version 1.18, 2001/03/15 13:20:46
|
version 1.45, 2004/01/09 17:43:40
|
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 |
# Dec 00 Scott Harrison |
|
# 12/23 Gerd Kortemeyer |
# 12/23 Gerd Kortemeyer |
# 02/12/2001 Scott Harrison |
# YEAR=2001 |
|
# 09/04,09/06,11/26 Gerd Kortemeyer |
|
|
|
$|=1; |
|
|
|
use lib '/home/httpd/lib/perl/'; |
|
use LONCAPA::Configuration; |
|
|
use IO::File; |
use IO::File; |
use IO::Socket; |
use IO::Socket; |
Line 44 sub errout {
|
Line 49 sub errout {
|
ENDERROUT |
ENDERROUT |
} |
} |
|
|
# ================================================================ Main Program |
sub start_daemon { |
|
my ($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 ...<p>"; |
|
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!<p>"; |
|
$errors++; |
|
return 0; |
|
} |
|
|
# ------------------------------------------------------------ Read access.conf |
sub checkon_daemon { |
{ |
my ($fh,$daemon,$maxsize,$sendusr1)=@_; |
my $config=IO::File->new("/etc/httpd/conf/access.conf"); |
|
|
|
while (my $configline=<$config>) { |
print $fh '<hr><a name="'.$daemon.'"><h2>'.$daemon.'</h2><h3>Log</h3><pre>'; |
if ($configline =~ /PerlSetVar/) { |
printf("%-10s ",$daemon); |
my ($dummy,$varname,$varvalue)=split(/\s+/,$configline); |
if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){ |
$perlvar{$varname}=$varvalue; |
open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.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 $pidfile="$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 $fh "<h3>$daemon at pid $daemonpid responding"; |
|
if ($sendusr1) { print $fh ", sending USR1"; } |
|
print $fh "</h3>"; |
|
if ($sendusr1) { kill USR1 => $daemonpid; } |
|
$restartflag=0; |
|
print "running\n"; |
|
} else { |
|
$errors++; |
|
print $fh "<h3>$daemon at pid $daemonpid not responding</h3>"; |
|
$restartflag=1; |
|
print $fh "<h3>Decided to clean up stale .pid file and restart $daemon</h3>"; |
|
} |
|
} |
|
if ($restartflag==1) { |
|
$simplestatus{$daemon}='off'; |
|
$errors++; |
|
print $fh '<br><font color="red">Killall '.$daemon.': '. |
|
`killall $daemon 2>&1`.' - '; |
|
sleep 2; |
|
print $fh unlink($pidfile).' - '. |
|
`killall -9 $daemon 2>&1`. |
|
'</font><br>'; |
|
print $fh "<h3>$daemon not running, trying to start</h3>"; |
|
|
|
if (&start_daemon($fh,$daemon,$pidfile)) { |
|
print $fh "<h3>$daemon at pid $daemonpid responding</h3>"; |
|
$simplestatus{$daemon}='restarted'; |
|
print "started\n"; |
|
} else { |
|
$errors++; |
|
print $fh "<h3>$daemon at pid $daemonpid not responding</h3>"; |
|
print $fh "Give it one more try ...<p>"; |
|
print " "; |
|
if (&start_daemon($fh,$daemon,$pidfile)) { |
|
print $fh "<h3>$daemon at pid $daemonpid responding</h3>"; |
|
$simplestatus{$daemon}='restarted'; |
|
print "started\n"; |
|
} else { |
|
print " failed\n"; |
|
$simplestatus{$daemon}='failed'; |
|
$errors++; $errors++; |
|
print $fh "<h3>$daemon at pid $daemonpid not responding</h3>"; |
|
print $fh "Unable to start $daemon<p>"; |
|
} |
|
} |
|
|
|
if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){ |
|
print $fh "<p><pre>"; |
|
open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|"); |
|
while ($line=<DFH>) { |
|
print $fh "$line"; |
|
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $notices++; } |
|
}; |
|
close (DFH); |
|
print $fh "</pre>"; |
|
} |
|
} |
|
|
|
$fname="$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 $fh "Rotating logs ...<p>"; |
|
rename("$fname.2","$fname.3"); |
|
rename("$fname.1","$fname.2"); |
|
rename("$fname","$fname.1"); |
} |
} |
|
|
|
&errout($fh); |
} |
} |
|
# ================================================================ Main Program |
|
|
|
# --------------------------------- Read loncapa_apache.conf and loncapa.conf |
|
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 |
# --------------------------------------- Make sure that LON-CAPA is configured |
# I only test for one thing here (lonHostID). This is just a safeguard. |
# I only test for one thing here (lonHostID). This is just a safeguard. |
if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) { |
if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) { |
print("Unconfigured machine.\n"); |
print("Unconfigured machine.\n"); |
$emailto=$perlvar{'lonSysEMail'}; |
$emailto=$perlvar{'lonSysEMail'}; |
$hostname=`/bin/hostname`; |
$hostname=`/bin/hostname`; |
chop $hostname; |
chop $hostname; |
$hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell |
$hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell |
$subj="LON: Unconfigured machine $hostname"; |
$subj="LON: Unconfigured machine $hostname"; |
system("echo 'Unconfigured machine $hostname.' |\ |
system("echo 'Unconfigured machine $hostname.' |\ |
mailto $emailto -s '$subj' > /dev/null"); |
mailto $emailto -s '$subj' > /dev/null"); |
exit 1; |
exit 1; |
} |
} |
Line 75 if ('{[[[[lonHostID]]]]}' eq $perlvar{'l
|
Line 202 if ('{[[[[lonHostID]]]]}' eq $perlvar{'l
|
# ----------------------------- Make sure this process is running from user=www |
# ----------------------------- Make sure this process is running from user=www |
my $wwwid=getpwnam('www'); |
my $wwwid=getpwnam('www'); |
if ($wwwid!=$<) { |
if ($wwwid!=$<) { |
print("User ID mismatch. This program must be run as user 'www'\n"); |
print("User ID mismatch. This program must be run as user 'www'\n"); |
$emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; |
$emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; |
$subj="LON: $perlvar{'lonHostID'} User ID mismatch"; |
$subj="LON: $perlvar{'lonHostID'} User ID mismatch"; |
system("echo 'User ID mismatch. loncron must be run as user www.' |\ |
system("echo 'User ID mismatch. loncron must be run as user www.' |\ |
mailto $emailto -s '$subj' > /dev/null"); |
mailto $emailto -s '$subj' > /dev/null"); |
exit 1; |
exit 1; |
} |
} |
|
|
# ------------------------------------------------------------- Read hosts file |
# ------------------------------------------------------------- Read hosts file |
{ |
{ |
my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab"); |
my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab"); |
|
|
while (my $configline=<$config>) { |
while (my $configline=<$config>) { |
my ($id,$domain,$role,$name,$ip)=split(/:/,$configline); |
my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline); |
$hostname{$id}=$name; |
if ($id && $domain && $role && $name && $ip) { |
$hostdom{$id}=$domain; |
$hostname{$id}=$name; |
$hostrole{$id}=$role; |
$hostdom{$id}=$domain; |
$hostip{$id}=$ip; |
$hostip{$id}=$ip; |
if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) { |
$hostrole{$id}=$role; |
$libserv{$id}=$name; |
if ($domdescr) { $domaindescription{$domain}=$domdescr; } |
} |
if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) { |
|
$libserv{$id}=$name; |
|
} |
|
} else { |
|
if ($configline) { |
|
# &logthis("Skipping hosts.tab line -$configline-"); |
|
} |
|
} |
} |
} |
} |
} |
|
|
# ------------------------------------------------------ Read spare server file |
# ------------------------------------------------------ Read spare server file |
{ |
{ |
my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab"); |
my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab"); |
|
|
while (my $configline=<$config>) { |
while (my $configline=<$config>) { |
chomp($configline); |
chomp($configline); |
if (($configline) && ($configline ne $perlvar{'lonHostID'})) { |
if (($configline) && ($configline ne $perlvar{'lonHostID'})) { |
$spareid{$configline}=1; |
$spareid{$configline}=1; |
} |
} |
} |
} |
} |
} |
|
|
Line 123 $now=time;
|
Line 257 $now=time;
|
$date=localtime($now); |
$date=localtime($now); |
|
|
{ |
{ |
my $fh=IO::File->new(">$statusdir/newstatus.html"); |
my $fh=IO::File->new(">$statusdir/newstatus.html"); |
|
my %simplestatus=(); |
print $fh (<<ENDHEADERS); |
|
|
print $fh (<<ENDHEADERS); |
<html> |
<html> |
<head> |
<head> |
<title>LON Status Report $perlvar{'lonHostID'}</title> |
<title>LON Status Report $perlvar{'lonHostID'}</title> |
Line 143 print $fh (<<ENDHEADERS);
|
Line 278 print $fh (<<ENDHEADERS);
|
<li><a href="#lonsql">lonsql</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="#lonhttpd">lonhttpd</a> |
<li><a href="#lonnet">lonnet</a> |
<li><a href="#lonnet">lonnet</a> |
<li><a href="#connections">Connections</a> |
<li><a href="#connections">Connections</a> |
<li><a href="#delayed">Delayed Messages</a> |
<li><a href="#delayed">Delayed Messages</a> |
Line 155 print $fh (<<ENDHEADERS);
|
Line 291 print $fh (<<ENDHEADERS);
|
<table border=2> |
<table border=2> |
ENDHEADERS |
ENDHEADERS |
|
|
foreach $varname (keys %perlvar) { |
foreach $varname (sort(keys(%perlvar))) { |
print $fh "<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n"; |
print $fh "<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n"; |
} |
} |
print $fh "</table><h3>Hosts</h3><table border=2>"; |
print $fh "</table><h3>Hosts</h3><table border=2>"; |
foreach $id (keys %hostname) { |
foreach $id (sort(keys(%hostname))) { |
print $fh |
print $fh |
"<tr><td>$id</td><td>$hostdom{$id}</td><td>$hostrole{$id}</td>"; |
"<tr><td>$id</td><td>$hostdom{$id}</td><td>$hostrole{$id}</td>"; |
print $fh "<td>$hostname{$id}</td><td>$hostip{$id}</td></tr>\n"; |
print $fh "<td>$hostname{$id}</td><td>$hostip{$id}</td></tr>\n"; |
} |
} |
print $fh "</table><h3>Spare Hosts</h3><ol>"; |
print $fh "</table><h3>Spare Hosts</h3><ol>"; |
foreach $id (keys %spareid) { |
foreach $id (sort(keys(%spareid))) { |
print $fh "<li>$id\n"; |
print $fh "<li>$id\n"; |
} |
} |
|
|
print $fh "</ol>\n"; |
print $fh "</ol>\n"; |
|
|
# --------------------------------------------------------------------- Machine |
# --------------------------------------------------------------------- Machine |
|
|
|
print $fh '<hr><a name="machine"><h2>Machine Information</h2>'; |
|
print $fh "<h3>loadavg</h3>"; |
|
|
|
open (LOADAVGH,"/proc/loadavg"); |
|
$loadavg=<LOADAVGH>; |
|
close (LOADAVGH); |
|
|
|
print $fh "<tt>$loadavg</tt>"; |
|
|
|
@parts=split(/\s+/,$loadavg); |
|
if ($parts[1]>4.0) { |
|
$errors++; |
|
} elsif ($parts[1]>2.0) { |
|
$warnings++; |
|
} elsif ($parts[1]>1.0) { |
|
$notices++; |
|
} |
|
|
print $fh '<hr><a name="machine"><h2>Machine Information</h2>'; |
print $fh "<h3>df</h3>"; |
print $fh "<h3>loadavg</h3>"; |
print $fh "<pre>"; |
|
|
open (LOADAVGH,"/proc/loadavg"); |
open (DFH,"df|"); |
$loadavg=<LOADAVGH>; |
while ($line=<DFH>) { |
close (LOADAVGH); |
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 "</pre>"; |
|
|
print $fh "<tt>$loadavg</tt>"; |
|
|
|
@parts=split(/\s+/,$loadavg); |
print $fh "<h3>ps</h3>"; |
if ($parts[1]>4.0) { |
print $fh "<pre>"; |
$errors++; |
$psproc=0; |
} elsif ($parts[1]>2.0) { |
|
$warnings++; |
open (PSH,"ps -aux --cols 140 |"); |
} elsif ($parts[1]>1.0) { |
while ($line=<PSH>) { |
$notices++; |
print $fh "$line"; |
} |
$psproc++; |
|
} |
print $fh "<h3>df</h3>"; |
close (PSH); |
print $fh "<pre>"; |
print $fh "</pre>"; |
|
|
open (DFH,"df|"); |
if ($psproc>200) { $notices++; } |
while ($line=<DFH>) { |
if ($psproc>250) { $notices++; } |
print $fh "$line"; |
|
@parts=split(/\s+/,$line); |
&errout($fh); |
$usage=$parts[4]; |
|
$usage=~s/\W//g; |
|
if ($usage>90) { |
|
$warnings++; |
|
} elsif ($usage>80) { |
|
$warnings++; |
|
} elsif ($usage>60) { |
|
$notices++; |
|
} |
|
if ($usage>95) { $warnings++; $warnings++ } |
|
} |
|
close (DFH); |
|
print $fh "</pre>"; |
|
&errout($fh); |
|
|
|
# --------------------------------------------------------------- clean out tmp |
# --------------------------------------------------------------- clean out tmp |
print $fh '<hr><a name="tmp"><h2>Temporary Files</h2>'; |
print $fh '<hr><a name="tmp"><h2>Temporary Files</h2>'; |
$cleaned=0; |
$cleaned=0; |
while ($fname=<$perlvar{'lonDaemons'}/tmp/*>) { |
$old=0; |
my ($dev,$ino,$mode,$nlink, |
while ($fname=<$perlvar{'lonDaemons'}/tmp/*>) { |
$uid,$gid,$rdev,$size, |
my ($dev,$ino,$mode,$nlink, |
$atime,$mtime,$ctime, |
$uid,$gid,$rdev,$size, |
$blksize,$blocks)=stat($fname); |
$atime,$mtime,$ctime, |
$now=time; |
$blksize,$blocks)=stat($fname); |
$since=$now-$mtime; |
$now=time; |
if ($since>$perlvar{'lonExpire'}) { |
$since=$now-$mtime; |
$cleaned++; |
if ($since>$perlvar{'lonExpire'}) { |
unlink("$fname"); |
$line=''; |
} |
if (open(PROBE,$fname)) { |
|
$line=<PROBE>; |
|
close(PROBE); |
|
} |
|
unless ($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 |
# ------------------------------------------------------------ clean out lonIDs |
print $fh '<hr><a name="tokens"><h2>Session Tokens</h2>'; |
print $fh '<hr><a name="tokens"><h2>Session Tokens</h2>'; |
$cleaned=0; |
$cleaned=0; |
$active=0; |
$active=0; |
while ($fname=<$perlvar{'lonIDsDir'}/*>) { |
while ($fname=<$perlvar{'lonIDsDir'}/*>) { |
my ($dev,$ino,$mode,$nlink, |
my ($dev,$ino,$mode,$nlink, |
$uid,$gid,$rdev,$size, |
$uid,$gid,$rdev,$size, |
$atime,$mtime,$ctime, |
$atime,$mtime,$ctime, |
$blksize,$blocks)=stat($fname); |
$blksize,$blocks)=stat($fname); |
$now=time; |
$now=time; |
$since=$now-$mtime; |
$since=$now-$mtime; |
if ($since>$perlvar{'lonExpire'}) { |
if ($since>$perlvar{'lonExpire'}) { |
$cleaned++; |
$cleaned++; |
print $fh "Unlinking $fname<br>"; |
print $fh "Unlinking $fname<br>"; |
unlink("$fname"); |
unlink("$fname"); |
} else { |
} else { |
$active++; |
$active++; |
} |
} |
|
|
} |
} |
print $fh "<p>Cleaned up ".$cleaned." stale session token(s)."; |
print $fh "<p>Cleaned up ".$cleaned." stale session token(s)."; |
print $fh "<h3>$active open session(s)</h3>"; |
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 $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>'; |
|
|
|
open (DFH,"tail -n25 /etc/httpd/logs/access_log|"); |
|
while ($line=<DFH>) { print $fh "$line" }; |
|
close (DFH); |
|
|
|
print $fh "</pre><h3>Error Log</h3><pre>"; |
|
|
open (DFH,"tail -n40 /etc/httpd/logs/access_log|"); |
open (DFH,"tail -n25 /etc/httpd/logs/error_log|"); |
while ($line=<DFH>) { print $fh "$line" }; |
while ($line=<DFH>) { |
close (DFH); |
print $fh "$line"; |
|
if ($line=~/\[error\]/) { $notices++; } |
print $fh "</pre><h3>Error Log</h3><pre>"; |
}; |
|
close (DFH); |
open (DFH,"tail -n50 /etc/httpd/logs/error_log|"); |
print $fh "</pre>"; |
while ($line=<DFH>) { |
&errout($fh); |
print $fh "$line"; |
|
if ($line=~/\[error\]/) { $notices++; } |
|
}; |
|
close (DFH); |
|
print $fh "</pre>"; |
|
&errout($fh); |
|
|
|
|
|
# ---------------------------------------------------------------------- lonsql |
# ---------------------------------------------------------------------- lonsql |
# |
|
# Do not run for now |
|
# |
|
if ($perlvar{'lonRole'} eq "library") { |
|
|
|
print $fh '<hr><a name="lonsql"><h2>lonsql</h2><h3>Log</h3><pre>'; |
&checkon_daemon($fh,'lonsql',200000); |
|
|
if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){ |
# ------------------------------------------------------------------------ lond |
open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|"); |
|
|
&checkon_daemon($fh,'lond',40000,1); |
|
|
|
# ------------------------------------------------------------------------ lonc |
|
|
|
&checkon_daemon($fh,'lonc',40000,1); |
|
|
|
# -------------------------------------------------------------------- lonhttpd |
|
|
|
&checkon_daemon($fh,'lonhttpd',40000); |
|
|
|
# ---------------------------------------------------------------------- lonnet |
|
|
|
print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>'; |
|
print "checking logs\n"; |
|
if (-e "$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=~/INFO/) { $notices++; } |
|
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $warnings++; } |
|
}; |
}; |
close (DFH); |
close (DFH); |
} |
} |
print $fh "</pre>"; |
print $fh "</pre><h3>Perm Log</h3><pre>"; |
|
|
my $lonsqlfile="$perlvar{'lonDaemons'}/logs/lonsql.pid"; |
if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") { |
|
open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|"); |
if (-e $lonsqlfile) { |
while ($line=<DFH>) { |
my $lfh=IO::File->new("$lonsqlfile"); |
print $fh "$line"; |
my $lonsqlpid=<$lfh>; |
}; |
chomp($lonsqlpid); |
close (DFH); |
if (kill 0 => $lonsqlpid) { |
} else { print $fh "No perm log\n" } |
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; |
|
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 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors"); |
|
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"; |
$fname="$perlvar{'lonDaemons'}/logs/lonnet.log"; |
|
|
my ($dev,$ino,$mode,$nlink, |
my ($dev,$ino,$mode,$nlink, |
$uid,$gid,$rdev,$size, |
$uid,$gid,$rdev,$size, |
Line 359 if ($perlvar{'lonRole'} eq "library") {
|
Line 492 if ($perlvar{'lonRole'} eq "library") {
|
rename("$fname","$fname.1"); |
rename("$fname","$fname.1"); |
} |
} |
|
|
|
print $fh "</pre>"; |
&errout($fh); |
&errout($fh); |
} |
# ----------------------------------------------------------------- Connections |
# ------------------------------------------------------------------------ lond |
|
|
|
print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>'; |
print $fh '<hr><a name="connections"><h2>Connections</h2>'; |
|
print "testing connections\n"; |
|
print $fh "<table border=2>"; |
|
foreach $tryserver (sort(keys(%hostname))) { |
|
print("."); |
|
$answer=reply("pong",$tryserver); |
|
if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { |
|
$result="<b>ok</b>"; |
|
} else { |
|
$result=$answer; |
|
$warnings++; |
|
if ($answer eq 'con_lost') { $warnings++; } |
|
} |
|
if ($answer =~ /con_lost/) { print(" $tryserver down\n"); } |
|
print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n"; |
|
|
if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){ |
} |
open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|"); |
print $fh "</table>"; |
while ($line=<DFH>) { |
|
print $fh "$line"; |
|
if ($line=~/INFO/) { $notices++; } |
|
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $warnings++; } |
|
}; |
|
close (DFH); |
|
} |
|
print $fh "</pre>"; |
|
|
|
my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid"; |
|
|
|
my $restartflag=1; |
|
if (-e $londfile) { |
|
my $lfh=IO::File->new("$londfile"); |
|
my $londpid=<$lfh>; |
|
chomp($londpid); |
|
if (kill 0 => $londpid) { |
|
print $fh "<h3>lond at pid $londpid responding</h3>"; |
|
$restartflag=0; |
|
} else { |
|
$errors++; |
|
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 $fh |
|
"<h3>Deciding to clean up stale .pid file and restart lond</h3>"; |
|
} |
|
} |
|
if ($restartflag==1) { |
|
$errors++; |
|
print $fh "<h3>lond not running, trying to start</h3>"; |
|
system( |
|
"$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); |
|
sleep 120; |
|
if (-e $londfile) { |
|
print $fh "Seems like it started ...<p>"; |
|
my $lfh=IO::File->new("$londfile"); |
|
my $londpid=<$lfh>; |
|
chomp($londpid); |
|
sleep 30; |
|
if (kill 0 => $londpid) { |
|
print $fh "<h3>lond at pid $londpid responding</h3>"; |
|
} else { |
|
$errors++; $errors++; |
|
print $fh "<h3>lond at pid $londpid not responding</h3>"; |
|
print $fh "Give it one more try ...<p>"; |
|
system( |
|
"$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); |
|
sleep 120; |
|
} |
|
} else { |
|
print $fh "Seems like that did not work!<p>"; |
|
$errors++; |
|
} |
|
if (-e "$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"; |
&errout($fh); |
|
# ------------------------------------------------------------ Delayed messages |
|
|
my ($dev,$ino,$mode,$nlink, |
print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>'; |
$uid,$gid,$rdev,$size, |
print "checking buffers\n"; |
$atime,$mtime,$ctime, |
|
$blksize,$blocks)=stat($fname); |
|
|
|
if ($size>40000) { |
print $fh '<h3>Scanning Permanent Log</h3>'; |
print $fh "Rotating logs ...<p>"; |
|
rename("$fname.2","$fname.3"); |
|
rename("$fname.1","$fname.2"); |
|
rename("$fname","$fname.1"); |
|
} |
|
|
|
&errout($fh); |
$unsend=0; |
# ------------------------------------------------------------------------ lonc |
{ |
|
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 "<b>Failed: $time, $dserv, $dcmd</b><br>"; |
|
$warnings++; |
|
} |
|
if ($sdf eq 'S') { $unsend--; } |
|
if ($sdf eq 'D') { $unsend++; } |
|
} |
|
} |
|
print $fh "Total unsend messages: <b>$unsend</b><p>\n"; |
|
$warnings=$warnings+5*$unsend; |
|
|
print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>'; |
if ($unsend) { $simplestatus{'unsend'}=$unsend; } |
|
print $fh "<h3>Outgoing Buffer</h3>"; |
|
|
if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){ |
open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|"); |
open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|"); |
|
while ($line=<DFH>) { |
|
print $fh "$line"; |
|
if ($line=~/INFO/) { $notices++; } |
|
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $warnings++; } |
|
}; |
|
close (DFH); |
|
} |
|
print $fh "</pre>"; |
|
|
|
my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid"; |
|
|
|
$restartflag=1; |
|
if (-e $loncfile) { |
|
my $lfh=IO::File->new("$loncfile"); |
|
my $loncpid=<$lfh>; |
|
chomp($loncpid); |
|
if (kill 0 => $loncpid) { |
|
print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>"; |
|
kill USR1 => $loncpid; |
|
$restartflag=0; |
|
} else { |
|
$errors++; |
|
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 $fh |
|
"<h3>Deciding to clean up stale .pid file and restart lonc</h3>"; |
|
} |
|
} |
|
if ($restartflag==1) { |
|
$errors++; |
|
print $fh "<h3>lonc not running, trying to start</h3>"; |
|
system( |
|
"$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors"); |
|
sleep 120; |
|
if (-e $loncfile) { |
|
print $fh "Seems like it started ...<p>"; |
|
my $lfh=IO::File->new("$loncfile"); |
|
my $loncpid=<$lfh>; |
|
chomp($loncpid); |
|
sleep 30; |
|
if (kill 0 => $loncpid) { |
|
print $fh "<h3>lonc at pid $loncpid responding</h3>"; |
|
} else { |
|
$errors++; $errors++; |
|
print $fh "<h3>lonc at pid $loncpid not responding</h3>"; |
|
print $fh "Give it one more try ...<p>"; |
|
system( |
|
"$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors"); |
|
sleep 120; |
|
} |
|
} else { |
|
print $fh "Seems like that did not work!<p>"; |
|
$errors++; |
|
} |
|
if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") { |
|
print $fh "<p><pre>"; |
|
open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|"); |
|
while ($line=<DFH>) { |
while ($line=<DFH>) { |
print $fh "$line"; |
print $fh "$line<br>"; |
if ($line=~/WARNING/) { $notices++; } |
|
if ($line=~/CRITICAL/) { $notices++; } |
|
}; |
}; |
close (DFH); |
close (DFH); |
print $fh "</pre>"; |
|
} |
|
} |
|
|
|
$fname="$perlvar{'lonDaemons'}/logs/lonc.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); |
|
# ---------------------------------------------------------------------- lonnet |
|
|
|
print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>'; |
|
if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){ |
|
open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|"); |
|
while ($line=<DFH>) { |
|
print $fh "$line"; |
|
}; |
|
close (DFH); |
|
} |
|
print $fh "</pre><h3>Perm Log</h3><pre>"; |
|
|
|
if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") { |
|
open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|"); |
|
while ($line=<DFH>) { |
|
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 $fh "Rotating logs ...<p>"; |
|
rename("$fname.2","$fname.3"); |
|
rename("$fname.1","$fname.2"); |
|
rename("$fname","$fname.1"); |
|
} |
|
|
|
print $fh "</pre>"; |
|
&errout($fh); |
|
# ----------------------------------------------------------------- Connections |
|
|
|
print $fh '<hr><a name="connections"><h2>Connections</h2>'; |
|
|
|
print $fh "<table border=2>"; |
|
foreach $tryserver (keys %hostname) { |
|
|
|
$answer=reply("pong",$tryserver); |
|
if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { |
|
$result="<b>ok</b>"; |
|
} else { |
|
$result=$answer; |
|
$warnings++; |
|
if ($answer eq 'con_lost') { $warnings++; } |
|
} |
|
print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n"; |
|
|
|
} |
|
print $fh "</table>"; |
|
|
|
&errout($fh); |
|
# ------------------------------------------------------------ Delayed messages |
|
|
|
print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>'; |
|
|
|
print $fh '<h3>Scanning Permanent Log</h3>'; |
|
|
|
$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 "<b>Failed: $time, $dserv, $dcmd</b><br>"; |
|
$warnings++; |
|
} |
|
if ($sdf eq 'S') { $unsend--; } |
|
if ($sdf eq 'D') { $unsend++; } |
|
} |
|
} |
|
print $fh "Total unsend messages: <b>$unsend</b><p>\n"; |
|
$warnings=$warnings+5*$unsend; |
|
|
|
print $fh "<h3>Outgoing Buffer</h3>"; |
|
|
|
open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|"); |
|
while ($line=<DFH>) { |
|
print $fh "$line<br>"; |
|
}; |
|
close (DFH); |
|
|
|
# ------------------------------------------------------------------------- End |
# ------------------------------------------------------------------------- End |
print $fh "<a name=errcount>\n"; |
print $fh "<a name=errcount>\n"; |
$totalcount=$notices+4*$warnings+100*$errors; |
$totalcount=$notices+4*$warnings+100*$errors; |
&errout($fh); |
&errout($fh); |
print $fh "<h1>Total Error Count: $totalcount</h1>"; |
print $fh "<h1>Total Error Count: $totalcount</h1>"; |
$now=time; |
$now=time; |
$date=localtime($now); |
$date=localtime($now); |
print $fh "<hr>$date ($now)</body></html>\n"; |
print $fh "<hr>$date ($now)</body></html>\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","$statusdir/index.html"); |
rename ("$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) { |
if ($totalcount>200) { |
$emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; |
print "sending mail\n"; |
$subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; |
$emailto="$perlvar{'lonAdmEMail'}"; |
system( |
if ($totalcount>1000) { |
"metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html"); |
$emailto.=",$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; |
1; |
|
|