File:  [LON-CAPA] / loncom / loncron
Revision 1.44: download - view: text, annotated - select for diffs
Tue Oct 14 15:36:21 2003 UTC (20 years, 5 months ago) by albertel
Branches: MAIN
CVS tags: version_1_1_X, version_1_1_3, version_1_1_2, version_1_1_1, version_1_1_0, version_1_0_99_3, version_1_0_99_2, version_1_0_99_1, version_1_0_99, HEAD
- making it easier to run loncnew,
   /etc/init.d/loncontrol startnew
   /etc/init.d/loncontrol restartnew
  will now start loncnew in place of lonc

#!/usr/bin/perl

# The LearningOnline Network
# Housekeeping program, started by cron
#
# (TCP networking package
# 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)
#
# 7/14,7/15,7/19,7/21,7/22,11/18,
# 2/8 Gerd Kortemeyer
# 12/23 Gerd Kortemeyer
# YEAR=2001
# 09/04,09/06,11/26 Gerd Kortemeyer

$|=1;

use lib '/home/httpd/lib/perl/';
use LONCAPA::Configuration;

use IO::File;
use IO::Socket;

# -------------------------------------------------- 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 errout {
   my $fh=shift;
   print $fh (<<ENDERROUT);
     <p><table border=2 bgcolor="#CCCCCC">
     <tr><td>Notices</td><td>$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>
ENDERROUT
}

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) {
	print $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;
	}
    }
    print $fh "Seems like that did not work!<p>";
    $errors++;
    return 0;
}

sub checkon_daemon {
    my ($fh,$daemon,$maxsize,$sendusr1)=@_;

    print $fh '<hr><a name="'.$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 ($line=<DFH>) { 
	    print $fh "$line";
	    if ($line=~/INFO/) { $notices++; }
	    if ($line=~/WARNING/) { $notices++; }
	    if ($line=~/CRITICAL/) { $warnings++; }
	};
	close (DFH);
    }
    print $fh "</pre>";
    
    my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid";
    
    my $restartflag=1;
    
    if (-e $pidfile) {
	my $lfh=IO::File->new("$pidfile");
	my $daemonpid=<$lfh>;
	chomp($daemonpid);
	if (kill 0 => $daemonpid) {
	    print $fh "<h3>$daemon at pid $daemonpid responding";
	    if ($sendusr1) { print $fh ", sending USR1"; }
	    print $fh "</h3>";
	    if ($sendusr1) { kill USR1 => $daemonpid; }
	    $restartflag=0;
	    print "running\n";
	} else {
	    $errors++;
	    print $fh "<h3>$daemon at pid $daemonpid not responding</h3>";
	    $restartflag=1;
	    print $fh "<h3>Decided to clean up stale .pid file and restart $daemon</h3>";
	}
    }
    if ($restartflag==1) {
	$simplestatus{$daemon}='off';
	$errors++;
	print $fh '<br><font color="red">Killall '.$daemon.': '.
	    `killall $daemon 2>&1`.' - ';
	sleep 2;
	print $fh unlink($pidfile).' - '.
	    `killall -9 $daemon 2>&1`.
	    '</font><br>';
	print $fh "<h3>$daemon not running, trying to start</h3>";
	
	if (&start_daemon($fh,$daemon,$pidfile)) {
	    print $fh "<h3>$daemon at pid $daemonpid responding</h3>";
	    $simplestatus{$daemon}='restarted';
	    print "started\n";
	} else {
	    $errors++;
	    print $fh "<h3>$daemon at pid $daemonpid not responding</h3>";
	    print $fh "Give it one more try ...<p>";
	    print " ";
	    if (&start_daemon($fh,$daemon,$pidfile)) {
		print $fh "<h3>$daemon at pid $daemonpid responding</h3>";
		$simplestatus{$daemon}='restarted';
		print "started\n";
	    } else {
		print " failed\n";
		$simplestatus{$daemon}='failed';
		$errors++; $errors++;
		print $fh "<h3>$daemon at pid $daemonpid not responding</h3>";
		print $fh "Unable to start $daemon<p>";
	    }
	}

	if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
	    print $fh "<p><pre>";
	    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|");
	    while ($line=<DFH>) { 
		print $fh "$line";
		if ($line=~/WARNING/) { $notices++; }
		if ($line=~/CRITICAL/) { $notices++; }
	    };
	    close (DFH);
	    print $fh "</pre>";
	}
    }
    
    $fname="$perlvar{'lonDaemons'}/logs/$daemon.log";
    
    my ($dev,$ino,$mode,$nlink,
	$uid,$gid,$rdev,$size,
	$atime,$mtime,$ctime,
	$blksize,$blocks)=stat($fname);
    
    if ($size>$maxsize) {
	print $fh "Rotating logs ...<p>";
	rename("$fname.2","$fname.3");
	rename("$fname.1","$fname.2");
	rename("$fname","$fname.1");
    }

    &errout($fh);
}
# ================================================================ Main Program

# --------------------------------- 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

# --------------------------------------- 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
{
    my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
    
    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;
	    $hostip{$id}=$ip;
	    $hostrole{$id}=$role;
	    if ($domdescr) { $domaindescription{$domain}=$domdescr; }
	    if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) {
		$libserv{$id}=$name;
	    }
	} else {
	    if ($configline) {
#		&logthis("Skipping hosts.tab line -$configline-");
	    }
	}
    }
}

# ------------------------------------------------------ Read spare server file
{
    my $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
    
    while (my $configline=<$config>) {
	chomp($configline);
	if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
	    $spareid{$configline}=1;
	}
    }
}

# ---------------------------------------------------------------- Start report

$statusdir="/home/httpd/html/lon-status";

$errors=0;
$warnings=0;
$notices=0;

$now=time;
$date=localtime($now);

{
    my $fh=IO::File->new(">$statusdir/newstatus.html");
    my %simplestatus=();
    
    print $fh (<<ENDHEADERS);
<html>
<head>
<title>LON Status Report $perlvar{'lonHostID'}</title>
</head>
<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="#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>
</ol>
<hr>
<a name="configuration">
<h2>Configuration</h2>
<h3>PerlVars</h3>
<table border=2>
ENDHEADERS

    foreach $varname (sort(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 (sort(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 (sort(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++; $simplestatus{'diskfull'}++; }
    }
    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;
    $old=0;
    while ($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 ($since>365*$perlvar{'lonExpire'}) {
		    $cleaned++;
		    unlink("$fname");
		} else { $old++; }
	    }
	}
    
    }
    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++;
	}

    }
    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 -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

    &checkon_daemon($fh,'lonsql',200000);

# ------------------------------------------------------------------------ lond

    &checkon_daemon($fh,'lond',40000,1);

# ------------------------------------------------------------------------ lonc

    &checkon_daemon($fh,'lonc',40000,1);

# -------------------------------------------------------------------- lonhttpd

    &checkon_daemon($fh,'lonhttpd',40000);

# ---------------------------------------------------------------------- lonnet

    print $fh '<hr><a name="lonnet"><h2>lonnet</h2><h3>Temp Log</h3><pre>';
    print "checking logs\n";
    if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
	open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
	while ($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");
    }

    print $fh "</pre>";
    &errout($fh);
# ----------------------------------------------------------------- Connections

    print $fh '<hr><a name="connections"><h2>Connections</h2>';
    print "testing connections\n";
    print $fh "<table border=2>";
    foreach $tryserver (sort(keys(%hostname))) {
	print(".");
	$answer=reply("pong",$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"); }
	print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n";

    }
    print $fh "</table>";

    &errout($fh);
# ------------------------------------------------------------ Delayed messages

    print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>';
    print "checking buffers\n";

    print $fh '<h3>Scanning Permanent Log</h3>';

    $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 $fh "<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;

    if ($unsend) { $simplestatus{'unsend'}=$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 "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;

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";
    $sfh->close();
}
if ($totalcount>200) {
    print "sending mail\n";
    $emailto="$perlvar{'lonAdmEMail'}";
    if ($totalcount>1000) {
	$emailto.=",$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");
}
1;









FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>