--- loncom/loncron 2011/04/23 16:18:49 1.88 +++ loncom/loncron 2013/02/02 14:42:01 1.97 @@ -2,7 +2,7 @@ # Housekeeping program, started by cron, loncontrol and loncron.pl # -# $Id: loncron,v 1.88 2011/04/23 16:18:49 raeburn Exp $ +# $Id: loncron,v 1.97 2013/02/02 14:42:01 raeburn Exp $ # # Copyright Michigan State University Board of Trustees # @@ -32,6 +32,8 @@ use strict; use lib '/home/httpd/lib/perl/'; use LONCAPA::Configuration; +use LONCAPA::Checksumming; +use LONCAPA; use Apache::lonnet; use Apache::loncommon; @@ -269,12 +271,7 @@ sub log_machine_info { &log($fh,"

distprobe

"); &log($fh,"
");
-    open(DSH,"$perlvar{'lonDaemons'}/distprobe |");
-    while (my $line=) { 
-	&log($fh,&encode_entities($line,'<>&"')); 
-	$psproc++;
-    }
-    close(DSH);
+    &get_distro($perlvar{'lonDaemons'},$fh);
     &log($fh,"
"); &errout($fh); @@ -514,21 +511,13 @@ sub clean_sockets { # ----------------------------------------------------------------------- httpd sub check_httpd_logs { my ($fh)=@_; - &log($fh,'

httpd

Access Log

');
-    
-    open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
-    while (my $line=) { &log($fh,&encode_entities($line,'<>&"')) };
-    close (DFH);
-	
-    &log($fh,"

Error Log

");
-	
-    open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
-    while (my $line=) { 
-	&log($fh,"$line");
-	if ($line=~/\[error\]/) { $notices++; } 
+    if (open(PIPE,"./lchttpdlogs|")) {
+        while (my $line=) {
+            &log($fh,$line);
+            if ($line=~/\[error\]/) { $notices++; }
+        }
+        close(PIPE);
     }
-    close (DFH);
-    &log($fh,"
"); &errout($fh); } @@ -634,7 +623,9 @@ sub check_delayed_msg { } &log($fh,"

Total unsend messages: $unsend

\n"); - $warnings=$warnings+5*$unsend; + if ($unsend > 0) { + $warnings=$warnings+5*$unsend; + } if ($unsend) { $simplestatus{'unsend'}=$unsend; } &log($fh,"

Outgoing Buffer

\n
");
@@ -648,11 +639,28 @@ sub check_delayed_msg {
     }
     &log($fh,"
\n"); close (DFH); + my %hostname = &Apache::lonnet::all_hostnames(); + my $numhosts = scalar(keys(%hostname)); # pong to all servers that have delayed messages # this will trigger a reverse connection, which should flush the buffers - foreach my $tryserver (keys %servers) { - my $answer=&Apache::lonnet::reply("pong",$tryserver); - &log($fh,"Pong to $tryserver: $answer
"); + foreach my $tryserver (sort(keys(%servers))) { + if ($hostname{$tryserver} || !$numhosts) { + my $answer; + eval { + local $SIG{ ALRM } = sub { die "TIMEOUT" }; + alarm(20); + $answer = &Apache::lonnet::reply("pong",$tryserver); + alarm(0); + }; + if ($@ && $@ =~ m/TIMEOUT/) { + &log($fh,"Attempted pong to $tryserver timed out
"); + print "time out while contacting: $tryserver for pong\n"; + } else { + &log($fh,"Pong to $tryserver: $answer
"); + } + } else { + &log($fh,"$tryserver has delayed messages, but is not part of the cluster -- skipping 'Pong'.
"); + } } } @@ -686,11 +694,22 @@ sub log_simplestatus { } sub write_loncaparevs { + print "Retrieving LON-CAPA version information\n"; if (open(my $fh,">$perlvar{'lonTabDir'}/loncaparevs.tab")) { my %hostname = &Apache::lonnet::all_hostnames(); foreach my $id (sort(keys(%hostname))) { if ($id ne '') { - my $loncaparev = &Apache::lonnet::get_server_loncaparev('',$id,1,'loncron'); + my $loncaparev; + eval { + local $SIG{ ALRM } = sub { die "TIMEOUT" }; + alarm(10); + $loncaparev = + &Apache::lonnet::get_server_loncaparev('',$id,1,'loncron'); + alarm(0); + }; + if ($@ && $@ =~ m/TIMEOUT/) { + print "time out while contacting lonHost: $id for version\n"; + } if ($loncaparev =~ /^[\w.\-]+$/) { print $fh $id.':'.$loncaparev."\n"; } @@ -702,12 +721,23 @@ sub write_loncaparevs { } sub write_serverhomeIDs { + print "Retrieving LON-CAPA lonHostID information\n"; if (open(my $fh,">$perlvar{'lonTabDir'}/serverhomeIDs.tab")) { my %name_to_host = &Apache::lonnet::all_names(); foreach my $name (sort(keys(%name_to_host))) { if ($name ne '') { if (ref($name_to_host{$name}) eq 'ARRAY') { - my $serverhomeID = &Apache::lonnet::get_server_homeID($name,1,'loncron'); + my $serverhomeID; + eval { + local $SIG{ ALRM } = sub { die "TIMEOUT" }; + alarm(10); + $serverhomeID = + &Apache::lonnet::get_server_homeID($name,1,'loncron'); + alarm(0); + }; + if ($@ && $@ =~ m/TIMEOUT/) { + print "Time out while contacting server: $name\n"; + } if ($serverhomeID ne '') { print $fh $name.':'.$serverhomeID."\n"; } else { @@ -721,6 +751,26 @@ sub write_serverhomeIDs { return; } +sub write_checksums { + my $distro = &get_distro($perlvar{'lonDaemons'}); + if ($distro) { + print "Retrieving file version and checksumming.\n"; + my $numchksums = 0; + my ($chksumsref,$versionsref) = + &LONCAPA::Checksumming::get_checksums($distro,$perlvar{'lonDaemons'}, + $perlvar{'lonLib'}, + $perlvar{'lonIncludes'}, + $perlvar{'lonTabDir'}); + if (ref($chksumsref) eq 'HASH') { + $numchksums = scalar(keys(%{$chksumsref})); + } + print "File version retrieved and checksumming completed for $numchksums files.\n"; + } else { + print "File version retrieval and checksumming skipped - could not determine Linux distro.\n"; + } + return; +} + sub send_mail { print "sending mail\n"; my $defdom = $perlvar{'lonDefDomain'}; @@ -738,6 +788,21 @@ sub send_mail { } } +sub get_distro { + my ($dir,$fh) = @_; + my $distro; + if (open(my $disth,"$dir/distprobe |")) { + while (my $line=<$disth>) { + if ($fh) { + &log($fh,&encode_entities($line,'<>&"')); + } + $distro .= $line; + } + close($disth); + } + return $distro; +} + sub usage { print(<200 && !$noemail) { &send_mail(); } } } @@ -876,10 +944,3 @@ sub main () { &main(); 1; - - - - - - -