--- loncom/loncron 1999/11/18 19:52:46 1.2 +++ loncom/loncron 2001/03/15 13:20:46 1.18 @@ -7,7 +7,11 @@ # 6/1/99,6/2,6/10,6/11,6/12,6/14,6/26,6/28,6/29,6/30, # 7/1,7/2,7/9,7/10,7/12 Gerd Kortemeyer) # -# 7/14,7/15,7/19,7/21,7/22,11/18 Gerd Kortemeyer +# 7/14,7/15,7/19,7/21,7/22,11/18, +# 2/8 Gerd Kortemeyer +# Dec 00 Scott Harrison +# 12/23 Gerd Kortemeyer +# 02/12/2001 Scott Harrison use IO::File; use IO::Socket; @@ -42,7 +46,6 @@ ENDERROUT # ================================================================ Main Program - # ------------------------------------------------------------ Read access.conf { my $config=IO::File->new("/etc/httpd/conf/access.conf"); @@ -55,6 +58,31 @@ ENDERROUT } } +# --------------------------------------- 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 { my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab"); @@ -102,14 +130,17 @@ print $fh (< LON Status Report $perlvar{'lonHostID'} - +

LON Status Report $perlvar{'lonHostID'}

$date ($now)

  1. Configuration
  2. Machine Information +
  3. Temporary Files +
  4. Session Tokens
  5. httpd +
  6. lonsql
  7. lond
  8. lonc
  9. lonnet @@ -152,7 +183,7 @@ close (LOADAVGH); print $fh "$loadavg"; @parts=split(/\s+/,$loadavg); -if ($parts[1]>3.0) { +if ($parts[1]>4.0) { $errors++; } elsif ($parts[1]>2.0) { $warnings++; @@ -170,17 +201,59 @@ while ($line=) { $usage=$parts[4]; $usage=~s/\W//g; if ($usage>90) { - $errors++; + $warnings++; } elsif ($usage>80) { $warnings++; } elsif ($usage>60) { $notices++; } - if ($usage>95) { $errors++; } + if ($usage>95) { $warnings++; $warnings++ } } close (DFH); print $fh ""; &errout($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

    ';
@@ -199,15 +272,106 @@ while ($line=) {
 close (DFH);
 print $fh "
    "; &errout($fh); + + +# ---------------------------------------------------------------------- lonsql +# +# Do not run for now +# +if ($perlvar{'lonRole'} eq "library") { + + print $fh '

    lonsql

    Log

    ';
+    
+    if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
+	open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
+	while ($line=) { 
+	    print $fh "$line";
+	    if ($line=~/INFO/) { $notices++; }
+	    if ($line=~/WARNING/) { $notices++; }
+	    if ($line=~/CRITICAL/) { $warnings++; }
+	};
+	close (DFH);
+    }
+    print $fh "
    "; + + my $lonsqlfile="$perlvar{'lonDaemons'}/logs/lonsql.pid"; + + 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

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

    lonsql at pid $lonsqlpid not responding

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

    lonsql not running, trying to start

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

    "; + my $lfh=IO::File->new("$lonsqlfile"); + my $lonsqlpid=<$lfh>; + chomp($lonsqlpid); + sleep 30; + 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( + "$perlvar{'lonDaemons'}/lonsql 2>>$perlvar{'lonDaemons'}/logs/lonsql_errors"); + sleep 120; + } + } else { + print $fh "Seems like that did not work!

    "; + $errors++; + } + if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){ + print $fh "

    ";
+	    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
+	    while ($line=) { 
+		print $fh "$line";
+		if ($line=~/WARNING/) { $notices++; }
+		if ($line=~/CRITICAL/) { $notices++; }
+	    };
+	    close (DFH);
+	    print $fh "
    "; + } + } + + $fname="$perlvar{'lonDaemons'}/logs/lonsql.log"; + + my ($dev,$ino,$mode,$nlink, + $uid,$gid,$rdev,$size, + $atime,$mtime,$ctime, + $blksize,$blocks)=stat($fname); + + if ($size>40000) { + print $fh "Rotating logs ...

    "; + rename("$fname.2","$fname.3"); + rename("$fname.1","$fname.2"); + rename("$fname","$fname.1"); + } + + &errout($fh); +} # ------------------------------------------------------------------------ lond print $fh '

    lond

    Log

    ';
 
 if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
-open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lond.log|");
+open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
 while ($line=) { 
    print $fh "$line";
-   if ($line=~/giving up/) { $notices++; }
+   if ($line=~/INFO/) { $notices++; }
+   if ($line=~/WARNING/) { $notices++; }
+   if ($line=~/CRITICAL/) { $warnings++; }
 };
 close (DFH);
 }
@@ -215,20 +379,41 @@ print $fh "
    "; my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid"; -if (-e $londfile) { +my $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

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

    Deciding to clean up stale .pid file and restart lond

    "; } -} else { +} +if ($restartflag==1) { $errors++; print $fh "

    lond not running, trying to start

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

    "; @@ -242,13 +427,25 @@ if (-e $londfile) { $errors++; $errors++; print $fh "

    lond at pid $londpid not responding

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

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

    "; $errors++; } + if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){ + print $fh "

    ";
+    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
+    while ($line=) { 
+      print $fh "$line";
+      if ($line=~/WARNING/) { $notices++; }
+      if ($line=~/CRITICAL/) { $notices++; }
+    };
+    close (DFH);
+    print $fh "
    "; + } } $fname="$perlvar{'lonDaemons'}/logs/lond.log"; @@ -271,10 +468,12 @@ if ($size>40000) { print $fh '

    lonc

    Log

    ';
 
 if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
-open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonc.log|");
+open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
 while ($line=) { 
    print $fh "$line";
-   if ($line=~/died/) { $notices++; }
+   if ($line=~/INFO/) { $notices++; }
+   if ($line=~/WARNING/) { $notices++; }
+   if ($line=~/CRITICAL/) { $warnings++; }
 };
 close (DFH);
 }
@@ -282,6 +481,7 @@ print $fh "
    "; my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid"; +$restartflag=1; if (-e $loncfile) { my $lfh=IO::File->new("$loncfile"); my $loncpid=<$lfh>; @@ -289,14 +489,34 @@ if (-e $loncfile) { if (kill 0 => $loncpid) { print $fh "

    lonc at pid $loncpid responding, sending USR1

    "; kill USR1 => $loncpid; + $restartflag=0; } else { - $errors++; $errors++; + $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

    "; } -} else { +} +if ($restartflag==1) { $errors++; print $fh "

    lonc not running, trying to start

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

    "; @@ -310,13 +530,25 @@ if (-e $loncfile) { $errors++; $errors++; print $fh "

    lonc at pid $loncpid not responding

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

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

    "; $errors++; } + if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") { + print $fh "

    ";
+    open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
+    while ($line=) { 
+      print $fh "$line";
+      if ($line=~/WARNING/) { $notices++; }
+      if ($line=~/CRITICAL/) { $notices++; }
+    };
+    close (DFH);
+    print $fh "
    "; + } } $fname="$perlvar{'lonDaemons'}/logs/lonc.log"; @@ -342,13 +574,10 @@ if (-e "$perlvar{'lonDaemons'}/logs/lonn open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|"); while ($line=) { print $fh "$line"; - if ($line=~/Delayed/) { $warnings++; } - if ($line=~/giving up/) { $warnings++; } - if ($line=~/FAILED/) { $errors++; } }; 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|");