--- loncom/loncron 2000/12/23 16:56:46 1.11 +++ loncom/loncron 2001/09/04 17:58:44 1.23 @@ -9,12 +9,20 @@ # # 7/14,7/15,7/19,7/21,7/22,11/18, # 2/8 Gerd Kortemeyer -# Dec 00 Scott Harrison +# 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 use IO::File; use IO::Socket; +my $qflag=0; +if (@ARGV) { + my $arg=shift @ARGV; + $qflag=1 if $arg eq 'quick'; +} + # -------------------------------------------------- Non-critical communication sub reply { my ($cmd,$server)=@_; @@ -45,7 +53,6 @@ ENDERROUT # ================================================================ Main Program - # ------------------------------------------------------------ Read access.conf { my $config=IO::File->new("/etc/httpd/conf/access.conf"); @@ -56,6 +63,33 @@ ENDERROUT $perlvar{$varname}=$varvalue; } } + 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 @@ -233,13 +267,13 @@ print $fh "

$active open session(s)

httpd

Access Log

';
 
-open (DFH,"tail -n40 /etc/httpd/logs/access_log|");
+open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
 while ($line=) { print $fh "$line" };
 close (DFH);
 
 print $fh "

Error Log

";
 
-open (DFH,"tail -n50 /etc/httpd/logs/error_log|");
+open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
 while ($line=) { 
    print $fh "$line";
    if ($line=~/\[error\]/) { $notices++; } 
@@ -250,13 +284,12 @@ print $fh "
"; # ---------------------------------------------------------------------- lonsql -# -# Do not run for now -# -if ($perlvar{'lonRole'} eq "library" && 1==0) { + +my $restartflag=1; +if ($perlvar{'lonRole'} eq "library") { print $fh '

lonsql

Log

';
-    
+    print "lonsql\n";
     if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
 	open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
 	while ($line=) { 
@@ -270,36 +303,51 @@ if ($perlvar{'lonRole'} eq "library" &&
     print $fh "
"; 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 "

lonsql at pid $lonsqlpid responding

"; + $restartflag=0; } else { $errors++; $errors++; print $fh "

lonsql at pid $lonsqlpid not responding

"; + $restartflag=1; + print $fh + "

Decided to clean up stale .pid file and restart lonsql

"; } - } else { + } + if ($restartflag==1) { $errors++; + print $fh '
Killall lonsql: '. + system('killall lonsql').' - '; + sleep 60; + print $fh unlink($lonsqlfile).' - '. + system('killall -9 lonsql'). + '
'; print $fh "

lonsql not running, trying to start

"; - system("$perlvar{'lonDaemons'}/lonsql"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors"); + sleep 120 unless $qflag; if (-e $lonsqlfile) { print $fh "Seems like it started ...

"; my $lfh=IO::File->new("$lonsqlfile"); my $lonsqlpid=<$lfh>; chomp($lonsqlpid); - sleep 30; + sleep 30 unless $qflag; if (kill 0 => $lonsqlpid) { print $fh "

lonsql at pid $lonsqlpid responding

"; } else { $errors++; $errors++; print $fh "

lonsql at pid $lonsqlpid not responding

"; print $fh "Give it one more try ...

"; - system("$perlvar{'lonDaemons'}/lonsql"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors"); + sleep 120 unless $qflag; } } else { print $fh "Seems like that did not work!

"; @@ -337,9 +385,10 @@ if ($perlvar{'lonRole'} eq "library" && # ------------------------------------------------------------------------ lond print $fh '


lond

Log

';
+print "lond\n";
 
 if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
-open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
+open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lond.log|");
 while ($line=) { 
    print $fh "$line";
    if ($line=~/INFO/) { $notices++; }
@@ -352,7 +401,7 @@ print $fh "
"; my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid"; -my $restartflag=1; +$restartflag=1; if (-e $londfile) { my $lfh=IO::File->new("$londfile"); my $londpid=<$lfh>; @@ -363,44 +412,37 @@ if (-e $londfile) { } else { $errors++; print $fh "

lond at pid $londpid not responding

"; - # 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; - } + $restartflag=1; print $fh - "

Deciding to clean up stale .pid file and restart lond

"; + "

Decided to clean up stale .pid file and restart lond

"; } } if ($restartflag==1) { $errors++; + print $fh '
Killall lond: '. + system('killall lond').' - '; + sleep 60; + print $fh unlink($londfile).' - '.system('killall -9 lond'). + '
'; print $fh "

lond not running, trying to start

"; - system("$perlvar{'lonDaemons'}/lond"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); + sleep 120 unless $qflag; if (-e $londfile) { print $fh "Seems like it started ...

"; my $lfh=IO::File->new("$londfile"); my $londpid=<$lfh>; chomp($londpid); - sleep 30; + sleep 30 unless $qflag; if (kill 0 => $londpid) { print $fh "

lond at pid $londpid responding

"; } else { $errors++; $errors++; print $fh "

lond at pid $londpid not responding

"; print $fh "Give it one more try ...

"; - system("$perlvar{'lonDaemons'}/lond"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); + sleep 120 unless $qflag; } } else { print $fh "Seems like that did not work!

"; @@ -437,9 +479,10 @@ if ($size>40000) { # ------------------------------------------------------------------------ lonc print $fh '


lonc

Log

';
+print "lonc\n";
 
 if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
-open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
+open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonc.log|");
 while ($line=) { 
    print $fh "$line";
    if ($line=~/INFO/) { $notices++; }
@@ -464,44 +507,38 @@ if (-e $loncfile) {
    } else {
       $errors++;
       print $fh "

lonc at pid $loncpid not responding

"; - # 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 - "

Deciding to clean up stale .pid file and restart lonc

"; + "

Decided to clean up stale .pid file and restart lonc

"; } } if ($restartflag==1) { $errors++; + print $fh '
Killall lonc: '. + system('killall lonc').' - '; + sleep 60; + print $fh unlink($loncfile).' - '.system('killall -9 lonc'). + '
'; print $fh "

lonc not running, trying to start

"; - system("$perlvar{'lonDaemons'}/lonc"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors"); + sleep 120 unless $qflag; if (-e $loncfile) { print $fh "Seems like it started ...

"; my $lfh=IO::File->new("$loncfile"); my $loncpid=<$lfh>; chomp($loncpid); - sleep 30; + sleep 30 unless $qflag; if (kill 0 => $loncpid) { print $fh "

lonc at pid $loncpid responding

"; } else { $errors++; $errors++; print $fh "

lonc at pid $loncpid not responding

"; print $fh "Give it one more try ...

"; - system("$perlvar{'lonDaemons'}/lonc"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors"); + sleep 120 unless $qflag; } } else { print $fh "Seems like that did not work!

"; @@ -539,6 +576,7 @@ if ($size>40000) { # ---------------------------------------------------------------------- lonnet print $fh '


lonnet

Temp Log

';
+print "lonnet\n";
 if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
 open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
 while ($line=) { 
@@ -596,6 +634,7 @@ print $fh "";
 # ------------------------------------------------------------ Delayed messages
 
 print $fh '

Delayed Messages

'; +print "buffers\n"; print $fh '

Scanning Permanent Log

'; @@ -632,16 +671,18 @@ print $fh "

Total Error Count: $total $now=time; $date=localtime($now); print $fh "
$date ($now)\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"); + "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html") + unless $qflag; } 1;