--- loncom/loncron 2004/05/11 21:08:55 1.49 +++ loncom/loncron 2005/06/13 19:54:28 1.60 @@ -2,7 +2,7 @@ # Housekeeping program, started by cron, loncontrol and loncron.pl # -# $Id: loncron,v 1.49 2004/05/11 21:08:55 albertel Exp $ +# $Id: loncron,v 1.60 2005/06/13 19:54:28 albertel Exp $ # # Copyright Michigan State University Board of Trustees # @@ -77,20 +77,27 @@ 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 => $daemonpid) { return 1; } else { @@ -103,10 +110,10 @@ sub start_daemon { } sub checkon_daemon { - my ($fh,$daemon,$maxsize,$sendusr1)=@_; + my ($fh,$daemon,$maxsize,$send,$args)=@_; &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 $line=) { @@ -129,11 +136,16 @@ sub checkon_daemon { chomp($daemonpid); if (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') { + print "reloaded\n"; + } else { + print "running\n"; + } } else { $errors++; &log($fh,"

$daemon at pid $daemonpid not responding

"); @@ -144,15 +156,17 @@ 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'; print "started\n"; @@ -161,7 +175,7 @@ sub checkon_daemon { &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'; print "started\n"; @@ -252,7 +266,7 @@ 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++;
@@ -455,7 +469,7 @@ sub test_connections {
     foreach my $tryserver (sort(keys(%{$hostname}))) {
 	print(".");
 	my $result;
-	my $answer=reply("pong",$tryserver);
+	my $answer=reply("ping",$tryserver);
 	if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
 	    $result="ok";
 	    $good++;
@@ -465,6 +479,8 @@ sub test_connections {
 	    if ($answer eq 'con_lost') {
 		$bad++;
 		$warnings++;
+	    } else {
+		$good++; #self connection
 	    }
 	}
 	if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }
@@ -544,11 +560,15 @@ 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 {
@@ -567,18 +587,24 @@ 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
@@ -619,8 +645,9 @@ 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 && $ip) {
+	my ($id,$domain,$role,$name)=split(/:/,$configline);
+	if ($id && $domain && $role && $name) {
+	    $name=~s/\s//g;
 	    $hostname{$id}=$name;
 	    $hostdom{$id}=$domain;
 	    $hostrole{$id}=$role;
@@ -647,7 +674,7 @@ sub main () {
 
 	
     my $fh;
-    if (!$justcheckdaemons && !$justcheckconnections) {
+    if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
 	$fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);
 
 	&log_machine_info($fh);
@@ -656,16 +683,25 @@ sub main () {
 	&check_httpd_logs($fh);
 	&rotate_lonnet_logs($fh);
     }
-    if (!$justcheckconnections) {
+    if (!$justcheckconnections && !$justreload) {
 	&checkon_daemon($fh,'lonsql',200000);
-	&checkon_daemon($fh,'lond',40000,1);
-	&checkon_daemon($fh,'lonc',40000,1);
+	&checkon_daemon($fh,'lond',40000,'USR1');
+	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',$args);
     }
-    if (!$justcheckdaemons) {
+    if (!$justcheckdaemons && !$justreload) {
 	&test_connections($fh,\%hostname);
     }
-    if (!$justcheckdaemons && !$justcheckconnections) {
+    if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
 	&check_delayed_msg($fh);
 	&finish_logging($fh);
 	&log_simplestatus();