--- loncom/loncron 1999/10/13 17:48:51 1.1
+++ loncom/loncron 2000/12/08 23:17:26 1.10
@@ -7,7 +7,8 @@
# 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
use IO::File;
use IO::Socket;
@@ -40,43 +41,6 @@ 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
@@ -139,7 +103,7 @@ 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>
@@ -189,7 +153,7 @@ 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) {
$warnings++;
@@ -207,13 +171,13 @@ while ($line=<DFH>) {
$usage=$parts[4];
$usage=~s/\W//g;
if ($usage>90) {
- $errors++;
+ $warnings++;
} elsif ($usage>80) {
$warnings++;
} elsif ($usage>60) {
$notices++;
}
- if ($usage>95) { $errors++; }
+ if ($usage>95) { $warnings++; $warnings++ }
}
close (DFH);
print $fh "</pre>";
@@ -236,15 +200,102 @@ while ($line=<DFH>) {
close (DFH);
print $fh "</pre>";
&errout($fh);
+
+
+# ------------------------------------------------------------------------ lonsql
+
+if ($perlvar{'lonRole'} eq "library" && 1==0) {
+
+ print $fh '<hr><a name="lond"><h2>lonsql</h2><h3>Log</h3><pre>';
+
+ if (-e "$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";
+
+ 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>";
+ } else {
+ $errors++; $errors++;
+ print $fh "<h3>lonsql at pid $lonsqlpid not responding</h3>";
+ }
+ } else {
+ $errors++;
+ print $fh "<h3>lonsql not running, trying to start</h3>";
+ system("$perlvar{'lonDaemons'}/lonsql");
+ sleep 120;
+ 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");
+ sleep 120;
+ }
+ } 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 $fh '<hr><a name="lond"><h2>lond</h2><h3>Log</h3><pre>';
if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
-open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lond.log|");
+open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
while ($line=<DFH>) {
print $fh "$line";
- if ($line=~/giving up/) { $notices++; }
+ if ($line=~/INFO/) { $notices++; }
+ if ($line=~/WARNING/) { $notices++; }
+ if ($line=~/CRITICAL/) { $warnings++; }
};
close (DFH);
}
@@ -252,17 +303,37 @@ print $fh "</pre>";
my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
-if (-e $londfile) {
+my $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>";
+ $restartflag=0;
} else {
- $errors++; $errors++;
+ $errors++;
print $fh "<h3>lond at pid $londpid not responding</h3>";
+ # Intelligently handle this.
+ # Possibility #1: there is no process
+ # Solution: remove .pid file and restart
+ if (getpgrp($londpid)==-1) {
+ unlink($londfile);
+ $restartflag=1;
+ }
+ else {
+ # Possibility #2: there is a live process that is not responding
+ # for an unknown reason
+ # Solution: kill parent and children processes, remove .pid and restart
+ `killall -9 lond`;
+ unlink($londfile);
+ $restartflag=1;
+ }
+ print $fh
+ "<h3>Deciding to clean up stale .pid file and restart lond</h3>";
}
-} else {
+}
+if ($restartflag==1) {
$errors++;
print $fh "<h3>lond not running, trying to start</h3>";
system("$perlvar{'lonDaemons'}/lond");
@@ -286,6 +357,17 @@ if (-e $londfile) {
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";
@@ -308,10 +390,12 @@ if ($size>40000) {
print $fh '<hr><a name="lonc"><h2>lonc</h2><h3>Log</h3><pre>';
if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
-open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonc.log|");
+open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
while ($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 +403,7 @@ print $fh "</pre>";
my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
+$restartflag=1;
if (-e $loncfile) {
my $lfh=IO::File->new("$loncfile");
my $loncpid=<$lfh>;
@@ -326,11 +411,30 @@ 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 $loncpid not responding</h3>";
+ # Intelligently handle this.
+ # Possibility #1: there is no process
+ # Solution: remove .pid file and restart
+ if (getpgrp($loncpid)==-1) {
+ unlink($loncfile);
+ $restartflag=1;
+ }
+ else {
+ # Possibility #2: there is a live process that is not responding
+ # for an unknown reason
+ # Solution: kill parent and children processes, remove .pid and restart
+ `killall -9 lonc`;
+ unlink($loncfile);
+ $restartflag=1;
+ }
+ print $fh
+ "<h3>Deciding to clean up stale .pid file and restart lonc</h3>";
}
-} else {
+}
+if ($restartflag==1) {
$errors++;
print $fh "<h3>lonc not running, trying to start</h3>";
system("$perlvar{'lonDaemons'}/lonc");
@@ -354,6 +458,17 @@ if (-e $loncfile) {
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";
@@ -379,9 +494,6 @@ if (-e "$perlvar{'lonDaemons'}/logs/lonn
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);
}
@@ -436,8 +548,6 @@ print $fh "</table>";
print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
-&senddelayed($fh);
-
print $fh '<h3>Scanning Permanent Log</h3>';
$unsend=0;