--- loncom/loncron 2005/05/26 20:25:00 1.59
+++ loncom/loncron 2007/05/31 05:20:18 1.74
@@ -2,7 +2,7 @@
# Housekeeping program, started by cron, loncontrol and loncron.pl
#
-# $Id: loncron,v 1.59 2005/05/26 20:25:00 albertel Exp $
+# $Id: loncron,v 1.74 2007/05/31 05:20:18 albertel Exp $
#
# Copyright Michigan State University Board of Trustees
#
@@ -32,6 +32,7 @@ use strict;
use lib '/home/httpd/lib/perl/';
use LONCAPA::Configuration;
+use Apache::lonnet;
use IO::File;
use IO::Socket;
@@ -43,21 +44,6 @@ use vars qw (%perlvar %simplestatus $err
my $statusdir="/home/httpd/html/lon-status";
-# -------------------------------------------------- Non-critical communication
-sub reply {
- my ($cmd,$server)=@_;
- my $peerfile="$perlvar{'lonSockDir'}/$server";
- my $client=IO::Socket::UNIX->new(Peer =>"$peerfile",
- Type => SOCK_STREAM,
- Timeout => 10)
- or return "con_lost";
- print $client "$cmd\n";
- my $answer=<$client>;
- chomp($answer);
- if (!$answer) { $answer="con_lost"; }
- return $answer;
-}
-
# --------------------------------------------------------- Output error status
sub log {
@@ -76,30 +62,36 @@ sub errout {
ENDERROUT
}
+sub rotate_logfile {
+ my ($file,$fh,$description) = @_;
+ my $size=(stat($file))[7];
+ if ($size>40000) {
+ &log($fh,"<p>Rotating $description ...</p>");
+ rename("$file.2","$file.3");
+ rename("$file.1","$file.2");
+ rename("$file","$file.1");
+ }
+}
+
sub start_daemon {
my ($fh,$daemon,$pidfile,$args) = @_;
my $progname=$daemon;
- if ($daemon eq 'lonc' && $args eq 'new') {
+ if ($daemon eq 'lonc') {
$progname='loncnew';
- print "new ";
}
my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors";
- 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");
+ &rotate_logfile($error_fname,$fh,'error logs');
+ if ($daemon eq 'lonc') {
+ &clean_sockets($fh);
}
system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors");
- sleep 2;
+ sleep 1;
if (-e $pidfile) {
&log($fh,"<p>Seems like it started ...</p>");
my $lfh=IO::File->new("$pidfile");
my $daemonpid=<$lfh>;
chomp($daemonpid);
- sleep 2;
- if (kill 0 => $daemonpid) {
+ if ($daemonpid =~ /^\d+$/ && kill 0 => $daemonpid) {
return 1;
} else {
return 0;
@@ -113,6 +105,7 @@ sub start_daemon {
sub checkon_daemon {
my ($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>');
printf("%-15s ",$daemon);
if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
@@ -135,7 +128,7 @@ 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,"<h3>$daemon at pid $daemonpid responding");
if ($send) { &log($fh,", sending $send"); }
&log($fh,"</h3>");
@@ -143,8 +136,10 @@ sub checkon_daemon {
if ($send eq 'USR2') { kill USR2 => $daemonpid; }
$restartflag=0;
if ($send eq 'USR2') {
+ $result = 'reloaded';
print "reloaded\n";
} else {
+ $result = 'running';
print "running\n";
}
} else {
@@ -161,7 +156,7 @@ sub checkon_daemon {
if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; }
&log($fh,'<br><font color="red">Killall '.$daemon.': '.
`killall $kadaemon 2>&1`.' - ');
- sleep 2;
+ sleep 1;
&log($fh,unlink($pidfile).' - '.
`killall -9 $kadaemon 2>&1`.
'</font><br>');
@@ -170,6 +165,7 @@ sub checkon_daemon {
if (&start_daemon($fh,$daemon,$pidfile,$args)) {
&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
$simplestatus{$daemon}='restarted';
+ $result = 'started';
print "started\n";
} else {
$errors++;
@@ -179,8 +175,10 @@ sub checkon_daemon {
if (&start_daemon($fh,$daemon,$pidfile,$args)) {
&log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
$simplestatus{$daemon}='restarted';
+ $result = 'started';
print "started\n";
} else {
+ $result = 'failed';
print " failed\n";
$simplestatus{$daemon}='failed';
$errors++; $errors++;
@@ -203,20 +201,10 @@ sub checkon_daemon {
}
my $fname="$perlvar{'lonDaemons'}/logs/$daemon.log";
-
- my ($dev,$ino,$mode,$nlink,
- $uid,$gid,$rdev,$size,
- $atime,$mtime,$ctime,
- $blksize,$blocks)=stat($fname);
-
- if ($size>$maxsize) {
- &log($fh,"<p>Rotating logs ...</p>");
- rename("$fname.2","$fname.3");
- rename("$fname.1","$fname.2");
- rename("$fname","$fname.1");
- }
+ &rotate_logfile($fname,$fh,'logs');
&errout($fh);
+ return $result;
}
# --------------------------------------------------------------------- Machine
@@ -278,11 +266,20 @@ sub log_machine_info {
if ($psproc>200) { $notices++; }
if ($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($fh);
}
sub start_logging {
- my ($hostdom,$hostrole,$hostname,$spareid)=@_;
my $fh=IO::File->new(">$statusdir/newstatus.html");
my %simplestatus=();
my $now=time;
@@ -325,17 +322,24 @@ ENDHEADERS
&encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n");
}
&log($fh,"</table><h3>Hosts</h3><table border='2'>");
- foreach my $id (sort(keys(%{$hostname}))) {
+ my %hostname = &Apache::lonnet::all_hostnames();
+ foreach my $id (sort(keys(%hostname))) {
+ my $role = (&Apache::lonnet::is_library($id) ? 'library'
+ : 'access');
&log($fh,
- "<tr><td>$id</td><td>".$hostdom->{$id}.
- "</td><td>".$hostrole->{$id}.
- "</td><td>".$hostname->{$id}."</td></tr>\n");
- }
- &log($fh,"</table><h3>Spare Hosts</h3><ol>");
- foreach my $id (sort(keys(%{$spareid}))) {
- &log($fh,"<li>$id\n</li>");
+ "<tr><td>$id</td><td>".&Apache::lonnet::host_domain($id).
+ "</td><td>".$role.
+ "</td><td>".&Apache::lonnet::hostname($id)."</td></tr>\n");
+ }
+ &log($fh,"</table><h3>Spare Hosts</h3><ul>");
+ foreach my $type (sort(keys(%Apache::lonnet::spareid))) {
+ &log($fh,"<li>$type\n<ol>");
+ foreach my $id (@{ $Apache::lonnet::spareid{$type} }) {
+ &log($fh,"<li>$id</li>\n");
+ }
+ &log($fh,"</ol>\n</li>\n");
}
- &log($fh,"</ol>\n");
+ &log($fh,"</ul>\n");
return $fh;
}
@@ -397,6 +401,21 @@ sub clean_lonIDs {
&log($fh,"<h3>$active open session(s)</h3>");
}
+# ----------------------------------------------------------- clean out sockets
+sub clean_sockets {
+ my ($fh)=@_;
+ my $cleaned=0;
+ opendir(SOCKETS,$perlvar{'lonSockDir'});
+ while (my $fname=readdir(SOCKETS)) {
+ next if (-d $fname
+ || $fname=~/(mysqlsock|maximasock|\Q$perlvar{'lonSockDir'}\E)/);
+ $cleaned++;
+ &log($fh,"Unlinking $fname<br />");
+ unlink("/home/httpd/sockets/$fname");
+ }
+ &log($fh,"<p>Cleaned up ".$cleaned." stale sockets.</p>");
+}
+
# ----------------------------------------------------------------------- httpd
sub check_httpd_logs {
@@ -443,34 +462,34 @@ sub rotate_lonnet_logs {
} else { &log($fh,"No perm log\n") }
my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
-
- my ($dev,$ino,$mode,$nlink,
- $uid,$gid,$rdev,$size,
- $atime,$mtime,$ctime,
- $blksize,$blocks)=stat($fname);
-
- if ($size>40000) {
- &log($fh,"<p>Rotating logs ...</p>");
- rename("$fname.2","$fname.3");
- rename("$fname.1","$fname.2");
- rename("$fname","$fname.1");
- }
+ &rotate_logfile($fname,$fh,'lonnet log');
&log($fh,"</pre>");
&errout($fh);
}
+sub rotate_other_logs {
+ my ($fh) = @_;
+ my $fname="$perlvar{'lonDaemons'}/logs/autoenroll.log";
+ &rotate_logfile($fname,$fh,'Auto Enroll log');
+ $fname="$perlvar{'lonDaemons'}/logs/autocreate.log";
+ &rotate_logfile($fname,$fh,'Create Course log');
+ $fname="$perlvar{'lonDaemons'}/logs/searchcat.log";
+ &rotate_logfile($fname,$fh,'Search Cataloguing log');
+}
+
# ----------------------------------------------------------------- Connections
sub test_connections {
- my ($fh,$hostname)=@_;
+ my ($fh)=@_;
&log($fh,'<hr /><a name="connections" /><h2>Connections</h2>');
print "testing connections\n";
&log($fh,"<table border='2'>");
my ($good,$bad)=(0,0);
- foreach my $tryserver (sort(keys(%{$hostname}))) {
+ my %hostname = &Apache::lonnet::all_hostnames();
+ foreach my $tryserver (sort(keys(%hostname))) {
print(".");
my $result;
- my $answer=reply("ping",$tryserver);
+ my $answer=&Apache::lonnet::reply("ping",$tryserver);
if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
$result="<b>ok</b>";
$good++;
@@ -520,13 +539,22 @@ sub check_delayed_msg {
if ($unsend) { $simplestatus{'unsend'}=$unsend; }
&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|");
- while (my $line=<DFH>) {
+ while (my $line=<DFH>) {
+ my ($server)=($line=~/\.(\w+)$/);
+ if ($server) { $servers{$server}=1; }
&log($fh,&encode_entities($line,'<>&"'));
}
&log($fh,"</pre>\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 $answer=&Apache::lonnet::reply("pong",$tryserver);
+ &log($fh,"Pong to $tryserver: $answer<br />");
+ }
}
sub finish_logging {
@@ -548,7 +576,7 @@ sub finish_logging {
}
sub log_simplestatus {
- rename ("$statusdir/newstatus.html","$statusdir/index.html");
+ rename("$statusdir/newstatus.html","$statusdir/index.html");
my $sfh=IO::File->new(">$statusdir/loncron_simple.txt");
foreach (keys %simplestatus) {
@@ -577,8 +605,7 @@ sub usage {
loncron - housekeeping program that checks up on various parts of Lon-CAPA
Options:
- --help Display help
- --oldlonc When starting the lonc daemon use 'lonc' not 'loncnew'
+ --help Display
--noemail Do not send the status email
--justcheckconnections Only check the current status of the lonc/d
connections, do not send emails do not
@@ -598,10 +625,9 @@ USAGE
# ================================================================ Main Program
sub main () {
- my ($oldlonc,$help,$justcheckdaemons,$noemail,$justcheckconnections,
+ my ($help,$justcheckdaemons,$noemail,$justcheckconnections,
$justreload);
&GetOptions("help" => \$help,
- "oldlonc" => \$oldlonc,
"justcheckdaemons" => \$justcheckdaemons,
"noemail" => \$noemail,
"justcheckconnections" => \$justcheckconnections,
@@ -640,32 +666,10 @@ sub main () {
exit 1;
}
-# ------------------------------------------------------------- Read hosts file
- my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
-
- my (%hostname,%hostdom,%hostrole,%spareid);
- while (my $configline=<$config>) {
- 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;
- }
- }
- undef $config;
-
-# ------------------------------------------------------ Read spare server file
- $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
-
- while (my $configline=<$config>) {
- chomp($configline);
- if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
- $spareid{$configline}=1;
- }
- }
- undef $config;
+# -------------------------------------------- Force reload of host information
+ &Apache::lonnet::load_hosts_tab(1);
+ &Apache::lonnet::load_domain_tab(1);
+ &Apache::lonnet::get_iphost(1);
# ---------------------------------------------------------------- Start report
@@ -676,31 +680,31 @@ sub main () {
my $fh;
if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
- $fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);
+ $fh=&start_logging();
&log_machine_info($fh);
&clean_tmp($fh);
&clean_lonIDs($fh);
&check_httpd_logs($fh);
&rotate_lonnet_logs($fh);
+ &rotate_other_logs($fh);
}
if (!$justcheckconnections && !$justreload) {
&checkon_daemon($fh,'lonsql',200000);
- &checkon_daemon($fh,'lond',40000,'USR1');
- my $args='new';
- if ($oldlonc) { $args = ''; }
- &checkon_daemon($fh,'lonc',40000,'USR1',$args);
+ if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') {
+ &checkon_daemon($fh,'lond',40000,'USR2');
+ }
+ &checkon_daemon($fh,'lonc',40000,'USR1');
&checkon_daemon($fh,'lonhttpd',40000);
&checkon_daemon($fh,'lonmemcached',40000);
+ &checkon_daemon($fh,'lonmaxima',40000);
}
if ($justreload) {
&checkon_daemon($fh,'lond',40000,'USR2');
- my $args='new';
- if ($oldlonc) { $args = ''; }
- &checkon_daemon($fh,'lonc',40000,'USR2',$args);
+ &checkon_daemon($fh,'lonc',40000,'USR2');
}
- if (!$justcheckdaemons && !$justreload) {
- &test_connections($fh,\%hostname);
+ if ($justcheckconnections) {
+ &test_connections($fh);
}
if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
&check_delayed_msg($fh);