--- loncom/loncron 2010/08/21 22:23:05 1.87
+++ loncom/loncron 2014/03/17 14:47:38 1.100
@@ -2,7 +2,7 @@
# Housekeeping program, started by cron, loncontrol and loncron.pl
#
-# $Id: loncron,v 1.87 2010/08/21 22:23:05 raeburn Exp $
+# $Id: loncron,v 1.100 2014/03/17 14:47:38 bisitz 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($fh,"<h3>distprobe</h3>");
&log($fh,"<pre>");
- open(DSH,"$perlvar{'lonDaemons'}/distprobe |");
- while (my $line=<DSH>) {
- &log($fh,&encode_entities($line,'<>&"'));
- $psproc++;
- }
- close(DSH);
+ &log($fh,&encode_entities(&LONCAPA::distro(),'<>&"'));
&log($fh,"</pre>");
&errout($fh);
@@ -514,21 +511,13 @@ sub clean_sockets {
# ----------------------------------------------------------------------- httpd
sub check_httpd_logs {
my ($fh)=@_;
- &log($fh,'<hr /><a name="httpd" /><h2>httpd</h2><h3>Access Log</h3><pre>');
-
- open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
- while (my $line=<DFH>) { &log($fh,&encode_entities($line,'<>&"')) };
- close (DFH);
-
- &log($fh,"</pre><h3>Error Log</h3><pre>");
-
- open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
- while (my $line=<DFH>) {
- &log($fh,"$line");
- if ($line=~/\[error\]/) { $notices++; }
+ if (open(PIPE,"./lchttpdlogs|")) {
+ while (my $line=<PIPE>) {
+ &log($fh,$line);
+ if ($line=~/\[error\]/) { $notices++; }
+ }
+ close(PIPE);
}
- close (DFH);
- &log($fh,"</pre>");
&errout($fh);
}
@@ -537,7 +526,7 @@ sub check_httpd_logs {
sub rotate_lonnet_logs {
my ($fh)=@_;
&log($fh,'<hr /><a name="lonnet" /><h2>lonnet</h2><h3>Temp Log</h3><pre>');
- print "checking logs\n";
+ print "Checking logs.\n";
if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
while (my $line=<DFH>) {
@@ -581,7 +570,7 @@ sub rotate_other_logs {
sub test_connections {
my ($fh)=@_;
&log($fh,'<hr /><a name="connections" /><h2>Connections</h2>');
- print "testing connections\n";
+ print "Testing connections.\n";
&log($fh,"<table border='2'>");
my ($good,$bad)=(0,0);
my %hostname = &Apache::lonnet::all_hostnames();
@@ -615,7 +604,7 @@ sub test_connections {
sub check_delayed_msg {
my ($fh)=@_;
&log($fh,'<hr /><a name="delayed" /><h2>Delayed Messages</h2>');
- print "checking buffers\n";
+ print "Checking buffers.\n";
&log($fh,'<h3>Scanning Permanent Log</h3>');
@@ -634,7 +623,9 @@ sub check_delayed_msg {
}
&log($fh,"<p>Total unsend messages: <b>$unsend</b></p>\n");
- $warnings=$warnings+5*$unsend;
+ if ($unsend > 0) {
+ $warnings=$warnings+5*$unsend;
+ }
if ($unsend) { $simplestatus{'unsend'}=$unsend; }
&log($fh,"<h3>Outgoing Buffer</h3>\n<pre>");
@@ -648,11 +639,28 @@ sub check_delayed_msg {
}
&log($fh,"</pre>\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<br />");
+ 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<br />");
+ print "Time out while contacting: $tryserver for pong.\n";
+ } else {
+ &log($fh,"Pong to $tryserver: $answer<br />");
+ }
+ } else {
+ &log($fh,"$tryserver has delayed messages, but is not part of the cluster -- skipping 'Pong'.<br />");
+ }
}
}
@@ -665,7 +673,7 @@ sub finish_logging {
my $now=time;
my $date=localtime($now);
&log($fh,"<hr />$date ($now)</body></html>\n");
- print "lon-status webpage updated\n";
+ print "lon-status webpage updated.\n";
$fh->close();
if ($errors) { $simplestatus{'errors'}=$errors; }
@@ -686,43 +694,95 @@ sub log_simplestatus {
}
sub write_loncaparevs {
- 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');
- if ($loncaparev =~ /^[\w.\-]+$/) {
- print $fh $id.':'.$loncaparev."\n";
- }
+ print "Retrieving LON-CAPA version information.\n";
+ my %hostname = &Apache::lonnet::all_hostnames();
+ my $output;
+ foreach my $id (sort(keys(%hostname))) {
+ if ($id ne '') {
+ 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.\-]+$/) {
+ $output .= $id.':'.$loncaparev."\n";
}
}
- close($fh);
+ }
+ if ($output) {
+ if (open(my $fh,">$perlvar{'lonTabDir'}/loncaparevs.tab")) {
+ print $fh $output;
+ close($fh);
+ &Apache::lonnet::load_loncaparevs();
+ }
}
return;
}
sub write_serverhomeIDs {
- 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');
- if ($serverhomeID ne '') {
- print $fh $name.':'.$serverhomeID."\n";
- } else {
- print $fh $name.':'.$name_to_host{$name}->[0]."\n";
- }
+ print "Retrieving LON-CAPA lonHostID information.\n";
+ my %name_to_host = &Apache::lonnet::all_names();
+ my $output;
+ foreach my $name (sort(keys(%name_to_host))) {
+ if ($name ne '') {
+ if (ref($name_to_host{$name}) eq 'ARRAY') {
+ 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 '') {
+ $output .= $name.':'.$serverhomeID."\n";
+ } else {
+ $output .= $name.':'.$name_to_host{$name}->[0]."\n";
}
}
}
- close($fh);
+ }
+ if ($output) {
+ if (open(my $fh,">$perlvar{'lonTabDir'}/serverhomeIDs.tab")) {
+ print $fh $output;
+ close($fh);
+ &Apache::lonnet::load_serverhomeIDs();
+ }
+ }
+ return;
+}
+
+sub write_checksums {
+ my $distro = &LONCAPA::distro();
+ if ($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";
+ print "Sending mail.\n";
my $defdom = $perlvar{'lonDefDomain'};
my $origmail = $perlvar{'lonAdmEMail'};
my $emailto = &Apache::loncommon::build_recipient_list(undef,
@@ -740,7 +800,7 @@ sub send_mail {
sub usage {
print(<<USAGE);
-loncron - housekeeping program that checks up on various parts of Lon-CAPA
+loncron - housekeeping program that checks up on various parts of LON-CAPA
Options:
--help Display
@@ -796,10 +856,10 @@ sub main () {
# ----------------------------- 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");
+ print("User ID mismatch. This program must be run as user 'www'.\n");
my $emailto="$perlvar{'lonAdmEMail'},$perlvar{'lonSysEMail'}";
my $subj="LON: $perlvar{'lonHostID'} User ID mismatch";
- system("echo 'User ID mismatch. loncron must be run as user www.' |\
+ system("echo 'User ID mismatch. loncron must be run as user www.' |\
mailto $emailto -s '$subj' > /dev/null");
exit 1;
}
@@ -821,9 +881,12 @@ sub main () {
print $fh "$key\n";
}
close($fh);
- my $execpath = $perlvar{'lonDaemons'}.'/lciptables';
- system("$execpath $tmpfile");
- unlink($fh);
+ if (&LONCAPA::try_to_lock('/tmp/lock_lciptables')) {
+ my $execpath = $perlvar{'lonDaemons'}.'/lciptables';
+ system("$execpath $tmpfile");
+ unlink('/tmp/lock_lciptables'); # Remove the lock file.
+ }
+ unlink($tmpfile);
}
}
@@ -868,7 +931,7 @@ sub main () {
&log_simplestatus();
&write_loncaparevs();
&write_serverhomeIDs();
-
+ &write_checksums();
if ($totalcount>200 && !$noemail) { &send_mail(); }
}
}
@@ -876,10 +939,3 @@ sub main () {
&main();
1;
-
-
-
-
-
-
-