"; + my $lfh=IO::File->new("$pidfile"); + my $daemonpid=<$lfh>; + chomp($daemonpid); + sleep 2; + if (kill 0 => $daemonpid) { + return 1; + } else { + return 0; + } + } + print $fh "Seems like that did not work!
"; + $errors++; + return 0; +} + +sub checkon_daemon { + my ($fh,$daemon,$maxsize,$sendusr1)=@_; + + print $fh '
';
+ printf("%-10s ",$daemon);
+ if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){
+ open (DFH,"tail -n25 $perlvar{'lonDaemons'}/logs/$daemon.log|");
+ while ($line=) {
+ print $fh "$line";
+ if ($line=~/INFO/) { $notices++; }
+ if ($line=~/WARNING/) { $notices++; }
+ if ($line=~/CRITICAL/) { $warnings++; }
+ };
+ close (DFH);
+ }
+ print $fh " ";
+
+ my $pidfile="$perlvar{'lonDaemons'}/logs/$daemon.pid";
+
+ my $restartflag=1;
+
+ if (-e $pidfile) {
+ my $lfh=IO::File->new("$pidfile");
+ my $daemonpid=<$lfh>;
+ chomp($daemonpid);
+ if (kill 0 => $daemonpid) {
+ print $fh ""; + print " "; + if (&start_daemon($fh,$daemon,$pidfile)) { + print $fh "
"; + } + } + + if (-e "$perlvar{'lonDaemons'}/logs/$daemon.log"){ + print $fh "
";
+ open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/$daemon.log|");
+ while ($line=) {
+ print $fh "$line";
+ if ($line=~/WARNING/) { $notices++; }
+ if ($line=~/CRITICAL/) { $notices++; }
+ };
+ close (DFH);
+ print $fh " ";
+ }
+ }
+
+ $fname="$perlvar{'lonDaemons'}/logs/$daemon.log";
+
+ my ($dev,$ino,$mode,$nlink,
+ $uid,$gid,$rdev,$size,
+ $atime,$mtime,$ctime,
+ $blksize,$blocks)=stat($fname);
+
+ if ($size>$maxsize) {
+ print $fh "Rotating logs ...";
+ rename("$fname.2","$fname.3");
+ rename("$fname.1","$fname.2");
+ rename("$fname","$fname.1");
+ }
+
+ &errout($fh);
+}
# ================================================================ Main Program
-# ----------------------------- 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") unless $noprint;
+# --------------------------------- Read loncapa_apache.conf and loncapa.conf
+my $perlvarref=LONCAPA::Configuration::read_conf('loncapa.conf');
+%perlvar=%{$perlvarref};
+undef $perlvarref;
+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;
}
-# ------------------------------------------------------------ Read access.conf
-{
- my $config=IO::File->new("/etc/httpd/conf/access.conf");
-
- while (my $configline=<$config>) {
- if ($configline =~ /PerlSetVar/) {
- my ($dummy,$varname,$varvalue)=split(/\s+/,$configline);
- $perlvar{$varname}=$varvalue;
- }
- }
+# ----------------------------- 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
@@ -69,14 +210,21 @@ if ($wwwid!=$<) {
my $config=IO::File->new("$perlvar{'lonTabDir'}/hosts.tab");
while (my $configline=<$config>) {
- my ($id,$domain,$role,$name,$ip)=split(/:/,$configline);
- $hostname{$id}=$name;
- $hostdom{$id}=$domain;
- $hostrole{$id}=$role;
- $hostip{$id}=$ip;
- if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) {
- $libserv{$id}=$name;
- }
+ my ($id,$domain,$role,$name,$ip,$domdescr)=split(/:/,$configline);
+ if ($id && $domain && $role && $name && $ip) {
+ $hostname{$id}=$name;
+ $hostdom{$id}=$domain;
+ $hostip{$id}=$ip;
+ $hostrole{$id}=$role;
+ if ($domdescr) { $domaindescription{$domain}=$domdescr; }
+ if (($role eq 'library') && ($id ne $perlvar{'lonHostID'})) {
+ $libserv{$id}=$name;
+ }
+ } else {
+ if ($configline) {
+# &logthis("Skipping hosts.tab line -$configline-");
+ }
+ }
}
}
@@ -105,6 +253,7 @@ $date=localtime($now);
{
my $fh=IO::File->new(">$statusdir/newstatus.html");
+my %simplestatus=();
print $fh (< ";
- my $lfh=IO::File->new("$lonsqlfile");
- my $lonsqlpid=<$lfh>;
- chomp($lonsqlpid);
- sleep 30;
- if (kill 0 => $lonsqlpid) {
- print $fh " ";
- system("$perlvar{'lonDaemons'}/lonsql");
- sleep 120;
- }
- } else {
- print $fh "Seems like that did not work! ";
- $errors++;
- }
- if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
- print $fh " ";
- rename("$fname.2","$fname.3");
- rename("$fname.1","$fname.2");
- rename("$fname","$fname.1");
- }
+&checkon_daemon($fh,'lonsql',200000);
- &errout($fh);
-}
# ------------------------------------------------------------------------ lond
-print $fh ' ";
- my $lfh=IO::File->new("$londfile");
- my $londpid=<$lfh>;
- chomp($londpid);
- sleep 30;
- if (kill 0 => $londpid) {
- print $fh " ";
- system("$perlvar{'lonDaemons'}/lond");
- sleep 120;
- }
- } else {
- print $fh "Seems like that did not work! ";
- $errors++;
- }
- if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
- print $fh " ";
- rename("$fname.2","$fname.3");
- rename("$fname.1","$fname.2");
- rename("$fname","$fname.1");
-}
+&checkon_daemon($fh,'lond',40000,1);
-&errout($fh);
# ------------------------------------------------------------------------ lonc
-print $fh ' ";
- my $lfh=IO::File->new("$loncfile");
- my $loncpid=<$lfh>;
- chomp($loncpid);
- sleep 30;
- if (kill 0 => $loncpid) {
- print $fh " ";
- system("$perlvar{'lonDaemons'}/lonc");
- sleep 120;
- }
- } else {
- print $fh "Seems like that did not work! ";
- $errors++;
- }
- if (-e "$perlvar{'lonDaemons'}/logs/lonc.log") {
- print $fh " ";
- rename("$fname.2","$fname.3");
- rename("$fname.1","$fname.2");
- rename("$fname","$fname.1");
-}
+&checkon_daemon($fh,'lonhttpd',40000);
-
-&errout($fh);
# ---------------------------------------------------------------------- lonnet
print $fh ' \n";
$warnings=$warnings+5*$unsend;
+if ($unsend) { $simplestatus{'unsend'}=$unsend; }
print $fh "$varname $perlvar{$varname} Hosts
";
-foreach $id (keys %hostname) {
-print $fh
- "
$id $hostdom{$id} $hostrole{$id} ";
-print $fh "$hostname{$id} $hostip{$id} $id $hostdom{$id} $hostrole{$id} ";
+ print $fh "$hostname{$id} $hostip{$id} Spare Hosts
";
-foreach $id (keys %spareid) {
+foreach $id (sort(keys(%spareid))) {
print $fh "
ps
";
+print $fh "";
+$psproc=0;
+
+open (PSH,"ps -aux|");
+while ($line=
";
+
+if ($psproc>200) { $notices++; }
+if ($psproc>250) { $notices++; }
+
&errout($fh);
# --------------------------------------------------------------- clean out tmp
print $fh 'Temporary Files
';
$cleaned=0;
+$old=0;
while ($fname=<$perlvar{'lonDaemons'}/tmp/*>) {
my ($dev,$ino,$mode,$nlink,
$uid,$gid,$rdev,$size,
@@ -205,12 +374,24 @@ while ($fname=<$perlvar{'lonDaemons'}/tm
$now=time;
$since=$now-$mtime;
if ($since>$perlvar{'lonExpire'}) {
- $cleaned++;
- unlink("$fname");
+ $line='';
+ if (open(PROBE,$fname)) {
+ $line=Session Tokens
';
@@ -239,13 +420,13 @@ print $fh "$active open session(s)
httpd
Access Log
';
-open (DFH,"tail -n40 /etc/httpd/logs/access_log|");
+open (DFH,"tail -n25 /etc/httpd/logs/access_log|");
while ($line=
Error Log
";
-open (DFH,"tail -n50 /etc/httpd/logs/error_log|");
+open (DFH,"tail -n25 /etc/httpd/logs/error_log|");
while ($line=
";
# ---------------------------------------------------------------------- lonsql
-#
-# Do not run for now
-#
-if ($perlvar{'lonRole'} eq "library" && 1==0) {
-
- print $fh 'lonsql
Log
';
-
- if (-e "$perlvar{'lonDaemons'}/logs/lonsql.log"){
- open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
- while ($line=";
-
- 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");
- sleep 120;
- if (-e $lonsqlfile) {
- print $fh "Seems like it started ...lonsql at pid $lonsqlpid responding
";
- } else {
- $errors++; $errors++;
- print $fh "lonsql at pid $lonsqlpid not responding
";
- print $fh "Give it one more try ...";
- open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonsql.log|");
- while ($line=";
- }
- }
-
- $fname="$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 $fh "Rotating logs ...lond
Log
';
-
-if (-e "$perlvar{'lonDaemons'}/logs/lond.log"){
-open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
-while ($line=";
-
-my $londfile="$perlvar{'lonDaemons'}/logs/lond.pid";
-
-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++;
- 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 $fh
- "Deciding to clean up stale .pid file and restart lond
";
- }
-}
-if ($restartflag==1) {
- $errors++;
- print $fh "lond not running, trying to start
";
- system("$perlvar{'lonDaemons'}/lond");
- sleep 120;
- if (-e $londfile) {
- print $fh "Seems like it started ...lond at pid $londpid responding
";
- } else {
- $errors++; $errors++;
- print $fh "lond at pid $londpid not responding
";
- print $fh "Give it one more try ...";
- open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lond.log|");
- while ($line=";
- }
-}
-
-$fname="$perlvar{'lonDaemons'}/logs/lond.log";
-
- my ($dev,$ino,$mode,$nlink,
- $uid,$gid,$rdev,$size,
- $atime,$mtime,$ctime,
- $blksize,$blocks)=stat($fname);
-
-if ($size>40000) {
- print $fh "Rotating logs ...lonc
Log
';
-
-if (-e "$perlvar{'lonDaemons'}/logs/lonc.log"){
-open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
-while ($line=";
-
-my $loncfile="$perlvar{'lonDaemons'}/logs/lonc.pid";
-
-$restartflag=1;
-if (-e $loncfile) {
- my $lfh=IO::File->new("$loncfile");
- my $loncpid=<$lfh>;
- chomp($loncpid);
- if (kill 0 => $loncpid) {
- print $fh "lonc at pid $loncpid responding, sending USR1
";
- kill USR1 => $loncpid;
- $restartflag=0;
- } 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);
- $restartflag=1;
- }
- else {
- # Possibility #2: there is a live process that is not responding
- # for an unknown reason
- # Solution: kill parent and children processes, remove .pid and restart
- `killall -9 lonc`;
- unlink($loncfile);
- $restartflag=1;
- }
- print $fh
- "Deciding to clean up stale .pid file and restart lonc
";
- }
-}
-if ($restartflag==1) {
- $errors++;
- print $fh "lonc not running, trying to start
";
- system("$perlvar{'lonDaemons'}/lonc");
- sleep 120;
- if (-e $loncfile) {
- print $fh "Seems like it started ...lonc at pid $loncpid responding
";
- } else {
- $errors++; $errors++;
- print $fh "lonc at pid $loncpid not responding
";
- print $fh "Give it one more try ...";
- open (DFH,"tail -n100 $perlvar{'lonDaemons'}/logs/lonc.log|");
- while ($line=";
- }
-}
-
-$fname="$perlvar{'lonDaemons'}/logs/lonc.log";
+&checkon_daemon($fh,'lonc',40000,1);
- my ($dev,$ino,$mode,$nlink,
- $uid,$gid,$rdev,$size,
- $atime,$mtime,$ctime,
- $blksize,$blocks)=stat($fname);
+# -------------------------------------------------------------------- lonhttpd
-if ($size>40000) {
- print $fh "Rotating logs ...lonnet
Temp Log
';
+print "checking logs\n";
if (-e "$perlvar{'lonDaemons'}/logs/lonnet.log"){
open (DFH,"tail -n50 $perlvar{'lonDaemons'}/logs/lonnet.log|");
while ($line=";
# ----------------------------------------------------------------- Connections
print $fh 'Connections
';
-
+print "testing connections\n";
print $fh "";
-foreach $tryserver (keys %hostname) {
-
+foreach $tryserver (sort(keys(%hostname))) {
+ print(".");
$answer=reply("pong",$tryserver);
if ($answer eq "$tryserver:$perlvar{'lonHostID'}") {
$result="ok";
@@ -593,6 +504,7 @@ foreach $tryserver (keys %hostname) {
$warnings++;
if ($answer eq 'con_lost') { $warnings++; }
}
+ if ($answer =~ /con_lost/) { print(" $tryserver down\n"); }
print $fh "
";
# ------------------------------------------------------------ Delayed messages
print $fh '$tryserver $result Delayed Messages
';
+print "checking buffers\n";
print $fh 'Scanning Permanent Log
';
@@ -612,7 +525,7 @@ $unsend=0;
($time,$sdf,$dserv,$dcmd)=split(/:/,$line);
if ($sdf eq 'F') {
$local=localtime($time);
- print "Failed: $time, $dserv, $dcmd
";
+ print $fh "Failed: $time, $dserv, $dcmd
";
$warnings++;
}
if ($sdf eq 'S') { $unsend--; }
@@ -622,6 +535,7 @@ $unsend=0;
print $fh "Total unsend messages: $unsendOutgoing Buffer
";
open (DFH,"ls -lF $perlvar{'lonSockDir'}/delayed|");
@@ -638,13 +552,29 @@ print $fh "Total Error Count: $total
$now=time;
$date=localtime($now);
print $fh "
$date ($now)