loncom/loncron - diff

Return to loncron CVS log

Up to [LON-CAPA] / loncom

Diff for /loncom/loncron between versions 1.50 and 1.71

version 1.50, 2004/05/11 21:14:23	version 1.71, 2007/04/03 00:47:26
Line 45 my $statusdir="/home/httpd/html/lon-stat	Line 45 my $statusdir="/home/httpd/html/lon-stat

# -------------------------------------------------- Non-critical communication	# -------------------------------------------------- Non-critical communication
sub reply {	sub reply {
my ($cmd,$server)=@_;	my ($cmd,$server,$hostname)=@_;
my $peerfile="$perlvar{'lonSockDir'}/$server";	my $peerfile="$perlvar{'lonSockDir'}/".$hostname->{$server};
my $client=IO::Socket::UNIX->new(Peer =>"$peerfile",	my $client=IO::Socket::UNIX->new(Peer =>"$peerfile",
Type => SOCK_STREAM,	Type => SOCK_STREAM,
Timeout => 10)	Timeout => 10)
or return "con_lost";	or return "con_lost";
print $client "$cmd\n";	print $client "sethost:$server:$cmd\n";
my $answer=<$client>;	my $answer=<$client>;
chomp($answer);	chomp($answer);
if (!$answer) { $answer="con_lost"; }	if (!$answer) { $answer="con_lost"; }
Line 79 ENDERROUT	Line 79 ENDERROUT
sub start_daemon {	sub start_daemon {
my ($fh,$daemon,$pidfile,$args) = @_;	my ($fh,$daemon,$pidfile,$args) = @_;
my $progname=$daemon;	my $progname=$daemon;
if ($daemon eq 'lonc' && $args eq 'new') {	if ($daemon eq 'lonc') {
$progname='loncnew';	$progname='loncnew';
print "new ";
}	}
system("$perlvar{'lonDaemons'}/$progname 2>>$perlvar{'lonDaemons'}/logs/${daemon}_errors");	my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors";
sleep 2;	my $size=(stat($error_fname))[7];
	if ($size>40000) {
	&log($fh,"<p>Rotating error logs ...</p>");
	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) {	if (-e $pidfile) {
&log($fh,"<p>Seems like it started ...</p>");	&log($fh,"<p>Seems like it started ...</p>");
my $lfh=IO::File->new("$pidfile");	my $lfh=IO::File->new("$pidfile");
my $daemonpid=<$lfh>;	my $daemonpid=<$lfh>;
chomp($daemonpid);	chomp($daemonpid);
sleep 2;	if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
if (kill 0 => $daemonpid) {
return 1;	return 1;
} else {	} else {
return 0;	return 0;
Line 103 sub start_daemon {	Line 109 sub start_daemon {
}	}

sub checkon_daemon {	sub checkon_daemon {
my ($fh,$daemon,$maxsize,$sendusr1,$args)=@_;	my ($fh,$daemon,$maxsize,$send,$args)=@_;

	my $result;
&log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>');	&log($fh,'<hr /><a name="'.$daemon.'" /><h2>'.$daemon.'</h2><h3>Log</h3><p style="white-space: pre;"><tt>');
printf("%-10s ",$daemon);	printf("%-15s ",$daemon);
if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){	if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log\|");	open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log\|");
while (my $line=<DFH>) {	while (my $line=<DFH>) {
Line 127 sub checkon_daemon {	Line 134 sub checkon_daemon {
my $lfh=IO::File->new("$pidfile");	my $lfh=IO::File->new("$pidfile");
$daemonpid=<$lfh>;	$daemonpid=<$lfh>;
chomp($daemonpid);	chomp($daemonpid);
if (kill 0 => $daemonpid) {	if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
&log($fh,"<h3>$daemon at pid $daemonpid responding");	&log($fh,"<h3>$daemon at pid $daemonpid responding");
if ($sendusr1) { &log($fh,", sending USR1"); }	if ($send) { &log($fh,", sending $send"); }
&log($fh,"</h3>");	&log($fh,"</h3>");
if ($sendusr1) { kill USR1 => $daemonpid; }	if ($send eq 'USR1') { kill USR1 => $daemonpid; }
	if ($send eq 'USR2') { kill USR2 => $daemonpid; }
$restartflag=0;	$restartflag=0;
print "running\n";	if ($send eq 'USR2') {
	$result = 'reloaded';
	print "reloaded\n";
	} else {
	$result = 'running';
	print "running\n";
	}
} else {	} else {
$errors++;	$errors++;
&log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");	&log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
Line 144 sub checkon_daemon {	Line 158 sub checkon_daemon {
if ($restartflag==1) {	if ($restartflag==1) {
$simplestatus{$daemon}='off';	$simplestatus{$daemon}='off';
$errors++;	$errors++;
	my $kadaemon=$daemon;
	if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; }
&log($fh,'<br><font color="red">Killall '.$daemon.': '.	&log($fh,'<br><font color="red">Killall '.$daemon.': '.
`killall $daemon 2>&1`.' - ');	`killall $kadaemon 2>&1`.' - ');
sleep 2;	sleep 1;
&log($fh,unlink($pidfile).' - '.	&log($fh,unlink($pidfile).' - '.
`killall -9 $daemon 2>&1`.	`killall -9 $kadaemon 2>&1`.
'</font><br>');	'</font><br>');
&log($fh,"<h3>$daemon not running, trying to start</h3>");	&log($fh,"<h3>$daemon not running, trying to start</h3>");

if (&start_daemon($fh,$daemon,$pidfile,$args)) {	if (&start_daemon($fh,$daemon,$pidfile,$args)) {
&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");	&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
$simplestatus{$daemon}='restarted';	$simplestatus{$daemon}='restarted';
	$result = 'started';
print "started\n";	print "started\n";
} else {	} else {
$errors++;	$errors++;
Line 164 sub checkon_daemon {	Line 181 sub checkon_daemon {
if (&start_daemon($fh,$daemon,$pidfile,$args)) {	if (&start_daemon($fh,$daemon,$pidfile,$args)) {
&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");	&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
$simplestatus{$daemon}='restarted';	$simplestatus{$daemon}='restarted';
	$result = 'started';
print "started\n";	print "started\n";
} else {	} else {
	$result = 'failed';
print " failed\n";	print " failed\n";
$simplestatus{$daemon}='failed';	$simplestatus{$daemon}='failed';
$errors++; $errors++;	$errors++; $errors++;
Line 202 sub checkon_daemon {	Line 221 sub checkon_daemon {
}	}

&errout($fh);	&errout($fh);
	return $result;
}	}

# --------------------------------------------------------------------- Machine	# --------------------------------------------------------------------- Machine
Line 252 sub log_machine_info {	Line 272 sub log_machine_info {
&log($fh,"<pre>");	&log($fh,"<pre>");
my $psproc=0;	my $psproc=0;

open (PSH,"ps -aux --cols 140 \|");	open (PSH,"ps aux --cols 140 \|");
while (my $line=<PSH>) {	while (my $line=<PSH>) {
&log($fh,&encode_entities($line,'<>&"'));	&log($fh,&encode_entities($line,'<>&"'));
$psproc++;	$psproc++;
Line 263 sub log_machine_info {	Line 283 sub log_machine_info {
if ($psproc>200) { $notices++; }	if ($psproc>200) { $notices++; }
if ($psproc>250) { $notices++; }	if ($psproc>250) { $notices++; }

	&log($fh,"<h3>distprobe</h3>");
	&log($fh,"<pre>");
	open(DSH,"$perlvar{'lonDaemons'}/distprobe \|");
	while (my $line=<DSH>) {
	&log($fh,&encode_entities($line,'<>&"'));
	$psproc++;
	}
	close(DSH);
	&log($fh,"</pre>");

&errout($fh);	&errout($fh);
}	}

Line 455 sub test_connections {	Line 485 sub test_connections {
foreach my $tryserver (sort(keys(%{$hostname}))) {	foreach my $tryserver (sort(keys(%{$hostname}))) {
print(".");	print(".");
my $result;	my $result;
my $answer=reply("pong",$tryserver);	my $answer=&reply("ping",$tryserver,$hostname);
if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {	if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
$result="<b>ok</b>";	$result="<b>ok</b>";
$good++;	$good++;
Line 480 sub test_connections {	Line 510 sub test_connections {

# ------------------------------------------------------------ Delayed messages	# ------------------------------------------------------------ Delayed messages
sub check_delayed_msg {	sub check_delayed_msg {
my ($fh)=@_;	my ($fh,$hostname)=@_;
&log($fh,'<hr /><a name="delayed" /><h2>Delayed Messages</h2>');	&log($fh,'<hr /><a name="delayed" /><h2>Delayed Messages</h2>');
print "checking buffers\n";	print "checking buffers\n";

Line 505 sub check_delayed_msg {	Line 535 sub check_delayed_msg {

if ($unsend) { $simplestatus{'unsend'}=$unsend; }	if ($unsend) { $simplestatus{'unsend'}=$unsend; }
&log($fh,"<h3>Outgoing Buffer</h3>\n<pre>");	&log($fh,"<h3>Outgoing Buffer</h3>\n<pre>");
	# list directory with delayed messages and remember offline servers
	my %servers=();
open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed\|");	open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed\|");
while (my $line=<DFH>) {	while (my $line=<DFH>) {
	my ($server)=($line=~/\.(\w+)$/);
	if ($server) { $servers{$server}=1; }
&log($fh,&encode_entities($line,'<>&"'));	&log($fh,&encode_entities($line,'<>&"'));
}	}
&log($fh,"</pre>\n");	&log($fh,"</pre>\n");
close (DFH);	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<br />");
	}
}	}

sub finish_logging {	sub finish_logging {
Line 546 sub log_simplestatus {	Line 585 sub log_simplestatus {
sub send_mail {	sub send_mail {
print "sending mail\n";	print "sending mail\n";
my $emailto="$perlvar{'lonAdmEMail'}";	my $emailto="$perlvar{'lonAdmEMail'}";
if ($totalcount>1000) {	if ($totalcount>2500) {
$emailto.=",$perlvar{'lonSysEMail'}";	$emailto.=",$perlvar{'lonSysEMail'}";
}	}
my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";	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 {	sub usage {
Line 558 sub usage {	Line 601 sub usage {
loncron - housekeeping program that checks up on various parts of Lon-CAPA	loncron - housekeeping program that checks up on various parts of Lon-CAPA

Options:	Options:
--help Display help	--help Display
--oldlonc When starting the lonc daemon use 'lonc' not 'loncnew'
--noemail Do not send the status email	--noemail Do not send the status email
--justcheckconnections Only check the current status of the lonc/d	--justcheckconnections Only check the current status of the lonc/d
connections, do not send emails do not	connections, do not send emails do not
Line 569 Options:	Line 611 Options:
running, do not send emails do not	running, do not send emails do not
check the lonc/d connections, do not	check the lonc/d connections, do not
generate lon-status	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	USAGE
}	}

# ================================================================ Main Program	# ================================================================ Main Program
sub main () {	sub main () {
my ($oldlonc,$help,$justcheckdaemons,$noemail,$justcheckconnections);	my ($help,$justcheckdaemons,$noemail,$justcheckconnections,
	$justreload);
&GetOptions("help" => \$help,	&GetOptions("help" => \$help,
"oldlonc" => \$oldlonc,
"justcheckdaemons" => \$justcheckdaemons,	"justcheckdaemons" => \$justcheckdaemons,
"noemail" => \$noemail,	"noemail" => \$noemail,
"justcheckconnections" => \$justcheckconnections	"justcheckconnections" => \$justcheckconnections,
	"justreload" => \$justreload
);	);
if ($help) { &usage(); return; }	if ($help) { &usage(); return; }
# --------------------------------- Read loncapa_apache.conf and loncapa.conf	# --------------------------------- Read loncapa_apache.conf and loncapa.conf
Line 621 sub main () {	Line 668 sub main () {
my (%hostname,%hostdom,%hostrole,%spareid);	my (%hostname,%hostdom,%hostrole,%spareid);
while (my $configline=<$config>) {	while (my $configline=<$config>) {
next if ($configline =~ /^(\#\|\s*\$)/);	next if ($configline =~ /^(\#\|\s*\$)/);
my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline);	my ($id,$domain,$role,$name)=split(/:/,$configline);
if ($id && $domain && $role && $name && $ip) {	if ($id && $domain && $role && $name) {
	$name=~s/\s//g;
$hostname{$id}=$name;	$hostname{$id}=$name;
$hostdom{$id}=$domain;	$hostdom{$id}=$domain;
$hostrole{$id}=$role;	$hostrole{$id}=$role;
Line 649 sub main () {	Line 697 sub main () {


my $fh;	my $fh;
if (!$justcheckdaemons && !$justcheckconnections) {	if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
$fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);	$fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);

&log_machine_info($fh);	&log_machine_info($fh);
Line 658 sub main () {	Line 706 sub main () {
&check_httpd_logs($fh);	&check_httpd_logs($fh);
&rotate_lonnet_logs($fh);	&rotate_lonnet_logs($fh);
}	}
if (!$justcheckconnections) {	if (!$justcheckconnections && !$justreload) {
&checkon_daemon($fh,'lonsql',200000);	&checkon_daemon($fh,'lonsql',200000);
&checkon_daemon($fh,'lond',40000,1);	if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') {
my $args='new';	&checkon_daemon($fh,'lond',40000,'USR2');
if ($oldlonc) { $args = ''; }	}
&checkon_daemon($fh,'lonc',40000,1,$args);	&checkon_daemon($fh,'lonc',40000,'USR1');
&checkon_daemon($fh,'lonhttpd',40000);	&checkon_daemon($fh,'lonhttpd',40000);
	&checkon_daemon($fh,'lonmemcached',40000);
	&checkon_daemon($fh,'lonmaxima',40000);
	}
	if ($justreload) {
	&checkon_daemon($fh,'lond',40000,'USR2');
	&checkon_daemon($fh,'lonc',40000,'USR2');
}	}
if (!$justcheckdaemons) {	if ($justcheckconnections) {
&test_connections($fh,\%hostname);	&test_connections($fh,\%hostname);
}	}
if (!$justcheckdaemons && !$justcheckconnections) {	if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
&check_delayed_msg($fh);	&check_delayed_msg($fh,\%hostname);
&finish_logging($fh);	&finish_logging($fh);
&log_simplestatus();	&log_simplestatus();

FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>

Removed from v.1.50
changed lines
	Added in v.1.71