--- loncom/loncron 1999/10/13 17:48:51 1.1
+++ loncom/loncron 2002/05/11 21:18:14 1.26
@@ -7,7 +7,20 @@
# 6/1/99,6/2,6/10,6/11,6/12,6/14,6/26,6/28,6/29,6/30,
# 7/1,7/2,7/9,7/10,7/12 Gerd Kortemeyer)
#
-# 7/14,7/15,7/19,7/21,7/22 Gerd Kortemeyer
+# 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;
+
+use lib '/home/httpd/lib/perl/';
+use LONCAPA::Configuration;
use IO::File;
use IO::Socket;
@@ -40,56 +53,38 @@ sub errout {
ENDERROUT
}
-# -------------------------------------------------------------- Permanent logs
-sub logperm {
- my $message=shift;
- my $execdir=$perlvar{'lonDaemons'};
- my $now=time;
- my $local=localtime($now);
- my $fh=Apache::File->new(">>$execdir/logs/lonnet.perm.log");
- print $fh "$now:$message:$local\n";
- return 1;
-}
-
-# ------------------------------------------------ Try to send delayed messages
-sub senddelayed {
- my $fh=shift;
- my $dfname;
- my $path="$perlvar{'lonSockDir'}/delayed";
- print $fh "<h3>Attempting to send delayed messages</h3>";
- while ($dfname=<$path/*>) {
- my $wcmd;
- {
- my $dfh=IO::File->new($dfname);
- $wcmd=<$dfh>;
- }
- my ($server,$cmd)=split(/:/,$wcmd);
- chomp($cmd);
- my $answer=reply($cmd,$server);
- if ($answer ne 'con_lost') {
- unlink("$dfname");
- print $fh "Send $cmd to $server: $answer<br>\n";
- &logperm("S:$server:$cmd");
- } else {
- print $fh "Failed to deliver $cmd to $server<br>\n";
- $warnings++;
- }
- }
-}
-
# ================================================================ Main Program
-
-# ------------------------------------------------------------ Read access.conf
-{
- my $config=IO::File->new("/etc/httpd/conf/access.conf");
-
- while (my $configline=<$config>) {
- if ($configline =~ /PerlSetVar/) {
- my ($dummy,$varname,$varvalue)=split(/\s+/,$configline);
- $perlvar{$varname}=$varvalue;
- }
- }
+# ------------------------------------------- Read access.conf and loncapa.conf
+my $perlvarref=LONCAPA::Configuration::read_conf('access.conf','loncapa.conf');
+my %perlvar=%{$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;
+}
+
+# ----------------------------- 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;
}
# ------------------------------------------------------------- Read hosts file
@@ -139,14 +134,17 @@ print $fh (<<ENDHEADERS);
<head>
<title>LON Status Report $perlvar{'lonHostID'}</title>
</head>
-<body bgcolor="#FFFFFF">
+<body bgcolor="#AAAAAA">
<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="#lonnet">lonnet</a>
@@ -189,7 +187,7 @@ close (LOADAVGH);
print $fh "<tt>$loadavg</tt>";
@parts=split(/\s+/,$loadavg);
-if ($parts[1]>3.0) {
+if ($parts[1]>4.0) {
$errors++;
} elsif ($parts[1]>2.0) {
$warnings++;
@@ -207,28 +205,88 @@ while ($line=<DFH>) {
$usage=$parts[4];
$usage=~s/\W//g;
if ($usage>90) {
- $errors++;
+ $warnings++;
+ $notices++;
} elsif ($usage>80) {
$warnings++;
} elsif ($usage>60) {
$notices++;
}
- if ($usage>95) { $errors++; }
+ if ($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++;
+}
+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;
+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'}) {
+ $cleaned++;
+ unlink("$fname");
+ }
+
+}
+print $fh "Cleaned up ".$cleaned." files.";
+
+# ------------------------------------------------------------ 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++;
+ }
+
+}
+print $fh "<p>Cleaned up ".$cleaned." stale session token(s).";
+print $fh "<h3>$active open session(s)</h3>";
+
# ----------------------------------------------------------------------- httpd
print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>';
-open (DFH,"tail -n40 /etc/httpd/logs/access_log|");
+open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
while ($line=<DFH>) { print $fh "$line" };
close (DFH);
print $fh "</pre><h3>Error Log</h3><pre>";
-open (DFH,"tail -n50 /etc/httpd/logs/error_log|");
+open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
while ($line=<DFH>) {
print $fh "$line";
if ($line=~/\[error\]/) { $notices++; }
@@ -236,15 +294,119 @@ while ($line=<DFH>) {
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++; }
+ };
+ close (DFH);
+ }
+ print $fh "</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 60;
+ 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 10;
+ if (-e $lonsqlfile) {
+ print $fh "Seems like it started ...<p>";
+ my $lfh=IO::File->new("$lonsqlfile");
+ my $lonsqlpid=<$lfh>;
+ chomp($lonsqlpid);
+ sleep 30;
+ 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 10;
+ }
+ } else {
+ print $fh "Seems like that did not work!<p>";
+ $errors++;
+ }
+ if (-e "$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>";
+ }
+ }
+
+ $fname="$perlvar{'lonDaemons'}/logs/lonsql.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);
+}
# ------------------------------------------------------------------------ lond
print $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>';
+print "lond\n";
if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
-open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lond.log|");
+open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lond.log|");
while ($line=<DFH>) {
print $fh "$line";
- if ($line=~/giving up/) { $notices++; }
+ if ($line=~/INFO/) { $notices++; }
+ if ($line=~/WARNING/) { $notices++; }
+ if ($line=~/CRITICAL/) { $warnings++; }
};
close (DFH);
}
@@ -252,21 +414,34 @@ print $fh "</pre>";
my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
-if (-e $londfile) {
+$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</h3>";
+ print $fh "<h3>lond at pid $londpid responding, sending USR1</h3>";
+ kill USR1 => $londpid;
+ $restartflag=0;
} else {
- $errors++; $errors++;
+ $errors++;
print $fh "<h3>lond at pid $londpid not responding</h3>";
+ $restartflag=1;
+ print $fh
+ "<h3>Decided to clean up stale .pid file and restart lond</h3>";
}
-} else {
+}
+if ($restartflag==1) {
$errors++;
+ print $fh '<br><font color="red">Killall lond: '.
+ system('killall lond').' - ';
+ sleep 60;
+ print $fh unlink($londfile).' - '.system('killall -9 lond').
+ '</font><br>';
print $fh "<h3>lond not running, trying to start</h3>";
- system("$perlvar{'lonDaemons'}/lond");
- sleep 120;
+ system(
+ "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
+ sleep 10;
if (-e $londfile) {
print $fh "Seems like it started ...<p>";
my $lfh=IO::File->new("$londfile");
@@ -279,13 +454,25 @@ if (-e $londfile) {
$errors++; $errors++;
print $fh "<h3>lond at pid $londpid not responding</h3>";
print $fh "Give it one more try ...<p>";
- system("$perlvar{'lonDaemons'}/lond");
- sleep 120;
+ system(
+ "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
+ sleep 10;
}
} 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>";
+ }
}
$fname="$perlvar{'lonDaemons'}/logs/lond.log";
@@ -306,12 +493,15 @@ if ($size>40000) {
# ------------------------------------------------------------------------ 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 -n50 $perlvar{'lonDaemons'}/logs/lonc.log|");
+open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonc.log|");
while ($line=<DFH>) {
print $fh "$line";
- if ($line=~/died/) { $notices++; }
+ if ($line=~/INFO/) { $notices++; }
+ if ($line=~/WARNING/) { $notices++; }
+ if ($line=~/CRITICAL/) { $warnings++; }
};
close (DFH);
}
@@ -319,6 +509,7 @@ print $fh "</pre>";
my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
+$restartflag=1;
if (-e $loncfile) {
my $lfh=IO::File->new("$loncfile");
my $loncpid=<$lfh>;
@@ -326,15 +517,27 @@ if (-e $loncfile) {
if (kill 0 => $loncpid) {
print $fh "<h3>lonc at pid $loncpid responding, sending USR1</h3>";
kill USR1 => $loncpid;
+ $restartflag=0;
} else {
- $errors++; $errors++;
+ $errors++;
print $fh "<h3>lonc at pid $loncpid not responding</h3>";
+ # Solution: kill parent and children processes, remove .pid and restart
+ $restartflag=1;
+ print $fh
+ "<h3>Decided to clean up stale .pid file and restart lonc</h3>";
}
-} else {
+}
+if ($restartflag==1) {
$errors++;
+ print $fh '<br><font color="red">Killall lonc: '.
+ system('killall lonc').' - ';
+ sleep 60;
+ print $fh unlink($loncfile).' - '.system('killall -9 lonc').
+ '</font><br>';
print $fh "<h3>lonc not running, trying to start</h3>";
- system("$perlvar{'lonDaemons'}/lonc");
- sleep 120;
+ system(
+ "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
+ sleep 10;
if (-e $loncfile) {
print $fh "Seems like it started ...<p>";
my $lfh=IO::File->new("$loncfile");
@@ -347,13 +550,25 @@ if (-e $loncfile) {
$errors++; $errors++;
print $fh "<h3>lonc at pid $loncpid not responding</h3>";
print $fh "Give it one more try ...<p>";
- system("$perlvar{'lonDaemons'}/lonc");
- sleep 120;
+ system(
+ "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
+ sleep 10;
}
} 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>";
+ }
}
$fname="$perlvar{'lonDaemons'}/logs/lonc.log";
@@ -375,17 +590,15 @@ if ($size>40000) {
# ---------------------------------------------------------------------- lonnet
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";
- if ($line=~/Delayed/) { $warnings++; }
- if ($line=~/giving up/) { $warnings++; }
- if ($line=~/FAILED/) { $errors++; }
};
close (DFH);
}
-print $fh "</pre><h3>Perm Log</h3>";
+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|");
@@ -435,8 +648,7 @@ print $fh "</table>";
# ------------------------------------------------------------ Delayed messages
print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
-
-&senddelayed($fh);
+print "buffers\n";
print $fh '<h3>Scanning Permanent Log</h3>';
@@ -473,12 +685,13 @@ print $fh "<h1>Total Error Count: $total
$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(