--- loncom/loncron 2002/06/08 14:59:04 1.30
+++ loncom/loncron 2004/05/11 19:12:50 1.46
@@ -9,13 +9,9 @@
#
# 7/14,7/15,7/19,7/21,7/22,11/18,
# 2/8 Gerd Kortemeyer
-# 12/6/2000,12/8 Scott Harrison
# 12/23 Gerd Kortemeyer
# YEAR=2001
-# 1/10/2001, 2/12/, 2/26, 3/15, 04/11, 04/21,8/27 Scott Harrison
# 09/04,09/06,11/26 Gerd Kortemeyer
-# YEAR=2002
-# 5/11/2002 Scott Harrison
$|=1;
@@ -24,6 +20,12 @@ use LONCAPA::Configuration;
use IO::File;
use IO::Socket;
+use strict;
+#globals
+use vars qw (%perlvar %simplestatus $errors $warnings $notices $totalcount);
+
+my $statusdir="/home/httpd/html/lon-status";
+
# -------------------------------------------------- Non-critical communication
sub reply {
@@ -42,9 +44,14 @@ sub reply {
# --------------------------------------------------------- Output error status
+sub log {
+ my $fh=shift;
+ if ($fh) { print $fh @_ }
+}
+
sub errout {
my $fh=shift;
- print $fh (<<ENDERROUT);
+ &log($fh,(<<ENDERROUT));
<p><table border=2 bgcolor="#CCCCCC">
<tr><td>Notices</td><td>$notices</td></tr>
<tr><td>Warnings</td><td>$warnings</td></tr>
@@ -53,84 +60,205 @@ sub errout {
ENDERROUT
}
-# ================================================================ Main Program
-
-# --------------------------------- Read loncapa_apache.conf and loncapa.conf
-my $perlvarref=LONCAPA::Configuration::read_conf('loncapa_apache.conf',
- 'loncapa.conf');
-%perlvar=%{$perlvarref};
-undef $perlvarref;
-delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
-delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
-
-# --------------------------------------- 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'}) {
- print("Unconfigured machine.\n");
- $emailto=$perlvar{'lonSysEMail'};
- $hostname=`/bin/hostname`;
- chop $hostname;
- $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
- $subj="LON: Unconfigured machine $hostname";
- system("echo 'Unconfigured machine $hostname.' |\
- mailto $emailto -s '$subj' > /dev/null");
- exit 1;
+sub start_daemon {
+ my ($fh,$daemon,$pidfile) = @_;
+ my $progname=$daemon;
+ if ($daemon eq 'lonc' && $ARGV[0] eq 'new') {
+ $progname='loncnew';
+ print "new ";
+ }
+ system("$perlvar{'lonDaemons'}/$progname 2>>$perlvar{'lonDaemons'}/logs/${daemon}_errors");
+ sleep 2;
+ if (-e $pidfile) {
+ &log($fh,"Seems like it started ...<p>");
+ my $lfh=IO::File->new("$pidfile");
+ my $daemonpid=<$lfh>;
+ chomp($daemonpid);
+ sleep 2;
+ if (kill 0 => $daemonpid) {
+ return 1;
+ } else {
+ return 0;
+ }
+ }
+ &log($fh,"Seems like that did not work!<p>");
+ $errors++;
+ return 0;
}
-# ----------------------------- Make sure this process is running from user=www
-my $wwwid=getpwnam('www');
-if ($wwwid!=$<) {
- print("User ID mismatch. This program must be run as user 'www'\n");
- $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
- $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
- system("echo 'User ID mismatch. loncron must be run as user www.' |\
- mailto $emailto -s '$subj' > /dev/null");
- exit 1;
-}
+sub checkon_daemon {
+ my ($fh,$daemon,$maxsize,$sendusr1)=@_;
-# ------------------------------------------------------------- Read hosts file
-{
- my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
+ &log($fh,'<hr><a name="'.$daemon.'"><h2>'.$daemon.'</h2><h3>Log</h3><pre>');
+ printf("%-10s ",$daemon);
+ if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
+ open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
+ while (my $line=<DFH>) {
+ &log($fh,"$line");
+ if ($line=~/INFO/) { $notices++; }
+ if ($line=~/WARNING/) { $notices++; }
+ if ($line=~/CRITICAL/) { $warnings++; }
+ };
+ close (DFH);
+ }
+ &log($fh,"</pre>");
+
+ my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid";
+
+ my $restartflag=1;
+ my $daemonpid;
+ if (-e $pidfile) {
+ my $lfh=IO::File->new("$pidfile");
+ $daemonpid=<$lfh>;
+ chomp($daemonpid);
+ if (kill 0 => $daemonpid) {
+ &log($fh,"<h3>$daemon at pid $daemonpid responding");
+ if ($sendusr1) { &log($fh,", sending USR1"); }
+ &log($fh,"</h3>");
+ if ($sendusr1) { kill USR1 => $daemonpid; }
+ $restartflag=0;
+ print "running\n";
+ } else {
+ $errors++;
+ &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
+ $restartflag=1;
+ &log($fh,"<h3>Decided to clean up stale .pid file and restart $daemon</h3>");
+ }
+ }
+ if ($restartflag==1) {
+ $simplestatus{$daemon}='off';
+ $errors++;
+ &log($fh,'<br><font color="red">Killall '.$daemon.': '.
+ `killall $daemon 2>&1`.' - ');
+ sleep 2;
+ &log($fh,unlink($pidfile).' - '.
+ `killall -9 $daemon 2>&1`.
+ '</font><br>');
+ &log($fh,"<h3>$daemon not running, trying to start</h3>");
+
+ if (&start_daemon($fh,$daemon,$pidfile)) {
+ &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
+ $simplestatus{$daemon}='restarted';
+ print "started\n";
+ } else {
+ $errors++;
+ &log($fh,"<h3>$daemon at pid $daemonpid not responding</h3>");
+ &log($fh,"Give it one more try ...<p>");
+ print " ";
+ if (&start_daemon($fh,$daemon,$pidfile)) {
+ &log($fh,"<h3>$daemon at pid $daemonpid responding</h3>");
+ $simplestatus{$daemon}='restarted';
+ print "started\n";
+ } else {
+ 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>");
+ }
+ }
- while (my $configline=<$config>) {
- my ($id,$domain,$role,$name,$ip)=split(/:/,$configline);
- $hostname{$id}=$name;
- $hostdom{$id}=$domain;
- $hostrole{$id}=$role;
- $hostip{$id}=$ip;
- if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) {
- $libserv{$id}=$name;
- }
+ if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
+ &log($fh,"<p><pre>");
+ open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|");
+ while (my $line=<DFH>) {
+ &log($fh,"$line");
+ if ($line=~/WARNING/) { $notices++; }
+ if ($line=~/CRITICAL/) { $notices++; }
+ };
+ close (DFH);
+ &log($fh,"</pre>");
+ }
+ }
+
+ 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");
}
+
+ &errout($fh);
}
-# ------------------------------------------------------ Read spare server file
-{
- my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
+# --------------------------------------------------------------------- Machine
+sub log_machine_info {
+ my ($fh)=@_;
+ &log($fh,'<hr><a name="machine"><h2>Machine Information</h2>');
+ &log($fh,"<h3>loadavg</h3>");
+
+ open (LOADAVGH,"/proc/loadavg");
+ my $loadavg=<LOADAVGH>;
+ close (LOADAVGH);
+
+ &log($fh,"<tt>$loadavg</tt>");
+
+ my @parts=split(/\s+/,$loadavg);
+ if ($parts[1]>4.0) {
+ $errors++;
+ } elsif ($parts[1]>2.0) {
+ $warnings++;
+ } elsif ($parts[1]>1.0) {
+ $notices++;
+ }
- while (my $configline=<$config>) {
- chomp($configline);
- if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
- $spareid{$configline}=1;
- }
+ &log($fh,"<h3>df</h3>");
+ &log($fh,"<pre>");
+
+ open (DFH,"df|");
+ while (my $line=<DFH>) {
+ &log($fh,"$line");
+ @parts=split(/\s+/,$line);
+ my $usage=$parts[4];
+ $usage=~s/\W//g;
+ if ($usage>90) {
+ $warnings++;
+ $notices++;
+ } elsif ($usage>80) {
+ $warnings++;
+ } elsif ($usage>60) {
+ $notices++;
+ }
+ if ($usage>95) { $warnings++; $warnings++; $simplestatus{'diskfull'}++; }
}
-}
+ close (DFH);
+ &log($fh,"</pre>");
-# ---------------------------------------------------------------- Start report
-$statusdir="/home/httpd/html/lon-status";
+ &log($fh,"<h3>ps</h3>");
+ &log($fh,"<pre>");
+ my $psproc=0;
-$errors=0;
-$warnings=0;
-$notices=0;
+ open (PSH,"ps -aux --cols 140 |");
+ while (my $line=<PSH>) {
+ &log($fh,"$line");
+ $psproc++;
+ }
+ close (PSH);
+ &log($fh,"</pre>");
+
+ if ($psproc>200) { $notices++; }
+ if ($psproc>250) { $notices++; }
-$now=time;
-$date=localtime($now);
+ &errout($fh);
+}
-{
-my $fh=IO::File->new(">$statusdir/newstatus.html");
+sub start_logging {
+ my ($hostdom,$hostrole,$hostname,$spareid)=@_;
+ my $fh=IO::File->new(">$statusdir/newstatus.html");
+ my %simplestatus=();
+ my $now=time;
+ my $date=localtime($now);
+
-print $fh (<<ENDHEADERS);
+ &log($fh,(<<ENDHEADERS));
<html>
<head>
<title>LON Status Report $perlvar{'lonHostID'}</title>
@@ -148,10 +276,11 @@ print $fh (<<ENDHEADERS);
<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 Count</a>
+<li><a href="#errcount">Error Coindex.html.unt</a>
</ol>
<hr>
<a name="configuration">
@@ -160,240 +289,128 @@ print $fh (<<ENDHEADERS);
<table border=2>
ENDHEADERS
-foreach $varname (keys %perlvar) {
- print $fh "<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n";
-}
-print $fh "</table><h3>Hosts</h3><table border=2>";
-foreach $id (keys %hostname) {
-print $fh
- "<tr><td>$id</td><td>$hostdom{$id}</td><td>$hostrole{$id}</td>";
-print $fh "<td>$hostname{$id}</td><td>$hostip{$id}</td></tr>\n";
-}
-print $fh "</table><h3>Spare Hosts</h3><ol>";
-foreach $id (keys %spareid) {
- print $fh "<li>$id\n";
-}
-
-print $fh "</ol>\n";
-
-# --------------------------------------------------------------------- Machine
-
-print $fh '<hr><a name="machine"><h2>Machine Information</h2>';
-print $fh "<h3>loadavg</h3>";
-
-open (LOADAVGH,"/proc/loadavg");
-$loadavg=<LOADAVGH>;
-close (LOADAVGH);
-
-print $fh "<tt>$loadavg</tt>";
-
-@parts=split(/\s+/,$loadavg);
-if ($parts[1]>4.0) {
- $errors++;
-} elsif ($parts[1]>2.0) {
- $warnings++;
-} elsif ($parts[1]>1.0) {
- $notices++;
-}
-
-print $fh "<h3>df</h3>";
-print $fh "<pre>";
-
-open (DFH,"df|");
-while ($line=<DFH>) {
- print $fh "$line";
- @parts=split(/\s+/,$line);
- $usage=$parts[4];
- $usage=~s/\W//g;
- if ($usage>90) {
- $warnings++;
- $notices++;
- } elsif ($usage>80) {
- $warnings++;
- } elsif ($usage>60) {
- $notices++;
- }
- if ($usage>95) { $warnings++; $warnings++ }
-}
-close (DFH);
-print $fh "</pre>";
-
-
-print $fh "<h3>ps</h3>";
-print $fh "<pre>";
-$psproc=0;
-
-open (PSH,"ps -aux|");
-while ($line=<PSH>) {
- print $fh "$line";
- $psproc++;
+ foreach my $varname (sort(keys(%perlvar))) {
+ &log($fh,"<tr><td>$varname</td><td>$perlvar{$varname}</td></tr>\n");
+ }
+ &log($fh,"</table><h3>Hosts</h3><table border=2>");
+ foreach my $id (sort(keys(%{$hostname}))) {
+ &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");
+ }
+ &log($fh,"</ol>\n");
+ return $fh;
}
-close (PSH);
-print $fh "</pre>";
-
-if ($psproc>200) { $notices++; }
-if ($psproc>250) { $notices++; }
-
-&errout($fh);
# --------------------------------------------------------------- clean out tmp
-print $fh '<hr><a name="tmp"><h2>Temporary Files</h2>';
-$cleaned=0;
-$old=0;
-while ($fname=<$perlvar{'lonDaemons'}/tmp/*>) {
- my ($dev,$ino,$mode,$nlink,
- $uid,$gid,$rdev,$size,
- $atime,$mtime,$ctime,
- $blksize,$blocks)=stat($fname);
- $now=time;
- $since=$now-$mtime;
- if ($since>$perlvar{'lonExpire'}) {
- $line='';
- if (open(PROBE,$fname)) {
- $line=<PROBE>;
- close(PROBE);
- }
- unless ($line=~/^CHECKOUTTOKEN\&/) {
- $cleaned++;
- unlink("$fname");
- } else {
- if ($since>100*$perlvar{'lonExpire'}) {
- $cleaned++;
- unlink("$fname");
- } else { $old++; }
- }
- }
-
+sub clean_tmp {
+ my ($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,"Cleaned up ".$cleaned." files (".$old." old checkout tokens).");
}
-print $fh "Cleaned up ".$cleaned." files (".$old." old checkout tokens).";
# ------------------------------------------------------------ clean out lonIDs
-print $fh '<hr><a name="tokens"><h2>Session Tokens</h2>';
-$cleaned=0;
-$active=0;
-while ($fname=<$perlvar{'lonIDsDir'}/*>) {
- my ($dev,$ino,$mode,$nlink,
- $uid,$gid,$rdev,$size,
- $atime,$mtime,$ctime,
- $blksize,$blocks)=stat($fname);
- $now=time;
- $since=$now-$mtime;
- if ($since>$perlvar{'lonExpire'}) {
- $cleaned++;
- print $fh "Unlinking $fname<br>";
- unlink("$fname");
- } else {
- $active++;
- }
-
+sub clean_lonIDs {
+ my ($fh)=@_;
+ &log($fh,'<hr><a name="tokens"><h2>Session Tokens</h2>');
+ my $cleaned=0;
+ my $active=0;
+ while (my $fname=<$perlvar{'lonIDsDir'}/*>) {
+ 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'}) {
+ $cleaned++;
+ &log($fh,"Unlinking $fname<br>");
+ unlink("$fname");
+ } else {
+ $active++;
+ }
+ }
+ &log($fh,"<p>Cleaned up ".$cleaned." stale session token(s).");
+ &log($fh,"<h3>$active open session(s)</h3>");
}
-print $fh "<p>Cleaned up ".$cleaned." stale session token(s).";
-print $fh "<h3>$active open session(s)</h3>";
+
# ----------------------------------------------------------------------- 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=~/\[error\]/) { $notices++; }
+ }
+ close (DFH);
+ &log($fh,"</pre>");
+ &errout($fh);
+}
-print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>';
+# ---------------------------------------------------------------------- lonnet
-open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
-while ($line=<DFH>) { print $fh "$line" };
-close (DFH);
-
-print $fh "</pre><h3>Error Log</h3><pre>";
-
-open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
-while ($line=<DFH>) {
- print $fh "$line";
- if ($line=~/\[error\]/) { $notices++; }
-};
-close (DFH);
-print $fh "</pre>";
-&errout($fh);
-
-
-# ---------------------------------------------------------------------- lonsql
-
-my $restartflag=1;
-if ($perlvar{'lonRole'} eq "library") {
-
- print $fh '<hr><a name="lonsql"><h2>lonsql</h2><h3>Log</h3><pre>';
- print "lonsql\n";
- if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
- open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
- while ($line=<DFH>) {
- print $fh "$line";
- if ($line=~/INFO/) { $notices++; }
- if ($line=~/WARNING/) { $notices++; }
- if ($line=~/CRITICAL/) { $warnings++; }
- };
+sub rotate_logs {
+ my ($fh)=@_;
+ &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");
+ }
close (DFH);
}
- print $fh "</pre>";
+ &log($fh,"</pre><h3>Perm Log</h3><pre>");
- my $lonsqlfile="$perlvar{'lonDaemons'}/logs/lonsql.pid";
-
- $restartflag=1;
-
- if (-e $lonsqlfile) {
- my $lfh=IO::File->new("$lonsqlfile");
- my $lonsqlpid=<$lfh>;
- chomp($lonsqlpid);
- if (kill 0 => $lonsqlpid) {
- print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";
- $restartflag=0;
- } else {
- $errors++; $errors++;
- print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
- $restartflag=1;
- print $fh
- "<h3>Decided to clean up stale .pid file and restart lonsql</h3>";
- }
- }
- if ($restartflag==1) {
- $errors++;
- print $fh '<br><font color="red">Killall lonsql: '.
- system('killall lonsql').' - ';
- sleep 2;
- print $fh unlink($lonsqlfile).' - '.
- system('killall -9 lonsql').
- '</font><br>';
- print $fh "<h3>lonsql not running, trying to start</h3>";
- system(
- "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors");
- sleep 2;
- if (-e $lonsqlfile) {
- print $fh "Seems like it started ...<p>";
- my $lfh=IO::File->new("$lonsqlfile");
- my $lonsqlpid=<$lfh>;
- chomp($lonsqlpid);
- sleep 2;
- if (kill 0 => $lonsqlpid) {
- print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>";
- } else {
- $errors++; $errors++;
- print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
- print $fh "Give it one more try ...<p>";
- system(
- "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors");
- sleep 2;
- }
- } else {
- print $fh "Seems like that did not work!<p>";
- $errors++;
+ if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
+ open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
+ while (my $line=<DFH>) {
+ &log($fh,"$line");
}
- if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
- print $fh "<p><pre>";
- open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
- while ($line=<DFH>) {
- print $fh "$line";
- if ($line=~/WARNING/) { $notices++; }
- if ($line=~/CRITICAL/) { $notices++; }
- };
- close (DFH);
- print $fh "</pre>";
- }
- }
+ close (DFH);
+ } else { &log($fh,"No perm log\n") }
- $fname="$perlvar{'lonDaemons'}/logs/lonsql.log";
+ my $fname="$perlvar{'lonDaemons'}/logs/lonnet.log";
my ($dev,$ino,$mode,$nlink,
$uid,$gid,$rdev,$size,
@@ -401,316 +418,206 @@ if ($perlvar{'lonRole'} eq "library") {
$blksize,$blocks)=stat($fname);
if ($size>40000) {
- print $fh "Rotating logs ...<p>";
+ &log($fh,"Rotating logs ...<p>");
rename("$fname.2","$fname.3");
rename("$fname.1","$fname.2");
rename("$fname","$fname.1");
}
+ &log($fh,"</pre>");
&errout($fh);
}
-# ------------------------------------------------------------------------ lond
-print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>';
-print "lond\n";
+# ----------------------------------------------------------------- Connections
+sub test_connections {
+ my ($fh,$hostname)=@_;
+ &log($fh,'<hr><a name="connections"><h2>Connections</h2>');
+ print "testing connections\n";
+ &log($fh,"<table border=2>");
+ foreach my $tryserver (sort(keys(%{$hostname}))) {
+ print(".");
+ my $result;
+ my $answer=reply("pong",$tryserver);
+ if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
+ $result="<b>ok</b>";
+ } else {
+ $result=$answer;
+ $warnings++;
+ if ($answer eq 'con_lost') { $warnings++; }
+ }
+ if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }
+ &log($fh,"<tr><td>$tryserver</td><td>$result</td></tr>\n");
+ }
+ &log($fh,"</table>");
-if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
-open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lond.log|");
-while ($line=<DFH>) {
- print $fh "$line";
- if ($line=~/INFO/) { $notices++; }
- if ($line=~/WARNING/) { $notices++; }
- if ($line=~/CRITICAL/) { $warnings++; }
-};
-close (DFH);
-}
-print $fh "</pre>";
-
-my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
-
-$restartflag=1;
-if (-e $londfile) {
- my $lfh=IO::File->new("$londfile");
- my $londpid=<$lfh>;
- chomp($londpid);
- if (kill 0 => $londpid) {
- print $fh "<h3>lond at pid $londpid responding, sending USR1</h3>";
- kill USR1 => $londpid;
- $restartflag=0;
- } else {
- $errors++;
- print $fh "<h3>lond at pid $londpid not responding</h3>";
- $restartflag=1;
- print $fh
- "<h3>Decided to clean up stale .pid file and restart lond</h3>";
- }
-}
-if ($restartflag==1) {
- $errors++;
- print $fh '<br><font color="red">Killall lond: '.
- system('killall lond').' - ';
- sleep 2;
- print $fh unlink($londfile).' - '.system('killall -9 lond').
- '</font><br>';
- print $fh "<h3>lond not running, trying to start</h3>";
- system(
- "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
- sleep 2;
- if (-e $londfile) {
- print $fh "Seems like it started ...<p>";
- my $lfh=IO::File->new("$londfile");
- my $londpid=<$lfh>;
- chomp($londpid);
- sleep 2;
- if (kill 0 => $londpid) {
- print $fh "<h3>lond at pid $londpid responding</h3>";
- } else {
- $errors++; $errors++;
- print $fh "<h3>lond at pid $londpid not responding</h3>";
- print $fh "Give it one more try ...<p>";
- system(
- "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
- sleep 2;
- }
- } else {
- print $fh "Seems like that did not work!<p>";
- $errors++;
- }
- if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
- print $fh "<p><pre>";
- open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
- while ($line=<DFH>) {
- print $fh "$line";
- if ($line=~/WARNING/) { $notices++; }
- if ($line=~/CRITICAL/) { $notices++; }
- };
- close (DFH);
- print $fh "</pre>";
- }
+ &errout($fh);
}
-$fname="$perlvar{'lonDaemons'}/logs/lond.log";
- my ($dev,$ino,$mode,$nlink,
- $uid,$gid,$rdev,$size,
- $atime,$mtime,$ctime,
- $blksize,$blocks)=stat($fname);
-
-if ($size>40000) {
- print $fh "Rotating logs ...<p>";
- rename("$fname.2","$fname.3");
- rename("$fname.1","$fname.2");
- rename("$fname","$fname.1");
-}
-
-&errout($fh);
-# ------------------------------------------------------------------------ lonc
-
-print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>';
-print "lonc\n";
-
-if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
-open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonc.log|");
-while ($line=<DFH>) {
- print $fh "$line";
- if ($line=~/INFO/) { $notices++; }
- if ($line=~/WARNING/) { $notices++; }
- if ($line=~/CRITICAL/) { $warnings++; }
-};
-close (DFH);
-}
-print $fh "</pre>";
-
-my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
-
-$restartflag=1;
-if (-e $loncfile) {
- my $lfh=IO::File->new("$loncfile");
- my $loncpid=<$lfh>;
- chomp($loncpid);
- if (kill 0 => $loncpid) {
- print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";
- kill USR1 => $loncpid;
- $restartflag=0;
- } else {
- $errors++;
- print $fh "<h3>lonc at pid $loncpid not responding</h3>";
- # Solution: kill parent and children processes, remove .pid and restart
- $restartflag=1;
- print $fh
- "<h3>Decided to clean up stale .pid file and restart lonc</h3>";
- }
-}
-if ($restartflag==1) {
- $errors++;
- print $fh '<br><font color="red">Killall lonc: '.
- system('killall lonc').' - ';
- sleep 2;
- print $fh unlink($loncfile).' - '.system('killall -9 lonc').
- '</font><br>';
- print $fh "<h3>lonc not running, trying to start</h3>";
- system(
- "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
- sleep 2;
- if (-e $loncfile) {
- print $fh "Seems like it started ...<p>";
- my $lfh=IO::File->new("$loncfile");
- my $loncpid=<$lfh>;
- chomp($loncpid);
- sleep 2;
- if (kill 0 => $loncpid) {
- print $fh "<h3>lonc at pid $loncpid responding</h3>";
- } else {
- $errors++; $errors++;
- print $fh "<h3>lonc at pid $loncpid not responding</h3>";
- print $fh "Give it one more try ...<p>";
- system(
- "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
- sleep 2;
- }
- } else {
- print $fh "Seems like that did not work!<p>";
- $errors++;
- }
- if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") {
- print $fh "<p><pre>";
- open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
- while ($line=<DFH>) {
- print $fh "$line";
- if ($line=~/WARNING/) { $notices++; }
- if ($line=~/CRITICAL/) { $notices++; }
- };
- close (DFH);
- print $fh "</pre>";
- }
-}
+# ------------------------------------------------------------ Delayed messages
+sub check_delayed_msg {
+ my ($fh)=@_;
+ &log($fh,'<hr><a name="delayed"><h2>Delayed Messages</h2>');
+ print "checking buffers\n";
+
+ &log($fh,'<h3>Scanning Permanent Log</h3>');
+
+ my $unsend=0;
-$fname="$perlvar{'lonDaemons'}/logs/lonc.log";
+ my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
+ while (my $line=<$dfh>) {
+ my ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
+ if ($sdf eq 'F') {
+ my $local=localtime($time);
+ &log($fh,"<b>Failed: $time, $dserv, $dcmd</b><br>");
+ $warnings++;
+ }
+ if ($sdf eq 'S') { $unsend--; }
+ if ($sdf eq 'D') { $unsend++; }
+ }
- my ($dev,$ino,$mode,$nlink,
- $uid,$gid,$rdev,$size,
- $atime,$mtime,$ctime,
- $blksize,$blocks)=stat($fname);
+ &log($fh,"Total unsend messages: <b>$unsend</b><p>\n");
+ $warnings=$warnings+5*$unsend;
-if ($size>40000) {
- print $fh "Rotating logs ...<p>";
- rename("$fname.2","$fname.3");
- rename("$fname.1","$fname.2");
- rename("$fname","$fname.1");
+ if ($unsend) { $simplestatus{'unsend'}=$unsend; }
+ &log($fh,"<h3>Outgoing Buffer</h3>");
+
+ open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
+ while (my $line=<DFH>) {
+ &log($fh,"$line<br>");
+ }
+ close (DFH);
}
-
-&errout($fh);
-# ---------------------------------------------------------------------- lonnet
+sub finish_logging {
+ my ($fh)=@_;
+ &log($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");
+ print "lon-status webpage updated\n";
+ $fh->close();
+
+ if ($errors) { $simplestatus{'errors'}=$errors; }
+ if ($warnings) { $simplestatus{'warnings'}=$warnings; }
+ if ($notices) { $simplestatus{'notices'}=$notices; }
+ $simplestatus{'time'}=time;
+}
-print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>';
-print "lonnet\n";
-if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
-open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
-while ($line=<DFH>) {
- print $fh "$line";
-};
-close (DFH);
-}
-print $fh "</pre><h3>Perm Log</h3><pre>";
-
-if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
- open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
-while ($line=<DFH>) {
- print $fh "$line";
-};
-close (DFH);
-} else { print $fh "No perm log\n" }
-
-$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) {
- print $fh "Rotating logs ...<p>";
- rename("$fname.2","$fname.3");
- rename("$fname.1","$fname.2");
- rename("$fname","$fname.1");
+sub log_simplestatus {
+ rename ("$statusdir/newstatus.html","$statusdir/index.html");
+
+ my $sfh=IO::File->new(">$statusdir/loncron_simple.txt");
+ foreach (keys %simplestatus) {
+ print $sfh $_.'='.$simplestatus{$_}.'&';
+ }
+ print $sfh "\n";
+ $sfh->close();
}
-print $fh "</pre>";
-&errout($fh);
-# ----------------------------------------------------------------- Connections
+sub send_mail {
+ print "sending mail\n";
+ my $emailto="$perlvar{'lonAdmEMail'}";
+ if ($totalcount>1000) {
+ $emailto.=",$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");
+}
-print $fh '<hr><a name="connections"><h2>Connections</h2>';
+# ================================================================ Main Program
+sub main () {
+# --------------------------------- 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
-print $fh "<table border=2>";
-foreach $tryserver (keys %hostname) {
+# --------------------------------------- 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'}) {
+ print("Unconfigured machine.\n");
+ my $emailto=$perlvar{'lonSysEMail'};
+ my $hostname=`/bin/hostname`;
+ chop $hostname;
+ $hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
+ my $subj="LON: Unconfigured machine $hostname";
+ system("echo 'Unconfigured machine $hostname.' |\
+ mailto $emailto -s '$subj' > /dev/null");
+ exit 1;
+ }
- $answer=reply("pong",$tryserver);
- if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
- $result="<b>ok</b>";
- } else {
- $result=$answer;
- $warnings++;
- if ($answer eq 'con_lost') { $warnings++; }
+# ----------------------------- Make sure this process is running from user=www
+ my $wwwid=getpwnam('www');
+ if ($wwwid!=$<) {
+ print("User ID mismatch. This program must be run as user 'www'\n");
+ my $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
+ my $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
+ system("echo 'User ID mismatch. loncron must be run as user www.' |\
+ mailto $emailto -s '$subj' > /dev/null");
+ exit 1;
}
- print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n";
-}
-print $fh "</table>";
+# ------------------------------------------------------------- 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 $config;
-&errout($fh);
-# ------------------------------------------------------------ Delayed messages
+# ------------------------------------------------------ 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;
-print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
-print "buffers\n";
+# ---------------------------------------------------------------- Start report
-print $fh '<h3>Scanning Permanent Log</h3>';
+ $errors=0;
+ $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($fh);
-$unsend=0;
-{
- my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
- while ($line=<$dfh>) {
- ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
- if ($sdf eq 'F') {
- $local=localtime($time);
- print "<b>Failed: $time, $dserv, $dcmd</b><br>";
- $warnings++;
- }
- if ($sdf eq 'S') { $unsend--; }
- if ($sdf eq 'D') { $unsend++; }
- }
-}
-print $fh "Total unsend messages: <b>$unsend</b><p>\n";
-$warnings=$warnings+5*$unsend;
-
-print $fh "<h3>Outgoing Buffer</h3>";
-
-open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
-while ($line=<DFH>) {
- print $fh "$line<br>";
-};
-close (DFH);
-
-# ------------------------------------------------------------------------- End
-print $fh "<a name=errcount>\n";
-$totalcount=$notices+4*$warnings+100*$errors;
-&errout($fh);
-print $fh "<h1>Total Error Count: $totalcount</h1>";
-$now=time;
-$date=localtime($now);
-print $fh "<hr>$date ($now)</body></html>\n";
-print "writing done\n";
-}
-
-rename ("$statusdir/newstatus.html","$statusdir/index.html");
-
-if ($totalcount>200) {
- print "mailing\n";
- $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
- $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";
- system(
- "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");
+ &finish_logging($fh);
+ &log_simplestatus();
+
+ if ($totalcount>200) { &send_mail(); }
}
+
+&main();
1;