--- loncom/loncron 2005/04/13 18:56:07 1.58 +++ 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 1.58 2005/04/13 18:56:07 albertel Exp $ +# $Id: loncron,v 1.68 2006/10/25 21:33:54 www Exp $ # # Copyright Michigan State University Board of Trustees # @@ -45,13 +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) { $answer="con_lost"; } @@ -92,14 +92,13 @@ sub start_daemon { rename("$error_fname","$error_fname.1"); } system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors"); - sleep 2; + 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 => $daemonpid) { return 1; } else { return 0; @@ -111,8 +110,9 @@ sub start_daemon { } sub checkon_daemon { - my ($fh,$daemon,$maxsize,$sendusr1,$args)=@_; + my ($fh,$daemon,$maxsize,$send,$args)=@_; + my $result; &log($fh,'

'.$daemon.'

Log

'); printf("%-15s ",$daemon); if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){ @@ -135,13 +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

"); @@ -156,7 +163,7 @@ sub checkon_daemon { if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; } &log($fh,'
Killall '.$daemon.': '. `killall $kadaemon 2>&1`.' - '); - sleep 2; + sleep 1; &log($fh,unlink($pidfile).' - '. `killall -9 $kadaemon 2>&1`. '
'); @@ -165,6 +172,7 @@ sub checkon_daemon { if (&start_daemon($fh,$daemon,$pidfile,$args)) { &log($fh,"

$daemon at pid $daemonpid responding

"); $simplestatus{$daemon}='restarted'; + $result = 'started'; print "started\n"; } else { $errors++; @@ -174,8 +182,10 @@ sub checkon_daemon { if (&start_daemon($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++; $errors++; @@ -212,6 +222,7 @@ sub checkon_daemon { } &errout($fh); + return $result; } # --------------------------------------------------------------------- Machine @@ -273,6 +284,16 @@ sub log_machine_info { if ($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); } @@ -465,7 +486,7 @@ sub test_connections { foreach my $tryserver (sort(keys(%{$hostname}))) { print("."); my $result; - my $answer=reply("ping",$tryserver); + my $answer=&reply("ping",$tryserver,$hostname); if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { $result="ok"; $good++; @@ -490,7 +511,7 @@ sub test_connections { # ------------------------------------------------------------ Delayed messages sub check_delayed_msg { - my ($fh)=@_; + my ($fh,$hostname)=@_; &log($fh,'

Delayed Messages

'); print "checking buffers\n"; @@ -515,13 +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: $answer"); + } } sub finish_logging { @@ -583,18 +613,24 @@ Options: running, do not send emails do not check the lonc/d connections, do not generate lon-status + --justreload Only tell the daemons to reload the config files, + do not send emails do not + check if the daemons are running, do not + generate lon-status USAGE } # ================================================================ Main Program sub main () { - my ($oldlonc,$help,$justcheckdaemons,$noemail,$justcheckconnections); + my ($oldlonc,$help,$justcheckdaemons,$noemail,$justcheckconnections, + $justreload); &GetOptions("help" => \$help, "oldlonc" => \$oldlonc, "justcheckdaemons" => \$justcheckdaemons, "noemail" => \$noemail, - "justcheckconnections" => \$justcheckconnections + "justcheckconnections" => \$justcheckconnections, + "justreload" => \$justreload ); if ($help) { &usage(); return; } # --------------------------------- Read loncapa_apache.conf and loncapa.conf @@ -664,7 +700,7 @@ sub main () { my $fh; - if (!$justcheckdaemons && !$justcheckconnections) { + if (!$justcheckdaemons && !$justcheckconnections && !$justreload) { $fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid); &log_machine_info($fh); @@ -673,20 +709,28 @@ sub main () { &check_httpd_logs($fh); &rotate_lonnet_logs($fh); } - if (!$justcheckconnections) { + if (!$justcheckconnections && !$justreload) { &checkon_daemon($fh,'lonsql',200000); - &checkon_daemon($fh,'lond',40000,1); + if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') { + &checkon_daemon($fh,'lond',40000,'USR2'); + } my $args='new'; if ($oldlonc) { $args = ''; } - &checkon_daemon($fh,'lonc',40000,1,$args); + &checkon_daemon($fh,'lonc',40000,'USR1',$args); &checkon_daemon($fh,'lonhttpd',40000); &checkon_daemon($fh,'lonmemcached',40000); } - if (!$justcheckdaemons) { + 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 && !$justcheckconnections) { - &check_delayed_msg($fh); + if (!$justcheckdaemons && !$justcheckconnections && !$justreload) { + &check_delayed_msg($fh,\%hostname); &finish_logging($fh); &log_simplestatus();