Cleaned up ".$cleaned." stale webDAV session token(s).
");
+ &log($fh,"Cleaned up ".$cleaned." stale sockets.
");
+}
+
# ----------------------------------------------------------------------- httpd
sub check_httpd_logs {
my ($fh)=@_;
- &log($fh,'");
my ($good,$bad)=(0,0);
- foreach my $tryserver (sort(keys(%{$hostname}))) {
+ my %hostname = &Apache::lonnet::all_hostnames();
+ foreach my $tryserver (sort(keys(%hostname))) {
print(".");
my $result;
- my $answer=&reply("ping",$tryserver,$hostname);
+ my $answer=&Apache::lonnet::reply("ping",$tryserver);
if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
$result="ok";
$good++;
@@ -511,20 +641,26 @@ sub test_connections {
# ------------------------------------------------------------ Delayed messages
sub check_delayed_msg {
- my ($fh,$hostname)=@_;
+ my ($fh)=@_;
&log($fh,'
Delayed Messages
');
- print "checking buffers\n";
+ print "Checking buffers.\n";
&log($fh,'Scanning Permanent Log
');
my $unsend=0;
+ my %hostname = &Apache::lonnet::all_hostnames();
+ my $numhosts = scalar(keys(%hostname));
+
my $dfh=IO::File->new("$perlvar{'lonDaemons'}/logs/lonnet.perm.log");
while (my $line=<$dfh>) {
my ($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
+ if ($numhosts) {
+ next unless ($hostname{$dserv});
+ }
if ($sdf eq 'F') {
my $local=localtime($time);
- &log($fh,"Failed: $time, $dserv, $dcmd
");
+ &log($fh,"Failed: $time, $dserv, $dcmd
");
$warnings++;
}
if ($sdf eq 'S') { $unsend--; }
@@ -532,7 +668,9 @@ sub check_delayed_msg {
}
&log($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");
@@ -548,9 +686,24 @@ sub check_delayed_msg {
close (DFH);
# 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 $answer=&reply("pong",$tryserver,$hostname);
- &log($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'.
");
+ }
}
}
@@ -563,7 +716,7 @@ sub finish_logging {
my $now=time;
my $date=localtime($now);
&log($fh,"
$date ($now)\n");
- print "lon-status webpage updated\n";
+ print "lon-status webpage updated.\n";
$fh->close();
if ($errors) { $simplestatus{'errors'}=$errors; }
@@ -573,7 +726,7 @@ sub finish_logging {
}
sub log_simplestatus {
- rename ("$statusdir/newstatus.html","$statusdir/index.html");
+ rename("$statusdir/newstatus.html","$statusdir/index.html");
my $sfh=IO::File->new(">$statusdir/loncron_simple.txt");
foreach (keys %simplestatus) {
@@ -583,27 +736,139 @@ sub log_simplestatus {
$sfh->close();
}
+sub write_loncaparevs {
+ 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";
+ }
+ }
+ }
+ if ($output) {
+ if (open(my $fh,">$perlvar{'lonTabDir'}/loncaparevs.tab")) {
+ print $fh $output;
+ close($fh);
+ &Apache::lonnet::load_loncaparevs();
+ }
+ }
+ return;
+}
+
+sub write_serverhomeIDs {
+ 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{$name}->[0]."\n";
+ }
+ }
+ }
+ }
+ if ($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";
- my $emailto="$perlvar{'lonAdmEMail'}";
+ my $defdom = $perlvar{'lonDefDomain'};
+ my $origmail = $perlvar{'lonAdmEMail'};
+ my $emailto = &Apache::loncommon::build_recipient_list(undef,
+ 'lonstatusmail',$defdom,$origmail);
if ($totalcount>2500) {
$emailto.=",$perlvar{'lonSysEMail'}";
}
- my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";
-
- my $result=system("metasend -b -S 4000000 -t $emailto -s '$subj' -f $statusdir/index.html -m text/html >& /dev/null");
- if ($result != 0) {
- $result=system("mail -s '$subj' $emailto < $statusdir/index.html");
+ my $from;
+ my $hostname=`/bin/hostname`;
+ chop($hostname);
+ $hostname=~s/[^\w\.]//g;
+ if ($hostname) {
+ $from = 'www@'.$hostname;
+ }
+ my $subj="LON: $perlvar{'lonHostID'} E:$errors W:$warnings N:$notices";
+ my $loncronmail = "To: $emailto\n".
+ "From: $from\n".
+ "Subject: ".$subj."\n".
+ "Content-type: text/html\; charset=UTF-8\n".
+ "MIME-Version: 1.0\n\n";
+ if (open(my $fh,"<$statusdir/index.html")) {
+ while (<$fh>) {
+ $loncronmail .= $_;
+ }
+ close($fh);
+ } else {
+ $loncronmail .= "Failed to read from http://$hostname/lon-status/index.html\n";
+ }
+ $loncronmail .= "\n\n";
+ if (open(my $mailh, "|/usr/lib/sendmail -oi -t -odb")) {
+ print $mailh $loncronmail;
+ close($mailh);
+ print "Sending mail.\n";
+ } else {
+ print "Sending mail failed.\n";
}
}
sub usage {
print(< \$help,
- "oldlonc" => \$oldlonc,
"justcheckdaemons" => \$justcheckdaemons,
"noemail" => \$noemail,
"justcheckconnections" => \$justcheckconnections,
@@ -639,7 +903,7 @@ sub main () {
undef $perlvarref;
delete $perlvar{'lonReceipt'}; # remove since sensitive and not needed
delete $perlvar{'lonSqlAccess'}; # remove since sensitive and not needed
-
+ chdir($perlvar{'lonDaemons'});
# --------------------------------------- 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'}) {
@@ -649,48 +913,61 @@ sub main () {
chop $hostname;
$hostname=~s/[^\w\.]//g; # make sure is safe to pass through shell
my $subj="LON: Unconfigured machine $hostname";
- system("echo 'Unconfigured machine $hostname.' |\
- mailto $emailto -s '$subj' > /dev/null");
+ system("echo 'Unconfigured machine $hostname.' |".
+ " mail -s '$subj' $emailto > /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");
+ 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.' |\
- mailto $emailto -s '$subj' > /dev/null");
+ system("echo 'User ID mismatch. loncron must be run as user www.' |".
+ " mail -s '$subj' $emailto > /dev/null");
exit 1;
}
-# ------------------------------------------------------------- Read hosts file
- my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
-
- my (%hostname,%hostdom,%hostrole,%spareid);
- while (my $configline=<$config>) {
- next if ($configline =~ /^(\#|\s*\$)/);
- my ($id,$domain,$role,$name)=split(/:/,$configline);
- if ($id && $domain && $role && $name) {
- $name=~s/\s//g;
- $hostname{$id}=$name;
- $hostdom{$id}=$domain;
- $hostrole{$id}=$role;
- }
- }
- undef $config;
-
-# ------------------------------------------------------ Read spare server file
- $config=IO::File->new("$perlvar{'lonTabDir'}/spare.tab");
-
- while (my $configline=<$config>) {
- chomp($configline);
- if (($configline) && ($configline ne $perlvar{'lonHostID'})) {
- $spareid{$configline}=1;
- }
+# -------------------------------------------- Force reload of host information
+ my $nomemcache;
+ if ($justcheckdaemons) {
+ $nomemcache=1;
+ my $memcachepidfile="$perlvar{'lonDaemons'}/logs/memcached.pid";
+ my $memcachepid;
+ if (-e $memcachepidfile) {
+ my $memfh=IO::File->new($memcachepidfile);
+ $memcachepid=<$memfh>;
+ chomp($memcachepid);
+ if ($memcachepid =~ /^\d+$/ && kill 0 => $memcachepid) {
+ undef($nomemcache);
+ }
+ }
+ }
+ &Apache::lonnet::load_hosts_tab(1,$nomemcache);
+ &Apache::lonnet::load_domain_tab(1,$nomemcache);
+ &Apache::lonnet::get_iphost(1,$nomemcache);
+
+# ----------------------------------------- Force firewall update for lond port
+
+ if ((!$justcheckdaemons) && (!$justreload)) {
+ my $now = time;
+ my $tmpfile = $perlvar{'lonDaemons'}.'/tmp/lciptables_iphost_'.
+ $now.$$.int(rand(10000));
+ if (open(my $fh,">$tmpfile")) {
+ my %iphosts = &Apache::lonnet::get_iphost();
+ foreach my $key (keys(%iphosts)) {
+ print $fh "$key\n";
+ }
+ close($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);
+ }
}
- undef $config;
# ---------------------------------------------------------------- Start report
@@ -701,39 +978,40 @@ sub main () {
my $fh;
if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
- $fh=&start_logging(\%hostdom,\%hostrole,\%hostname,\%spareid);
+ $fh=&start_logging();
&log_machine_info($fh);
&clean_tmp($fh);
&clean_lonIDs($fh);
+ &clean_webDAV_sessionIDs($fh);
&check_httpd_logs($fh);
&rotate_lonnet_logs($fh);
+ &rotate_other_logs($fh);
}
if (!$justcheckconnections && !$justreload) {
+ &checkon_daemon($fh,'lonmemcached',40000);
&checkon_daemon($fh,'lonsql',200000);
if ( &checkon_daemon($fh,'lond',40000,'USR1') eq 'running') {
&checkon_daemon($fh,'lond',40000,'USR2');
}
- my $args='new';
- if ($oldlonc) { $args = ''; }
- &checkon_daemon($fh,'lonc',40000,'USR1',$args);
- &checkon_daemon($fh,'lonhttpd',40000);
- &checkon_daemon($fh,'lonmemcached',40000);
+ &checkon_daemon($fh,'lonc',40000,'USR1');
+ &checkon_daemon($fh,'lonmaxima',40000);
+ &checkon_daemon($fh,'lonr',40000);
}
if ($justreload) {
&checkon_daemon($fh,'lond',40000,'USR2');
- my $args='new';
- if ($oldlonc) { $args = ''; }
- &checkon_daemon($fh,'lonc',40000,'USR2',$args);
+ &checkon_daemon($fh,'lonc',40000,'USR2');
}
if ($justcheckconnections) {
- &test_connections($fh,\%hostname);
+ &test_connections($fh);
}
if (!$justcheckdaemons && !$justcheckconnections && !$justreload) {
- &check_delayed_msg($fh,\%hostname);
+ &check_delayed_msg($fh);
&finish_logging($fh);
&log_simplestatus();
-
+ &write_loncaparevs();
+ &write_serverhomeIDs();
+ &write_checksums();
if ($totalcount>200 && !$noemail) { &send_mail(); }
}
}
@@ -741,10 +1019,3 @@ sub main () {
&main();
1;
-
-
-
-
-
-
-