--- loncom/loncron 2004/05/11 20:19:46 1.48 +++ loncom/loncron 2005/02/14 00:29:48 1.57 @@ -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 1.57 2005/02/14 00:29:48 albertel 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 $totalcount); @@ -76,13 +77,21 @@ ENDERROUT } sub start_daemon { - my ($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"); + 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 2; if (-e $pidfile) { &log($fh,"

Seems like it started ...

"); @@ -102,10 +111,10 @@ sub start_daemon { } sub checkon_daemon { - my ($fh,$daemon,$maxsize,$sendusr1)=@_; + my ($fh,$daemon,$maxsize,$sendusr1,$args)=@_; &log($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 $line=) { @@ -143,15 +152,17 @@ sub checkon_daemon { if ($restartflag==1) { $simplestatus{$daemon}='off'; $errors++; + my $kadaemon=$daemon; + if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; } &log($fh,'
Killall '.$daemon.': '. - `killall $daemon 2>&1`.' - '); + `killall $kadaemon 2>&1`.' - '); sleep 2; &log($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"; @@ -160,7 +171,7 @@ 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"; @@ -251,7 +262,7 @@ 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,'<>&"')); 
 	$psproc++;
@@ -450,22 +461,29 @@ sub test_connections {
     &log($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);
 	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);
 }
 
@@ -538,15 +556,47 @@ 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 -t $emailto -s '$subj' -f $statusdir/index.html -m text/html >& /dev/null");
+    if ($result != 0) {
+	$result=system("mail -s '$subj' $emailto < $statusdir/index.html");
+    }
+}
+
+sub usage {
+    print(< \$help,
+		"oldlonc"              => \$oldlonc,
+		"justcheckdaemons"     => \$justcheckdaemons,
+		"noemail"              => \$noemail,
+		"justcheckconnections" => \$justcheckconnections
+		);
+    if ($help) { &usage(); return; }
 # --------------------------------- Read loncapa_apache.conf and loncapa.conf
     my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
     %perlvar=%{$perlvarref};
@@ -584,9 +634,10 @@ 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 +662,36 @@ 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) {
+	$fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);
+
+	&log_machine_info($fh);
+	&clean_tmp($fh);
+	&clean_lonIDs($fh);
+	&check_httpd_logs($fh);
+	&rotate_lonnet_logs($fh);
+    }
+    if (!$justcheckconnections) {
+	&checkon_daemon($fh,'lonsql',200000);
+	&checkon_daemon($fh,'lond',40000,1);
+	my $args='new';
+	if ($oldlonc) { $args = ''; }
+	&checkon_daemon($fh,'lonc',40000,1,$args);
+	&checkon_daemon($fh,'lonhttpd',40000);
+	&checkon_daemon($fh,'lonmemcached',40000);
+    }
+    if (!$justcheckdaemons) {
+	&test_connections($fh,\%hostname);
+    }
+    if (!$justcheckdaemons && !$justcheckconnections) {
+	&check_delayed_msg($fh);
+	&finish_logging($fh);
+	&log_simplestatus();
+	
+	if ($totalcount>200 && !$noemail) { &send_mail(); }
+    }
 }
 
 &main();