--- loncom/loncron 2000/12/08 22:08:57 1.8 +++ loncom/loncron 2001/11/26 21:00:42 1.25 @@ -9,6 +9,13 @@ # # 7/14,7/15,7/19,7/21,7/22,11/18, # 2/8 Gerd Kortemeyer +# 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 +# 09/04,09/06,11/26 Gerd Kortemeyer + +$|=1; use IO::File; use IO::Socket; @@ -43,7 +50,6 @@ ENDERROUT # ================================================================ Main Program - # ------------------------------------------------------------ Read access.conf { my $config=IO::File->new("/etc/httpd/conf/access.conf"); @@ -54,6 +60,33 @@ 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 @@ -110,7 +143,10 @@ print $fh (<
  • Configuration
  • Machine Information +
  • Temporary Files +
  • Session Tokens
  • httpd +
  • lonsql
  • lond
  • lonc
  • lonnet @@ -171,7 +207,8 @@ while ($line=) { $usage=$parts[4]; $usage=~s/\W//g; if ($usage>90) { - $warnings++; + $warnings++; + $notices++; } elsif ($usage>80) { $warnings++; } elsif ($usage>60) { @@ -181,18 +218,77 @@ 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

    '; +$cleaned=0; +while ($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'}) { + $cleaned++; + unlink("$fname"); + } + +} +print $fh "Cleaned up ".$cleaned." files."; + +# ------------------------------------------------------------ clean out lonIDs +print $fh '

    Session Tokens

    '; +$cleaned=0; +$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 $fname
    "; + unlink("$fname"); + } else { + $active++; + } + +} +print $fh "

    Cleaned up ".$cleaned." stale session token(s)."; +print $fh "

    $active open session(s)

    "; + # ----------------------------------------------------------------------- httpd print $fh '

    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++; } 
@@ -202,12 +298,13 @@ print $fh "
    "; &errout($fh); -# ------------------------------------------------------------------------ lonsql +# ---------------------------------------------------------------------- lonsql -if ($perlvar{'lonRole'} eq "library" && 1==0) { +my $restartflag=1; +if ($perlvar{'lonRole'} eq "library") { - print $fh '

    lonsql

    Log

    ';
-    
+    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=) { 
@@ -221,22 +318,36 @@ 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 10;
 	if (-e $lonsqlfile) {
 	    print $fh "Seems like it started ...
    ";
 	    my $lfh=IO::File->new("$lonsqlfile");
@@ -249,8 +360,9 @@ if ($perlvar{'lonRole'} eq "library" &&
 		$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 10;
 	    }
 	} else {
 	    print $fh "Seems like that did not work!
    ";
@@ -288,9 +400,10 @@ if ($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++; }
@@ -303,42 +416,34 @@ 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>;
    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 zombie process
-      # Possibility #3: there is a live process that is not responding
-      #                 for an unknown reason
-      # Solution: kill process, remove .pid and restart
-	  kill 2 => $londpid;
-	  unlink($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 10;
    if (-e $londfile) {
        print $fh "Seems like it started ...
    ";
        my $lfh=IO::File->new("$londfile");
@@ -351,8 +456,9 @@ if ($restartflag==1) {
           $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 10;
        }
    } else {
        print $fh "Seems like that did not work!
    ";
@@ -389,9 +495,10 @@ if ($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++; }
@@ -416,31 +523,23 @@ 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);
+      # Solution: kill parent and children processes, remove .pid and restart
 	  $restartflag=1;
-      }
-      else {
-      # Possibility #2: there is a zombie process
-      # Possibility #3: there is a live process that is not responding
-      #                 for an unknown reason
-      # Solution: kill process, remove .pid and restart
-	  kill 2 => $loncpid;
-	  unlink($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 10;
    if (-e $loncfile) {
        print $fh "Seems like it started ...
    ";
        my $lfh=IO::File->new("$loncfile");
@@ -453,8 +552,9 @@ if ($restartflag==1) {
           $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 10;
        }
    } else {
        print $fh "Seems like that did not work!
    ";
@@ -492,6 +592,7 @@ if ($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=) { 
@@ -499,7 +600,7 @@ while ($line=) {
 };
 close (DFH);
 }
-print $fh "
    Perm Log
    ";
+print $fh "
    Perm Log
    ";
 
 if (-e "$perlvar{'lonDaemons'}/logs/lonnet.perm.log") {
     open(DFH,"tail -n10 $perlvar{'lonDaemons'}/logs/lonnet.perm.log|");
@@ -549,6 +650,7 @@ print $fh "";
 # ------------------------------------------------------------ Delayed messages
 
 print $fh '
    Delayed Messages
    ';
+print "buffers\n";
 
 print $fh '
    Scanning Permanent Log
    ';
 
@@ -585,12 +687,13 @@ 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(