--- loncom/loncron 2001/08/27 13:54:50 1.22 +++ loncom/loncron 2002/07/25 18:24:27 1.31 @@ -13,16 +13,18 @@ # 12/23 Gerd Kortemeyer # YEAR=2001 # 1/10/2001, 2/12/, 2/26, 3/15, 04/11, 04/21,8/27 Scott Harrison +# 09/04,09/06,11/26 Gerd Kortemeyer +# YEAR=2002 +# 5/11/2002 Scott Harrison + +$|=1; + +use lib '/home/httpd/lib/perl/'; +use LONCAPA::Configuration; 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)=@_; @@ -53,19 +55,13 @@ 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; - } - } - delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed - delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed -} +# --------------------------------- Read loncapa_apache.conf and loncapa.conf +my $perlvarref=LONCAPA::Configuration::read_conf('loncapa_apache.conf', + 'loncapa.conf'); +%perlvar=%{$perlvarref}; +undef $perlvarref; +delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed +delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed # --------------------------------------- Make sure that LON-CAPA is configured # I only test for one thing here (lonHostID). This is just a safeguard. @@ -97,14 +93,21 @@ if ($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-"); + } + } } } @@ -164,17 +167,17 @@ print $fh (< ENDHEADERS -foreach $varname (keys %perlvar) { +foreach $varname (sort(keys(%perlvar))) { print $fh "$varname$perlvar{$varname}\n"; } print $fh "

Hosts

"; -foreach $id (keys %hostname) { -print $fh - ""; -print $fh "\n"; +foreach $id (sort(keys(%hostname))) { + print $fh + ""; + print $fh "\n"; } print $fh "
$id$hostdom{$id}$hostrole{$id}$hostname{$id}$hostip{$id}
$id$hostdom{$id}$hostrole{$id}$hostname{$id}$hostip{$id}

Spare Hosts

    "; -foreach $id (keys %spareid) { +foreach $id (sort(keys(%spareid))) { print $fh "
  1. $id\n"; } @@ -210,7 +213,8 @@ while ($line=) { $usage=$parts[4]; $usage=~s/\W//g; if ($usage>90) { - $warnings++; + $warnings++; + $notices++; } elsif ($usage>80) { $warnings++; } elsif ($usage>60) { @@ -220,11 +224,29 @@ while ($line=) { } close (DFH); print $fh ""; + + +print $fh "

    ps

    "; +print $fh "
    ";
+$psproc=0;
+
+open (PSH,"ps -aux|");
+while ($line=) { 
+   print $fh "$line"; 
+   $psproc++;
+}
+close (PSH);
+print $fh "
    "; + +if ($psproc>200) { $notices++; } +if ($psproc>250) { $notices++; } + &errout($fh); # --------------------------------------------------------------- clean out tmp print $fh '

    Temporary Files

    '; $cleaned=0; +$old=0; while ($fname=<$perlvar{'lonDaemons'}/tmp/*>) { my ($dev,$ino,$mode,$nlink, $uid,$gid,$rdev,$size, @@ -233,12 +255,24 @@ while ($fname=<$perlvar{'lonDaemons'}/tm $now=time; $since=$now-$mtime; if ($since>$perlvar{'lonExpire'}) { - $cleaned++; - unlink("$fname"); + $line=''; + if (open(PROBE,$fname)) { + $line=; + close(PROBE); + } + unless ($line=~/^CHECKOUTTOKEN\&/) { + $cleaned++; + unlink("$fname"); + } else { + if ($since>100*$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 '

    Session Tokens

    '; @@ -267,13 +301,13 @@ 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++; } 
@@ -289,7 +323,7 @@ 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=) { 
@@ -303,7 +337,9 @@ 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>;
@@ -314,38 +350,29 @@ if ($perlvar{'lonRole'} eq "library") {
 	} else {
 	    $errors++; $errors++;
 	    print $fh "
    lonsql at pid $lonsqlpid not responding
    ";
-	    # Intelligently handle this.
-	    # Possibility #1: there is no process
-	    # Solution: remove .pid file and restart
-	    if (getpgrp($lonsqlpid)==-1) {
-		unlink($lonsqlfile);
 		$restartflag=1;
-	    }
-	    else {
-		# Possibility #2: there is a live process that is not
-		# responding for an unknown reason
-		# Solution: kill parent and children processes, remove .pid
-		# and restart
-		`killall -9 lonsql`;
-		unlink($lonsqlfile);
-		$restartflag=1;
-	    }
-	}
 	print $fh 
-	    "
    Deciding to clean up stale .pid file and restart lonsql
    ";
+	    "
    Decided to clean up stale .pid file and restart lonsql
    ";
+	}
     }
     if ($restartflag==1) {
 	$errors++;
+	         print $fh '
    Killall lonsql: '.
+                    system('killall lonsql').' - ';
+                    sleep 2;
+                    print $fh unlink($lonsqlfile).' - '.
+                              system('killall -9 lonsql').
+                    '
    ';
 	print $fh "
    lonsql not running, trying to start
    ";
 	system(
  "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors");
-	sleep 120 unless $qflag;
+	sleep 2;
 	if (-e $lonsqlfile) {
 	    print $fh "Seems like it started ...
    ";
 	    my $lfh=IO::File->new("$lonsqlfile");
 	    my $lonsqlpid=<$lfh>;
 	    chomp($lonsqlpid);
-	    sleep 30 unless $qflag;
+	    sleep 2;
 	    if (kill 0 => $lonsqlpid) {
 		print $fh "
    lonsql at pid $lonsqlpid responding
    ";
 	    } else {
@@ -354,7 +381,7 @@ if ($perlvar{'lonRole'} eq "library") {
 		print $fh "Give it one more try ...
    ";
 		system(
  "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors");
-		sleep 120 unless $qflag;
+		sleep 2;
 	    }
 	} else {
 	    print $fh "Seems like that did not work!
    ";
@@ -392,9 +419,10 @@ if ($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++; }
@@ -413,42 +441,34 @@ if (-e $londfile) {
    my $londpid=<$lfh>;
    chomp($londpid);
    if (kill 0 => $londpid) {
-      print $fh "
    lond at pid $londpid responding
    ";
+      print $fh "
    lond at pid $londpid responding, sending USR1
    ";
+      kill USR1 => $londpid;
       $restartflag=0;
    } 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 2;
+          print $fh unlink($londfile).' - '.system('killall -9 lond').
+                    '
    ';
    print $fh "
    lond not running, trying to start
    ";
    system(
      "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
-   sleep 120 unless $qflag;
+   sleep 2;
    if (-e $londfile) {
        print $fh "Seems like it started ...
    ";
        my $lfh=IO::File->new("$londfile");
        my $londpid=<$lfh>;
        chomp($londpid);
-       sleep 30 unless $qflag;
+       sleep 2;
        if (kill 0 => $londpid) {
           print $fh "
    lond at pid $londpid responding
    ";
        } else {
@@ -457,7 +477,7 @@ if ($restartflag==1) {
           print $fh "Give it one more try ...
    ";
 	  system(
  "$perlvar{'lonDaemons'}/lond 2>>$perlvar{'lonDaemons'}/logs/lond_errors");
-          sleep 120 unless $qflag;
+          sleep 2;
        }
    } else {
        print $fh "Seems like that did not work!
    ";
@@ -494,9 +514,10 @@ if ($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++; }
@@ -521,37 +542,29 @@ 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 2;
+          print $fh unlink($loncfile).' - '.system('killall -9 lonc').
+                    '
    ';
    print $fh "
    lonc not running, trying to start
    ";
 	system(
  "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
-   sleep 120 unless $qflag;
+   sleep 2;
    if (-e $loncfile) {
        print $fh "Seems like it started ...
    ";
        my $lfh=IO::File->new("$loncfile");
        my $loncpid=<$lfh>;
        chomp($loncpid);
-       sleep 30 unless $qflag;
+       sleep 2;
        if (kill 0 => $loncpid) {
           print $fh "
    lonc at pid $loncpid responding
    ";
        } else {
@@ -560,7 +573,7 @@ if ($restartflag==1) {
           print $fh "Give it one more try ...
    ";
  	  system(
  "$perlvar{'lonDaemons'}/lonc 2>>$perlvar{'lonDaemons'}/logs/lonc_errors");
-          sleep 120 unless $qflag;
+          sleep 2;
        }
    } else {
        print $fh "Seems like that did not work!
    ";
@@ -598,6 +611,7 @@ if ($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=) { 
@@ -636,7 +650,7 @@ print $fh "
    ";
 print $fh '
    Connections
    ';
 
 print $fh "";
-foreach $tryserver (keys %hostname) {
+foreach $tryserver (sort(keys(%hostname))) {
 
     $answer=reply("pong",$tryserver);
     if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
@@ -655,6 +669,7 @@ print $fh "
    ";
 # ------------------------------------------------------------ Delayed messages
 
 print $fh '
    Delayed Messages
    ';
+print "buffers\n";
 
 print $fh '
    Scanning Permanent Log
    ';
 
@@ -691,17 +706,17 @@ 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")
-    unless $qflag;
+ "metasend -b -t $emailto -s '$subj' -f $statusdir/index.html -m text/html");
 }
 1;