--- loncom/loncron 2004/05/11 19:15:03 1.47
+++ loncom/loncron 2012/02/17 22:47:39 1.95.10.1
@@ -2,7 +2,7 @@
# Housekeeping program, started by cron, loncontrol and loncron.pl
#
-# $Id: loncron,v 1.47 2004/05/11 19:15:03 albertel Exp $
+# $Id: loncron,v 1.95.10.1 2012/02/17 22:47:39 raeburn Exp $
#
# Copyright Michigan State University Board of Trustees
#
@@ -28,34 +28,24 @@
#
$|=1;
+use strict;
use lib '/home/httpd/lib/perl/';
use LONCAPA::Configuration;
+use LONCAPA;
+use Apache::lonnet;
+use Apache::loncommon;
use IO::File;
use IO::Socket;
-use strict;
+use HTML::Entities;
+use Getopt::Long;
#globals
use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount);
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 {
@@ -66,45 +56,60 @@ sub log {
sub errout {
my $fh=shift;
&log($fh,(<<ENDERROUT));
- <p><table border=2 bgcolor="#CCCCCC">
+ <table border="2" bgcolor="#CCCCCC">
<tr><td>Notices</td><td>$notices</td></tr>
<tr><td>Warnings</td><td>$warnings</td></tr>
<tr><td>Errors</td><td>$errors</td></tr>
- </table><p><a href="#top">Top</a><p>
+ </table><p><a href="#top">Top</a></p>
ENDERROUT
}
+sub rotate_logfile {
+ my ($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) = @_;
+ my ($fh,$daemon,$pidfile,$args) = @_;
my $progname=$daemon;
- if ($daemon eq 'lonc' && $ARGV[0] eq 'new') {
+ if ($daemon eq 'lonc') {
$progname='loncnew';
- print "new ";
}
- system("$perlvar{'lonDaemons'}/$progname 2>>$perlvar{'lonDaemons'}/logs/${daemon}_errors");
- sleep 2;
+ my $error_fname="$perlvar{'lonDaemons'}/logs/${daemon}_errors";
+ &rotate_logfile($error_fname,$fh,'error logs');
+ if ($daemon eq 'lonc') {
+ &clean_sockets($fh);
+ }
+ system("$perlvar{'lonDaemons'}/$progname 2>$perlvar{'lonDaemons'}/logs/${daemon}_errors");
+ sleep 1;
if (-e $pidfile) {
- &log($fh,"Seems like it started ...<p>");
+ &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 => $daemonpid) {
return 1;
} else {
return 0;
}
}
- &log($fh,"Seems like that did not work!<p>");
+ &log($fh,"<p>Seems like that did not work!</p>");
$errors++;
return 0;
}
sub checkon_daemon {
- my ($fh,$daemon,$maxsize,$sendusr1)=@_;
+ my ($fh,$daemon,$maxsize,$send,$args)=@_;
- &log($fh,'<hr><a name="'.$daemon.'"><h2>'.$daemon.'</h2><h3>Log</h3><pre>');
- printf("%-10s ",$daemon);
+ 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"){
open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
while (my $line=<DFH>) {
@@ -115,7 +120,7 @@ sub checkon_daemon {
};
close (DFH);
}
- &log($fh,"</pre>");
+ &log($fh,"</tt></p>");
my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid";
@@ -125,13 +130,20 @@ 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,"<h3>$daemon at pid $daemonpid responding");
- if ($sendusr1) { &log($fh,", sending USR1"); }
+ if ($send) { &log($fh,", sending $send"); }
&log($fh,"</h3>");
- 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,"<h3>$daemon at pid $daemonpid not responding</h3>");
@@ -142,33 +154,38 @@ sub checkon_daemon {
if ($restartflag==1) {
$simplestatus{$daemon}='off';
$errors++;
+ my $kadaemon=$daemon;
+ if ($kadaemon eq 'lonmemcached') { $kadaemon='memcached'; }
&log($fh,'<br><font color="red">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 $kadaemon 2>&1`.
'</font><br>');
&log($fh,"<h3>$daemon not running, trying to start</h3>");
- if (&start_daemon($fh,$daemon,$pidfile)) {
+ 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 {
$errors++;
&log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
- &log($fh,"Give it one more try ...<p>");
+ &log($fh,"<p>Give it one more try ...</p>");
print " ";
- if (&start_daemon($fh,$daemon,$pidfile)) {
+ 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++; $errors++;
&log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
- &log($fh,"Unable to start $daemon<p>");
+ &log($fh,"<p>Unable to start $daemon</p>");
}
}
@@ -181,31 +198,21 @@ sub checkon_daemon {
if ($line=~/CRITICAL/) { $notices++; }
};
close (DFH);
- &log($fh,"</pre>");
+ &log($fh,"</pre></p>");
}
}
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,"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 $result;
}
# --------------------------------------------------------------------- Machine
sub log_machine_info {
my ($fh)=@_;
- &log($fh,'<hr><a name="machine"><h2>Machine Information</h2>');
+ &log($fh,'<hr /><a name="machine" /><h2>Machine Information</h2>');
&log($fh,"<h3>loadavg</h3>");
open (LOADAVGH,"/proc/loadavg");
@@ -228,7 +235,7 @@ sub log_machine_info {
open (DFH,"df|");
while (my $line=<DFH>) {
- &log($fh,"$line");
+ &log($fh,&encode_entities($line,'<>&"'));
@parts=split(/\s+/,$line);
my $usage=$parts[4];
$usage=~s/\W//g;
@@ -250,9 +257,9 @@ sub log_machine_info {
&log($fh,"<pre>");
my $psproc=0;
- open (PSH,"ps -aux --cols 140 |");
+ open (PSH,"ps aux --cols 140 |");
while (my $line=<PSH>) {
- &log($fh,"$line");
+ &log($fh,&encode_entities($line,'<>&"'));
$psproc++;
}
close (PSH);
@@ -261,11 +268,20 @@ sub log_machine_info {
if ($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;
@@ -278,86 +294,187 @@ sub start_logging {
<title>LON Status Report $perlvar{'lonHostID'}</title>
</head>
<body bgcolor="#AAAAAA">
-<a name="top">
+<a name="top" />
<h1>LON Status Report $perlvar{'lonHostID'}</h1>
<h2>$date ($now)</h2>
<ol>
-<li><a href="#configuration">Configuration</a>
-<li><a href="#machine">Machine Information</a>
-<li><a href="#tmp">Temporary Files</a>
-<li><a href="#tokens">Session Tokens</a>
-<li><a href="#httpd">httpd</a>
-<li><a href="#lonsql">lonsql</a>
-<li><a href="#lond">lond</a>
-<li><a href="#lonc">lonc</a>
-<li><a href="#lonhttpd">lonhttpd</a>
-<li><a href="#lonnet">lonnet</a>
-<li><a href="#connections">Connections</a>
-<li><a href="#delayed">Delayed Messages</a>
-<li><a href="#errcount">Error Coindex.html.unt</a>
+<li><a href="#configuration">Configuration</a></li>
+<li><a href="#machine">Machine Information</a></li>
+<li><a href="#tmp">Temporary Files</a></li>
+<li><a href="#tokens">Session Tokens</a></li>
+<li><a href="#httpd">httpd</a></li>
+<li><a href="#lonsql">lonsql</a></li>
+<li><a href="#lond">lond</a></li>
+<li><a href="#lonc">lonc</a></li>
+<li><a href="#lonnet">lonnet</a></li>
+<li><a href="#connections">Connections</a></li>
+<li><a href="#delayed">Delayed Messages</a></li>
+<li><a href="#errcount">Error Count</a></li>
</ol>
-<hr>
-<a name="configuration">
+<hr />
+<a name="configuration" />
<h2>Configuration</h2>
<h3>PerlVars</h3>
-<table border=2>
+<table border="2">
ENDHEADERS
foreach my $varname (sort(keys(%perlvar))) {
- &log($fh,"<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n");
+ &log($fh,"<tr><td>$varname</td><td>".
+ &encode_entities($perlvar{$varname},'<>&"')."</td></tr>\n");
}
- &log($fh,"</table><h3>Hosts</h3><table border=2>");
- foreach my $id (sort(keys(%{$hostname}))) {
+ &log($fh,"</table><h3>Hosts</h3><table border='2'>");
+ 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");
+ "<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 $fh;
}
# --------------------------------------------------------------- clean out tmp
sub clean_tmp {
my ($fh)=@_;
- &log($fh,'<hr><a name="tmp"><h2>Temporary Files</h2>');
- my $cleaned=0;
- my $old=0;
- while (my $fname=<$perlvar{'lonDaemons'}/tmp/*>) {
- my ($dev,$ino,$mode,$nlink,
- $uid,$gid,$rdev,$size,
- $atime,$mtime,$ctime,
- $blksize,$blocks)=stat($fname);
- my $now=time;
- my $since=$now-$mtime;
- if ($since>$perlvar{'lonExpire'}) {
- my $line='';
- if (open(PROBE,$fname)) {
- $line=<PROBE>;
- close(PROBE);
- }
- unless ($line=~/^CHECKOUTTOKEN\&/) {
- $cleaned++;
- unlink("$fname");
- } else {
- if ($since>365*$perlvar{'lonExpire'}) {
- $cleaned++;
- unlink("$fname");
- } else { $old++; }
- }
- }
+ &log($fh,'<hr /><a name="tmp" /><h2>Temporary Files</h2>');
+ my ($cleaned,$old,$removed) = (0,0,0);
+ my %errors = (
+ dir => [],
+ file => [],
+ failopen => [],
+ );
+ my %error_titles = (
+ dir => 'failed to remove empty directory:',
+ file => 'failed to unlike stale file',
+ failopen => 'failed to open file or directory'
+ );
+ ($cleaned,$old,$removed) = &recursive_clean_tmp('',$cleaned,$old,$removed,\%errors);
+ &log($fh,"Cleaned up: ".$cleaned." files; removed: $removed empty directories; (found: $old old checkout tokens)");
+ foreach my $key (sort(keys(%errors))) {
+ if (ref($errors{$key}) eq 'ARRAY') {
+ if (@{$errors{$key}} > 0) {
+ &log($fh,"Error during cleanup ($error_titles{$key}):<ul><li>".
+ join('</li><li><tt>',@{$errors{$key}}).'</tt></li></ul><br />');
+ }
+ }
+ }
+}
+
+sub recursive_clean_tmp {
+ my ($subdir,$cleaned,$old,$removed,$errors) = @_;
+ my $base = "$perlvar{'lonDaemons'}/tmp";
+ my $path = $base;
+ next if ($subdir =~ m{\.\./});
+ next unless (ref($errors) eq 'HASH');
+ unless ($subdir eq '') {
+ $path .= '/'.$subdir;
+ }
+ if (opendir(my $dh,"$path")) {
+ while (my $file = readdir($dh)) {
+ next if ($file =~ /^\.\.?$/);
+ my $fname = "$path/$file";
+ if (-d $fname) {
+ my $innerdir;
+ if ($subdir eq '') {
+ $innerdir = $file;
+ } else {
+ $innerdir = $subdir.'/'.$file;
+ }
+ ($cleaned,$old,$removed) =
+ &recursive_clean_tmp($innerdir,$cleaned,$old,$removed,$errors);
+ my @doms = &Apache::lonnet::current_machine_domains();
+
+ if (open(my $dirhandle,$fname)) {
+ unless (($innerdir eq 'helprequests') ||
+ (($innerdir =~ /^addcourse/) && ($innerdir !~ m{/\d+$}))) {
+ my @contents = grep {!/^\.\.?$/} readdir($dirhandle);
+ join('&&',@contents)."\n";
+ if (scalar(grep {!/^\.\.?$/} readdir($dirhandle)) == 0) {
+ closedir($dirhandle);
+ if ($fname =~ m{^\Q$perlvar{'lonDaemons'}\E/tmp/}) {
+ if (rmdir($fname)) {
+ $removed ++;
+ } elsif (ref($errors->{dir}) eq 'ARRAY') {
+ push(@{$errors->{dir}},$fname);
+ }
+ }
+ }
+ } else {
+ closedir($dirhandle);
+ }
+ }
+ } else {
+ my ($dev,$ino,$mode,$nlink,
+ $uid,$gid,$rdev,$size,
+ $atime,$mtime,$ctime,
+ $blksize,$blocks)=stat($fname);
+ my $now=time;
+ my $since=$now-$mtime;
+ if ($since>$perlvar{'lonExpire'}) {
+ if ($subdir eq '') {
+ my $line='';
+ if ($fname =~ /\.db$/) {
+ if (unlink($fname)) {
+ $cleaned++;
+ } elsif (ref($errors->{file}) eq 'ARRAY') {
+ push(@{$errors->{file}},$fname);
+ }
+ } elsif (open(PROBE,$fname)) {
+ my $line='';
+ $line=<PROBE>;
+ close(PROBE);
+ if ($line=~/^CHECKOUTTOKEN\&/) {
+ if ($since>365*$perlvar{'lonExpire'}) {
+ if (unlink($fname)) {
+ $cleaned++;
+ } elsif (ref($errors->{file}) eq 'ARRAY') {
+ push(@{$errors->{file}},$fname);
+ }
+ } else {
+ $old++;
+ }
+ } else {
+ if (unlink($fname)) {
+ $cleaned++;
+ } elsif (ref($errors->{file}) eq 'ARRAY') {
+ push(@{$errors->{file}},$fname);
+ }
+ }
+ } elsif (ref($errors->{failopen}) eq 'ARRAY') {
+ push(@{$errors->{failopen}},$fname);
+ }
+ } else {
+ if (unlink($fname)) {
+ $cleaned++;
+ } elsif (ref($errors->{file}) eq 'ARRAY') {
+ push(@{$errors->{file}},$fname);
+ }
+ }
+ }
+ }
+ }
+ closedir($dh);
+ } elsif (ref($errors->{failopen}) eq 'ARRAY') {
+ push(@{$errors->{failopen}},$path);
}
- &log($fh,"Cleaned up ".$cleaned." files (".$old." old checkout tokens).");
+ return ($cleaned,$old,$removed);
}
# ------------------------------------------------------------ clean out lonIDs
sub clean_lonIDs {
my ($fh)=@_;
- &log($fh,'<hr><a name="tokens"><h2>Session Tokens</h2>');
+ &log($fh,'<hr /><a name="tokens" /><h2>Session Tokens</h2>');
my $cleaned=0;
my $active=0;
while (my $fname=<$perlvar{'lonIDsDir'}/*>) {
@@ -375,42 +492,49 @@ sub clean_lonIDs {
$active++;
}
}
- &log($fh,"<p>Cleaned up ".$cleaned." stale session token(s).");
+ &log($fh,"<p>Cleaned up ".$cleaned." stale session token(s).</p>");
&log($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
+ || $fname=~/(mysqlsock|maximasock|rsock|\Q$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 {
my ($fh)=@_;
- &log($fh,'<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>');
-
- open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
- while (my $line=<DFH>) { &log($fh,"$line") };
- close (DFH);
-
- &log($fh,"</pre><h3>Error Log</h3><pre>");
-
- open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
- while (my $line=<DFH>) {
- &log($fh,"$line");
- if ($line=~/\[error\]/) { $notices++; }
+ if (open(PIPE,"./lchttpdlogs|")) {
+ while (my $line=<PIPE>) {
+ &log($fh,$line);
+ if ($line=~/\[error\]/) { $notices++; }
+ }
+ close(PIPE);
}
- close (DFH);
- &log($fh,"</pre>");
&errout($fh);
}
# ---------------------------------------------------------------------- lonnet
-sub rotate_logs {
+sub rotate_lonnet_logs {
my ($fh)=@_;
- &log($fh,'<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>');
+ &log($fh,'<hr /><a name="lonnet" /><h2>lonnet</h2><h3>Temp Log</h3><pre>');
print "checking logs\n";
if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
while (my $line=<DFH>) {
- &log($fh,"$line");
+ &log($fh,&encode_entities($line,'<>&"'));
}
close (DFH);
}
@@ -419,51 +543,63 @@ sub rotate_logs {
if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
while (my $line=<DFH>) {
- &log($fh,"$line");
+ &log($fh,&encode_entities($line,'<>&"'));
}
close (DFH);
} else { &log($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($fh,"Rotating logs ...<p>");
- rename("$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 %logs = (
+ autoenroll => 'Auto Enroll log',
+ autocreate => 'Create Course log',
+ searchcat => 'Search Cataloguing log',
+ autoupdate => 'Auto Update log',
+ refreshcourseids_db => 'Refresh CourseIDs db log',
+ );
+ foreach my $item (keys(%logs)) {
+ my $fname=$perlvar{'lonDaemons'}.'/logs/'.$item.'.log';
+ &rotate_logfile($fname,$fh,$logs{$item});
+ }
+}
+
# ----------------------------------------------------------------- Connections
sub test_connections {
- my ($fh,$hostname)=@_;
- &log($fh,'<hr><a name="connections"><h2>Connections</h2>');
+ my ($fh)=@_;
+ &log($fh,'<hr /><a name="connections" /><h2>Connections</h2>');
print "testing connections\n";
- &log($fh,"<table border=2>");
- foreach my $tryserver (sort(keys(%{$hostname}))) {
+ &log($fh,"<table border='2'>");
+ my ($good,$bad)=(0,0);
+ my %hostname = &Apache::lonnet::all_hostnames();
+ foreach my $tryserver (sort(keys(%hostname))) {
print(".");
my $result;
- my $answer=reply("pong",$tryserver);
+ my $answer=&Apache::lonnet::reply("ping",$tryserver);
if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
$result="<b>ok</b>";
+ $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,"<tr><td>$tryserver</td><td>$result</td></tr>\n");
}
&log($fh,"</table>");
-
+ print "\n$good good, $bad bad connections\n";
&errout($fh);
}
@@ -471,7 +607,7 @@ sub test_connections {
# ------------------------------------------------------------ Delayed messages
sub check_delayed_msg {
my ($fh)=@_;
- &log($fh,'<hr><a name="delayed"><h2>Delayed Messages</h2>');
+ &log($fh,'<hr /><a name="delayed" /><h2>Delayed Messages</h2>');
print "checking buffers\n";
&log($fh,'<h3>Scanning Permanent Log</h3>');
@@ -490,28 +626,57 @@ sub check_delayed_msg {
if ($sdf eq 'D') { $unsend++; }
}
- &log($fh,"Total unsend messages: <b>$unsend</b><p>\n");
- $warnings=$warnings+5*$unsend;
+ &log($fh,"<p>Total unsend messages: <b>$unsend</b></p>\n");
+ if ($unsend > 0) {
+ $warnings=$warnings+5*$unsend;
+ }
if ($unsend) { $simplestatus{'unsend'}=$unsend; }
- &log($fh,"<h3>Outgoing Buffer</h3>");
-
+ &log($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 $line=<DFH>) {
- &log($fh,"$line<br>");
+ while (my $line=<DFH>) {
+ my ($server)=($line=~/\.(\w+)$/);
+ if ($server) { $servers{$server}=1; }
+ &log($fh,&encode_entities($line,'<>&"'));
}
+ &log($fh,"</pre>\n");
close (DFH);
+ my %hostname = &Apache::lonnet::all_hostnames();
+ my $numhosts = scalar(keys(%hostname));
+# pong to all servers that have delayed messages
+# this will trigger a reverse connection, which should flush the buffers
+ foreach my $tryserver (sort(keys(%servers))) {
+ if ($hostname{$tryserver} || !$numhosts) {
+ my $answer;
+ eval {
+ local $SIG{ ALRM } = sub { die "TIMEOUT" };
+ alarm(20);
+ $answer = &Apache::lonnet::reply("pong",$tryserver);
+ alarm(0);
+ };
+ if ($@ && $@ =~ m/TIMEOUT/) {
+ &log($fh,"Attempted pong to $tryserver timed out<br />");
+ print "time out while contacting: $tryserver for pong\n";
+ } else {
+ &log($fh,"Pong to $tryserver: $answer<br />");
+ }
+ } else {
+ &log($fh,"$tryserver has delayed messages, but is not part of the cluster -- skipping 'Pong'.<br />");
+ }
+ }
}
sub finish_logging {
my ($fh)=@_;
- &log($fh,"<a name=errcount>\n");
+ &log($fh,"<a name='errcount' />\n");
$totalcount=$notices+4*$warnings+100*$errors;
&errout($fh);
&log($fh,"<h1>Total Error Count: $totalcount</h1>");
my $now=time;
my $date=localtime($now);
- &log($fh,"<hr>$date ($now)</body></html>\n");
+ &log($fh,"<hr />$date ($now)</body></html>\n");
print "lon-status webpage updated\n";
$fh->close();
@@ -522,7 +687,7 @@ sub finish_logging {
}
sub log_simplestatus {
- rename ("$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) {
@@ -532,25 +697,124 @@ sub log_simplestatus {
$sfh->close();
}
+sub write_loncaparevs {
+ print "Retrieving LON-CAPA version information\n";
+ if (open(my $fh,">$perlvar{'lonTabDir'}/loncaparevs.tab")) {
+ my %hostname = &Apache::lonnet::all_hostnames();
+ foreach my $id (sort(keys(%hostname))) {
+ if ($id ne '') {
+ my $loncaparev;
+ eval {
+ local $SIG{ ALRM } = sub { die "TIMEOUT" };
+ alarm(10);
+ $loncaparev =
+ &Apache::lonnet::get_server_loncaparev('',$id,1,'loncron');
+ alarm(0);
+ };
+ if ($@ && $@ =~ m/TIMEOUT/) {
+ print "time out while contacting lonHost: $id for version\n";
+ }
+ if ($loncaparev =~ /^[\w.\-]+$/) {
+ print $fh $id.':'.$loncaparev."\n";
+ }
+ }
+ }
+ close($fh);
+ }
+ return;
+}
+
+sub write_serverhomeIDs {
+ print "Retrieving LON-CAPA lonHostID information\n";
+ if (open(my $fh,">$perlvar{'lonTabDir'}/serverhomeIDs.tab")) {
+ my %name_to_host = &Apache::lonnet::all_names();
+ foreach my $name (sort(keys(%name_to_host))) {
+ if ($name ne '') {
+ if (ref($name_to_host{$name}) eq 'ARRAY') {
+ my $serverhomeID;
+ eval {
+ local $SIG{ ALRM } = sub { die "TIMEOUT" };
+ alarm(10);
+ $serverhomeID =
+ &Apache::lonnet::get_server_homeID($name,1,'loncron');
+ alarm(0);
+ };
+ if ($@ && $@ =~ m/TIMEOUT/) {
+ print "Time out while contacting server: $name\n";
+ }
+ if ($serverhomeID ne '') {
+ print $fh $name.':'.$serverhomeID."\n";
+ } else {
+ print $fh $name.':'.$name_to_host{$name}->[0]."\n";
+ }
+ }
+ }
+ }
+ close($fh);
+ }
+ return;
+}
+
sub send_mail {
print "sending mail\n";
- my $emailto="$perlvar{'lonAdmEMail'}";
- if ($totalcount>1000) {
+ my $defdom = $perlvar{'lonDefDomain'};
+ my $origmail = $perlvar{'lonAdmEMail'};
+ my $emailto = &Apache::loncommon::build_recipient_list(undef,
+ 'lonstatusmail',$defdom,$origmail);
+ if ($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 < $statusdir/index.html");
+ }
+}
+
+sub usage {
+ print(<<USAGE);
+loncron - housekeeping program that checks up on various parts of Lon-CAPA
+
+Options:
+ --help Display
+ --noemail Do not send the status email
+ --justcheckconnections Only check the current status of the lonc/d
+ connections, do not send emails do not
+ check if the daemons are running, do not
+ generate lon-status
+ --justcheckdaemons Only check that all of the Lon-CAPA daemons are
+ running, do not send emails do not
+ check the lonc/d connections, do not
+ generate lon-status
+ --justreload Only tell the daemons to reload the config files,
+ do not send emails do not
+ check if the daemons are running, do not
+ generate lon-status
+
+USAGE
}
# ================================================================ Main Program
sub main () {
+ my ($help,$justcheckdaemons,$noemail,$justcheckconnections,
+ $justreload);
+ &GetOptions("help" => \$help,
+ "justcheckdaemons" => \$justcheckdaemons,
+ "noemail" => \$noemail,
+ "justcheckconnections" => \$justcheckconnections,
+ "justreload" => \$justreload
+ );
+ if ($help) { &usage(); return; }
# --------------------------------- Read loncapa_apache.conf and loncapa.conf
my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
%perlvar=%{$perlvarref};
undef $perlvarref;
delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
-
+ delete $perlvar{'reCAPTCHA_PRIVATE'}; # remove since sensitive and not needed
+ delete $perlvar{'reCAPTCHA_PUBLIC'}; # remove since sensitive and not needed
+ chdir($perlvar{'lonDaemons'});
# --------------------------------------- Make sure that LON-CAPA is configured
# I only test for one thing here (lonHostID). This is just a safeguard.
if ('{[[[[lonHostID]]]]}' eq $perlvar{'lonHostID'}) {
@@ -576,34 +840,31 @@ 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>) {
- my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline);
- if ($id && $domain && $role && $name && $ip) {
- $hostname{$id}=$name;
- $hostdom{$id}=$domain;
- $hostrole{$id}=$role;
- } else {
- if ($configline) {
-# &logthis("Skipping hosts.tab line -$configline-");
- }
- }
- }
- 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;
- }
+# -------------------------------------------- Force reload of host information
+ &Apache::lonnet::load_hosts_tab(1);
+ &Apache::lonnet::load_domain_tab(1);
+ &Apache::lonnet::get_iphost(1);
+
+# ----------------------------------------- Force firewall update for lond port
+
+ if ((!$justcheckdaemons) && (!$justreload)) {
+ my $now = time;
+ my $tmpfile = $perlvar{'lonDaemons'}.'/tmp/lciptables_iphost_'.
+ $now.$$.int(rand(10000));
+ if (open(my $fh,">$tmpfile")) {
+ my %iphosts = &Apache::lonnet::get_iphost();
+ foreach my $key (keys(%iphosts)) {
+ print $fh "$key\n";
+ }
+ close($fh);
+ if (&LONCAPA::try_to_lock('/tmp/lock_lciptables')) {
+ my $execpath = $perlvar{'lonDaemons'}.'/lciptables';
+ system("$execpath $tmpfile");
+ unlink('/tmp/lock_lciptables'); # Remove the lock file.
+ }
+ unlink($tmpfile);
+ }
}
- undef $config;
# ---------------------------------------------------------------- Start report
@@ -611,24 +872,44 @@ 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($fh);
- &checkon_daemon($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 && !$justcheckconnections && !$justreload) {
+ $fh=&start_logging();
+
+ &log_machine_info($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,'lonmemcached',40000);
+ &checkon_daemon($fh,'lonsql',200000);
+ if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') {
+ &checkon_daemon($fh,'lond',40000,'USR2');
+ }
+ &checkon_daemon($fh,'lonc',40000,'USR1');
+ &checkon_daemon($fh,'lonmaxima',40000);
+ &checkon_daemon($fh,'lonr',40000);
+ }
+ if ($justreload) {
+ &checkon_daemon($fh,'lond',40000,'USR2');
+ &checkon_daemon($fh,'lonc',40000,'USR2');
+ }
+ if ($justcheckconnections) {
+ &test_connections($fh);
+ }
+ if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
+ &check_delayed_msg($fh);
+ &finish_logging($fh);
+ &log_simplestatus();
+ &write_loncaparevs();
+ &write_serverhomeIDs();
+
+ if ($totalcount>200 && !$noemail) { &send_mail(); }
+ }
}
&main();