--- loncom/Attic/lonc 2001/03/13 21:15:40 1.14
+++ loncom/Attic/lonc 2001/11/28 21:19:58 1.21
@@ -16,7 +16,7 @@
# 12/05 Scott Harrison
# 12/05 Gerd Kortemeyer
# 01/10/01 Scott Harrison
-# 03/14/01 Gerd Kortemeyer
+# 03/14/01,03/15,06/12,11/26,11/27,11/28 Gerd Kortemeyer
#
# based on nonforker from Perl Cookbook
# - server who multiplexes without forking
@@ -30,6 +30,9 @@ use Fcntl;
use Tie::RefHash;
use Crypt::IDEA;
+my $status='';
+my $lastlog='';
+
# grabs exception and records it to log before exiting
sub catchexception {
my ($signal)=@_;
@@ -41,7 +44,7 @@ sub catchexception {
die($@);
}
-$childmaxattempts=10;
+$childmaxattempts=5;
# -------------------------------- Set signal handlers to record abnormal exits
@@ -116,7 +119,13 @@ sub REAPER { # ta
sub HUNTSMAN { # signal handler for SIGINT
local($SIG{CHLD}) = 'IGNORE'; # we're going to kill our children
- kill 'INT' => keys %children;
+ map {
+ $wasserver=$children{$_};
+ &status("Closing $wasserver");
+ &logthis('Closing '.$wasserver.': '.&subreply('exit',$wasserver));
+ &status("Kill PID $_ for $wasserver");
+ kill ('INT',$_);
+ } keys %children;
my $execdir=$perlvar{'lonDaemons'};
unlink("$execdir/logs/lonc.pid");
&logthis("CRITICAL: Shutting down");
@@ -125,13 +134,32 @@ sub HUNTSMAN { # si
sub HUPSMAN { # signal handler for SIGHUP
local($SIG{CHLD}) = 'IGNORE'; # we're going to kill our children
- kill 'INT' => keys %children;
+ map {
+ $wasserver=$children{$_};
+ &status("Closing $wasserver");
+ &logthis('Closing '.$wasserver.': '.&subreply('exit',$wasserver));
+ &status("Kill PID $_ for $wasserver");
+ kill ('INT',$_);
+ } keys %children;
&logthis("CRITICAL: Restarting");
unlink("$execdir/logs/lonc.pid");
my $execdir=$perlvar{'lonDaemons'};
exec("$execdir/lonc"); # here we go again
}
+sub checkchildren {
+ &initnewstatus();
+ &logstatus();
+ &logthis('Going to check on the children');
+ map {
+ sleep 1;
+ unless (kill 'USR1' => $_) {
+ &logthis ('Child '.$_.' is dead');
+ &logstatus($$.' is dead');
+ }
+ } sort keys %children;
+}
+
sub USRMAN {
&logthis("USR1: Trying to establish connections again");
foreach $thisserver (keys %hostip) {
@@ -141,6 +169,7 @@ sub USRMAN {
." >$answer<");
}
%childatt=();
+ &checkchildren();
}
# -------------------------------------------------- Non-critical communication
@@ -153,10 +182,20 @@ sub subreply {
Type => SOCK_STREAM,
Timeout => 10)
or return "con_lost";
- print $sclient "$cmd\n";
- my $answer=<$sclient>;
- chomp($answer);
- if (!$answer) { $answer="con_lost"; }
+
+
+ $SIG{ALRM}=sub { die "timeout" };
+ $SIG{__DIE__}='DEFAULT';
+ eval {
+ alarm(10);
+ print $sclient "$cmd\n";
+ $answer=<$sclient>;
+ chomp($answer);
+ alarm(0);
+ };
+ if ((!$answer) || ($@=~/timeout/)) { $answer="con_lost"; }
+ $SIG{ALRM}='DEFAULT';
+ $SIG{__DIE__}=\&catchexception;
} else { $answer='self_reply'; }
return $answer;
}
@@ -169,6 +208,7 @@ sub logthis {
my $fh=IO::File->new(">>$execdir/logs/lonc.log");
my $now=time;
my $local=localtime($now);
+ $lastlog=$local.': '.$message;
print $fh "$local ($$): $message\n";
}
@@ -181,6 +221,31 @@ sub logperm {
my $fh=IO::File->new(">>$execdir/logs/lonnet.perm.log");
print $fh "$now:$message:$local\n";
}
+# ------------------------------------------------------------------ Log status
+
+sub logstatus {
+ my $docdir=$perlvar{'lonDocRoot'};
+ my $fh=IO::File->new(">>$docdir/lon-status/loncstatus.txt");
+ print $fh $$."\t".$status."\t".$lastlog."\n";
+}
+
+sub initnewstatus {
+ my $docdir=$perlvar{'lonDocRoot'};
+ my $fh=IO::File->new(">$docdir/lon-status/loncstatus.txt");
+ my $now=time;
+ my $local=localtime($now);
+ print $fh "LONC status $local - parent $$\n\n";
+}
+
+# -------------------------------------------------------------- Status setting
+
+sub status {
+ my $what=shift;
+ my $now=time;
+ my $local=localtime($now);
+ $status=$local.': '.$what;
+}
+
# ---------------------------------------------------- Fork once and dissociate
@@ -204,6 +269,8 @@ $SIG{HUP}=$SIG{USR1}='IGNORE';
# Fork off our children, one for every server
+&status("Forking ...");
+
foreach $thisserver (keys %hostip) {
make_new_child($thisserver);
}
@@ -218,11 +285,13 @@ $SIG{USR1} = \&USRMAN;
# And maintain the population.
while (1) {
+ &status("Sleeping");
sleep; # wait for a signal (i.e., child's death)
# See who died and start new one
+ &status("Woke up");
foreach $thisserver (keys %hostip) {
if (!$childpid{$thisserver}) {
- if ($childatt{$thisserver}<=$childmaxattempts) {
+ if ($childatt{$thisserver}<$childmaxattempts) {
$childatt{$thisserver}++;
&logthis(
"INFO: Trying to reconnect for $thisserver "
@@ -257,7 +326,8 @@ sub make_new_child {
} else {
# Child can *not* return from this subroutine.
$SIG{INT} = 'DEFAULT'; # make SIGINT kill us as it did before
-
+ $SIG{USR1}= \&logstatus;
+
# unblock signals
sigprocmask(SIG_UNBLOCK, $sigset)
or die "Can't unblock SIGINT for fork: $!\n";
@@ -267,7 +337,11 @@ sub make_new_child {
$port = "$perlvar{'lonSockDir'}/$conserver";
unlink($port);
+
# ---------------------------------------------------- Client to network server
+
+&status("Opening TCP: $conserver");
+
unless (
$remotesock = IO::Socket::INET->new(PeerAddr => $hostip{$conserver},
PeerPort => $perlvar{'londPort'},
@@ -280,20 +354,47 @@ unless (
sleep($st);
exit;
};
-# --------------------------------------- Send a ping to make other end do USR1
+# ----------------------------------------------------------------- Init dialog
+
+&status("Init dialogue: $conserver");
+
+ $SIG{ALRM}=sub { die "timeout" };
+ $SIG{__DIE__}='DEFAULT';
+ eval {
+ alarm(60);
print $remotesock "init\n";
$answer=<$remotesock>;
print $remotesock "$answer";
$answer=<$remotesock>;
chomp($answer);
+ alarm(0);
+ };
+ $SIG{ALRM}='DEFAULT';
+ $SIG{__DIE__}=\&catchexception;
+
+ if ($@=~/timeout/) {
+ &logthis("Timed out during init: $conserver");
+ exit;
+ }
+
+
&logthis("Init reply for $conserver: >$answer<");
+if ($answer ne 'ok') {
+ my $st=120+int(rand(240));
+ &logthis(
+"WARNING: Init failed $conserver ($st secs)");
+ sleep($st);
+ exit;
+}
sleep 5;
+&status("Ponging $conserver");
print $remotesock "pong\n";
$answer=<$remotesock>;
chomp($answer);
&logthis("Pong reply for $conserver: >$answer<");
# ----------------------------------------------------------- Initialize cipher
+&status("Initialize cipher: $conserver");
print $remotesock "ekey\n";
my $buildkey=<$remotesock>;
my $key=$conserver.$perlvar{'lonHostID'};
@@ -315,7 +416,7 @@ if ($cipher=new IDEA $cipherkey) {
}
# ----------------------------------------- We're online, send delayed messages
-
+ &status("Checking for delayed messages");
my @allbuffered;
my $path="$perlvar{'lonSockDir'}/delayed";
opendir(DIRHANDLE,$path);
@@ -323,6 +424,7 @@ if ($cipher=new IDEA $cipherkey) {
closedir(DIRHANDLE);
my $dfname;
map {
+ &status("Sending delayed $conserver $_");
$dfname="$path/$_";
&logthis($dfname);
my $wcmd;
@@ -345,11 +447,19 @@ if ($cipher=new IDEA $cipherkey) {
}
$cmd="enc:$cmdlength:$encrequest\n";
}
-
+ $SIG{ALRM}=sub { die "timeout" };
+ $SIG{__DIE__}='DEFAULT';
+ eval {
+ alarm(60);
print $remotesock "$cmd\n";
$answer=<$remotesock>;
chomp($answer);
- if ($answer ne '') {
+ alarm(0);
+ };
+ $SIG{ALRM}='DEFAULT';
+ $SIG{__DIE__}=\&catchexception;
+
+ if (($answer ne '') && ($@!~/timeout/)) {
unlink("$dfname");
&logthis("Delayed $cmd to $conserver: >$answer<");
&logperm("S:$conserver:$bcmd");
@@ -357,6 +467,7 @@ if ($cipher=new IDEA $cipherkey) {
} @allbuffered;
# ------------------------------------------------------- Listen to UNIX socket
+&status("Opening socket $conserver");
unless (
$server = IO::Socket::UNIX->new(Local => $port,
Type => SOCK_STREAM,
@@ -394,11 +505,11 @@ while (1) {
# check for new information on the connections we have
# anything to read or accept?
- foreach $client ($select->can_read(1)) {
+ foreach $client ($select->can_read(0.1)) {
if ($client == $server) {
# accept a new connection
-
+ &status("Accept new connection: $conserver");
$client = $server->accept();
$select->add($client);
nonblock($client);
@@ -413,6 +524,7 @@ while (1) {
delete $outbuffer{$client};
delete $ready{$client};
+ &status("Idle $conserver");
$select->remove($client);
close $client;
next;
@@ -443,16 +555,21 @@ while (1) {
$rv = $client->send($outbuffer{$client}, 0);
unless (defined $rv) {
# Whine, but move on.
- warn "I was told I could write, but I can't.\n";
+ &logthis("I was told I could write, but I can't.\n");
next;
}
+ $errno=$!;
if (($rv == length $outbuffer{$client}) ||
- ($! == POSIX::EWOULDBLOCK)) {
+ ($errno == POSIX::EWOULDBLOCK) || ($errno == 0)) {
substr($outbuffer{$client}, 0, $rv) = '';
delete $outbuffer{$client} unless length $outbuffer{$client};
} else {
# Couldn't write all the data, and it wasn't because
# it would have blocked. Shutdown and move on.
+
+ &logthis("Dropping data with ".$errno.": ".
+ length($outbuffer{$client}).", $rv");
+
delete $inbuffer{$client};
delete $outbuffer{$client};
delete $ready{$client};
@@ -492,8 +609,27 @@ sub handle {
}
$request="enc:$cmdlength:$encrequest\n";
}
+# --------------------------------------------------------------- Main exchange
+ $SIG{ALRM}=sub { die "timeout" };
+ $SIG{__DIE__}='DEFAULT';
+ eval {
+ alarm(300);
+ &status("Sending $conserver: $request");
print $remotesock "$request";
+ &status("Waiting for reply from $conserver: $request");
$answer=<$remotesock>;
+ &status("Received reply: $request");
+ alarm(0);
+ };
+ if ($@=~/timeout/) {
+ $answer='';
+ &logthis(
+ "CRITICAL: Timeout $conserver: $request");
+ }
+ $SIG{ALRM}='DEFAULT';
+ $SIG{__DIE__}=\&catchexception;
+
+
if ($answer) {
if ($answer =~ /^enc/) {
my ($cmd,$cmdlength,$encinput)=split(/:/,$answer);
@@ -515,6 +651,7 @@ sub handle {
# ===================================================== Done processing request
}
delete $ready{$client};
+ &status("Completed $conserver: $request");
# -------------------------------------------------------------- End non-forker
}
# ---------------------------------------------------------- End make_new_child