--- loncom/loncron 2004/05/11 20:19:46 1.48 +++ loncom/loncron 2006/10/25 21:33:54 1.68 @@ -2,7 +2,7 @@ # Housekeeping program, started by cron, loncontrol and loncron.pl # -# $Id: loncron,v 1.48 2004/05/11 20:19:46 albertel Exp $ +# $Id: loncron,v 1.68 2006/10/25 21:33:54 www Exp $ # # Copyright Michigan State University Board of Trustees # @@ -36,6 +36,7 @@ use LONCAPA::Configuration; use IO::File; use IO::Socket; use HTML::Entities; +use Getopt::Long; #globals use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount); @@ -44,13 +45,13 @@ my $statusdir="/home/httpd/html/lon-stat # -------------------------------------------------- Non-critical communication sub reply { - my ($cmd,$server)=@_; - my $peerfile="$perlvar{'lonSockDir'}/$server"; + my ($cmd,$server,$hostname)=@_; + my $peerfile="$perlvar{'lonSockDir'}/".$hostname->{$server}; my $client=IO::Socket::UNIX->new(Peer =>"$peerfile", Type => SOCK_STREAM, Timeout => 10) or return "con_lost"; - print $client "$cmd\n"; + print $client "sethost:$server:$cmd\n"; my $answer=<$client>; chomp($answer); if (!$answer) { $answer="con_lost"; } @@ -76,21 +77,28 @@ ENDERROUT } sub start_daemon { - my ($fh,$daemon,$pidfile) = @_; + my ($fh,$daemon,$pidfile,$args) = @_; my $progname=$daemon; - if ($daemon eq 'lonc' && $ARGV[0] eq 'new') { + if ($daemon eq 'lonc' && $args eq 'new') { $progname='loncnew'; print "new "; } - system("$perlvar{'lonDaemons'}/$progname 2>>$perlvar{'lonDaemons'}/logs/${daemon}_errors"); - sleep 2; + my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors"; + my $size=(stat($error_fname))[7]; + if ($size>40000) { + &log($fh,"

Rotating error logs ...

"); + rename("$error_fname.2","$error_fname.3"); + rename("$error_fname.1","$error_fname.2"); + rename("$error_fname","$error_fname.1"); + } + system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors"); + sleep 1; if (-e $pidfile) { &log($fh,"

Seems like it started ...

"); my $lfh=IO::File->new("$pidfile"); my $daemonpid=<$lfh>; chomp($daemonpid); - sleep 2; - if (kill 0 => $daemonpid) { + if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) { return 1; } else { return 0; @@ -102,10 +110,11 @@ sub start_daemon { } sub checkon_daemon { - my ($fh,$daemon,$maxsize,$sendusr1)=@_; + my ($fh,$daemon,$maxsize,$send,$args)=@_; + my $result; &log($fh,'

'.$daemon.'

Log

'); - printf("%-10s ",$daemon); + printf("%-15s ",$daemon); if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){ open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|"); while (my $line=) { @@ -126,13 +135,20 @@ sub checkon_daemon { my $lfh=IO::File->new("$pidfile"); $daemonpid=<$lfh>; chomp($daemonpid); - if (kill 0 => $daemonpid) { + if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) { &log($fh,"

$daemon at pid $daemonpid responding"); - if ($sendusr1) { &log($fh,", sending USR1"); } + if ($send) { &log($fh,", sending $send"); } &log($fh,"

"); - if ($sendusr1) { kill USR1 => $daemonpid; } + if ($send eq 'USR1') { kill USR1 => $daemonpid; } + if ($send eq 'USR2') { kill USR2 => $daemonpid; } $restartflag=0; - print "running\n"; + if ($send eq 'USR2') { + $result = 'reloaded'; + print "reloaded\n"; + } else { + $result = 'running'; + print "running\n"; + } } else { $errors++; &log($fh,"

$daemon at pid $daemonpid not responding

"); @@ -143,28 +159,33 @@ sub checkon_daemon { if ($restartflag==1) { $simplestatus{$daemon}='off'; $errors++; + my $kadaemon=$daemon; + if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; } &log($fh,'
Killall '.$daemon.': '. - `killall $daemon 2>&1`.' - '); - sleep 2; + `killall $kadaemon 2>&1`.' - '); + sleep 1; &log($fh,unlink($pidfile).' - '. - `killall -9 $daemon 2>&1`. + `killall -9 $kadaemon 2>&1`. '
'); &log($fh,"

$daemon not running, trying to start

"); - if (&start_daemon($fh,$daemon,$pidfile)) { + if (&start_daemon($fh,$daemon,$pidfile,$args)) { &log($fh,"

$daemon at pid $daemonpid responding

"); $simplestatus{$daemon}='restarted'; + $result = 'started'; print "started\n"; } else { $errors++; &log($fh,"

$daemon at pid $daemonpid not responding

"); &log($fh,"

Give it one more try ...

"); print " "; - if (&start_daemon($fh,$daemon,$pidfile)) { + if (&start_daemon($fh,$daemon,$pidfile,$args)) { &log($fh,"

$daemon at pid $daemonpid responding

"); $simplestatus{$daemon}='restarted'; + $result = 'started'; print "started\n"; } else { + $result = 'failed'; print " failed\n"; $simplestatus{$daemon}='failed'; $errors++; $errors++; @@ -201,6 +222,7 @@ sub checkon_daemon { } &errout($fh); + return $result; } # --------------------------------------------------------------------- Machine @@ -251,7 +273,7 @@ sub log_machine_info { &log($fh,"
");
     my $psproc=0;
 
-    open (PSH,"ps -aux --cols 140 |");
+    open (PSH,"ps aux --cols 140 |");
     while (my $line=) { 
 	&log($fh,&encode_entities($line,'<>&"')); 
 	$psproc++;
@@ -262,6 +284,16 @@ sub log_machine_info {
     if ($psproc>200) { $notices++; }
     if ($psproc>250) { $notices++; }
 
+    &log($fh,"

distprobe

"); + &log($fh,"
");
+    open(DSH,"$perlvar{'lonDaemons'}/distprobe |");
+    while (my $line=) { 
+	&log($fh,&encode_entities($line,'<>&"')); 
+	$psproc++;
+    }
+    close(DSH);
+    &log($fh,"
"); + &errout($fh); } @@ -450,29 +482,36 @@ sub test_connections { &log($fh,'

Connections

'); print "testing connections\n"; &log($fh,""); + my ($good,$bad)=(0,0); foreach my $tryserver (sort(keys(%{$hostname}))) { print("."); my $result; - my $answer=reply("pong",$tryserver); + my $answer=&reply("ping",$tryserver,$hostname); if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { $result="ok"; + $good++; } else { $result=$answer; $warnings++; - if ($answer eq 'con_lost') { $warnings++; } + if ($answer eq 'con_lost') { + $bad++; + $warnings++; + } else { + $good++; #self connection + } } if ($answer =~ /con_lost/) { print(" $tryserver down\n"); } &log($fh,"\n"); } &log($fh,"
$tryserver$result
"); - + print "\n$good good, $bad bad connections\n"; &errout($fh); } # ------------------------------------------------------------ Delayed messages sub check_delayed_msg { - my ($fh)=@_; + my ($fh,$hostname)=@_; &log($fh,'

Delayed Messages

'); print "checking buffers\n"; @@ -497,13 +536,22 @@ sub check_delayed_msg { if ($unsend) { $simplestatus{'unsend'}=$unsend; } &log($fh,"

Outgoing Buffer

\n
");
-
+# list directory with delayed messages and remember offline servers
+    my %servers=();
     open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
-    while (my $line=) { 
+    while (my $line=) {
+        my ($server)=($line=~/\.(\w+)$/);
+        if ($server) { $servers{$server}=1; }
 	&log($fh,&encode_entities($line,'<>&"'));
     }
     &log($fh,"
\n"); close (DFH); +# pong to all servers that have delayed messages +# this will trigger a reverse connection, which should flush the buffers + foreach my $tryserver (keys %servers) { + my $answer=&reply("pong",$tryserver,$hostname); + &log($fh,"Pong to $tryserver: $answer"); + } } sub finish_logging { @@ -538,15 +586,53 @@ sub log_simplestatus { sub send_mail { print "sending mail\n"; my $emailto="$perlvar{'lonAdmEMail'}"; - if ($totalcount>1000) { + if ($totalcount>2500) { $emailto.=",$perlvar{'lonSysEMail'}"; } 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 < $statusdir/index.html"); + } +} + +sub usage { + print(< \$help, + "oldlonc" => \$oldlonc, + "justcheckdaemons" => \$justcheckdaemons, + "noemail" => \$noemail, + "justcheckconnections" => \$justcheckconnections, + "justreload" => \$justreload + ); + if ($help) { &usage(); return; } # --------------------------------- Read loncapa_apache.conf and loncapa.conf my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf'); %perlvar=%{$perlvarref}; @@ -584,9 +670,10 @@ sub main () { my (%hostname,%hostdom,%hostrole,%spareid); while (my $configline=<$config>) { - next if ($configline =~ /^(\#|\s*$)/); - my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline); - if ($id && $domain && $role && $name && $ip) { + next if ($configline =~ /^(\#|\s*\$)/); + my ($id,$domain,$role,$name)=split(/:/,$configline); + if ($id && $domain && $role && $name) { + $name=~s/\s//g; $hostname{$id}=$name; $hostdom{$id}=$domain; $hostrole{$id}=$role; @@ -611,25 +698,44 @@ sub main () { $warnings=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) { + $fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid); + + &log_machine_info($fh); + &clean_tmp($fh); + &clean_lonIDs($fh); + &check_httpd_logs($fh); + &rotate_lonnet_logs($fh); + } + if (!$justcheckconnections && !$justreload) { + &checkon_daemon($fh,'lonsql',200000); + if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') { + &checkon_daemon($fh,'lond',40000,'USR2'); + } + my $args='new'; + if ($oldlonc) { $args = ''; } + &checkon_daemon($fh,'lonc',40000,'USR1',$args); + &checkon_daemon($fh,'lonhttpd',40000); + &checkon_daemon($fh,'lonmemcached',40000); + } + if ($justreload) { + &checkon_daemon($fh,'lond',40000,'USR2'); + my $args='new'; + if ($oldlonc) { $args = ''; } + &checkon_daemon($fh,'lonc',40000,'USR2',$args); + } + if ($justcheckconnections) { + &test_connections($fh,\%hostname); + } + if (!$justcheckdaemons && !$justcheckconnections && !$justreload) { + &check_delayed_msg($fh,\%hostname); + &finish_logging($fh); + &log_simplestatus(); + + if ($totalcount>200 && !$noemail) { &send_mail(); } + } } &main();