--- loncom/loncron 2003/09/10 19:13:09 1.42 +++ loncom/loncron 2004/01/09 17:43:40 1.45 @@ -51,7 +51,12 @@ ENDERROUT sub start_daemon { my ($fh,$daemon,$pidfile) = @_; - system("$perlvar{'lonDaemons'}/$daemon 2>>$perlvar{'lonDaemons'}/logs/${daemon}_errors"); + 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>"; @@ -183,13 +188,13 @@ delete $perlvar{'lonSqlAccess'}; # remov # --------------------------------------- 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.' |\ + 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; } @@ -197,18 +202,18 @@ if ('{[[[[lonHostID]]]]}' eq $perlvar{'l # ----------------------------- 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.' |\ + 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 { 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 && $ip) { @@ -231,12 +236,12 @@ if ($wwwid!=$<) { # ------------------------------------------------------ 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; - } + chomp($configline); + if (($configline) && ($configline ne $perlvar{'lonHostID'})) { + $spareid{$configline}=1; + } } } @@ -252,10 +257,10 @@ $now=time; $date=localtime($now); { -my $fh=IO::File->new(">$statusdir/newstatus.html"); -my %simplestatus=(); - -print $fh (<<ENDHEADERS); + my $fh=IO::File->new(">$statusdir/newstatus.html"); + my %simplestatus=(); + + print $fh (<<ENDHEADERS); <html> <head> <title>LON Status Report $perlvar{'lonHostID'}</title> @@ -286,274 +291,274 @@ print $fh (<<ENDHEADERS); <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"; + 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 '<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>"; + 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 $fh "</pre>"; + open (DFH,"df|"); + while ($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>"; + print $fh "<h3>ps</h3>"; + print $fh "<pre>"; + $psproc=0; + + open (PSH,"ps -aux --cols 140 |"); + while ($line=<PSH>) { + print $fh "$line"; + $psproc++; + } + close (PSH); + print $fh "</pre>"; -if ($psproc>200) { $notices++; } -if ($psproc>250) { $notices++; } + if ($psproc>200) { $notices++; } + if ($psproc>250) { $notices++; } -&errout($fh); + &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 '<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)."; + } + 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>"; + 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++; + } -# ----------------------------------------------------------------------- httpd + } + print $fh "<p>Cleaned up ".$cleaned." stale session token(s)."; + print $fh "<h3>$active open session(s)</h3>"; -print $fh '<hr><a name="httpd"><h2>httpd</h2><h3>Access Log</h3><pre>'; +# ----------------------------------------------------------------------- httpd -open (DFH,"tail -n25 /etc/httpd/logs/access_log|"); -while ($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); + 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($fh,'lonsql',200000); + &checkon_daemon($fh,'lonsql',200000); # ------------------------------------------------------------------------ lond -&checkon_daemon($fh,'lond',40000,1); + &checkon_daemon($fh,'lond',40000,1); # ------------------------------------------------------------------------ lonc -&checkon_daemon($fh,'lonc',40000,1); + &checkon_daemon($fh,'lonc',40000,1); # -------------------------------------------------------------------- lonhttpd -&checkon_daemon($fh,'lonhttpd',40000); + &checkon_daemon($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>"; + 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 $fh "No perm log\n" } -if (-e "$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"); -} + $fname="$perlvar{'lonDaemons'}/logs/lonnet.log"; -print $fh "</pre>"; -&errout($fh); -# ----------------------------------------------------------------- Connections + my ($dev,$ino,$mode,$nlink, + $uid,$gid,$rdev,$size, + $atime,$mtime,$ctime, + $blksize,$blocks)=stat($fname); -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') { $warnings++; } + if ($size>40000) { + print $fh "Rotating logs ...<p>"; + rename("$fname.2","$fname.3"); + rename("$fname.1","$fname.2"); + rename("$fname","$fname.1"); } - if ($answer =~ /con_lost/) { print(" $tryserver down\n"); } - print $fh "<tr><td>$tryserver</td><td>$result</td></tr>\n"; -} -print $fh "</table>"; + 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); + &errout($fh); # ------------------------------------------------------------ Delayed messages -print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>'; -print "checking buffers\n"; + print $fh '<hr><a name="delayed"><h2>Delayed Messages</h2>'; + print "checking buffers\n"; -print $fh '<h3>Scanning Permanent Log</h3>'; + 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++; } + $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; + print $fh "Total unsend messages: <b>$unsend</b><p>\n"; + $warnings=$warnings+5*$unsend; -if ($unsend) { $simplestatus{'unsend'}=$unsend; } -print $fh "<h3>Outgoing Buffer</h3>"; + if ($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); + 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(); + 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; } @@ -562,22 +567,21 @@ $simplestatus{'time'}=time; rename ("$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(); + 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"); + 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;