--- loncom/loncron 2001/01/10 17:39:36 1.13 +++ loncom/loncron 2003/07/30 16:49:27 1.41 @@ -9,8 +9,14 @@ # # 7/14,7/15,7/19,7/21,7/22,11/18, # 2/8 Gerd Kortemeyer -# Dec 00 Scott Harrison # 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; @@ -45,27 +51,36 @@ ENDERROUT # ================================================================ 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 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") unless $noprint; + 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; + exit 1; } # ------------------------------------------------------------- Read hosts file @@ -73,14 +88,21 @@ if ($wwwid!=$<) { my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab"); 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}=$name; - } + my ($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-"); + } + } } } @@ -109,6 +131,7 @@ $date=localtime($now); { my $fh=IO::File->new(">$statusdir/newstatus.html"); +my %simplestatus=(); print $fh (<<ENDHEADERS); <html> @@ -128,6 +151,7 @@ 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> @@ -140,17 +164,17 @@ print $fh (<<ENDHEADERS); <table border=2> ENDHEADERS -foreach $varname (keys %perlvar) { +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 (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"; +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 (keys %spareid) { +foreach $id (sort(keys(%spareid))) { print $fh "<li>$id\n"; } @@ -186,21 +210,40 @@ while ($line=<DFH>) { $usage=$parts[4]; $usage=~s/\W//g; if ($usage>90) { - $warnings++; + $warnings++; + $notices++; } elsif ($usage>80) { $warnings++; } elsif ($usage>60) { $notices++; } - if ($usage>95) { $warnings++; $warnings++ } + if ($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, @@ -209,12 +252,24 @@ while ($fname=<$perlvar{'lonDaemons'}/tm $now=time; $since=$now-$mtime; if ($since>$perlvar{'lonExpire'}) { - $cleaned++; - unlink("$fname"); + $line=''; + if (open(PROBE,$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."; +print $fh "Cleaned up ".$cleaned." files (".$old." old checkout tokens)."; # ------------------------------------------------------------ clean out lonIDs print $fh '<hr><a name="tokens"><h2>Session Tokens</h2>'; @@ -243,13 +298,13 @@ print $fh "<h3>$active open session(s)</ 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++; } @@ -260,13 +315,10 @@ print $fh "</pre>"; # ---------------------------------------------------------------------- lonsql -# -# Do not run for now -# -if ($perlvar{'lonRole'} eq "library" && 1==0) { +my $restartflag=1; 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>) { @@ -280,36 +332,53 @@ if ($perlvar{'lonRole'} eq "library" && 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 $fh + "<h3>Decided to clean up stale .pid file and restart lonsql</h3>"; } - } else { + } + if ($restartflag==1) { + $simplestatus{'lonsql'}='off'; $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"); - sleep 120; + 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 30; + sleep 2; if (kill 0 => $lonsqlpid) { print $fh "<h3>lonsql at pid $lonsqlpid responding</h3>"; + $simplestatus{'lonsql'}='restarted'; } 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; + system( + "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors"); + sleep 2; } } else { print $fh "Seems like that did not work!<p>"; @@ -335,7 +404,7 @@ if ($perlvar{'lonRole'} eq "library" && $atime,$mtime,$ctime, $blksize,$blocks)=stat($fname); - if ($size>40000) { + if ($size>200000) { print $fh "Rotating logs ...<p>"; rename("$fname.2","$fname.3"); rename("$fname.1","$fname.2"); @@ -343,13 +412,13 @@ if ($perlvar{'lonRole'} eq "library" && } &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 -n100 $perlvar{'lonDaemons'}/logs/lond.log|"); +open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lond.log|"); while ($line=<DFH>) { print $fh "$line"; if ($line=~/INFO/) { $notices++; } @@ -362,55 +431,51 @@ print $fh "</pre>"; my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid"; -my $restartflag=1; +$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++; 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; - } + $restartflag=1; print $fh - "<h3>Deciding to clean up stale .pid file and restart lond</h3>"; + "<h3>Decided to clean up stale .pid file and restart lond</h3>"; } } if ($restartflag==1) { + $simplestatus{'lond'}='off'; $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"); - sleep 120; + 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 30; + sleep 2; if (kill 0 => $londpid) { print $fh "<h3>lond at pid $londpid responding</h3>"; + $simplestatus{'lond'}='restarted'; } 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"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors"); + sleep 2; } } else { print $fh "Seems like that did not work!<p>"; @@ -447,9 +512,10 @@ if ($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 -n100 $perlvar{'lonDaemons'}/logs/lonc.log|"); +open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonc.log|"); while ($line=<DFH>) { print $fh "$line"; if ($line=~/INFO/) { $notices++; } @@ -474,44 +540,40 @@ if (-e $loncfile) { } else { $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 $fh - "<h3>Deciding to clean up stale .pid file and restart lonc</h3>"; + "<h3>Decided to clean up stale .pid file and restart lonc</h3>"; } } if ($restartflag==1) { + $simplestatus{'lonc'}='off'; $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"); - sleep 120; + 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 30; + sleep 2; if (kill 0 => $loncpid) { print $fh "<h3>lonc at pid $loncpid responding</h3>"; + $simplestatus{'lonc'}='restarted'; } 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"); - sleep 120; + system( + "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors"); + sleep 2; } } else { print $fh "Seems like that did not work!<p>"; @@ -546,9 +608,108 @@ if ($size>40000) { &errout($fh); +# -------------------------------------------------------------------- lonhttpd + +print $fh '<hr><a name="lonhttpd"><h2>lonhttpd</h2><h3>Log</h3><pre>'; +print "lonhttpd\n"; + +if (-e "$perlvar{'lonDaemons'}/logs/lonhttpd.log"){ +open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/lonhttpd.log|"); +while ($line=<DFH>) { + print $fh "$line"; + if ($line=~/INFO/) { $notices++; } + if ($line=~/WARNING/) { $notices++; } + if ($line=~/CRITICAL/) { $warnings++; } +}; +close (DFH); +} +print $fh "</pre>"; + +my $lonhttpdfile="$perlvar{'lonDaemons'}/logs/lonhttpd.pid"; + +$restartflag=1; +if (-e $lonhttpdfile) { + my $lfh=IO::File->new("$lonhttpdfile"); + my $lonhttpdpid=<$lfh>; + chomp($lonhttpdpid); + if (kill 0 => $lonhttpdpid) { + print $fh "<h3>lonhttpd at pid $lonhttpdpid responding</h3>"; + $restartflag=0; + } else { + $errors++; + print $fh "<h3>lonhttpd at pid $lonhttpdpid not responding</h3>"; + # Solution: kill parent and children processes, remove .pid and restart + $restartflag=1; + print $fh + "<h3>Decided to clean up stale .pid file and restart lonhttpd</h3>"; + } +} +if ($restartflag==1) { + $simplestatus{'lonhttpd'}='off'; + $errors++; + print $fh '<br><font color="red">Killall lonhttpd: '. + system('killall lonhttpd').' - '; + sleep 2; + print $fh unlink($lonhttpdfile).' - '.system('killall -9 lonhttpd'). + '</font><br>'; + print $fh "<h3>lonhttpd not running, trying to start</h3>"; + system( + "$perlvar{'lonDaemons'}/lonhttpd 2>>$perlvar{'lonDaemons'}/logs/lonhttpd_errors"); + sleep 2; + if (-e $lonhttpdfile) { + print $fh "Seems like it started ...<p>"; + my $lfh=IO::File->new("$lonhttpdfile"); + my $lonhttpdpid=<$lfh>; + chomp($lonhttpdpid); + sleep 2; + if (kill 0 => $lonhttpdpid) { + print $fh "<h3>lonhttpd at pid $lonhttpdpid responding</h3>"; + $simplestatus{'lonhttpd'}='restarted'; + } else { + $errors++; $errors++; + print $fh "<h3>lonhttpd at pid $lonhttpdpid not responding</h3>"; + print $fh "Give it one more try ...<p>"; + system( + "$perlvar{'lonDaemons'}/lonhttpd 2>>$perlvar{'lonDaemons'}/logs/lonhttpd_errors"); + sleep 2; + } + } else { + print $fh "Seems like that did not work!<p>"; + $errors++; + } + if (-e "$perlvar{'lonDaemons'}/logs/lonhttpd.log") { + print $fh "<p><pre>"; + open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonhttpd.log|"); + while ($line=<DFH>) { + print $fh "$line"; + if ($line=~/WARNING/) { $notices++; } + if ($line=~/CRITICAL/) { $notices++; } + }; + close (DFH); + print $fh "</pre>"; + } +} + +$fname="$perlvar{'lonDaemons'}/logs/lonhttpd.log"; + + my ($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); # ---------------------------------------------------------------------- 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>) { @@ -587,7 +748,7 @@ print $fh "</pre>"; print $fh '<hr><a name="connections"><h2>Connections</h2>'; print $fh "<table border=2>"; -foreach $tryserver (keys %hostname) { +foreach $tryserver (sort(keys(%hostname))) { $answer=reply("pong",$tryserver); if ($answer eq "$tryserver:$perlvar{'lonHostID'}") { @@ -606,6 +767,7 @@ print $fh "</table>"; # ------------------------------------------------------------ Delayed messages print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>'; +print "buffers\n"; print $fh '<h3>Scanning Permanent Log</h3>'; @@ -626,6 +788,7 @@ $unsend=0; 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|"); @@ -642,13 +805,29 @@ print $fh "<h1>Total Error Count: $total $now=time; $date=localtime($now); print $fh "<hr>$date ($now)</body></html>\n"; - +print "writing done\n"; +$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) { - $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}"; + print "mailing\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");