version 1.48, 2004/05/11 20:19:46
|
version 1.95, 2011/11/14 17:27:34
|
Line 32 use strict;
|
Line 32 use strict;
|
|
|
use lib '/home/httpd/lib/perl/'; |
use lib '/home/httpd/lib/perl/'; |
use LONCAPA::Configuration; |
use LONCAPA::Configuration; |
|
use LONCAPA; |
|
use Apache::lonnet; |
|
use Apache::loncommon; |
|
|
use IO::File; |
use IO::File; |
use IO::Socket; |
use IO::Socket; |
use HTML::Entities; |
use HTML::Entities; |
|
use Getopt::Long; |
#globals |
#globals |
use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount); |
use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount); |
|
|
my $statusdir="/home/httpd/html/lon-status"; |
my $statusdir="/home/httpd/html/lon-status"; |
|
|
|
|
# -------------------------------------------------- Non-critical communication |
|
sub reply { |
|
my ($cmd,$server)=@_; |
|
my $peerfile="$perlvar{'lonSockDir'}/$server"; |
|
my $client=IO::Socket::UNIX->new(Peer =>"$peerfile", |
|
Type => SOCK_STREAM, |
|
Timeout => 10) |
|
or return "con_lost"; |
|
print $client "$cmd\n"; |
|
my $answer=<$client>; |
|
chomp($answer); |
|
if (!$answer) { $answer="con_lost"; } |
|
return $answer; |
|
} |
|
|
|
# --------------------------------------------------------- Output error status |
# --------------------------------------------------------- Output error status |
|
|
sub log { |
sub log { |
Line 75 sub errout {
|
Line 64 sub errout {
|
ENDERROUT |
ENDERROUT |
} |
} |
|
|
|
sub rotate_logfile { |
|
my ($file,$fh,$description) = @_; |
|
my $size=(stat($file))[7]; |
|
if ($size>40000) { |
|
&log($fh,"<p>Rotating $description ...</p>"); |
|
rename("$file.2","$file.3"); |
|
rename("$file.1","$file.2"); |
|
rename("$file","$file.1"); |
|
} |
|
} |
|
|
sub start_daemon { |
sub start_daemon { |
my ($fh,$daemon,$pidfile) = @_; |
my ($fh,$daemon,$pidfile,$args) = @_; |
my $progname=$daemon; |
my $progname=$daemon; |
if ($daemon eq 'lonc' && $ARGV[0] eq 'new') { |
if ($daemon eq 'lonc') { |
$progname='loncnew'; |
$progname='loncnew'; |
print "new "; |
|
} |
} |
system("$perlvar{'lonDaemons'}/$progname 2>>$perlvar{'lonDaemons'}/logs/${daemon}_errors"); |
my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors"; |
sleep 2; |
&rotate_logfile($error_fname,$fh,'error logs'); |
|
if ($daemon eq 'lonc') { |
|
&clean_sockets($fh); |
|
} |
|
system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors"); |
|
sleep 1; |
if (-e $pidfile) { |
if (-e $pidfile) { |
&log($fh,"<p>Seems like it started ...</p>"); |
&log($fh,"<p>Seems like it started ...</p>"); |
my $lfh=IO::File->new("$pidfile"); |
my $lfh=IO::File->new("$pidfile"); |
my $daemonpid=<$lfh>; |
my $daemonpid=<$lfh>; |
chomp($daemonpid); |
chomp($daemonpid); |
sleep 2; |
if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) { |
if (kill 0 => $daemonpid) { |
|
return 1; |
return 1; |
} else { |
} else { |
return 0; |
return 0; |
Line 102 sub start_daemon {
|
Line 105 sub start_daemon {
|
} |
} |
|
|
sub checkon_daemon { |
sub checkon_daemon { |
my ($fh,$daemon,$maxsize,$sendusr1)=@_; |
my ($fh,$daemon,$maxsize,$send,$args)=@_; |
|
|
|
my $result; |
&log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>'); |
&log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>'); |
printf("%-10s ",$daemon); |
printf("%-15s ",$daemon); |
if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){ |
if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){ |
open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|"); |
open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|"); |
while (my $line=<DFH>) { |
while (my $line=<DFH>) { |
Line 126 sub checkon_daemon {
|
Line 130 sub checkon_daemon {
|
my $lfh=IO::File->new("$pidfile"); |
my $lfh=IO::File->new("$pidfile"); |
$daemonpid=<$lfh>; |
$daemonpid=<$lfh>; |
chomp($daemonpid); |
chomp($daemonpid); |
if (kill 0 => $daemonpid) { |
if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) { |
&log($fh,"<h3>$daemon at pid $daemonpid responding"); |
&log($fh,"<h3>$daemon at pid $daemonpid responding"); |
if ($sendusr1) { &log($fh,", sending USR1"); } |
if ($send) { &log($fh,", sending $send"); } |
&log($fh,"</h3>"); |
&log($fh,"</h3>"); |
if ($sendusr1) { kill USR1 => $daemonpid; } |
if ($send eq 'USR1') { kill USR1 => $daemonpid; } |
|
if ($send eq 'USR2') { kill USR2 => $daemonpid; } |
$restartflag=0; |
$restartflag=0; |
print "running\n"; |
if ($send eq 'USR2') { |
|
$result = 'reloaded'; |
|
print "reloaded\n"; |
|
} else { |
|
$result = 'running'; |
|
print "running\n"; |
|
} |
} else { |
} else { |
$errors++; |
$errors++; |
&log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>"); |
&log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>"); |
Line 143 sub checkon_daemon {
|
Line 154 sub checkon_daemon {
|
if ($restartflag==1) { |
if ($restartflag==1) { |
$simplestatus{$daemon}='off'; |
$simplestatus{$daemon}='off'; |
$errors++; |
$errors++; |
|
my $kadaemon=$daemon; |
|
if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; } |
&log($fh,'<br><font color="red">Killall '.$daemon.': '. |
&log($fh,'<br><font color="red">Killall '.$daemon.': '. |
`killall $daemon 2>&1`.' - '); |
`killall $kadaemon 2>&1`.' - '); |
sleep 2; |
sleep 1; |
&log($fh,unlink($pidfile).' - '. |
&log($fh,unlink($pidfile).' - '. |
`killall -9 $daemon 2>&1`. |
`killall -9 $kadaemon 2>&1`. |
'</font><br>'); |
'</font><br>'); |
&log($fh,"<h3>$daemon not running, trying to start</h3>"); |
&log($fh,"<h3>$daemon not running, trying to start</h3>"); |
|
|
if (&start_daemon($fh,$daemon,$pidfile)) { |
if (&start_daemon($fh,$daemon,$pidfile,$args)) { |
&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>"); |
&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>"); |
$simplestatus{$daemon}='restarted'; |
$simplestatus{$daemon}='restarted'; |
|
$result = 'started'; |
print "started\n"; |
print "started\n"; |
} else { |
} else { |
$errors++; |
$errors++; |
&log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>"); |
&log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>"); |
&log($fh,"<p>Give it one more try ...</p>"); |
&log($fh,"<p>Give it one more try ...</p>"); |
print " "; |
print " "; |
if (&start_daemon($fh,$daemon,$pidfile)) { |
if (&start_daemon($fh,$daemon,$pidfile,$args)) { |
&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>"); |
&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>"); |
$simplestatus{$daemon}='restarted'; |
$simplestatus{$daemon}='restarted'; |
|
$result = 'started'; |
print "started\n"; |
print "started\n"; |
} else { |
} else { |
|
$result = 'failed'; |
print " failed\n"; |
print " failed\n"; |
$simplestatus{$daemon}='failed'; |
$simplestatus{$daemon}='failed'; |
$errors++; $errors++; |
$errors++; $errors++; |
Line 187 sub checkon_daemon {
|
Line 203 sub checkon_daemon {
|
} |
} |
|
|
my $fname="$perlvar{'lonDaemons'}/logs/$daemon.log"; |
my $fname="$perlvar{'lonDaemons'}/logs/$daemon.log"; |
|
&rotate_logfile($fname,$fh,'logs'); |
my ($dev,$ino,$mode,$nlink, |
|
$uid,$gid,$rdev,$size, |
|
$atime,$mtime,$ctime, |
|
$blksize,$blocks)=stat($fname); |
|
|
|
if ($size>$maxsize) { |
|
&log($fh,"<p>Rotating logs ...</p>"); |
|
rename("$fname.2","$fname.3"); |
|
rename("$fname.1","$fname.2"); |
|
rename("$fname","$fname.1"); |
|
} |
|
|
|
&errout($fh); |
&errout($fh); |
|
return $result; |
} |
} |
|
|
# --------------------------------------------------------------------- Machine |
# --------------------------------------------------------------------- Machine |
Line 251 sub log_machine_info {
|
Line 257 sub log_machine_info {
|
&log($fh,"<pre>"); |
&log($fh,"<pre>"); |
my $psproc=0; |
my $psproc=0; |
|
|
open (PSH,"ps -aux --cols 140 |"); |
open (PSH,"ps aux --cols 140 |"); |
while (my $line=<PSH>) { |
while (my $line=<PSH>) { |
&log($fh,&encode_entities($line,'<>&"')); |
&log($fh,&encode_entities($line,'<>&"')); |
$psproc++; |
$psproc++; |
Line 262 sub log_machine_info {
|
Line 268 sub log_machine_info {
|
if ($psproc>200) { $notices++; } |
if ($psproc>200) { $notices++; } |
if ($psproc>250) { $notices++; } |
if ($psproc>250) { $notices++; } |
|
|
|
&log($fh,"<h3>distprobe</h3>"); |
|
&log($fh,"<pre>"); |
|
open(DSH,"$perlvar{'lonDaemons'}/distprobe |"); |
|
while (my $line=<DSH>) { |
|
&log($fh,&encode_entities($line,'<>&"')); |
|
$psproc++; |
|
} |
|
close(DSH); |
|
&log($fh,"</pre>"); |
|
|
&errout($fh); |
&errout($fh); |
} |
} |
|
|
sub start_logging { |
sub start_logging { |
my ($hostdom,$hostrole,$hostname,$spareid)=@_; |
|
my $fh=IO::File->new(">$statusdir/newstatus.html"); |
my $fh=IO::File->new(">$statusdir/newstatus.html"); |
my %simplestatus=(); |
my %simplestatus=(); |
my $now=time; |
my $now=time; |
Line 291 sub start_logging {
|
Line 306 sub start_logging {
|
<li><a href="#lonsql">lonsql</a></li> |
<li><a href="#lonsql">lonsql</a></li> |
<li><a href="#lond">lond</a></li> |
<li><a href="#lond">lond</a></li> |
<li><a href="#lonc">lonc</a></li> |
<li><a href="#lonc">lonc</a></li> |
<li><a href="#lonhttpd">lonhttpd</a></li> |
|
<li><a href="#lonnet">lonnet</a></li> |
<li><a href="#lonnet">lonnet</a></li> |
<li><a href="#connections">Connections</a></li> |
<li><a href="#connections">Connections</a></li> |
<li><a href="#delayed">Delayed Messages</a></li> |
<li><a href="#delayed">Delayed Messages</a></li> |
Line 309 ENDHEADERS
|
Line 323 ENDHEADERS
|
&encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n"); |
&encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n"); |
} |
} |
&log($fh,"</table><h3>Hosts</h3><table border='2'>"); |
&log($fh,"</table><h3>Hosts</h3><table border='2'>"); |
foreach my $id (sort(keys(%{$hostname}))) { |
my %hostname = &Apache::lonnet::all_hostnames(); |
|
foreach my $id (sort(keys(%hostname))) { |
|
my $role = (&Apache::lonnet::is_library($id) ? 'library' |
|
: 'access'); |
&log($fh, |
&log($fh, |
"<tr><td>$id</td><td>".$hostdom->{$id}. |
"<tr><td>$id</td><td>".&Apache::lonnet::host_domain($id). |
"</td><td>".$hostrole->{$id}. |
"</td><td>".$role. |
"</td><td>".$hostname->{$id}."</td></tr>\n"); |
"</td><td>".&Apache::lonnet::hostname($id)."</td></tr>\n"); |
} |
} |
&log($fh,"</table><h3>Spare Hosts</h3><ol>"); |
&log($fh,"</table><h3>Spare Hosts</h3><ul>"); |
foreach my $id (sort(keys(%{$spareid}))) { |
foreach my $type (sort(keys(%Apache::lonnet::spareid))) { |
&log($fh,"<li>$id\n</li>"); |
&log($fh,"<li>$type\n<ol>"); |
|
foreach my $id (@{ $Apache::lonnet::spareid{$type} }) { |
|
&log($fh,"<li>$id</li>\n"); |
|
} |
|
&log($fh,"</ol>\n</li>\n"); |
} |
} |
&log($fh,"</ol>\n"); |
&log($fh,"</ul>\n"); |
return $fh; |
return $fh; |
} |
} |
|
|
Line 327 ENDHEADERS
|
Line 348 ENDHEADERS
|
sub clean_tmp { |
sub clean_tmp { |
my ($fh)=@_; |
my ($fh)=@_; |
&log($fh,'<hr /><a name="tmp" /><h2>Temporary Files</h2>'); |
&log($fh,'<hr /><a name="tmp" /><h2>Temporary Files</h2>'); |
my $cleaned=0; |
my ($cleaned,$old,$removed) = (0,0,0); |
my $old=0; |
my %errors = ( |
while (my $fname=<$perlvar{'lonDaemons'}/tmp/*>) { |
dir => [], |
my ($dev,$ino,$mode,$nlink, |
file => [], |
$uid,$gid,$rdev,$size, |
failopen => [], |
$atime,$mtime,$ctime, |
); |
$blksize,$blocks)=stat($fname); |
my %error_titles = ( |
my $now=time; |
dir => 'failed to remove empty directory:', |
my $since=$now-$mtime; |
file => 'failed to unlike stale file', |
if ($since>$perlvar{'lonExpire'}) { |
failopen => 'failed to open file or directory' |
my $line=''; |
); |
if (open(PROBE,$fname)) { |
($cleaned,$old,$removed) = &recursive_clean_tmp('',$cleaned,$old,$removed,\%errors); |
$line=<PROBE>; |
&log($fh,"Cleaned up: ".$cleaned." files; removed: $removed empty directories; (found: $old old checkout tokens)"); |
close(PROBE); |
foreach my $key (sort(keys(%errors))) { |
} |
if (ref($errors{$key}) eq 'ARRAY') { |
unless ($line=~/^CHECKOUTTOKEN\&/) { |
if (@{$errors{$key}} > 0) { |
$cleaned++; |
&log($fh,"Error during cleanup ($error_titles{$key}):<ul><li>". |
unlink("$fname"); |
join('</li><li><tt>',@{$errors{$key}}).'</tt></li></ul><br />'); |
} else { |
} |
if ($since>365*$perlvar{'lonExpire'}) { |
} |
$cleaned++; |
} |
unlink("$fname"); |
} |
} else { $old++; } |
|
} |
sub recursive_clean_tmp { |
} |
my ($subdir,$cleaned,$old,$removed,$errors) = @_; |
|
my $base = "$perlvar{'lonDaemons'}/tmp"; |
|
my $path = $base; |
|
next if ($subdir =~ m{\.\./}); |
|
next unless (ref($errors) eq 'HASH'); |
|
unless ($subdir eq '') { |
|
$path .= '/'.$subdir; |
|
} |
|
if (opendir(my $dh,"$path")) { |
|
while (my $file = readdir($dh)) { |
|
next if ($file =~ /^\.\.?$/); |
|
my $fname = "$path/$file"; |
|
if (-d $fname) { |
|
my $innerdir; |
|
if ($subdir eq '') { |
|
$innerdir = $file; |
|
} else { |
|
$innerdir = $subdir.'/'.$file; |
|
} |
|
($cleaned,$old,$removed) = |
|
&recursive_clean_tmp($innerdir,$cleaned,$old,$removed,$errors); |
|
my @doms = &Apache::lonnet::current_machine_domains(); |
|
|
|
if (open(my $dirhandle,$fname)) { |
|
unless (($innerdir eq 'helprequests') || |
|
(($innerdir =~ /^addcourse/) && ($innerdir !~ m{/\d+$}))) { |
|
my @contents = grep {!/^\.\.?$/} readdir($dirhandle); |
|
join('&&',@contents)."\n"; |
|
if (scalar(grep {!/^\.\.?$/} readdir($dirhandle)) == 0) { |
|
closedir($dirhandle); |
|
if ($fname =~ m{^\Q$perlvar{'lonDaemons'}\E/tmp/}) { |
|
if (rmdir($fname)) { |
|
$removed ++; |
|
} elsif (ref($errors->{dir}) eq 'ARRAY') { |
|
push(@{$errors->{dir}},$fname); |
|
} |
|
} |
|
} |
|
} else { |
|
closedir($dirhandle); |
|
} |
|
} |
|
} else { |
|
my ($dev,$ino,$mode,$nlink, |
|
$uid,$gid,$rdev,$size, |
|
$atime,$mtime,$ctime, |
|
$blksize,$blocks)=stat($fname); |
|
my $now=time; |
|
my $since=$now-$mtime; |
|
if ($since>$perlvar{'lonExpire'}) { |
|
if ($subdir eq '') { |
|
my $line=''; |
|
if ($fname =~ /\.db$/) { |
|
if (unlink($fname)) { |
|
$cleaned++; |
|
} elsif (ref($errors->{file}) eq 'ARRAY') { |
|
push(@{$errors->{file}},$fname); |
|
} |
|
} elsif (open(PROBE,$fname)) { |
|
my $line=''; |
|
$line=<PROBE>; |
|
close(PROBE); |
|
if ($line=~/^CHECKOUTTOKEN\&/) { |
|
if ($since>365*$perlvar{'lonExpire'}) { |
|
if (unlink($fname)) { |
|
$cleaned++; |
|
} elsif (ref($errors->{file}) eq 'ARRAY') { |
|
push(@{$errors->{file}},$fname); |
|
} |
|
} else { |
|
$old++; |
|
} |
|
} else { |
|
if (unlink($fname)) { |
|
$cleaned++; |
|
} elsif (ref($errors->{file}) eq 'ARRAY') { |
|
push(@{$errors->{file}},$fname); |
|
} |
|
} |
|
} elsif (ref($errors->{failopen}) eq 'ARRAY') { |
|
push(@{$errors->{failopen}},$fname); |
|
} |
|
} else { |
|
if (unlink($fname)) { |
|
$cleaned++; |
|
} elsif (ref($errors->{file}) eq 'ARRAY') { |
|
push(@{$errors->{file}},$fname); |
|
} |
|
} |
|
} |
|
} |
|
} |
|
closedir($dh); |
|
} elsif (ref($errors->{failopen}) eq 'ARRAY') { |
|
push(@{$errors->{failopen}},$path); |
} |
} |
&log($fh,"Cleaned up ".$cleaned." files (".$old." old checkout tokens)."); |
return ($cleaned,$old,$removed); |
} |
} |
|
|
# ------------------------------------------------------------ clean out lonIDs |
# ------------------------------------------------------------ clean out lonIDs |
Line 381 sub clean_lonIDs {
|
Line 496 sub clean_lonIDs {
|
&log($fh,"<h3>$active open session(s)</h3>"); |
&log($fh,"<h3>$active open session(s)</h3>"); |
} |
} |
|
|
|
# ----------------------------------------------------------- clean out sockets |
|
sub clean_sockets { |
|
my ($fh)=@_; |
|
my $cleaned=0; |
|
opendir(SOCKETS,$perlvar{'lonSockDir'}); |
|
while (my $fname=readdir(SOCKETS)) { |
|
next if (-d $fname |
|
|| $fname=~/(mysqlsock|maximasock|rsock|\Q$perlvar{'lonSockDir'}\E)/); |
|
$cleaned++; |
|
&log($fh,"Unlinking $fname<br />"); |
|
unlink("/home/httpd/sockets/$fname"); |
|
} |
|
&log($fh,"<p>Cleaned up ".$cleaned." stale sockets.</p>"); |
|
} |
|
|
|
|
# ----------------------------------------------------------------------- httpd |
# ----------------------------------------------------------------------- httpd |
sub check_httpd_logs { |
sub check_httpd_logs { |
my ($fh)=@_; |
my ($fh)=@_; |
&log($fh,'<hr /><a name="httpd" /><h2>httpd</h2><h3>Access Log</h3><pre>'); |
if (open(PIPE,"./lchttpdlogs|")) { |
|
while (my $line=<PIPE>) { |
open (DFH,"tail -n25 /etc/httpd/logs/access_log|"); |
&log($fh,$line); |
while (my $line=<DFH>) { &log($fh,&encode_entities($line,'<>&"')) }; |
if ($line=~/\[error\]/) { $notices++; } |
close (DFH); |
} |
|
close(PIPE); |
&log($fh,"</pre><h3>Error Log</h3><pre>"); |
|
|
|
open (DFH,"tail -n25 /etc/httpd/logs/error_log|"); |
|
while (my $line=<DFH>) { |
|
&log($fh,"$line"); |
|
if ($line=~/\[error\]/) { $notices++; } |
|
} |
} |
close (DFH); |
|
&log($fh,"</pre>"); |
|
&errout($fh); |
&errout($fh); |
} |
} |
|
|
Line 427 sub rotate_lonnet_logs {
|
Line 549 sub rotate_lonnet_logs {
|
} else { &log($fh,"No perm log\n") } |
} else { &log($fh,"No perm log\n") } |
|
|
my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log"; |
my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log"; |
|
&rotate_logfile($fname,$fh,'lonnet log'); |
my ($dev,$ino,$mode,$nlink, |
|
$uid,$gid,$rdev,$size, |
|
$atime,$mtime,$ctime, |
|
$blksize,$blocks)=stat($fname); |
|
|
|
if ($size>40000) { |
|
&log($fh,"<p>Rotating logs ...</p>"); |
|
rename("$fname.2","$fname.3"); |
|
rename("$fname.1","$fname.2"); |
|
rename("$fname","$fname.1"); |
|
} |
|
|
|
&log($fh,"</pre>"); |
&log($fh,"</pre>"); |
&errout($fh); |
&errout($fh); |
} |
} |
|
|
|
sub rotate_other_logs { |
|
my ($fh) = @_; |
|
my %logs = ( |
|
autoenroll => 'Auto Enroll log', |
|
autocreate => 'Create Course log', |
|
searchcat => 'Search Cataloguing log', |
|
autoupdate => 'Auto Update log', |
|
refreshcourseids_db => 'Refresh CourseIDs db log', |
|
); |
|
foreach my $item (keys(%logs)) { |
|
my $fname=$perlvar{'lonDaemons'}.'/logs/'.$item.'.log'; |
|
&rotate_logfile($fname,$fh,$logs{$item}); |
|
} |
|
} |
|
|
# ----------------------------------------------------------------- Connections |
# ----------------------------------------------------------------- Connections |
sub test_connections { |
sub test_connections { |
my ($fh,$hostname)=@_; |
my ($fh)=@_; |
&log($fh,'<hr /><a name="connections" /><h2>Connections</h2>'); |
&log($fh,'<hr /><a name="connections" /><h2>Connections</h2>'); |
print "testing connections\n"; |
print "testing connections\n"; |
&log($fh,"<table border='2'>"); |
&log($fh,"<table border='2'>"); |
foreach my $tryserver (sort(keys(%{$hostname}))) { |
my ($good,$bad)=(0,0); |
|
my %hostname = &Apache::lonnet::all_hostnames(); |
|
foreach my $tryserver (sort(keys(%hostname))) { |
print("."); |
print("."); |
my $result; |
my $result; |
my $answer=reply("pong",$tryserver); |
my $answer=&Apache::lonnet::reply("ping",$tryserver); |
if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { |
if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { |
$result="<b>ok</b>"; |
$result="<b>ok</b>"; |
|
$good++; |
} else { |
} else { |
$result=$answer; |
$result=$answer; |
$warnings++; |
$warnings++; |
if ($answer eq 'con_lost') { $warnings++; } |
if ($answer eq 'con_lost') { |
|
$bad++; |
|
$warnings++; |
|
} else { |
|
$good++; #self connection |
|
} |
} |
} |
if ($answer =~ /con_lost/) { print(" $tryserver down\n"); } |
if ($answer =~ /con_lost/) { print(" $tryserver down\n"); } |
&log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n"); |
&log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n"); |
} |
} |
&log($fh,"</table>"); |
&log($fh,"</table>"); |
|
print "\n$good good, $bad bad connections\n"; |
&errout($fh); |
&errout($fh); |
} |
} |
|
|
Line 493 sub check_delayed_msg {
|
Line 627 sub check_delayed_msg {
|
} |
} |
|
|
&log($fh,"<p>Total unsend messages: <b>$unsend</b></p>\n"); |
&log($fh,"<p>Total unsend messages: <b>$unsend</b></p>\n"); |
$warnings=$warnings+5*$unsend; |
if ($unsend > 0) { |
|
$warnings=$warnings+5*$unsend; |
|
} |
|
|
if ($unsend) { $simplestatus{'unsend'}=$unsend; } |
if ($unsend) { $simplestatus{'unsend'}=$unsend; } |
&log($fh,"<h3>Outgoing Buffer</h3>\n<pre>"); |
&log($fh,"<h3>Outgoing Buffer</h3>\n<pre>"); |
|
# list directory with delayed messages and remember offline servers |
|
my %servers=(); |
open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|"); |
open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|"); |
while (my $line=<DFH>) { |
while (my $line=<DFH>) { |
|
my ($server)=($line=~/\.(\w+)$/); |
|
if ($server) { $servers{$server}=1; } |
&log($fh,&encode_entities($line,'<>&"')); |
&log($fh,&encode_entities($line,'<>&"')); |
} |
} |
&log($fh,"</pre>\n"); |
&log($fh,"</pre>\n"); |
close (DFH); |
close (DFH); |
|
my %hostname = &Apache::lonnet::all_hostnames(); |
|
my $numhosts = scalar(keys(%hostname)); |
|
# pong to all servers that have delayed messages |
|
# this will trigger a reverse connection, which should flush the buffers |
|
foreach my $tryserver (sort(keys(%servers))) { |
|
if ($hostname{$tryserver} || !$numhosts) { |
|
my $answer; |
|
eval { |
|
local $SIG{ ALRM } = sub { die "TIMEOUT" }; |
|
alarm(20); |
|
$answer = &Apache::lonnet::reply("pong",$tryserver); |
|
alarm(0); |
|
}; |
|
if ($@ && $@ =~ m/TIMEOUT/) { |
|
&log($fh,"Attempted pong to $tryserver timed out<br />"); |
|
print "time out while contacting: $tryserver for pong\n"; |
|
} else { |
|
&log($fh,"Pong to $tryserver: $answer<br />"); |
|
} |
|
} else { |
|
&log($fh,"$tryserver has delayed messages, but is not part of the cluster -- skipping 'Pong'.<br />"); |
|
} |
|
} |
} |
} |
|
|
sub finish_logging { |
sub finish_logging { |
Line 525 sub finish_logging {
|
Line 687 sub finish_logging {
|
} |
} |
|
|
sub log_simplestatus { |
sub log_simplestatus { |
rename ("$statusdir/newstatus.html","$statusdir/index.html"); |
rename("$statusdir/newstatus.html","$statusdir/index.html"); |
|
|
my $sfh=IO::File->new(">$statusdir/loncron_simple.txt"); |
my $sfh=IO::File->new(">$statusdir/loncron_simple.txt"); |
foreach (keys %simplestatus) { |
foreach (keys %simplestatus) { |
Line 535 sub log_simplestatus {
|
Line 697 sub log_simplestatus {
|
$sfh->close(); |
$sfh->close(); |
} |
} |
|
|
|
sub write_loncaparevs { |
|
print "Retrieving LON-CAPA version information\n"; |
|
if (open(my $fh,">$perlvar{'lonTabDir'}/loncaparevs.tab")) { |
|
my %hostname = &Apache::lonnet::all_hostnames(); |
|
foreach my $id (sort(keys(%hostname))) { |
|
if ($id ne '') { |
|
my $loncaparev; |
|
eval { |
|
local $SIG{ ALRM } = sub { die "TIMEOUT" }; |
|
alarm(10); |
|
$loncaparev = |
|
&Apache::lonnet::get_server_loncaparev('',$id,1,'loncron'); |
|
alarm(0); |
|
}; |
|
if ($@ && $@ =~ m/TIMEOUT/) { |
|
print "time out while contacting lonHost: $id for version\n"; |
|
} |
|
if ($loncaparev =~ /^[\w.\-]+$/) { |
|
print $fh $id.':'.$loncaparev."\n"; |
|
} |
|
} |
|
} |
|
close($fh); |
|
} |
|
return; |
|
} |
|
|
|
sub write_serverhomeIDs { |
|
print "Retrieving LON-CAPA lonHostID information\n"; |
|
if (open(my $fh,">$perlvar{'lonTabDir'}/serverhomeIDs.tab")) { |
|
my %name_to_host = &Apache::lonnet::all_names(); |
|
foreach my $name (sort(keys(%name_to_host))) { |
|
if ($name ne '') { |
|
if (ref($name_to_host{$name}) eq 'ARRAY') { |
|
my $serverhomeID; |
|
eval { |
|
local $SIG{ ALRM } = sub { die "TIMEOUT" }; |
|
alarm(10); |
|
$serverhomeID = |
|
&Apache::lonnet::get_server_homeID($name,1,'loncron'); |
|
alarm(0); |
|
}; |
|
if ($@ && $@ =~ m/TIMEOUT/) { |
|
print "Time out while contacting server: $name\n"; |
|
} |
|
if ($serverhomeID ne '') { |
|
print $fh $name.':'.$serverhomeID."\n"; |
|
} else { |
|
print $fh $name.':'.$name_to_host{$name}->[0]."\n"; |
|
} |
|
} |
|
} |
|
} |
|
close($fh); |
|
} |
|
return; |
|
} |
|
|
sub send_mail { |
sub send_mail { |
print "sending mail\n"; |
print "sending mail\n"; |
my $emailto="$perlvar{'lonAdmEMail'}"; |
my $defdom = $perlvar{'lonDefDomain'}; |
if ($totalcount>1000) { |
my $origmail = $perlvar{'lonAdmEMail'}; |
|
my $emailto = &Apache::loncommon::build_recipient_list(undef, |
|
'lonstatusmail',$defdom,$origmail); |
|
if ($totalcount>2500) { |
$emailto.=",$perlvar{'lonSysEMail'}"; |
$emailto.=",$perlvar{'lonSysEMail'}"; |
} |
} |
my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; |
my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices"; |
system("metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html"); |
|
|
my $result=system("metasend -b -S 4000000 -t $emailto -s '$subj' -f $statusdir/index.html -m text/html >& /dev/null"); |
|
if ($result != 0) { |
|
$result=system("mail -s '$subj' $emailto < $statusdir/index.html"); |
|
} |
|
} |
|
|
|
sub usage { |
|
print(<<USAGE); |
|
loncron - housekeeping program that checks up on various parts of Lon-CAPA |
|
|
|
Options: |
|
--help Display |
|
--noemail Do not send the status email |
|
--justcheckconnections Only check the current status of the lonc/d |
|
connections, do not send emails do not |
|
check if the daemons are running, do not |
|
generate lon-status |
|
--justcheckdaemons Only check that all of the Lon-CAPA daemons are |
|
running, do not send emails do not |
|
check the lonc/d connections, do not |
|
generate lon-status |
|
--justreload Only tell the daemons to reload the config files, |
|
do not send emails do not |
|
check if the daemons are running, do not |
|
generate lon-status |
|
|
|
USAGE |
} |
} |
|
|
# ================================================================ Main Program |
# ================================================================ Main Program |
sub main () { |
sub main () { |
|
my ($help,$justcheckdaemons,$noemail,$justcheckconnections, |
|
$justreload); |
|
&GetOptions("help" => \$help, |
|
"justcheckdaemons" => \$justcheckdaemons, |
|
"noemail" => \$noemail, |
|
"justcheckconnections" => \$justcheckconnections, |
|
"justreload" => \$justreload |
|
); |
|
if ($help) { &usage(); return; } |
# --------------------------------- Read loncapa_apache.conf and loncapa.conf |
# --------------------------------- Read loncapa_apache.conf and loncapa.conf |
my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf'); |
my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf'); |
%perlvar=%{$perlvarref}; |
%perlvar=%{$perlvarref}; |
undef $perlvarref; |
undef $perlvarref; |
delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed |
delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed |
delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed |
delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed |
|
chdir($perlvar{'lonDaemons'}); |
# --------------------------------------- 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'}) { |
Line 579 sub main () {
|
Line 838 sub main () {
|
exit 1; |
exit 1; |
} |
} |
|
|
# ------------------------------------------------------------- Read hosts file |
# -------------------------------------------- Force reload of host information |
my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab"); |
&Apache::lonnet::load_hosts_tab(1); |
|
&Apache::lonnet::load_domain_tab(1); |
my (%hostname,%hostdom,%hostrole,%spareid); |
&Apache::lonnet::get_iphost(1); |
while (my $configline=<$config>) { |
|
next if ($configline =~ /^(\#|\s*$)/); |
# ----------------------------------------- Force firewall update for lond port |
my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline); |
|
if ($id && $domain && $role && $name && $ip) { |
if ((!$justcheckdaemons) && (!$justreload)) { |
$hostname{$id}=$name; |
my $now = time; |
$hostdom{$id}=$domain; |
my $tmpfile = $perlvar{'lonDaemons'}.'/tmp/lciptables_iphost_'. |
$hostrole{$id}=$role; |
$now.$$.int(rand(10000)); |
} |
if (open(my $fh,">$tmpfile")) { |
|
my %iphosts = &Apache::lonnet::get_iphost(); |
|
foreach my $key (keys(%iphosts)) { |
|
print $fh "$key\n"; |
|
} |
|
close($fh); |
|
if (&LONCAPA::try_to_lock('/tmp/lock_lciptables')) { |
|
my $execpath = $perlvar{'lonDaemons'}.'/lciptables'; |
|
system("$execpath $tmpfile"); |
|
unlink('/tmp/lock_lciptables'); # Remove the lock file. |
|
} |
|
unlink($tmpfile); |
|
} |
} |
} |
undef $config; |
|
|
|
# ------------------------------------------------------ Read spare server file |
|
$config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab"); |
|
|
|
while (my $configline=<$config>) { |
|
chomp($configline); |
|
if (($configline) && ($configline ne $perlvar{'lonHostID'})) { |
|
$spareid{$configline}=1; |
|
} |
|
} |
|
undef $config; |
|
|
|
# ---------------------------------------------------------------- Start report |
# ---------------------------------------------------------------- Start report |
|
|
Line 611 sub main () {
|
Line 870 sub main () {
|
$warnings=0; |
$warnings=0; |
$notices=0; |
$notices=0; |
|
|
my $fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid); |
|
|
|
&log_machine_info($fh); |
|
&clean_tmp($fh); |
|
&clean_lonIDs($fh); |
|
&check_httpd_logs($fh); |
|
&rotate_lonnet_logs($fh); |
|
&checkon_daemon($fh,'lonsql',200000); |
|
&checkon_daemon($fh,'lond',40000,1); |
|
&checkon_daemon($fh,'lonc',40000,1); |
|
&checkon_daemon($fh,'lonhttpd',40000); |
|
|
|
&test_connections($fh,\%hostname); |
|
&check_delayed_msg($fh); |
|
|
|
&finish_logging($fh); |
|
&log_simplestatus(); |
|
|
|
if ($totalcount>200) { &send_mail(); } |
my $fh; |
|
if (!$justcheckdaemons && !$justcheckconnections && !$justreload) { |
|
$fh=&start_logging(); |
|
|
|
&log_machine_info($fh); |
|
&clean_tmp($fh); |
|
&clean_lonIDs($fh); |
|
&check_httpd_logs($fh); |
|
&rotate_lonnet_logs($fh); |
|
&rotate_other_logs($fh); |
|
} |
|
if (!$justcheckconnections && !$justreload) { |
|
&checkon_daemon($fh,'lonmemcached',40000); |
|
&checkon_daemon($fh,'lonsql',200000); |
|
if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') { |
|
&checkon_daemon($fh,'lond',40000,'USR2'); |
|
} |
|
&checkon_daemon($fh,'lonc',40000,'USR1'); |
|
&checkon_daemon($fh,'lonmaxima',40000); |
|
&checkon_daemon($fh,'lonr',40000); |
|
} |
|
if ($justreload) { |
|
&checkon_daemon($fh,'lond',40000,'USR2'); |
|
&checkon_daemon($fh,'lonc',40000,'USR2'); |
|
} |
|
if ($justcheckconnections) { |
|
&test_connections($fh); |
|
} |
|
if (!$justcheckdaemons && !$justcheckconnections && !$justreload) { |
|
&check_delayed_msg($fh); |
|
&finish_logging($fh); |
|
&log_simplestatus(); |
|
&write_loncaparevs(); |
|
&write_serverhomeIDs(); |
|
|
|
if ($totalcount>200 && !$noemail) { &send_mail(); } |
|
} |
} |
} |
|
|
&main(); |
&main(); |