version 1.8, 2003/07/30 21:20:16
|
version 1.12, 2003/07/31 19:55:37
|
Line 2
|
Line 2
|
$|=1; |
$|=1; |
# The LearningOnline Network with CAPA |
# The LearningOnline Network with CAPA |
# Cluster Status |
# Cluster Status |
# (Versions |
# |
# (Running loncron |
# $Id$ |
# 09/06/01 Gerd Kortemeyer) |
|
# 02/18/02,02/19/02 Gerd Kortemeyer) |
|
|
|
use lib '/home/httpd/lib/perl/'; |
use lib '/home/httpd/lib/perl/'; |
use LONCAPA::Configuration; |
use LONCAPA::Configuration; |
Line 18 my %host=();
|
Line 16 my %host=();
|
my $oneday=60*60*24; |
my $oneday=60*60*24; |
|
|
my %connectionstatus=(); |
my %connectionstatus=(); |
|
my %perlvar=(); |
|
|
|
my $mode; |
|
|
|
sub select_form { |
|
my ($def,$name,%hash) = @_; |
|
my $selectform = "<select name=\"$name\" size=\"1\">\n"; |
|
foreach (sort keys %hash) { |
|
$selectform.="<option value=\"$_\" ". |
|
($_ eq $def ? 'selected' : ''). |
|
">".$hash{$_}."</option>\n"; |
|
} |
|
$selectform.="</select>"; |
|
return $selectform; |
|
} |
|
|
|
|
sub key { |
sub key { |
my ($local,$url)=@_; |
my ($local,$url)=@_; |
Line 86 sub connected {
|
Line 100 sub connected {
|
# Actually do the query |
# Actually do the query |
# |
# |
&statuslist($local,'connecting '.$remote); |
&statuslist($local,'connecting '.$remote); |
my $reply=&request($local,$url,1800); |
my $reply=&request($local,$url,3600); |
$reply=(split("\n",$reply))[0]; |
$reply=(split("\n",$reply))[0]; |
$reply=~s/\W//g; |
$reply=~s/\W//g; |
if ($reply ne $remote) { return $reply; } |
if ($reply ne $remote) { return $reply; } |
Line 106 sub replyhash {
|
Line 120 sub replyhash {
|
return %returnhash; |
return %returnhash; |
} |
} |
|
|
# ========================================================== Show server status |
# ================================================================ Link to host |
|
|
sub otherwindow { |
sub otherwindow { |
my ($local,$url,$label)=@_; |
my ($local,$url,$label)=@_; |
return |
return |
"<a href='http://$hostname{$local}$url' target='newwin$local'>$label</a>"; |
" <a href='http://$hostname{$local}$url' target='newwin$local'>$label</a> "; |
} |
} |
|
|
sub serverstatus { |
sub login { |
|
my $local=shift; |
|
print &otherwindow($local,'/adm/login?domain='.$perlvar{'lonDefDomain'}, |
|
'Login'); |
|
} |
|
|
|
sub runloncron { |
|
my $local=shift; |
|
print &otherwindow($local,'/cgi-bin/loncron.pl','Run loncron'); |
|
} |
|
|
|
sub loncron { |
|
my $local=shift; |
|
print &otherwindow($local,'/lon-status','loncron'); |
|
} |
|
|
|
sub lonc { |
|
my $local=shift; |
|
print &otherwindow($local,'/lon-status/loncstatus.txt','lonc'); |
|
} |
|
|
|
sub lond { |
|
my $local=shift; |
|
print &otherwindow($local,'/lon-status/londstatus.txt','lond'); |
|
} |
|
|
|
sub users { |
|
my $local=shift; |
|
print &otherwindow($local,'/cgi-bin/userstatus.pl','Users'); |
|
} |
|
|
|
sub versions { |
|
my $local=shift; |
|
print &otherwindow($local,'/cgi-bin/lonversions.pl','Versions'); |
|
} |
|
|
|
sub server { |
my $local=shift; |
my $local=shift; |
print "\n<hr /><h3>$local $hostdom{$local} ($hostname{$local}; $hostrole{$local})</h3>\n"; |
print &otherwindow($local,'/server-status','Server Status'); |
|
} |
|
|
|
# ========================================================= Produce a green bar |
|
sub bar { |
|
my $parm=shift; |
|
my $number=int($parm+0.5); |
|
print "<table><tr><td bgcolor='#225522'><font color='#225522'>"; |
|
for (my $i=0;$i<$number;$i++) { |
|
print "+"; |
|
} |
|
print "</font></table>"; |
|
} |
|
|
|
# ========================================================== Show server status |
|
|
|
sub serverstatus { |
|
my ($local,$trouble)=@_; |
|
print (<<ENDHEADER); |
|
<a name="$local" /> |
|
<table width="100%" bgcolor="#225522" cellspacing="2" cellpadding="2" border="0"> |
|
<tr><td bgcolor="#BBDDBB"><font color="#225522" face="arial"><b> |
|
$local $hostdom{$local}</b> <tt>($hostname{$local}; $hostrole{$local})</tt> |
|
<br />$domaindescription{$hostdom{$local}} |
|
</font></th></tr><tr><td bgcolor="#DDDDBB"><font color="#225522"> |
|
ENDHEADER |
|
&login($local);&server($local);&users($local);&versions($local); |
|
&loncron($local);&lond($local);&lonc($local);&runloncron($local); |
|
print "</font></td></tr>"; |
|
if ($trouble) { |
|
print ("<tr><td bgcolor='#DDBBBB'><font color='#552222' size='+2'>$trouble</font></td></tr>"); |
|
} |
|
print "<tr><td bgcolor='#BBBBBB'>"; |
|
# load |
|
if (($host{$local.'_load_doomed'}>0.5) || ($mode eq 'load_doomed')) { |
|
print "<br />Load: ".$host{$local.'_load'} |
|
} |
|
# users |
|
if (($host{$local.'_users_doomed'}>10) || ($mode eq 'users_doomed')) { |
|
print "<br />Active Users: ".$host{$local.'_users'} |
|
} |
|
|
# checkrpms |
# checkrpms |
if ($host{$local.'_checkrpms'}) { |
if ($host{$local.'_checkrpms'}) { |
print "<br />RPMs: ".$host{$local.'_checkrpms'} |
print "<br />RPMs: ".$host{$local.'_checkrpms'} |
Line 125 sub serverstatus {
|
Line 216 sub serverstatus {
|
if ($host{$local.'_mysql'}) { |
if ($host{$local.'_mysql'}) { |
print "<br />MySQL Database: ".$host{$local.'_mysql'} |
print "<br />MySQL Database: ".$host{$local.'_mysql'} |
} |
} |
|
# connections |
|
if ($host{$local.'_notconnected'}) { |
|
print "<br />Not connected: "; |
|
foreach (split(/ /,$host{$local.'_notconnected'})) { |
|
if ($_) { |
|
print " <a href='#$_'>$_</a>"; |
|
} |
|
} |
|
} |
|
# errors |
|
if ($host{$local.'_errors'}) { |
|
print "<br />loncron errors: ".$host{$local.'_errors'}; |
|
} |
|
print "</td></tr></table><br />"; |
|
} |
|
|
|
# =========================================================== Doomedness sorted |
|
|
|
sub doomedness { |
|
my $crit=shift; |
|
my %alldoomed=(); |
|
my @allhosts=(); |
|
foreach (keys %host) { |
|
if ($_=~/^(\w+)\_$crit$/) { |
|
if ($host{$_}) { |
|
push (@allhosts,$1); |
|
$alldoomed{$1}=$host{$_}; |
|
} |
|
} |
|
} |
|
return sort { $alldoomed{$b} <=> $alldoomed{$a} } @allhosts; |
} |
} |
|
|
# ====================================================================== Status |
# ====================================================================== Status |
Line 163 foreach $pair (@pairs) {
|
Line 285 foreach $pair (@pairs) {
|
|
|
# ====================================================== Determine refresh rate |
# ====================================================== Determine refresh rate |
|
|
my $refresh=(($FORM{'refresh'}=~/^\d+$/)?$FORM{'refresh'}:60); |
my $refresh=(($FORM{'refresh'}=~/^\d+$/)?$FORM{'refresh'}:120); |
if ($refresh<30) { $refresh=30; } |
if ($refresh<30) { $refresh=30; } |
my $starttime=time; |
my $starttime=time; |
|
|
|
# ============================================================== Determine mode |
|
|
|
my %modes=('trouble' => 'Trouble', |
|
'users_doomed' => 'Doomed: Users', |
|
'loncron_doomed' => 'Doomed: General (loncron)', |
|
'mysql_doomed' => 'Doomed: Database (mysql)', |
|
'notconnected_doomed' => 'Doomed: Connections', |
|
'checkrpms_doomed' => 'Doomed: RPMs', |
|
'load_doomed' => 'Doomed: Load', |
|
'unresponsive_doomed' => 'Doomed: Status could not be determined', |
|
'users' => 'User Report', |
|
'load' => 'Load Report', |
|
'connections' => 'Connections Matrix'); |
|
|
|
$mode=$FORM{'mode'}; |
|
unless ($modes{$mode}) { $mode='trouble'; } |
# ================================================================ Send Headers |
# ================================================================ Send Headers |
print "Content-type: text/html\n\n". |
print "Content-type: text/html\n\n". |
"<html><body bgcolor=#FFFFFF>\n"; |
"<html><body bgcolor='#FFFFFF'>\n"; |
# -------------------- Read loncapa.conf (and by default, loncapa_apache.conf). |
# -------------------- Read loncapa.conf (and by default, loncapa_apache.conf). |
my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf'); |
my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf'); |
my %perlvar=%{$perlvarref}; |
%perlvar=%{$perlvarref}; |
undef $perlvarref; # remove since sensitive and not needed |
undef $perlvarref; # remove since sensitive and not needed |
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 |
Line 195 delete $perlvar{'lonSqlAccess'}; # remov
|
Line 334 delete $perlvar{'lonSqlAccess'}; # remov
|
} |
} |
} |
} |
} |
} |
|
# ------------------------------------------------------------ Read domain file |
|
{ |
|
my $fh=IO::File->new($perlvar{'lonTabDir'}.'/domain.tab'); |
|
%domaindescription = (); |
|
%domain_auth_def = (); |
|
%domain_auth_arg_def = (); |
|
if ($fh) { |
|
while (<$fh>) { |
|
next if (/^(\#|\s*$)/); |
|
chomp; |
|
my ($domain, $domain_description, $def_auth, $def_auth_arg) |
|
= split(/:/,$_,4); |
|
$domain_auth_def{$domain}=$def_auth; |
|
$domain_auth_arg_def{$domain}=$def_auth_arg; |
|
$domaindescription{$domain}=$domain_description; |
|
} |
|
} |
|
} |
|
|
print "<h1>Cluster Status ".localtime()."</h1>"; |
print "<img src='/adm/lonIcons/lonlogos.gif' align='right' /><h1>LON-CAPA Cluster Status ".localtime()."</h1>"; |
print "<form name='prgstat'>\n". |
print "<form name='prgstat'>\n". |
"<input type='text' name='progress' value='Starting ...' size='100' /><br />". |
"<input type='text' name='progress' value='Starting ...' size='100' /><br />". |
"</form>\n";; |
"</form>\n";; |
print "<form name='status' method='post'>\n"; |
print "<form name='status' method='post'>\n"; |
|
print 'Choose next report: '.&select_form($mode,'mode',%modes).'<hr />'; |
&hidden('refresh',$refresh); |
&hidden('refresh',$refresh); |
|
|
# ==================================================== Main Loop over all Hosts |
# ==================================================== Main Loop over all Hosts |
|
|
|
my $maxusers=0; |
|
my $maxload=0; |
|
my $totalusers=0; |
|
|
foreach $local (sort keys %hostname) { |
foreach $local (sort keys %hostname) { |
|
$host{$local.'_unresponsive_doomed'}=0; |
# -- Check general status |
# -- Check general status |
&statuslist($local,'General'); |
&statuslist($local,'General'); |
my %loncron=&replyhash($local,'/lon-status/loncron_simple.txt',1200); |
my %loncron=&replyhash($local,'/lon-status/loncron_simple.txt',1200); |
if (defined($loncron{'local_error'})) { |
if (defined($loncron{'local_error'})) { |
$host{$local.'_loncron'}='Could not determine.'; |
$host{$local.'_loncron'}='Could not determine.'; |
|
$host{$local.'_unresponsive_doomed'}++; |
} else { |
} else { |
if ((time-$loncron{'time'})>$oneday) { |
if ((time-$loncron{'time'})>$oneday) { |
$host{$local.'_loncron'}='Stale.'; |
$host{$local.'_loncron'}='Stale.'; |
|
$host{$local.'_unresponsive_doomed'}++; |
} else { |
} else { |
|
$host{$local.'_loncron_doomed'}=$loncron{'notices'} |
|
+4*$loncron{'warnings'} |
|
+100*$loncron{'errors'}; |
|
$host{$local.'_errors'}=$loncron{'errors'}; |
} |
} |
} |
} |
# -- Check user status |
# -- Check user status |
Line 222 foreach $local (sort keys %hostname) {
|
Line 391 foreach $local (sort keys %hostname) {
|
my %userstatus=&replyhash($local,'/cgi-bin/userstatus.pl?simple',600); |
my %userstatus=&replyhash($local,'/cgi-bin/userstatus.pl?simple',600); |
if (defined($userstatus{'local_error'})) { |
if (defined($userstatus{'local_error'})) { |
$host{$local.'_userstatus'}='Could not determine.'; |
$host{$local.'_userstatus'}='Could not determine.'; |
|
$host{$local.'_unresponsive_doomed'}++; |
} else { |
} else { |
|
$host{$local.'_users_doomed'}=$userstatus{'Active'}; |
|
$host{$local.'_users'}=$userstatus{'Active'}; |
|
unless ($host{$local.'_users'}) { $host{$local.'_users'}=0; } |
|
if ($host{$local.'_users'}>$maxusers) { |
|
$maxusers=$host{$local.'_users'}; |
|
} |
|
$totalusers+=$host{$local.'_users'}; |
|
my ($sload,$mload,$lload)=split(/ /,$userstatus{'loadavg'}); |
|
$host{$local.'_load_doomed'}=$mload; |
|
if ($mload>$maxload) { |
|
$maxload=$mload; |
|
} |
|
$host{$local.'_load'}=$userstatus{'loadavg'}; |
} |
} |
# -- Check mysql status |
# -- Check mysql status |
&statuslist($local,'Database'); |
&statuslist($local,'Database'); |
my %mysql=&replyhash($local,'/lon-status/mysql.txt',1200); |
my %mysql=&replyhash($local,'/lon-status/mysql.txt',3600); |
if (defined($mysql{'local_error'})) { |
if (defined($mysql{'local_error'})) { |
$host{$local.'_mysql'}='Could not determine.'; |
$host{$local.'_mysql'}='Could not determine.'; |
|
$host{$local.'_unresponsive_doomed'}++; |
} else { |
} else { |
if ((time-$mysql{'time'})>(7*$oneday)) { |
if ((time-$mysql{'time'})>(7*$oneday)) { |
if ($hostrole{$local} eq 'library') { |
if ($hostrole{$local} eq 'library') { |
Line 246 foreach $local (sort keys %hostname) {
|
Line 430 foreach $local (sort keys %hostname) {
|
} |
} |
# -- Check rpm status |
# -- Check rpm status |
&statuslist($local,'RPMs'); |
&statuslist($local,'RPMs'); |
my %checkrpms=&replyhash($local,'/lon-status/checkrpms.txt',2400); |
my %checkrpms=&replyhash($local,'/lon-status/checkrpms.txt',7200); |
if (defined($checkrpms{'local_error'})) { |
if (defined($checkrpms{'local_error'})) { |
$host{$local.'_checkrpms'}='Could not determine.'; |
$host{$local.'_checkrpms'}='Could not determine.'; |
|
$host{$local.'_unresponsive_doomed'}++; |
} else { |
} else { |
if ((time-$checkrpms{'time'})>(4*$oneday)) { |
if ((time-$checkrpms{'time'})>(4*$oneday)) { |
$host{$local.'_checkrpms'}='Stale.'; |
$host{$local.'_checkrpms'}='Stale.'; |
$host{$local.'_checkrpms_doomed'}=50; |
$host{$local.'_checkrpms_doomed'}=50; |
|
$host{$local.'_unresponsive_doomed'}++; |
} elsif ($checkrpms{'status'} eq 'fail') { |
} elsif ($checkrpms{'status'} eq 'fail') { |
$host{$local.'_checkrpms'}='Could not checked RPMs.'; |
$host{$local.'_checkrpms'}='Could not checked RPMs.'; |
$host{$local.'_checkrpms_doomed'}=100; |
$host{$local.'_checkrpms_doomed'}=100; |
Line 274 foreach $local (sort keys %hostname) {
|
Line 460 foreach $local (sort keys %hostname) {
|
$host{$local.'_notconnected_doomed'}++; |
$host{$local.'_notconnected_doomed'}++; |
} |
} |
} |
} |
# Eventually, use doomed count |
|
&serverstatus($local); |
|
} |
|
|
|
# =============================================================== End Mail Loop |
# =============================================================== End Mail Loop |
|
} |
|
&statuslist('Done.'); |
|
# ====================================================================== Output |
|
if ($mode=~/\_doomed$/) { |
|
# Output by doomedness |
|
foreach (&doomedness($mode)) { |
|
&serverstatus($_); |
|
} |
|
} elsif ($mode eq 'connections') { |
|
print |
|
"<table cellspacing='3' cellpadding='3' border='0' bgcolor='#225522'>". |
|
"<tr><td bgcolor='#225522'> </td>"; |
|
foreach my $remote (sort keys %hostname) { |
|
print '<th bgcolor="#DDDDBB">'.$remote.'</th>'; |
|
} |
|
print "</tr>\n"; |
|
# connection matrix |
|
foreach my $local (sort keys %hostname) { |
|
print '<tr><th bgcolor="#DDDDBB">'.$local.'</th>'; |
|
foreach my $remote (sort keys %hostname) { |
|
if ($connectionstatus{$local.'_TO_'.$remote} eq 'not_yet') { |
|
print '<td bgcolor="#FFFFFF"> </td>'; |
|
} elsif ($connectionstatus{$local.'_TO_'.$remote} eq 'ok') { |
|
print |
|
'<td bgcolor="#BBDDBB"><font color="#225522" face="arial"><b>ok</b></td>'; |
|
} else { |
|
print |
|
'<td bgcolor="#DDBBBB"><font color="#552222" size="-2">'. |
|
$connectionstatus{$local.'_TO_'.$remote}.'<br />'; |
|
&lonc($local); &lond($remote); |
|
print '</td>'; |
|
} |
|
} |
|
print "</tr>\n"; |
|
} |
|
print "</table>"; |
|
} elsif ($mode eq 'users') { |
|
# Users |
|
if ($maxusers) { |
|
my $factor=50/$maxusers; |
|
print "<h3>Total active user(s): $totalusers</h3>". |
|
"<table cellspacing='3' cellpadding='3' border='0' bgcolor='#225522'>"; |
|
|
|
foreach $local (sort keys %hostname) { |
|
if (defined($host{$local.'_users'})) { |
|
print |
|
'<tr><th bgcolor="#BBDDBB"><font face="arial" color="#225522" size="+1">'.$local. |
|
'</font></th><td bgcolor="#DDDDBB">'; |
|
&users($local); |
|
print |
|
'</td><td bgcolor="#DDDDBB"><font face="arial" color="#225522">'. |
|
$host{$local.'_users'}.'</font></td><td bgcolor="#DDDDBB"'; |
|
&bar($factor*$host{$local.'_users'}); |
|
print "</td></tr>\n"; |
|
} |
|
} |
|
print "</table>"; |
|
} else { |
|
print "No active users logged in."; |
|
} |
|
} elsif ($mode eq 'load') { |
|
# Load |
|
if ($maxload) { |
|
my $factor=50/$maxload; |
|
print |
|
"<table cellspacing='3' cellpadding='3' border='0' bgcolor='#225522'>"; |
|
foreach $local (sort keys %hostname) { |
|
if (defined($host{$local.'_load_doomed'})) { |
|
print |
|
'<tr><th bgcolor="#BBDDBB"><font face="arial" color="#225522" size="+1">'. |
|
$local. |
|
'</font></th><td bgcolor="#DDDDBB">'; |
|
&server($local); |
|
print |
|
'</td><td bgcolor="#DDDDBB"><font face="arial" color="#225522">'. |
|
$host{$local.'_load_doomed'}.'</font></td><td bgcolor="#DDDDBB"'; |
|
&bar($factor*$host{$local.'_load_doomed'}); |
|
print "</td></tr>\n"; |
|
} |
|
} |
|
print "</table>"; |
|
} else { |
|
print "No workload."; |
|
} |
|
} elsif ($mode eq 'trouble') { |
|
my $count=0; |
|
foreach $local (sort keys %hostname) { |
|
my $trouble=''; |
|
if ($host{$local.'_errors'}) { |
|
$trouble='Has loncron errors.<br />'; |
|
} elsif ($host{$local.'_loncron_doomed'}>600) { |
|
$trouble='High loncron count.<br />'; |
|
} |
|
if ($host{$local.'_load_doomed'}>5) { |
|
$trouble='High load.<br />'; |
|
} |
|
if ($host{$local.'_users_doomed'}>200) { |
|
$trouble='High user volume.<br />'; |
|
} |
|
if ($host{$local.'_mysql_doomed'}>1) { |
|
$trouble='MySQL database apparently offline.<br />'; |
|
} |
|
if ($host{$local.'_checkrpms_doomed'}>100) { |
|
$trouble='RPMs outdated.<br />'; |
|
} |
|
if ($trouble) { $count++; &serverstatus($local,$trouble); } |
|
} |
|
unless ($count) { print "No mayor trouble."; } |
|
} |
|
# ============================================================== Close, refresh |
print "</form><script>"; |
print "</form><script>"; |
$runtime=time-$starttime; |
$runtime=time-$starttime; |
if ($runtime>=$refresh) { |
if ($runtime>=$refresh) { |
print 'document.status.submit();'; |
print 'document.status.submit();'; |
} else { |
} else { |
$refreshtime=int(1000*($refresh-$runtime)); |
$refreshtime=int(1000*($refresh-$runtime)); |
print "setTimeout('document.status.submit()',$refreshtime);"; |
print "setTimeout('document.status.submit()',$refreshtime);\n". |
|
"document.prgstat.progress.value='Will automatically refresh ($refresh secs refresh cycle)'"; |
} |
} |
print "</script></body></html>"; |
print "</script></body></html>"; |
exit 0; |
exit 0; |