'; |
|
|
|
if (-e "$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 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors"); |
|
sleep 120 unless $qflag; |
|
if (-e $lonsqlfile) { |
|
print $fh "Seems like it started ...<p>"; |
|
my $lfh=IO::File->new("$lonsqlfile"); |
|
my $lonsqlpid=<$lfh>; |
|
chomp($lonsqlpid); |
|
sleep 30 unless $qflag; |
|
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 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors"); |
|
sleep 120 unless $qflag; |
|
} |
|
} 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>'; |
Line 209 open (DFH,"tail -n100 $perlvar{'lonDaemo
|
Line 378 open (DFH,"tail -n100 $perlvar{'lonDaemo
|
while ($line=<DFH>) { |
while ($line=<DFH>) { |
print $fh "$line"; |
print $fh "$line"; |
if ($line=~/INFO/) { $notices++; } |
if ($line=~/INFO/) { $notices++; } |
if ($line=~/WARNING/) { $warnings++; } |
if ($line=~/WARNING/) { $notices++; } |
if ($line=~/CRITICAL/) { $errors++; } |
if ($line=~/CRITICAL/) { $warnings++; } |
}; |
}; |
close (DFH); |
close (DFH); |
} |
} |
Line 218 print $fh "</pre>";
|
Line 387 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++; |
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); |
|
$restartflag=1; |
|
} |
|
else { |
|
# Possibility #2: there is a live process that is not responding |
|
# for an unknown reason |
|
# Solution: kill parent and children processes, remove .pid and restart |
|
`killall -9 lond`; |
|
unlink($londfile); |
|
$restartflag=1; |
|
} |
|
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( |
sleep 120; |
"$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); |
|
sleep 120 unless $qflag; |
if (-e $londfile) { |
if (-e $londfile) { |
print $fh "Seems like it started ...<p>"; |
print $fh "Seems like it started ...<p>"; |
my $lfh=IO::File->new("$londfile"); |
my $lfh=IO::File->new("$londfile"); |
my $londpid=<$lfh>; |
my $londpid=<$lfh>; |
chomp($londpid); |
chomp($londpid); |
sleep 30; |
sleep 30 unless $qflag; |
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>"; |
} 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>"; |
print $fh "Give it one more try ...<p>"; |
print $fh "Give it one more try ...<p>"; |
system("$perlvar{'lonDaemons'}/lond"); |
system( |
sleep 120; |
"$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); |
|
sleep 120 unless $qflag; |
} |
} |
} else { |
} else { |
print $fh "Seems like that did not work!<p>"; |
print $fh "Seems like that did not work!<p>"; |
Line 257 if (-e $londfile) {
|
Line 448 if (-e $londfile) {
|
open (DFH,"tail -n100 $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=~/INFO/) { $notices++; } |
if ($line=~/WARNING/) { $notices++; } |
if ($line=~/WARNING/) { $warnings++; } |
if ($line=~/CRITICAL/) { $notices++; } |
if ($line=~/CRITICAL/) { $errors++; } |
|
}; |
}; |
close (DFH); |
close (DFH); |
print $fh "</pre>"; |
print $fh "</pre>"; |
Line 290 open (DFH,"tail -n100 $perlvar{'lonDaemo
|
Line 480 open (DFH,"tail -n100 $perlvar{'lonDaemo
|
while ($line=<DFH>) { |
while ($line=<DFH>) { |
print $fh "$line"; |
print $fh "$line"; |
if ($line=~/INFO/) { $notices++; } |
if ($line=~/INFO/) { $notices++; } |
if ($line=~/WARNING/) { $warnings++; } |
if ($line=~/WARNING/) { $notices++; } |
if ($line=~/CRITICAL/) { $errors++; } |
if ($line=~/CRITICAL/) { $warnings++; } |
}; |
}; |
close (DFH); |
close (DFH); |
} |
} |
Line 299 print $fh "</pre>";
|
Line 489 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 306 if (-e $loncfile) {
|
Line 497 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++; |
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); |
|
$restartflag=1; |
|
} |
|
else { |
|
# Possibility #2: there is a live process that is not responding |
|
# for an unknown reason |
|
# Solution: kill parent and children processes, remove .pid and restart |
|
`killall -9 lonc`; |
|
unlink($loncfile); |
|
$restartflag=1; |
|
} |
|
print $fh |
|
"<h3>Deciding to clean up stale .pid file and restart lonc</h3>"; |
} |
} |
} else { |
} |
|
if ($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( |
sleep 120; |
"$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors"); |
|
sleep 120 unless $qflag; |
if (-e $loncfile) { |
if (-e $loncfile) { |
print $fh "Seems like it started ...<p>"; |
print $fh "Seems like it started ...<p>"; |
my $lfh=IO::File->new("$loncfile"); |
my $lfh=IO::File->new("$loncfile"); |
my $loncpid=<$lfh>; |
my $loncpid=<$lfh>; |
chomp($loncpid); |
chomp($loncpid); |
sleep 30; |
sleep 30 unless $qflag; |
if (kill 0 => $loncpid) { |
if (kill 0 => $loncpid) { |
print $fh "<h3>lonc at pid $loncpid responding</h3>"; |
print $fh "<h3>lonc at pid $loncpid responding</h3>"; |
} 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>"; |
print $fh "Give it one more try ...<p>"; |
print $fh "Give it one more try ...<p>"; |
system("$perlvar{'lonDaemons'}/lonc"); |
system( |
sleep 120; |
"$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors"); |
|
sleep 120 unless $qflag; |
} |
} |
} else { |
} else { |
print $fh "Seems like that did not work!<p>"; |
print $fh "Seems like that did not work!<p>"; |
Line 339 if (-e $loncfile) {
|
Line 551 if (-e $loncfile) {
|
open (DFH,"tail -n100 $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=~/INFO/) { $notices++; } |
if ($line=~/WARNING/) { $notices++; } |
if ($line=~/WARNING/) { $warnings++; } |
if ($line=~/CRITICAL/) { $notices++; } |
if ($line=~/CRITICAL/) { $errors++; } |
|
}; |
}; |
close (DFH); |
close (DFH); |
print $fh "</pre>"; |
print $fh "</pre>"; |
Line 374 while ($line=<DFH>) {
|
Line 585 while ($line=<DFH>) {
|
}; |
}; |
close (DFH); |
close (DFH); |
} |
} |
print $fh "</pre><h3>Perm Log</h3>"; |
print $fh "</pre><h3>Perm Log</h3><pre>"; |
|
|
if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") { |
if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") { |
open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|"); |
open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|"); |
Line 469 if ($totalcount>200) {
|
Line 680 if ($totalcount>200) {
|
$emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; |
$emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; |
$subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; |
$subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; |
system( |
system( |
"metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html"); |
"metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html") |
|
unless $qflag; |
} |
} |
1; |
1; |
|
|