--- loncom/cgi/clusterstatus.pl 2003/07/30 21:20:16 1.8 +++ loncom/cgi/clusterstatus.pl 2003/08/01 19:20:26 1.14 @@ -2,10 +2,8 @@ $|=1; # The LearningOnline Network with CAPA # Cluster Status -# (Versions -# (Running loncron -# 09/06/01 Gerd Kortemeyer) -# 02/18/02,02/19/02 Gerd Kortemeyer) +# +# $Id: clusterstatus.pl,v 1.14 2003/08/01 19:20:26 www Exp $ use lib '/home/httpd/lib/perl/'; use LONCAPA::Configuration; @@ -18,6 +16,23 @@ my %host=(); my $oneday=60*60*24; my %connectionstatus=(); +my %perlvar=(); + +my $mode; +my $concount=0; + +sub select_form { + my ($def,$name,%hash) = @_; + my $selectform = ""; + return $selectform; +} + sub key { my ($local,$url)=@_; @@ -33,6 +48,7 @@ sub hidden { sub request { my ($local,$url,$cachetime)=@_; + $cachetime*=(0.5+rand); my $key=&key($local,$url); my $reply=''; if ($FORM{$key.'_time'}) { @@ -47,7 +63,7 @@ sub request { $reply='local_unknown'; } else { - my $ua=new LWP::UserAgent(timeout => 20); + my $ua=new LWP::UserAgent(timeout => 10); my $request=new HTTP::Request('GET', "http://".$hostname{$local}.$url); @@ -77,16 +93,21 @@ sub connected { unless ($hostname{$remote}) { return 'remote_unknown'; } my $url='/cgi-bin/ping.pl?'.$remote; # -# Slowly phase this in: if not cached, only do 10 percent of the cases +# Slowly phase this in: if not cached, only do 5 percent of the cases, +# but always do the first five. # unless ($FORM{&key($local,$url)}) { - unless (rand>0.9) { return 'not_yet'; } + unless (($concount<=5) || (rand>0.95)) { + return 'not_yet'; + } else { + $concount++; + } } # # Actually do the query # &statuslist($local,'connecting '.$remote); - my $reply=&request($local,$url,1800); + my $reply=&request($local,$url,3600); $reply=(split("\n",$reply))[0]; $reply=~s/\W//g; if ($reply ne $remote) { return $reply; } @@ -106,17 +127,94 @@ sub replyhash { return %returnhash; } -# ========================================================== Show server status +# ================================================================ Link to host sub otherwindow { my ($local,$url,$label)=@_; return - "$label"; + " $label "; } -sub serverstatus { +sub login { my $local=shift; - print "\n

$local $hostdom{$local} ($hostname{$local}; $hostrole{$local})

\n"; + 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 $local=shift; + print &otherwindow($local,'/server-status','Server Status'); +} + +# ========================================================= Produce a green bar +sub bar { + my $parm=shift; + my $number=int($parm+0.5); + print "
"; + for (my $i=0;$i<$number;$i++) { + print "+"; + } + print "
"; +} + +# ========================================================== Show server status + +sub serverstatus { + my ($local,$trouble)=@_; + print (< + +"; + if ($trouble) { + print (""); + } + print "
+$local $hostdom{$local} ($hostname{$local}; $hostrole{$local}) +
$domaindescription{$hostdom{$local}} +
+ENDHEADER + &login($local);&server($local);&users($local);&versions($local); + &loncron($local);&lond($local);&lonc($local);&runloncron($local); + print "
$trouble
"; +# load + if (($host{$local.'_load_doomed'}>0.5) || ($mode eq 'load_doomed')) { + print "
Load: ".$host{$local.'_load'} + } +# users + if (($host{$local.'_users_doomed'}>10) || ($mode eq 'users_doomed')) { + print "
Active Users: ".$host{$local.'_users'} + } + # checkrpms if ($host{$local.'_checkrpms'}) { print "
RPMs: ".$host{$local.'_checkrpms'} @@ -125,6 +223,37 @@ sub serverstatus { if ($host{$local.'_mysql'}) { print "
MySQL Database: ".$host{$local.'_mysql'} } +# connections + if ($host{$local.'_notconnected'}) { + print "
Not connected: "; + foreach (split(/ /,$host{$local.'_notconnected'})) { + if ($_) { + print " $_"; + } + } + } +# errors + if ($host{$local.'_errors'}) { + print "
loncron errors: ".$host{$local.'_errors'}; + } + print "

"; +} + +# =========================================================== 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 @@ -163,15 +292,32 @@ foreach $pair (@pairs) { # ====================================================== Determine refresh rate -my $refresh=(($FORM{'refresh'}=~/^\d+$/)?$FORM{'refresh'}:60); +my $refresh=(($FORM{'refresh'}=~/^\d+$/)?$FORM{'refresh'}:120); if ($refresh<30) { $refresh=30; } 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 print "Content-type: text/html\n\n". - "\n"; + "\n"; # -------------------- Read loncapa.conf (and by default, loncapa_apache.conf). my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf'); -my %perlvar=%{$perlvarref}; +%perlvar=%{$perlvarref}; undef $perlvarref; # remove since sensitive and not needed delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed @@ -195,26 +341,56 @@ 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 "

Cluster Status ".localtime()."

"; +print "

LON-CAPA Cluster Status ".localtime()."

"; print "
\n". "
". "
\n";; print "
\n"; +print 'Choose next report: '.&select_form($mode,'mode',%modes).'
'; &hidden('refresh',$refresh); # ==================================================== Main Loop over all Hosts +my $maxusers=0; +my $maxload=0; +my $totalusers=0; + foreach $local (sort keys %hostname) { + $host{$local.'_unresponsive_doomed'}=0; # -- Check general status &statuslist($local,'General'); my %loncron=&replyhash($local,'/lon-status/loncron_simple.txt',1200); if (defined($loncron{'local_error'})) { $host{$local.'_loncron'}='Could not determine.'; + $host{$local.'_unresponsive_doomed'}++; } else { if ((time-$loncron{'time'})>$oneday) { $host{$local.'_loncron'}='Stale.'; + $host{$local.'_unresponsive_doomed'}++; } else { + $host{$local.'_loncron_doomed'}=$loncron{'notices'} + +4*$loncron{'warnings'} + +100*$loncron{'errors'}; + $host{$local.'_errors'}=$loncron{'errors'}; } } # -- Check user status @@ -222,13 +398,28 @@ foreach $local (sort keys %hostname) { my %userstatus=&replyhash($local,'/cgi-bin/userstatus.pl?simple',600); if (defined($userstatus{'local_error'})) { $host{$local.'_userstatus'}='Could not determine.'; + $host{$local.'_unresponsive_doomed'}++; } 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'}=$userstatus{'loadavg'}; } # -- Check mysql status &statuslist($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'})) { $host{$local.'_mysql'}='Could not determine.'; + $host{$local.'_unresponsive_doomed'}++; } else { if ((time-$mysql{'time'})>(7*$oneday)) { if ($hostrole{$local} eq 'library') { @@ -246,13 +437,15 @@ foreach $local (sort keys %hostname) { } # -- Check rpm status &statuslist($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'})) { $host{$local.'_checkrpms'}='Could not determine.'; + $host{$local.'_unresponsive_doomed'}++; } else { if ((time-$checkrpms{'time'})>(4*$oneday)) { $host{$local.'_checkrpms'}='Stale.'; $host{$local.'_checkrpms_doomed'}=50; + $host{$local.'_unresponsive_doomed'}++; } elsif ($checkrpms{'status'} eq 'fail') { $host{$local.'_checkrpms'}='Could not checked RPMs.'; $host{$local.'_checkrpms_doomed'}=100; @@ -274,18 +467,140 @@ foreach $local (sort keys %hostname) { $host{$local.'_notconnected_doomed'}++; } } -# Eventually, use doomed count - &serverstatus($local); -} - # =============================================================== End Mail Loop +} +&statuslist('Done.'); +# ====================================================================== Output + if ($mode=~/\_doomed$/) { +# Output by doomedness + foreach (&doomedness($mode)) { + &serverstatus($_); + } + } elsif ($mode eq 'connections') { + print + "". + ""; + foreach my $remote (sort keys %hostname) { + print ''; + } + print "\n"; +# connection matrix + foreach my $local (sort keys %hostname) { + print ''; + foreach my $remote (sort keys %hostname) { + if ($connectionstatus{$local.'_TO_'.$remote} eq 'not_yet') { + my $cellcolor='#FFFFFF'; + if ($local eq $remote) { $cellcolor='#DDDDDD'; } + print ''; + } elsif ($connectionstatus{$local.'_TO_'.$remote} eq 'ok') { + my $cellcolor='#BBDDBB'; + if ($local eq $remote) { $cellcolor='#99DD99'; } + print +''; + } else { + my $cellcolor='#DDBBBB'; + if ($connectionstatus{$local.'_TO_'.$remote} eq 'local_error') { + if ($local eq $remote) { + $cellcolor='#DD88AA'; + } else { + $cellcolor='#DDAACC'; + } + } else { + if ($local eq $remote) { $cellcolor='#DD9999'; } + } + print + ''; + } + } + print "\n"; + } + print "
 '.$remote.'
'.$local.'not yet testedok'. + $connectionstatus{$local.'_TO_'.$remote}.'
'; + &lonc($local); &lond($remote); + print '
"; + } elsif ($mode eq 'users') { +# Users + if ($maxusers) { + my $factor=50/$maxusers; + print "

Total active user(s): $totalusers

". + ""; + + foreach $local (sort keys %hostname) { + if (defined($host{$local.'_users'})) { + print +'\n"; + } + } + print "
'.$local. + ''; + &users($local); + print + ''. + $host{$local.'_users'}.'
"; + } else { + print "No active users logged in."; + } + } elsif ($mode eq 'load') { +# Load + if ($maxload) { + my $factor=50/$maxload; + print + ""; + foreach $local (sort keys %hostname) { + if (defined($host{$local.'_load_doomed'})) { + print +'\n"; + } + } + print "
'. + $local. + ''; + &server($local); + print + ''. + $host{$local.'_load_doomed'}.'
"; + } else { + print "No workload."; + } + } elsif ($mode eq 'trouble') { + my $count=0; + foreach $local (sort keys %hostname) { + my $trouble=''; + if ($host{$local.'_errors'}) { + $trouble='Has loncron errors.
'; + } elsif ($host{$local.'_loncron_doomed'}>600) { + $trouble='High loncron count.
'; + } + if ($host{$local.'_load_doomed'}>5) { + $trouble='High load.
'; + } + if ($host{$local.'_users_doomed'}>200) { + $trouble='High user volume.
'; + } + if ($host{$local.'_mysql_doomed'}>1) { + $trouble='MySQL database apparently offline.
'; + } + if ($host{$local.'_checkrpms_doomed'}>100) { + $trouble='RPMs outdated.
'; + } + if ($trouble) { $count++; &serverstatus($local,$trouble); } + } + unless ($count) { print "No mayor trouble."; } + } +# ============================================================== Close, refresh print "
"; exit 0;