--- loncom/cgi/clusterstatus.pl 2002/02/19 20:23:59 1.2 +++ loncom/cgi/clusterstatus.pl 2003/07/30 21:20:16 1.8 @@ -6,70 +6,184 @@ $|=1; # (Running loncron # 09/06/01 Gerd Kortemeyer) # 02/18/02,02/19/02 Gerd Kortemeyer) -# + +use lib '/home/httpd/lib/perl/'; +use LONCAPA::Configuration; + use LWP::UserAgent(); use HTTP::Headers; use IO::File; -use Net::Ping; -sub online { - my $host=shift; - my $p=Net::Ping->new("tcp",10); - my $online=$p->ping("$host"); - $p->close(); - undef ($p); - return $online; +my %host=(); +my $oneday=60*60*24; + +my %connectionstatus=(); + +sub key { + my ($local,$url)=@_; + my $key=$local.'_'.$url; + $key=~s/\W/\_/gs; + return $key; +} + +sub hidden { + my ($name,$value)=@_; + print "\n"; +} + +sub request { + my ($local,$url,$cachetime)=@_; + my $key=&key($local,$url); + my $reply=''; + if ($FORM{$key.'_time'}) { + if ((time-$FORM{$key.'_time'})<$cachetime) { + $reply=$FORM{$key}; + &hidden($key.'_time',$FORM{$key.'_time'}); + &hidden($key.'_fromcache',1); + } + } + unless ($reply) { + unless ($hostname{$local}) { + $reply='local_unknown'; + } else { + + my $ua=new LWP::UserAgent(timeout => 20); + + my $request=new HTTP::Request('GET', + "http://".$hostname{$local}.$url); + $request->authorization_basic('lonadm','litelite'); + + my $response=$ua->request($request); + + unless ($response->is_success) { + $reply='local_error'; + } else { + $reply=$response->content; + chomp($reply); + } + } + &hidden($key.'_time',time); + } + &hidden($key,$reply); + return $reply; } +# ============================================= Are local and remote connected? sub connected { my ($local,$remote)=@_; $local=~s/\W//g; $remote=~s/\W//g; - unless ($hostname{$local}) { return 'local_unknown'; } unless ($hostname{$remote}) { return 'remote_unknown'; } - - unless (&online($hostname{$local})) { return 'local_offline'; } - - my $ua=new LWP::UserAgent; - - my $request=new HTTP::Request('GET', - "http://".$hostname{$local}.'/cgi-bin/ping.pl?'.$remote); - - my $response=$ua->request($request); - - unless ($response->is_success) { return 'local_error'; } - - my $reply=$response->content; + my $url='/cgi-bin/ping.pl?'.$remote; +# +# Slowly phase this in: if not cached, only do 10 percent of the cases +# + unless ($FORM{&key($local,$url)}) { + unless (rand>0.9) { return 'not_yet'; } + } +# +# Actually do the query +# + &statuslist($local,'connecting '.$remote); + my $reply=&request($local,$url,1800); $reply=(split("\n",$reply))[0]; $reply=~s/\W//g; if ($reply ne $remote) { return $reply; } return 'ok'; } +# ============================================================ Get a reply hash - -print "Content-type: text/html\n\n". - "\n"; -# ------------------------------------------------------------ Read access.conf -{ - my $config=IO::File->new("/etc/httpd/conf/access.conf"); +sub replyhash { + my %returnhash=(); + foreach (split(/\&/,&request(@_))) { + my ($name,$value)=split(/\=/,$_); + if ($name) { + unless ($value) { $value=''; } + $returnhash{$name}=$value; + } + } + return %returnhash; +} - while (my $configline=<$config>) { - if ($configline =~ /PerlSetVar/) { - my ($dummy,$varname,$varvalue)=split(/\s+/,$configline); - $perlvar{$varname}=$varvalue; - } +# ========================================================== Show server status + +sub otherwindow { + my ($local,$url,$label)=@_; + return + "$label"; +} + +sub serverstatus { + my $local=shift; + print "\n

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

\n"; +# checkrpms + if ($host{$local.'_checkrpms'}) { + print "
RPMs: ".$host{$local.'_checkrpms'} + } +# mysql + if ($host{$local.'_mysql'}) { + print "
MySQL Database: ".$host{$local.'_mysql'} } - delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed - delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed } +# ====================================================================== Status +sub statuslist { + my ($local,$what)=@_; + print +"\n"; +} + +# +# Main program +# +# ========================================================= Get form parameters +my $buffer; + +read(STDIN, $buffer, $ENV{'CONTENT_LENGTH'}); +my @pairs=split(/&/,$buffer); +my $pair; my $name; my $value; +undef %FORM; +%FORM=(); +foreach $pair (@pairs) { + ($name,$value) = split(/=/,$pair); + $value =~ tr/+/ /; + $value =~ s/%([a-fA-F0-9][a-fA-F0-9])/pack("C",hex($1))/eg; + $FORM{$name}=$value; +} + +$buffer=$ENV{'QUERY_STRING'}; +@pairs=split(/&/,$buffer); +foreach $pair (@pairs) { + ($name,$value) = split(/=/,$pair); + $value =~ tr/+/ /; + $value =~ s/%([a-fA-F0-9][a-fA-F0-9])/pack("C",hex($1))/eg; + $FORM{$name}=$value; +} + +# ====================================================== Determine refresh rate + +my $refresh=(($FORM{'refresh'}=~/^\d+$/)?$FORM{'refresh'}:60); +if ($refresh<30) { $refresh=30; } +my $starttime=time; +# ================================================================ Send Headers +print "Content-type: text/html\n\n". + "\n"; +# -------------------- Read loncapa.conf (and by default, loncapa_apache.conf). +my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf'); +my %perlvar=%{$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 + # ------------------------------------------------------------- Read hosts file { my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab"); $total=0; while (my $configline=<$config>) { + $configline=~s/#.*$//; + unless ($configline=~/\w/) { next; } my ($id,$domain,$role,$name,$ip)=split(/:/,$configline); $hostname{$id}=$name; $hostdom{$id}=$domain; @@ -82,77 +196,96 @@ print "Content-type: text/html\n\n". } } -print -"

Cluster Status

"; -print "Please be patient while building cluster status ...

\n"; -$table=''; - -$table.="". -""; -foreach $remote (sort keys %hostname) { - $table.="$remote"; -} -$table.=""; +print "

Cluster Status ".localtime()."

"; +print "\n". +"
". +"\n";; +print "\n"; +&hidden('refresh',$refresh); + +# ==================================================== Main Loop over all Hosts + foreach $local (sort keys %hostname) { - print "Checking $local "; - $table.="$local
". - $hostrole{$local}.' '.$hostdom{$local}.'
'. - "". - $hostname{$local}."
". - "New Report". - ""; - if (&online($hostname{$local})) { - foreach $remote (sort keys %hostname) { - $status=&connected($local,$remote); - if ($status eq 'ok') { - $table.="
". - "lonc"; - if (&online($hostname{$remote})) { - $table.=" lond". - ""; - } else { - $table.=' offline'; - } - - } - } - } else { - print "offline"; - $table.=''; - } - $table.="\n"; - print "
\n"; -} -$table.="
From (local)To (remote)
Offline
"; -print $table; -print ""; +# -- Check general status + &statuslist($local,'General'); + my %loncron=&replyhash($local,'/lon-status/loncron_simple.txt',1200); + if (defined($loncron{'local_error'})) { + $host{$local.'_loncron'}='Could not determine.'; + } else { + if ((time-$loncron{'time'})>$oneday) { + $host{$local.'_loncron'}='Stale.'; + } else { + } + } +# -- Check user status + &statuslist($local,'Users'); + my %userstatus=&replyhash($local,'/cgi-bin/userstatus.pl?simple',600); + if (defined($userstatus{'local_error'})) { + $host{$local.'_userstatus'}='Could not determine.'; + } else { + } +# -- Check mysql status + &statuslist($local,'Database'); + my %mysql=&replyhash($local,'/lon-status/mysql.txt',1200); + if (defined($mysql{'local_error'})) { + $host{$local.'_mysql'}='Could not determine.'; + } else { + if ((time-$mysql{'time'})>(7*$oneday)) { + if ($hostrole{$local} eq 'library') { + $host{$local.'_mysql'}='Stale.'; + $host{$local.'_mysql_doomed'}=1; + } + if ($mysql{'mysql'} eq 'defunct') { + $host{$local.'_mysql'}='Defunct (maybe stale).'; + $host{$local.'_mysql_doomed'}=2; + } + } elsif ($mysql{'mysql'} eq 'defunct') { + $host{$local.'_mysql'}='Defunct.'; + $host{$local.'_mysql_doomed'}=3; + } + } +# -- Check rpm status + &statuslist($local,'RPMs'); + my %checkrpms=&replyhash($local,'/lon-status/checkrpms.txt',2400); + if (defined($checkrpms{'local_error'})) { + $host{$local.'_checkrpms'}='Could not determine.'; + } else { + if ((time-$checkrpms{'time'})>(4*$oneday)) { + $host{$local.'_checkrpms'}='Stale.'; + $host{$local.'_checkrpms_doomed'}=50; + } elsif ($checkrpms{'status'} eq 'fail') { + $host{$local.'_checkrpms'}='Could not checked RPMs.'; + $host{$local.'_checkrpms_doomed'}=100; + } elsif ($checkrpms{'rpmcount'}) { + $host{$local.'_checkrpms'}='Outdated RPMs: '. + $checkrpms{'rpmcount'}; + $host{$local.'_checkrpms_doomed'}=$checkrpms{'rpmcount'}; + } + } +# -- Check connections + &statuslist($local,'Connections'); + $host{$local.'_notconnected'}=''; + $host{$local.'_notconnected_doomed'}=0; + foreach $remote (sort keys %hostname) { + my $status=&connected($local,$remote); + $connectionstatus{$local.'_TO_'.$remote}=$status; + unless (($status eq 'ok') || ($status eq 'not_yet')) { + $host{$local.'_notconnected'}.=' '.$remote; + $host{$local.'_notconnected_doomed'}++; + } + } +# Eventually, use doomed count + &serverstatus($local); +} + +# =============================================================== End Mail Loop +print ""; +exit 0;