loncom/loncnew - diff

Return to loncnew CVS log

Up to [LON-CAPA] / loncom

Diff for /loncom/loncnew between versions 1.2 and 1.110

-version 1.2, 2003/04/18 02:59:31
+version 1.110, 2024/06/14 18:49:41
  Line 7
  # Copyright Michigan State University Board of Trustees
  #
  # This file is part of the LearningOnline Network with CAPA (LON-CAPA).
- #
+ ## LON-CAPA is free software; you can redistribute it and/or modify
- # LON-CAPA is free software; you can redistribute it and/or modify
  # it under the terms of the GNU General Public License as published by
  # the Free Software Foundation; either version 2 of the License, or
  # (at your option) any later version.
- Line 27
+ Line 26
  # http://www.lon-capa.org/
  #
  #
- # new lonc handles n requestors spread out bver m connections to londs.
+ # new lonc handles n request out over m connections to londs.
  # This module is based on the Event class.
  #   Development iterations:
  #    - Setup basic event loop.   (done)
  #    - Add timer dispatch.       (done)
  #    - Add ability to accept lonc UNIX domain sockets.  (done)
  #    - Add ability to create/negotiate lond connections (done).
- #    - Add general logic for dispatching requests and timeouts.
+ #    - Add general logic for dispatching requests and timeouts. (done).
- #    - Add support for the lonc/lond requests.
+ #    - Add support for the lonc/lond requests.          (done).
- #    - Add logging/status monitoring.
+ #    - Add logging/status monitoring.                    (done)
- #    - Add Signal handling - HUP restarts. USR1 status report.
+ #    - Add Signal handling - HUP restarts. USR1 status report. (done)
- #    - Add Configuration file I/O
+ #    - Add Configuration file I/O                       (done).
- #    - Add Pending request processing on startup.
+ #    - Add management/status request interface.         (done)
- #    - Add management/status request interface.
+ #    - Add deferred request capability.                  (done)
+ #    - Detect transmission timeouts.                     (done)
+ #
+ use strict;
  use lib "/home/httpd/lib/perl/";
- use lib "/home/foxr/newloncapa/types";
  use Event qw(:DEFAULT );
  use POSIX qw(:signal_h);
+ use POSIX;
  use IO::Socket;
  use IO::Socket::INET;
  use IO::Socket::UNIX;
+ use IO::File;
+ use IO::Handle;
  use Socket;
  use Crypt::IDEA;
  use LONCAPA::Queue;
  use LONCAPA::Stack;
  use LONCAPA::LondConnection;
+ use LONCAPA::LondTransaction;
  use LONCAPA::Configuration;
- use LONCAPA::HashIterator;
+ use Fcntl qw(:flock);
- print "Loncnew starting\n";
- #
- #   Disable all signals we might receive from outside for now.
- #
- $SIG{QUIT}  = IGNORE;
- $SIG{HUP}   = IGNORE;
- $SIG{USR1}  = IGNORE;
- $SIG{INT}   = IGNORE;
- $SIG{CHLD}  = IGNORE;
- $SIG{__DIE__}  = IGNORE;
  # Read the httpd configuration file to get perl variables
- Line 79  my %perlvar    = %{$perlvarref};
+ Line 72  my %perlvar    = %{$perlvarref};
  #
  #  parent and shared variables.
- my %ChildHash;			# by pid -> host.
+ my %ChildPid;			# by pid -> host.
+ my %ChildHost;			# by host.
+ my %ChildKeyMode;               # by pid -> keymode
+ my %listening_to;		# Socket->host table for who the parent
+                                 # is listening to.
+ my %parent_dispatchers;         # host-> listener watcher events.
+ my %parent_handlers;		# Parent signal handlers...
- my $MaxConnectionCount = 5;	# Will get from config later.
+ my $MaxConnectionCount = 10;	# Will get from config later.
  my $ClientConnection = 0;	# Uniquifier for client events.
- my $DebugLevel = 5;
+ my $DebugLevel = 0;
- my $IdleTimeout= 3600;		# Wait an hour before pruning connections.
+ my $NextDebugLevel= 2;		# So Sigint can toggle this.
+ my $IdleTimeout= 5*60;		# Seconds to wait prior to pruning connections.
+ my $LogTransactions = 0;	# When True, all transactions/replies get logged.
+ my $executable      = $0;	# Get the full path to me.
  #
  #  The variables below are only used by the child processes.
  #
- my $RemoteHost;			# Name of host child is talking to.
+ my $RemoteHost;			# Hostname of host child is talking to.
- my $UnixSocketDir= "/home/httpd/sockets";
+ my $RemoteHostId;		# lonid of host child is talking to.
+ my $RemoteDefHostId;		# default lonhostID of host child is talking to.
+ my $RemoteLoncapaRev;           # LON-CAPA version of host child is talking to,
+                                 # if 2.12.0 or newer, format: X.Y.Z
+ my @all_host_ids;
+ my $UnixSocketDir= $perlvar{'lonSockDir'};
  my $IdleConnections = Stack->new(); # Set of idle connections
  my %ActiveConnections;		# Connections to the remote lond.
- my %ActiveTransactions;		# Transactions in flight.
+ my %ActiveTransactions;		# LondTransactions in flight.
  my %ActiveClients;		# Serial numbers of active clients by socket.
  my $WorkQueue       = Queue->new(); # Queue of pending transactions.
- my $ClientQueue     = Queue->new(); # Queue of clients causing xactinos.
  my $ConnectionCount = 0;
+ my $IdleSeconds     = 0;	# Number of seconds idle.
+ my $Status          = "";	# Current status string.
+ my $RecentLogEntry  = "";
+ my $ConnectionRetries=5;	# Number of connection retries allowed.
+ my $ConnectionRetriesLeft=5;	# Number of connection retries remaining.
+ my $LondVersion     = "unknown"; # Version of lond we talk with.
+ my $KeyMode         = "";       # e.g. ssl, local, insecure from last connect.
+ my $LondConnecting  = 0;       # True when a connection is being built.
+ my $I_am_child      = 0;	# True if this is the child process.
+ #
+ #   The hash below gives the HTML format for log messages
+ #   given a severity.
+ #
+ my %LogFormats;
+ $LogFormats{"CRITICAL"} = "<font color='red'>CRITICAL: %s</font>";
+ $LogFormats{"SUCCESS"}  = "<font color='green'>SUCCESS: %s</font>";
+ $LogFormats{"INFO"}     = "<font color='yellow'>INFO: %s</font>";
+ $LogFormats{"WARNING"}  = "<font color='blue'>WARNING: %s</font>";
+ $LogFormats{"DEFAULT"}  = " %s ";
+ #  UpdateStatus;
+ #    Update the idle status display to show how many connections
+ #    are left, retries and other stuff.
  #
+ sub UpdateStatus {
+     if ($ConnectionRetriesLeft > 0) {
+ 	ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount
+ 		   ." Retries remaining: ".$ConnectionRetriesLeft
+ 		   ." ($KeyMode)");
+     } else {
+ 	ShowStatus(GetServerHost()." >> DEAD <<");
+     }
+ }
+ =pod
+ =head2 LogPerm
+ Makes an entry into the permanent log file.
+ =cut
+ sub LogPerm {
+     my $message=shift;
+     my $execdir=$perlvar{'lonDaemons'};
+     my $now=time;
+     my $local=localtime($now);
+     my $fh=IO::File->new(">>$execdir/logs/lonnet.perm.log");
+     chomp($message);
+     print $fh "$now:$message:$local\n";
+ }
  =pod
- =head 2 GetPeerName
-    Returns the name of the host that a socket object is connected
+ =head2 Log
-    to.
+ Logs a message to the log file.
+ Parameters:
+ =item severity
+ One of CRITICAL, WARNING, INFO, SUCCESS used to select the
+ format string used to format the message.  if the severity is
+ not a defined severity the Default format string is used.
+ =item message
+ The base message.  In addtion to the format string, the message
+ will be appended to a string containing the name of our remote
+ host and the time will be formatted into the message.
+ =cut
+ sub Log {
+     my ($severity, $message) = @_;
+     if(!$LogFormats{$severity}) {
+ 	$severity = "DEFAULT";
+     }
+     my $format = $LogFormats{$severity};
+     #  Put the window dressing in in front of the message format:
+     my $now   = time;
+     my $local = localtime($now);
+     my $finalformat = "$local ($$) [$RemoteHost] [$Status] ";
+     $finalformat = $finalformat.$format."\n";
+     # open the file and put the result.
+     my $execdir = $perlvar{'lonDaemons'};
+     my $fh      = IO::File->new(">>$execdir/logs/lonc.log");
+     my $msg = sprintf($finalformat, $message);
+     $RecentLogEntry = $msg;
+     print $fh $msg;
+ }
+ =pod
+ =head2 GetPeerName
+ Returns the name of the host that a socket object is connected to.
  =cut
  sub GetPeername {
-     my $connection = shift;
-     my $AdrFamily  = shift;
+     my ($connection, $AdrFamily) = @_;
      my $peer       = $connection->peername();
      my $peerport;
      my $peerip;
      if($AdrFamily == AF_INET) {
  	($peerport, $peerip) = sockaddr_in($peer);
- 	my $peername    = gethostbyaddr($iaddr, $AdrFamily);
+ 	my $peername    = gethostbyaddr($peerip, $AdrFamily);
  	return $peername;
      } elsif ($AdrFamily == AF_UNIX) {
  	my $peerfile;
- Line 125  sub GetPeername {
+ Line 242  sub GetPeername {
  	return $peerfile;
      }
  }
- #----------------------------- Timer management ------------------------
  =pod
  =head2 Debug
-    Invoked to issue a debug message.
+ Invoked to issue a debug message.
  =cut
  sub Debug {
-     my $level   = shift;
-     my $message = shift;
+     my ($level, $message) = @_;
      if ($level <= $DebugLevel) {
- 	print $message." host = ".$RemoteHost."\n";
+ 	Log("INFO", "-Debug- $message host = $RemoteHost");
      }
  }
  sub SocketDump {
-     my $level = shift;
-     my $socket= shift;
+     my ($level, $socket) = @_;
      if($level <= $DebugLevel) {
- 	$socket->Dump();
+ 	$socket->Dump(-1);	# Ensure it will get dumped.
      }
  }
+ =pod
+ =head2 ShowStatus
+  Place some text as our pid status.
+  and as what we return in a SIGUSR1
+ =cut
+ sub ShowStatus {
+     my $state = shift;
+     my $now = time;
+     my $local = localtime($now);
+     $Status   = $local.": ".$state;
+     $0='lonc: '.$state.' '.$local;
+ }
  =pod
+ =head2 SocketTimeout
+     Called when an action on the socket times out.  The socket is
+    destroyed and any active transaction is failed.
+ =cut
+ sub SocketTimeout {
+     my $Socket = shift;
+     Log("WARNING", "A socket timeout was detected");
+     Debug(5, " SocketTimeout called: ");
+     $Socket->Dump(0);
+     if(exists($ActiveTransactions{$Socket})) {
+ 	FailTransaction($ActiveTransactions{$Socket});
+     }
+     KillSocket($Socket);	# A transaction timeout also counts as
+                                 # a connection failure:
+     $ConnectionRetriesLeft--;
+     if($ConnectionRetriesLeft <= 0) {
+ 	Log("CRITICAL", "Host marked DEAD: ".GetServerHost());
+ 	$LondConnecting = 0;
+     }
+ }
+ #
+ #   This function should be called by the child in all cases where it must
+ #   exit.  The child process must create a lock file for the AF_UNIX socket
+ #   in order to prevent connection requests from lonnet in the time between
+ #   process exit and the parent picking up the listen again.
+ #
+ # Parameters:
+ #     exit_code           - Exit status value, however see the next parameter.
+ #     message             - If this optional parameter is supplied, the exit
+ #                           is via a die with this message.
+ #
+ sub child_exit {
+     my ($exit_code, $message) = @_;
+     # Regardless of how we exit, we may need to do the lock thing:
+     #
+     #  Create a lock file since there will be a time window
+     #  between our exit and the parent's picking up the listen
+     #  during which no listens will be done on the
+     #  lonnet client socket.
+     #
+     my $lock_file = &GetLoncSocketPath().".lock";
+     open(LOCK,">$lock_file");
+     print LOCK "Contents not important";
+     close(LOCK);
+     unlink(&GetLoncSocketPath());
+     if ($message) {
+ 	die($message);
+     } else {
+ 	exit($exit_code);
+     }
+ }
+ #----------------------------- Timer management ------------------------
+ =pod
  =head2 Tick
-     Invoked  each timer tick.
+ Invoked each timer tick.
  =cut
  sub Tick {
+     my ($Event)       = @_;
+     my $clock_watcher = $Event->w;
      my $client;
-     Debug(6, "Tick");
+     UpdateStatus();
-     Debug(6, "    Current connection count: ".$ConnectionCount);
-     foreach $client (keys %ActiveClients) {
+     # Is it time to prune connection count:
- 	Debug(7, "    Have client:  with id: ".$ActiveClients{$client});
+     if($IdleConnections->Count()  &&
+        ($WorkQueue->Count() == 0)) { # Idle connections and nothing to do?
+ 	$IdleSeconds++;
+ 	if($IdleSeconds > $IdleTimeout) { # Prune a connection...
+ 	    my $Socket = $IdleConnections->pop();
+ 	    KillSocket($Socket);
+ 	    $IdleSeconds = 0;	# Otherwise all connections get trimmed to fast.
+ 	    UpdateStatus();
+ 	    if(($ConnectionCount == 0)) {
+ 		&child_exit(0);
+ 	    }
+ 	}
+     } else {
+ 	$IdleSeconds = 0;	# Reset idle count if not idle.
+     }
+     #
+     #  For each inflight transaction, tick down its timeout counter.
+     #
+     foreach my $item (keys %ActiveConnections) {
+ 	my $State = $ActiveConnections{$item}->data->GetState();
+ 	if ($State ne 'Idle') {
+ 	    Debug(5,"Ticking Socket $State $item");
+ 	    $ActiveConnections{$item}->data->Tick();
+ 	}
      }
+     # Do we have work in the queue, but no connections to service them?
+     # If so, try to make some new connections to get things going again.
+     #
+     #   Note this code is dead now...
+     #
+     my $Requests = $WorkQueue->Count();
+     if (($ConnectionCount == 0)  && ($Requests > 0) && (!$LondConnecting)) {
+ 	if ($ConnectionRetriesLeft > 0) {
+ 	    Debug(5,"Work but no connections, Make a new one");
+ 	    my $success;
+ 	    $success    = &MakeLondConnection;
+ 	    if($success == 0) { # All connections failed:
+ 		Debug(5,"Work in queue failed to make any connectiouns\n");
+ 		EmptyQueue();	# Fail pending transactions with con_lost.
+ 		CloseAllLondConnections(); # Should all be closed but....
+ 	    }
+ 	} else {
+ 	    $LondConnecting = 0;
+ 	    ShowStatus(GetServerHost()." >>> DEAD!!! <<<");
+ 	    Debug(5,"Work in queue, but gave up on connections..flushing\n");
+ 	    EmptyQueue();	# Connections can't be established.
+ 	    CloseAllLondConnections(); # Should all already be closed but...
+ 	}
+     }
+     if ($ConnectionCount == 0) {
+ 	$KeyMode = "";
+ 	$clock_watcher->cancel();
+     }
+     &UpdateStatus();
  }
  =pod
  =head2 SetupTimer
-    Sets up a 1 per sec recurring timer event.  The event handler is used to:
- =item Trigger timeouts on communications along active sockets.
+ Sets up a 1 per sec recurring timer event.  The event handler is used to:
- =item Trigger disconnections of idle sockets.
+ =item
+ Trigger timeouts on communications along active sockets.
+ =item
+ Trigger disconnections of idle sockets.
  =cut
  sub SetupTimer {
      Debug(6, "SetupTimer");
-     Event->timer(interval => 1, debug => 1, cb => \&Tick );
+     Event->timer(interval => 1, cb => \&Tick,
+ 	hard => 1);
  }
  =pod
  =head2 ServerToIdle
-    This function is called when a connection to the server is
- ready for more work.
+ This function is called when a connection to the server is
-    If there is work in the Work queue the top element is dequeued
+ ready for more work.
+ If there is work in the Work queue the top element is dequeued
  and the connection will start to work on it.  If the work queue is
  empty, the connection is pushed on the idle connection stack where
  it will either get another work unit, or alternatively, if it sits there
  long enough, it will be shut down and released.
  =cut
  sub ServerToIdle {
      my $Socket   = shift;	# Get the socket.
+     $KeyMode = $Socket->{AuthenticationMode};
+     delete($ActiveTransactions{$Socket}); # Server has no transaction
-     &Debug(6, "Server to idle");
+     &Debug(5, "Server to idle");
      #  If there's work to do, start the transaction:
-     $reqdata = $WorkQueue->dequeue();
+     my $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction
-     Debug(9, "Queue gave request data: ".$reqdata);
+     if ($reqdata ne undef)  {
-     unless($reqdata eq undef)  {
+ 	Debug(5, "Queue gave request data: ".$reqdata->getRequest());
- 	my $unixSocket = $ClientQueue->dequeue();
+ 	&StartRequest($Socket,  $reqdata);
- 	&Debug(6, "Starting new work request");
- 	&Debug(7, "Request: ".$reqdata);
- 	&StartRequest($Socket, $unixSocket, $reqdata);
      } else {
      #  There's no work waiting, so push the server to idle list.
- 	&Debug(8, "No new work requests, server connection going idle");
+ 	&Debug(5, "No new work requests, server connection going idle");
- 	delete($ActiveTransactions{$Socket});
  	$IdleConnections->push($Socket);
      }
  }
  =pod
  =head2 ClientWritable
-   Event callback for when a client socket is writable.
-   This callback is established when a transaction reponse is
+ Event callback for when a client socket is writable.
-   avaiable from lond.  The response is forwarded to the unix socket
-   as it becomes writable in this sub.
+ This callback is established when a transaction response is
+ available from lond.  The response is forwarded to the unix socket
+ as it becomes writable in this sub.
  Parameters:
- =item Event - The event that has been triggered. Event->w->data is
+ =item Event
-               the data and Event->w->fd is the socket to write.
+ The event that has been triggered. Event->w->data is
+ the data and Event->w->fd is the socket to write.
  =cut
  sub ClientWritable {
      my $Event    = shift;
      my $Watcher  = $Event->w;
+     if (!defined($Watcher)) {
+ 	&child_exit(-1,'No watcher for event in ClientWritable');
+     }
      my $Data     = $Watcher->data;
      my $Socket   = $Watcher->fd;
- Line 231  sub ClientWritable {
+ Line 519  sub ClientWritable {
      &Debug(6, "ClientWritable writing".$Data);
      &Debug(9, "Socket is: ".$Socket);
-     my $result = $Socket->send($Data, 0);
+     if($Socket->connected) {
+ 	my $result = $Socket->send($Data, 0);
-     # $result undefined: the write failed.
-     # otherwise $result is the number of bytes written.
-     # Remove that preceding string from the data.
-     # If the resulting data is empty, destroy the watcher
-     # and set up a read event handler to accept the next
-     # request.
-     &Debug(9,"Send result is ".$result." Defined: ".defined($result));
-     if(defined($result)) {
- 	&Debug(9, "send result was defined");
- 	if($result == length($Data)) { # Entire string sent.
- 	    &Debug(9, "ClientWritable data all written");
- 	    $Watcher->cancel();
- 	    #
- 	    #  Set up to read next request from socket:
- 	    my $descr     = sprintf("Connection to lonc client %d",
- 				    $ActiveClients{$Socket});
- 	    Event->io(cb    => \&ClientRequest,
- 		      poll  => 'r',
- 		      desc  => $descr,
- 		      data  => "",
- 		      fd    => $Socket);
- 	} else {		# Partial string sent.
- 	    $Watcher->data(substr($Data, $result));
- 	}
-     } else {			# Error of some sort...
+ 	# $result undefined: the write failed.
+ 	# otherwise $result is the number of bytes written.
- 	# Some errnos are possible:
+ 	# Remove that preceding string from the data.
- 	my $errno = $!;
+ 	# If the resulting data is empty, destroy the watcher
- 	if($errno == POSIX::EWOULDBLOCK   ||
+ 	# and set up a read event handler to accept the next
- 	   $errno == POSIX::EAGAIN        ||
+ 	# request.
- 	   $errno == POSIX::EINTR) {
- 	    # No action taken?
- 	} else {		# Unanticipated errno.
- 	    &Debug(5,"ClientWritable error or peer shutdown".$RemoteHost);
- 	    $Watcher->cancel;	# Stop the watcher.
- 	    $Socket->shutdown(2); # Kill connection
- 	    $Socket->close();	# Close the socket.
- 	}
+ 	&Debug(9,"Send result is ".$result." Defined: ".defined($result));
+ 	if($result ne undef) {
+ 	    &Debug(9, "send result was defined");
+ 	    if($result == length($Data)) { # Entire string sent.
+ 		&Debug(9, "ClientWritable data all written");
+ 		$Watcher->cancel();
+ 		#
+ 		#  Set up to read next request from socket:
+ 		my $descr     = sprintf("Connection to lonc client %d",
+ 					$ActiveClients{$Socket});
+ 		Event->io(cb    => \&ClientRequest,
+ 			  poll  => 'r',
+ 			  desc  => $descr,
+ 			  data  => "",
+ 			  fd    => $Socket);
+ 	    } else {		# Partial string sent.
+ 		$Watcher->data(substr($Data, $result));
+ 		if($result == 0) {    # client hung up on us!!
+ 		    # Log("INFO", "lonc pipe client hung up on us!");
+ 		    $Watcher->cancel;
+ 		    $Socket->shutdown(2);
+ 		    $Socket->close();
+ 		}
+ 	    }
+ 	} else {			# Error of some sort...
+ 	    # Some errnos are possible:
+ 	    my $errno = $!;
+ 	    if($errno == POSIX::EWOULDBLOCK   ||
+ 	       $errno == POSIX::EAGAIN        ||
+ 	       $errno == POSIX::EINTR) {
+ 		# No action taken...the socket will be writable firing the event again
+ 		# which will result in a retry of the write.
+ 	    } else {		# Unanticipated errno.
+ 		&Debug(5,"ClientWritable error or peer shutdown".$RemoteHost);
+ 		$Watcher->cancel;	# Stop the watcher.
+ 		$Socket->shutdown(2); # Kill connection
+ 		$Socket->close();	# Close the socket.
+ 	    }
+ 	}
+     } else {
+ 	$Watcher->cancel();	# A delayed request...just cancel.
+ 	return;
      }
  }
  =pod
  =head2 CompleteTransaction
-   Called when the reply data has been received for a lond
+ Called when the reply data has been received for a lond
  transaction.   The reply data must now be sent to the
  ultimate client on the other end of the Unix socket.  This is
  done by setting up a writable event for the socket with the
  data the reply data.
  Parameters:
- =item Socket - Socket on which the lond transaction occured.  This
-       is a LondConnection. The data received is in the
+ =item Socket
- 	TransactionReply member.
- =item Client - Unix domain socket open on the ultimate client.
+ Socket on which the lond transaction occurred.  This is a
+ LondConnection. The data received are in the TransactionReply member.
+ =item Transaction
+ The transaction that is being completed.
  =cut
  sub CompleteTransaction {
-     &Debug(6,"Complete transaction");
+     &Debug(5,"Complete transaction");
-     my $Socket = shift;
-     my $Client = shift;
+     my ($Socket, $Transaction) = @_;
+     if (!$Transaction->isDeferred()) { # Normal transaction
+ 	my $data   = $Socket->GetReply(); # Data to send.
+ 	if($LogTransactions) {
+ 	    Log("SUCCESS", "Reply from lond: '$data'");
+ 	}
+ 	StartClientReply($Transaction, $data);
+     } else {			# Delete deferred transaction file.
+ 	Log("SUCCESS", "A delayed transaction was completed");
+ 	LogPerm("S:".$Transaction->getClient().":".$Transaction->getRequest());
+ 	unlink($Transaction->getFile());
+     }
+ }
+ =pod
+ =head1 StartClientReply
-     my $data   = $Socket->GetReply(); # Data to send.
+    Initiates a reply to a client where the reply data is a parameter.
+ =head2  parameters:
+ =item Transaction
+     The transaction for which we are responding to the client.
+ =item data
+     The data to send to apache client.
+ =cut
+ sub StartClientReply {
+     my ($Transaction, $data) = @_;
+     my $Client   = $Transaction->getClient();
      &Debug(8," Reply was: ".$data);
      my $Serial         = $ActiveClients{$Client};
- Line 311  sub CompleteTransaction {
+ Line 655  sub CompleteTransaction {
  	      data     => $data);
  }
+ =pod
+ =head2 FailTransaction
+   Finishes a transaction with failure because the associated lond socket
+   disconnected.  There are two possibilities:
+   - The transaction is deferred: in which case we just quietly
+     delete the transaction since there is no client connection.
+   - The transaction is 'live' in which case we initiate the sending
+     of "con_lost" to the client.
+ Deleting the transaction means killing it from the %ActiveTransactions hash.
+ Parameters:
+ =item client
+    The LondTransaction we are failing.
+ =cut
+ sub FailTransaction {
+     my $transaction = shift;
+     #  If the socket is dead, that's already logged.
+     if ($ConnectionRetriesLeft > 0) {
+ 	Log("WARNING", "Failing transaction "
+ 	    .$transaction->getLoggableRequest());
+     }
+     Debug(1, "Failing transaction: ".$transaction->getLoggableRequest());
+     if (!$transaction->isDeferred()) { # If the transaction is deferred we'll get to it.
+ 	my $client  = $transaction->getClient();
+ 	Debug(1," Replying con_lost to ".$transaction->getRequest());
+ 	StartClientReply($transaction, "con_lost\n");
+     }
+ }
+ =pod
+ =head1  EmptyQueue
+   Fails all items in the work queue with con_lost.
+   Note that each item in the work queue is a transaction.
+ =cut
+ sub EmptyQueue {
+     $ConnectionRetriesLeft--;	# Counts as connection failure too.
+     while($WorkQueue->Count()) {
+ 	my $request = $WorkQueue->dequeue(); # This is a transaction
+ 	FailTransaction($request);
+     }
+ }
  =pod
+ =head2 CloseAllLondConnections
+ Close all connections open on lond prior to exit e.g.
+ =cut
+ sub CloseAllLondConnections {
+     foreach my $Socket (keys %ActiveConnections) {
+       if(exists($ActiveTransactions{$Socket})) {
+ 	FailTransaction($ActiveTransactions{$Socket});
+       }
+       KillSocket($Socket);
+     }
+ }
+ =pod
+ =head2 KillSocket
+ Destroys a socket.  This function can be called either when a socket
+ has died of 'natural' causes or because a socket needs to be pruned due to
+ idleness.  If the socket has died naturally, if there are no longer any
+ live connections a new connection is created (in case there are transactions
+ in the queue).  If the socket has been pruned, it is never re-created.
+ Parameters:
+ =item Socket
+   The socket to kill off.
+ =item restart
+ non-zero if we are allowed to create a new connection.
+ =cut
+ sub KillSocket {
+     my $Socket = shift;
+     my $restart = shift;
+     Log("WARNING", "Shutting down a socket");
+     $Socket->Shutdown();
+     #  If the socket came from the active connection set,
+     #  delete its transaction... note that FailTransaction should
+     #  already have been called!!!
+     #  otherwise it came from the idle set.
+     #
+     if(exists($ActiveTransactions{$Socket})) {
+ 	delete ($ActiveTransactions{$Socket});
+     }
+     if(exists($ActiveConnections{$Socket})) {
+ 	$ActiveConnections{$Socket}->cancel;
+ 	delete($ActiveConnections{$Socket});
+         # Decrement ConnectionCount unless we will immediately
+         # re-connect (i.e., $restart is true), because this was
+         # a connection where the SSL channel for exchange of the
+         # shared key failed, and we may use an insecure channel.
+         unless ($restart) {
+ 	    $ConnectionCount--;
+         }
+ 	if ($ConnectionCount < 0) { $ConnectionCount = 0; }
+     }
+     #  If the connection count has gone to zero and there is work in the
+     #  work queue, the work all gets failed with con_lost.
+     #
+     if($ConnectionCount == 0) {
+ 	$LondConnecting = 0;	# No connections so also not connecting.
+ 	EmptyQueue();
+ 	CloseAllLondConnections(); # Should all already be closed but...
+         &clear_childpid($$);
+     }
+     UpdateStatus();
+ }
+ =pod
  =head2 LondReadable
  This function is called whenever a lond connection
  is readable.  The action is state dependent:
- =head3 State = Initialized
+ =head3 State=Initialized
-     We''re waiting for the challenge, this is a no-op until the
+ We are waiting for the challenge, this is a no-op until the
  state changes.
  =head3 State=Challenged
-     The challenge has arrived we need to transition to Writable.
+ The challenge has arrived we need to transition to Writable.
  The connection must echo the challenge back.
  =head3 State=ChallengeReplied
-     The challenge has been replied to.  The we are receiveing the
+ The challenge has been replied to.  Then we are receiving the
  'ok' from the partner.
+ =head3  State=ReadingVersionString
+ We have requested the lond version and are reading the
+ version back.  Upon completion, we'll store the version away
+ for future use(?).
+ =head3 State=HostSet
+ We have selected the domain name of our peer (multhomed hosts)
+ and are getting the reply (presumably ok) back.
  =head3 State=RequestingKey
-     The ok has been received and we need to send the request for
+ The ok has been received and we need to send the request for
  an encryption key.  Transition to writable for that.
  =head3 State=ReceivingKey
-     The the key has been requested, now we are reading the new key.
+ The the key has been requested, now we are reading the new key.
  =head3 State=Idle
-     The encryption key has been negotiated or we have finished
- reading data from the a transaction.   If the callback data has
+ The encryption key has been negotiated or we have finished
- a client as well as the socket iformation, then we are
+ reading data from the a transaction.   If the callback data have
- doing a transaction and the data received is relayed to the client
+ a client as well as the socket information, then we are
+ doing a transaction and the data received are relayed to the client
  before the socket is put on the idle list.
  =head3 State=SendingRequest
-     I do not think this state can be received here, but if it is,
+ I do not think this state can be received here, but if it is,
  the appropriate thing to do is to transition to writable, and send
  the request.
  =head3 State=ReceivingReply
-     We finished sending the request to the server and now transition
+ We finished sending the request to the server and now transition
  to readable to receive the reply.
  The parameter to this function are:
  The event. Implicit in this is the watcher and its data.  The data
- contains at least the lond connection object and, if a
+ contain at least the lond connection object and, if a
  transaction is in progress, the socket attached to the local client.
  =cut
  sub LondReadable {
      my $Event      = shift;
      my $Watcher    = $Event->w;
      my $Socket     = $Watcher->data;
      my $client     = undef;
+     &Debug(6,"LondReadable called state = ".$Socket->GetState());
      my $State = $Socket->GetState(); # All action depends on the state.
-     &Debug(6,"LondReadable called state = ".$State);
      SocketDump(6, $Socket);
+     my $status = $Socket->Readable();
+     &Debug(2, "Socket->Readable returned: $status");
-     if($Socket->Readable() != 0) {
+     if($status != 0) {
- 	 # bad return from socket read.
+ 	# bad return from socket read. Currently this means that
+ 	# The socket has become disconnected. We fail the transaction.
+ 	Log("WARNING",
+ 	    "Lond connection lost.");
+         my $state_on_exit = $Socket->GetState();
+ 	if(exists($ActiveTransactions{$Socket})) {
+ 	    FailTransaction($ActiveTransactions{$Socket});
+ 	} else {
+ 	    #  Socket is connecting and failed... need to mark
+ 	    #  no longer connecting.
+ 	    $LondConnecting = 0;
+ 	}
+ 	$Watcher->cancel();
+         if ($state_on_exit eq 'ReInitNoSSL') {
+             # SSL certificate verification failed, and insecure connection
+             # allowed. Send restart arg to KillSocket(), so EmptyQueue()
+             # is not called, as we still hope to process queued request.
+             KillSocket($Socket,1);
+             # Re-initiate creation of Lond Connection for use with queued
+             # request.
+             ShowStatus("Connected to ".$RemoteHost);
+             Log("WARNING","No SSL channel (verification failed), will try with insecure channel");
+             &MakeLondConnection(1);
+         } else {
+ 	    KillSocket($Socket);
+ 	    $ConnectionRetriesLeft--;       # Counts as connection failure
+         }
+ 	return;
      }
      SocketDump(6,$Socket);
      $State = $Socket->GetState(); # Update in case of transition.
      &Debug(6, "After read, state is ".$State);
-    if($State eq "Initialized") {
+     if($State eq "Initialized") {
+     } elsif ($State eq "ReInitNoSSL") {
      } elsif ($State eq "ChallengeReceived") {
  	#  The challenge must be echoed back;  The state machine
  	# in the connection takes care of setting that up.  Just
  	# need to transition to writable:
- 	$Watcher->poll("w");
  	$Watcher->cb(\&LondWritable);
+ 	$Watcher->poll("w");
      } elsif ($State eq "ChallengeReplied") {
+     } elsif ($State eq "RequestingVersion") {
+ 	# Need to ask for the version... that is writiability:
+ 	$Watcher->cb(\&LondWritable);
+ 	$Watcher->poll("w");
+     } elsif ($State eq "ReadingVersionString") {
+ 	# Read the rest of the version string...
+     } elsif ($State eq "SetHost") {
+ 	# Need to request the actual domain get set...
+ 	$Watcher->cb(\&LondWritable);
+ 	$Watcher->poll("w");
+     } elsif ($State eq "HostSet") {
+ 	# Reading the 'ok' from the peer.
      } elsif ($State eq "RequestingKey") {
  	#  The ok was received.  Now we need to request the key
  	#  That requires us to be writable:
- 	$Watcher->poll("w");
  	$Watcher->cb(\&LondWritable);
+ 	$Watcher->poll("w");
      } elsif ($State eq "ReceivingKey") {
      } elsif ($State eq "Idle") {
+         if ($ConnectionCount == 1) {
+             # Write child Pid file to keep track of ssl and insecure
+             # connections
+             &record_childpid($Socket);
+         }
+ 	# This is as good a spot as any to get the peer version
+ 	# string:
+ 	if($LondVersion eq "unknown") {
+ 	    $LondVersion = $Socket->PeerVersion();
+ 	    Log("INFO", "Connected to lond version: $LondVersion");
+ 	}
  	# If necessary, complete a transaction and then go into the
  	# idle queue.
+ 	#  Note that a trasition to idle indicates a live lond
+ 	# on the other end so reset the connection retries.
+ 	#
+ 	$ConnectionRetriesLeft = $ConnectionRetries; # success resets the count
+ 	$Watcher->cancel();
  	if(exists($ActiveTransactions{$Socket})) {
- 	    Debug(8,"Completing transaction!!");
+ 	    Debug(5,"Completing transaction!!");
  	    CompleteTransaction($Socket,
  				$ActiveTransactions{$Socket});
+ 	} else {
+ 	    my $count = $Socket->GetClientData();
+ 	    Log("SUCCESS", "Connection ".$count." to "
+ 		.$RemoteHost." now ready for action");
  	}
- 	$Watcher->cancel();
  	ServerToIdle($Socket);	# Next work unit or idle.
+ 	#
+ 	$LondConnecting = 0;	# Best spot I can think of for this.
+ 	#
      } elsif ($State eq "SendingRequest") {
  	#  We need to be writable for this and probably don't belong
  	#  here inthe first place.
- 	Deubg(6, "SendingRequest state encountered in readable");
+ 	Debug(6, "SendingRequest state encountered in readable");
  	$Watcher->poll("w");
  	$Watcher->cb(\&LondWritable);
- Line 418  sub LondReadable {
+ Line 1007  sub LondReadable {
      } else {
- 	 # Invalid state.
+ 	# Invalid state.
  	Debug(4, "Invalid state in LondReadable");
      }
  }
  =pod
  =head2 LondWritable
  This function is called whenever a lond connection
  becomes writable while there is a writeable monitoring
  event.  The action taken is very state dependent:
  =head3 State = Connected
-   The connection is in the process of sending the
-   'init' hailing to the lond on the remote end.
+ The connection is in the process of sending the 'init' hailing to the
-    The connection object''s Writable member is called.
+ lond on the remote end.  The Writable member of the connection object
-    On error, ConnectionError is called to destroy
+ is called.  On error, call ConnectionError to destroy the connection
-    the connection and remove it from the ActiveConnections
+ and remove it from the ActiveConnections hash.
-    hash
  =head3 Initialized
-    'init' has been sent, writability monitoring is
-     removed and readability monitoring is started
+ 'init' has been sent, writability monitoring is removed and
-     with LondReadable as the callback.
+ readability monitoring is started with LondReadable as the callback.
  =head3 ChallengeReceived
-     The connection has received the who are you
-    challenge from the remote system, and is in the
+ The connection has received the who are you challenge from the remote
-     process of sending the challenge response.
+ system, and is in the process of sending the challenge
-     Writable is called.
+ response. Writable is called.
  =head3 ChallengeReplied
-     The connection has replied to the initial challenge
-     The we switch to monitoring readability looking
+ The connection has replied to the initial challenge The we switch to
-     for the server to reply with 'ok'.
+ monitoring readability looking for the server to reply with 'ok'.
  =head3 RequestingKey
-     The connection is in the process of requesting its
-     encryption key.  Writable is called.
+ The connection is in the process of requesting its encryption key.
+ Writable is called.
  =head3 ReceivingKey
-     The connection has sent the request for a key.
-     Switch to readability monitoring to accept the key
+ The connection has sent the request for a key.  Switch to readability
+ monitoring to accept the key
  =head3 SendingRequest
-     The connection is in the process of sending a
-     request to the server.  This request is part of
+ The connection is in the process of sending a request to the server.
-     a client transaction.  All the states until now
+ This request is part of a client transaction.  All the states until
-     represent the client setup protocol. Writable
+ now represent the client setup protocol. Writable is called.
-     is called.
  =head3 ReceivingReply
-     The connection has sent a request.  Now it must
-     receive a reply.  Readability monitoring is
-     requested.
-  This function is an event handler and therefore receives as
+ The connection has sent a request.  Now it must receive a reply.
+ Readability monitoring is requested.
+ This function is an event handler and therefore receives as
  a parameter the event that has fired.  The data for the watcher
  of this event is a reference to a list of one or two elements,
  depending on state. The first (and possibly only) element is the
- Line 472  socket.  The second (present only if a r
+ Line 1072  socket.  The second (present only if a r
  is the socket on which to return a reply to the caller.
  =cut
  sub LondWritable {
      my $Event   = shift;
      my $Watcher = $Event->w;
-     my @data    = $Watcher->data;
+     my $Socket  = $Watcher->data;
-     Debug(6,"LondWritable State = ".$State." data has ".@data." elts.\n");
+     my $State   = $Socket->GetState();
-     my $Socket  = $data[0];	# I know there's at least a socket.
+     Debug(6,"LondWritable State = ".$State."\n");
      #  Figure out what to do depending on the state of the socket:
-     my $State   = $Socket->GetState();
      SocketDump(6,$Socket);
+     #  If the socket is writable, we must always write.
+     # Only by writing will we undergo state transitions.
+     # Old logic wrote in state specific code below, however
+     # That forces us at least through another invocation of
+     # this function after writability is possible again.
+     # This logic also factors out common code for handling
+     # write failures... in all cases, write failures
+     # Kill the socket.
+     #  This logic makes the branches of the >big< if below
+     # so that the writing states are actually NO-OPs.
+     if ($Socket->Writable() != 0) {
+ 	#  The write resulted in an error.
+ 	# We'll treat this as if the socket got disconnected:
+ 	Log("WARNING", "Connection to ".$RemoteHost.
+ 	    " has been disconnected");
+ 	if(exists($ActiveTransactions{$Socket})) {
+ 	    FailTransaction($ActiveTransactions{$Socket});
+ 	} else {
+ 	    #  In the process of conneting, so need to turn that off.
+ 	    $LondConnecting = 0;
+ 	}
+ 	$Watcher->cancel();
+ 	KillSocket($Socket);
+ 	return;
+     }
      if      ($State eq "Connected")         {
- 	#  "init" is being sent...
- 	if ($Socket->Writable() != 0) {
+ 	#  "init" is being sent...
- 	    #  The write resulted in an error.
- 	}
      } elsif ($State eq "Initialized")       {
  	# Now that init was sent, we switch
  	# to watching for readability:
- 	$Watcher->poll("r");
  	$Watcher->cb(\&LondReadable);
+ 	$Watcher->poll("r");
+     } elsif ($State eq "ReInitNoSSL") {
      } elsif ($State eq "ChallengeReceived") {
  	# We received the challenge, now we
  	# are echoing it back. This is a no-op,
  	# we're waiting for the state to change
- 	if($Socket->Writable() != 0) {
- 	    # Write of the next chunk resulted in an error.
- 	}
      } elsif ($State eq "ChallengeReplied")  {
  	# The echo was sent back, so we switch
  	# to watching readability.
+ 	$Watcher->cb(\&LondReadable);
  	$Watcher->poll("r");
+     } elsif ($State eq "RequestingVersion") {
+ 	# Sending the peer a version request...
+     } elsif ($State eq "ReadingVersionString") {
+ 	# Transition to read since we have sent the
+ 	# version command and now just need to read the
+ 	# version string from the peer:
  	$Watcher->cb(\&LondReadable);
+ 	$Watcher->poll("r");
+     } elsif ($State eq "SetHost") {
+ 	#  Setting the remote domain...
+     } elsif ($State eq "HostSet") {
+ 	# Back to readable to get the ok.
+ 	$Watcher->cb(\&LondReadable);
+ 	$Watcher->poll("r");
      } elsif ($State eq "RequestingKey")     {
  	# At this time we're requesting the key.
  	# again, this is essentially a no-op.
- 	# we'll write the next chunk until the
- 	# state changes.
- 	if($Socket->Writable() != 0) {
- 	    # Write resulted in an error.
- 	}
      } elsif ($State eq "ReceivingKey")      {
  	# Now we need to wait for the key
  	# to come back from the peer:
- 	$Watcher->poll("r");
  	$Watcher->cb(\&LondReadable);
+ 	$Watcher->poll("r");
      } elsif ($State eq "SendingRequest")    {
  	# At this time we are sending a request to the
  	# peer... write the next chunk:
- 	if($Socket->Writable() != 0) {
- 	    # Write resulted in an error.
- 	}
      } elsif ($State eq "ReceivingReply")    {
  	# The send has completed.  Wait for the
  	# data to come in for a reply.
  	Debug(8,"Writable sent request/receiving reply");
- 	$Watcher->poll("r");
  	$Watcher->cb(\&LondReadable);
+ 	$Watcher->poll("r");
      } else {
  	#  Control only passes here on an error:
- Line 564  sub LondWritable {
+ Line 1201  sub LondWritable {
  }
  =pod
+ =cut
+ sub QueueDelayed {
+     Debug(3,"QueueDelayed called");
+     my $path = "$perlvar{'lonSockDir'}/delayed";
+     Debug(4, "Delayed path: ".$path);
+     opendir(DIRHANDLE, $path);
+     my $host_id_re = '(?:'.join('|',map {quotemeta($_)} (@all_host_ids)).')';
+     my @alldelayed = grep(/\.$host_id_re$/, readdir(DIRHANDLE));
+     closedir(DIRHANDLE);
+     foreach my $dfname (sort(@alldelayed)) {
+ 	my $reqfile = "$path/$dfname";
+ 	my ($host_id) = ($dfname =~ /\.([^.]*)$/);
+ 	Debug(4, "queueing ".$reqfile." for $host_id");
+ 	my $Handle = IO::File->new($reqfile);
+ 	my $cmd    = <$Handle>;
+ 	chomp $cmd;		# There may or may not be a newline...
+ 	$cmd = $cmd."\n";	# now for sure there's exactly one newline.
+ 	my $Transaction = LondTransaction->new("sethost:$host_id:$cmd");
+ 	$Transaction->SetDeferred($reqfile);
+ 	QueueTransaction($Transaction);
+     }
+ }
+ =pod
  =head2 MakeLondConnection
-     Create a new lond connection object, and start it towards
-     its initial idleness.  Once idle, it becomes elligible to
+ Create a new lond connection object, and start it towards its initial
-     receive transactions from the work queue.  If the work queue
+ idleness.  Once idle, it becomes eligible to receive transactions
-     is not empty when the connection is completed and becomes idle,
+ from the work queue.  If the work queue is not empty when the
-     it will dequeue an entry and start off on it.
+ connection is completed and becomes idle, it will dequeue an entry and
+ start off on it.
  =cut
- sub MakeLondConnection {
+ sub MakeLondConnection {
+     my ($restart) = @_;
      Debug(4,"MakeLondConnection to ".GetServerHost()." on port "
  	  .GetServerPort());
      my $Connection = LondConnection->new(&GetServerHost(),
- 					 &GetServerPort());
+ 					 &GetServerPort(),
+ 					 &GetHostId(),
-     if($Connection == undef) {	# Needs to be more robust later.
+ 					 &GetDefHostId(),
- 	die "Failed to make a connection!!".$!."\n";
+ 					 &GetLoncapaRev());
+     if($Connection eq undef) {
+ 	Log("CRITICAL","Failed to make a connection with lond.");
+ 	$ConnectionRetriesLeft--;
+ 	return 0;		# Failure.
+     }  else {
+ 	$LondConnecting = 1;	# Connection in progress.
+ 	# The connection needs to have writability
+ 	# monitored in order to send the init sequence
+ 	# that starts the whole authentication/key
+ 	# exchange underway.
+ 	#
+ 	my $Socket = $Connection->GetSocket();
+ 	if($Socket eq undef) {
+ 	    &child_exit(-1, "did not get a socket from the connection");
+ 	} else {
+ 	    &Debug(9,"MakeLondConnection got socket: ".$Socket);
+ 	}
-     }
+ 	$Connection->SetTimeoutCallback(\&SocketTimeout);
-     # The connection needs to have writability
-     # monitored in order to send the init sequence
-     # that starts the whole authentication/key
-     # exchange underway.
-     #
-     my $Socket = $Connection->GetSocket();
-     if($Socket == undef) {
- 	die "did not get a socket from the connection";
-     } else {
- 	&Debug(9,"MakeLondConnection got socket: ".$Socket);
-     }
-     $event = Event->io(fd       => $Socket,
- 		       poll     => 'w',
- 		       cb       => \&LondWritable,
- 		       data     => ($Connection, undef),
- 		       desc => 'Connection to lond server');
-     $ActiveConnections{$Lond} = $event;
-     $ConnectionCount++;
+ 	my $event = Event->io(fd       => $Socket,
+ 			   poll     => 'w',
+ 			   cb       => \&LondWritable,
+ 			   data     => $Connection,
+ 			   desc => 'Connection to lond server');
+ 	$ActiveConnections{$Connection} = $event;
+ 	if ($ConnectionCount == 0) {
+ 	    &SetupTimer;	# Need to handle timeouts with connections...
+ 	}
+         unless ($restart) {
+ 	    $ConnectionCount++;
+         }
+ 	$Connection->SetClientData($ConnectionCount);
+ 	Debug(4, "Connection count = ".$ConnectionCount);
+ 	if($ConnectionCount == 1) { # First Connection:
+ 	    QueueDelayed;
+ 	}
+ 	Log("SUCCESS", "Created connection ".$ConnectionCount
+ 	    ." to host ".GetServerHost());
+ 	return 1;		# Return success.
+     }
  }
  =pod
  =head2 StartRequest
-    Starts a lond request going on a specified lond connection.
-     parameters are:
+ Starts a lond request going on a specified lond connection.
- =item $Lond - Connection to the lond that will send the transaction
+ parameters are:
-     and receive the reply.
- =item $Client - Connection to the client that is making this request
+ =item $Lond
-     We got the request from this socket, and when the request has
-     been relayed to lond and we get a reply back from lond it will
+ Connection to the lond that will send the transaction and receive the
-     get sent to this socket.
+ reply.
- =item $Request - The text of the request to send.
+ =item $Client
+ Connection to the client that is making this request. We got the
+ request from this socket, and when the request has been relayed to
+ lond and we get a reply back from lond it will get sent to this
+ socket.
+ =item $Request
+ The text of the request to send.
  =cut
  sub StartRequest {
-     my $Lond     = shift;
-     my $Client   = shift;
+     my ($Lond, $Request) = @_;
-     my $Request  = shift;
-     Debug(6, "StartRequest: ".$Request);
+     Debug(6, "StartRequest: ".$Request->getRequest());
      my $Socket = $Lond->GetSocket();
-     $ActiveTransactions{$Lond} = $Client; # Socket to relay to client.
+     $Request->Activate($Lond);
+     $ActiveTransactions{$Lond} = $Request;
-     $Lond->InitiateTransaction($Request);
+     $Lond->InitiateTransaction($Request->getRequest());
-     $event = Event->io(fd      => $Lond->GetSocket(),
+     my $event = Event->io(fd      => $Socket,
  		       poll    => "w",
  		       cb      => \&LondWritable,
  		       data    => $Lond,
- Line 641  sub StartRequest {
+ Line 1345  sub StartRequest {
  }
  =pod
  =head2 QueueTransaction
-     - If there is an idle lond connection, it is put to
-     work doing this transaction.  Otherwise, the transaction is
+ If there is an idle lond connection, it is put to work doing this
-     placed in the work queue.  If placed in the work queue and the
+ transaction.  Otherwise, the transaction is placed in the work queue.
-     maximum number of connections has not yet been created, a new
+ If placed in the work queue and the maximum number of connections has
-     connection will be started.  Our goal is to eventually have
+ not yet been created, a new connection will be started.  Our goal is
-     a sufficient number of connections that the work queue will
+ to eventually have a sufficient number of connections that the work
-     typically be empty.
+ queue will typically be empty.  parameters are:
-     parameters are:
- =item Socket open on the lonc client.
+ =item Socket
- =item Request data to send to the lond.
+ open on the lonc client.
+ =item Request
+ data to send to the lond.
  =cut
  sub QueueTransaction {
-     my $requestSocket = shift;
-     my $requestData   = shift;
-     Debug(6,"QueueTransaction: ".$requestData);
+     my $requestData   = shift;	# This is a LondTransaction.
+     my $cmd           = $requestData->getRequest();
+     Debug(6,"QueueTransaction: ".$cmd);
      my $LondSocket    = $IdleConnections->pop();
      if(!defined $LondSocket) {	# Need to queue request.
- 	Debug(8,"Must queue...");
+ 	Debug(5,"Must queue...");
- 	$ClientQueue->enqueue($requestSocket);
  	$WorkQueue->enqueue($requestData);
- 	if($ConnectionCount < $MaxConnectionCount) {
+ 	Debug(5, "Queue Transaction startnew $ConnectionCount $LondConnecting");
- 	    Debug(4,"Starting additional lond connection");
+ 	if(($ConnectionCount < $MaxConnectionCount)   && (! $LondConnecting)) {
- 	    MakeLondConnection();
+ 	    if($ConnectionRetriesLeft > 0) {
+ 		Debug(5,"Starting additional lond connection");
+ 		if(&MakeLondConnection() == 0) {
+ 		    EmptyQueue();	# Fail transactions, can't make connection.
+ 		    CloseAllLondConnections; # Should all be closed but...
+ 		}
+ 	    } else {
+ 		ShowStatus(GetServerHost()." >>> DEAD !!!! <<<");
+ 		$LondConnecting = 0;
+ 		EmptyQueue();	# It's worse than that ... he's dead Jim.
+ 		CloseAllLondConnections; # Should all be closed but..
+ 	    }
  	}
      } else {			# Can start the request:
  	Debug(8,"Can start...");
- 	StartRequest($LondSocket, $requestSocket, $requestData);
+ 	StartRequest($LondSocket,  $requestData);
      }
  }
- #-------------------------- Lonc UNIX socket handling ---------------------
+ #-------------------------- Lonc UNIX socket handling -------------------
  =pod
  =head2 ClientRequest
-    Callback that is called when data can be read from the
-    UNIX domain socket connecting us with an apache server process.
+ Callback that is called when data can be read from the UNIX domain
+ socket connecting us with an apache server process.
  =cut
- Line 696  sub ClientRequest {
+ Line 1421  sub ClientRequest {
      my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
      Debug(8, "rcv:  data length = ".length($thisread)
  	  ." read =".$thisread);
-     unless (defined $rv && length($thisread)) {
+     unless (defined $rv  && length($thisread)) {
  	 # Likely eof on socket.
  	Debug(5,"Client Socket closed on lonc for ".$RemoteHost);
  	close($socket);
  	$watcher->cancel();
  	delete($ActiveClients{$socket});
+ 	return;
      }
      Debug(8,"Data: ".$data." this read: ".$thisread);
      $data = $data.$thisread;	# Append new data.
      $watcher->data($data);
-     if($data =~ /(.*\n)/) {	# Request entirely read.
+     if($data =~ /\n$/) {	# Request entirely read.
+ 	if ($data eq "close_connection_exit\n") {
+ 	    Log("CRITICAL",
+ 		"Request Close Connection ... exiting");
+ 	    CloseAllLondConnections();
+ 	    exit;
+ 	} elsif ($data eq "reset_retries\n") {
+ 	    Log("INFO", "Resetting Connection Retries.");
+ 	    $ConnectionRetriesLeft = $ConnectionRetries;
+ 	    &UpdateStatus();
+ 	    my $Transaction = LondTransaction->new($data);
+ 	    $Transaction->SetClient($socket);
+ 	    StartClientReply($Transaction, "ok\n");
+ 	    $watcher->cancel();
+ 	    return;
+ 	}
  	Debug(8, "Complete transaction received: ".$data);
- 	QueueTransaction($socket, $data);
+ 	if ($LogTransactions) {
+ 	    Log("SUCCESS", "Transaction: '$data'"); # Transaction has \n.
+ 	}
+ 	my $Transaction = LondTransaction->new($data);
+ 	$Transaction->SetClient($socket);
+ 	QueueTransaction($Transaction);
  	$watcher->cancel();	# Done looking for input data.
      }
  }
+ #
+ #     Accept a connection request for a client (lonc child) and
+ #    start up an event watcher to keep an eye on input from that
+ #    Event.  This can be called both from NewClient and from
+ #    ChildProcess.
+ # Parameters:
+ #    $socket       - The listener socket.
+ # Returns:
+ #   NONE
+ # Side Effects:
+ #    An event is made to watch the accepted connection.
+ #    Active clients hash is updated to reflect the new connection.
+ #    The client connection count is incremented.
+ #
+ sub accept_client {
+     my ($socket) = @_;
- =pod
+     Debug(8, "Entering accept for lonc UNIX socket\n");
- =head2  NewClient
-     Callback that is called when a connection is received on the
-     unix socket for a new client of lonc.  The callback is parameterized
-     by the event.. which is a-priori assumed to be an  io event, and therefore
-     has an fd member that is the Listener socket.   We Accept the connection
-     and register a new event on the readability of that socket:
- =cut
- sub NewClient {
-     Debug(6, "NewClient");
-     my $event      = shift;		# Get the event parameters.
-     my $watcher    = $event->w;
-     my $socket     = $watcher->fd;	# Get the event' socket.
      my $connection = $socket->accept();	# Accept the client connection.
      Debug(8,"Connection request accepted from "
  	  .GetPeername($connection, AF_UNIX));
- Line 744  sub NewClient {
+ Line 1494  sub NewClient {
      $ActiveClients{$connection} = $ClientConnection;
      $ClientConnection++;
  }
- =pod GetLoncSocketPath
-     Returns the name of the UNIX socket on which to listen for client
+ =pod
-     connections.
+ =head2  NewClient
+ Callback that is called when a connection is received on the unix
+ socket for a new client of lonc.  The callback is parameterized by the
+ event.. which is a-priori assumed to be an io event, and therefore has
+ an fd member that is the Listener socket.  We accept the connection
+ and register a new event on the readability of that socket:
+ =cut
+ sub NewClient {
+     Debug(6, "NewClient");
+     my $event      = shift;		# Get the event parameters.
+     my $watcher    = $event->w;
+     my $socket     = $watcher->fd;	# Get the event' socket.
+     &accept_client($socket);
+ }
+ =pod
+ =head2 GetLoncSocketPath
+ Returns the name of the UNIX socket on which to listen for client
+ connections.
+ =head2 Parameters:
+     host (optional)  - Name of the host socket to return.. defaults to
+                        the return from GetServerHost().
  =cut
  sub GetLoncSocketPath {
-     return $UnixSocketDir."/".GetServerHost();
+     my $host = GetServerHost();	# Default host.
+     if (@_) {
+ 	($host)  = @_;		# Override if supplied.
+     }
+     return $UnixSocketDir."/".$host;
  }
- =pod GetServerHost
+ =pod
-    Returns the host whose lond we talk with.
+ =head2 GetServerHost
+ Returns the host whose lond we talk with.
  =cut
- sub GetServerHost {		# Stub - get this from config.
+ sub GetServerHost {
      return $RemoteHost;		# Setup by the fork.
  }
- =pod GetServerPort
-    Returns the lond port number.
+ =pod
+ =head2 GetHostId
+ Returns the hostid whose lond we talk with.
+ =cut
+ sub GetHostId {
+     return $RemoteHostId;		# Setup by the fork.
+ }
+ =pod
+ =head2 GetDefHostId
+ Returns the default hostid for the node whose lond we talk with.
+ =cut
+ sub GetDefHostId {                      # Setup by the fork.
+     return $RemoteDefHostId;
+ }
+ =pod
+ =head2 GetLoncapaRev
+ Returns the LON-CAPA version for the node whose lond we talk with.
  =cut
- sub GetServerPort {		# Stub - get this from config.
+ sub GetLoncapaRev {
+     return $RemoteLoncapaRev;           # Setup by the fork.
+ }
+ =pod
+ =head2 GetServerPort
+ Returns the lond port number.
+ =cut
+ sub GetServerPort {
      return $perlvar{londPort};
  }
- =pod SetupLoncListener
-    Setup a lonc listener event.  The event is called when
+ =pod
-    the socket becomes readable.. that corresponds to the
-    receipt of a new connection.  The event handler established
+ =head2 SetupLoncListener
-    will accept the connection (creating a communcations channel), that
-    int turn will establish another event handler to subess requests.
+ Setup a lonc listener event.  The event is called when the socket
+ becomes readable.. that corresponds to the receipt of a new
+ connection.  The event handler established will accept the connection
+ (creating a communications channel), that in turn will establish
+ another event handler to subess requests.
+ =head2  Parameters:
+    host (optional)   Name of the host to set up a unix socket to.
  =cut
  sub SetupLoncListener {
+     my ($host,$SocketName) = @_;
+     if (!$host) { $host = &GetServerHost(); }
+     if (!$SocketName) { $SocketName = &GetLoncSocketPath($host); }
-     my $socket;
-     my $SocketName = GetLoncSocketPath();
      unlink($SocketName);
-     unless ($socket = IO::Socket::UNIX->new(Local  => $SocketName,
- 					    Listen => 10,
+     my $socket;
+     unless ($socket =IO::Socket::UNIX->new(Local  => $SocketName,
+ 					    Listen => 250,
  					    Type   => SOCK_STREAM)) {
- 	die "Failed to create a lonc listner socket";
+ 	if($I_am_child) {
+ 	    &child_exit(-1, "Failed to create a lonc listener socket");
+ 	} else {
+ 	    die "Failed to create a lonc listner socket";
+ 	}
      }
-     Event->io(cb     => \&NewClient,
+     return $socket;
- 	      poll   => 'r',
+ }
- 	      desc   => 'Lonc listener Unix Socket',
- 	      fd     => $socket);
+ #
+ #   Toggle transaction logging.
+ #  Implicit inputs:
+ #     LogTransactions
+ #  Implicit Outputs:
+ #     LogTransactions
+ sub ToggleTransactionLogging {
+     print STDERR "Toggle transaction logging...\n";
+     if(!$LogTransactions) {
+ 	$LogTransactions = 1;
+     } else {
+ 	$LogTransactions = 0;
+     }
+     Log("SUCCESS", "Toggled transaction logging: $LogTransactions \n");
+ }
+ =pod
+ =head2 ChildStatus
+ Child USR1 signal handler to report the most recent status
+ into the status file.
+ We also use this to reset the retries count in order to allow the
+ client to retry connections with a previously dead server.
+ =cut
+ sub ChildStatus {
+     my $event = shift;
+     my $watcher = $event->w;
+     Debug(2, "Reporting child status because : ".$watcher->data);
+     my $docdir = $perlvar{'lonDocRoot'};
+     open(LOG,">>$docdir/lon-status/loncstatus.txt");
+     flock(LOG,LOCK_EX);
+     print LOG $$."\t".$RemoteHost."\t".$Status."\t".
+ 	$RecentLogEntry."\n";
+     #
+     #  Write out information about each of the connections:
+     #
+     if ($DebugLevel > 2) {
+ 	print LOG "Active connection statuses: \n";
+ 	my $i = 1;
+ 	print STDERR  "================================= Socket Status Dump:\n";
+ 	foreach my $item (keys %ActiveConnections) {
+ 	    my $Socket = $ActiveConnections{$item}->data;
+ 	    my $state  = $Socket->GetState();
+ 	    print LOG "Connection $i State: $state\n";
+ 	    print STDERR "---------------------- Connection $i \n";
+ 	    $Socket->Dump(-1);	# Ensure it gets dumped..
+ 	    $i++;
+ 	}
+     }
+     flock(LOG,LOCK_UN);
+     close(LOG);
+     $ConnectionRetriesLeft = $ConnectionRetries;
+     UpdateStatus();
  }
  =pod
+ =head2 SignalledToDeath
+ Called in response to a signal that causes a chid process to die.
+ =cut
+ sub SignalledToDeath {
+     my $event  = shift;
+     my $watcher= $event->w;
+     Debug(2,"Signalled to death! via ".$watcher->data);
+     my ($signal) = $watcher->data;
+     chomp($signal);
+     Log("CRITICAL", "Abnormal exit.  Child $$ for $RemoteHost "
+ 	."died through "."\"$signal\"");
+     #LogPerm("F:lonc: $$ on $RemoteHost signalled to death: "
+ #	    ."\"$signal\"");
+     &clear_childpid($$);
+     exit 0;
+ }
+ =pod
+ =head2 ToggleDebug
+ This sub toggles trace debugging on and off.
+ =cut
+ sub ToggleDebug {
+     my $Current    = $DebugLevel;
+        $DebugLevel = $NextDebugLevel;
+        $NextDebugLevel = $Current;
+     Log("SUCCESS", "New debugging level for $RemoteHost now $DebugLevel");
+ }
+ =pod
  =head2 ChildProcess
  This sub implements a child process for a single lonc daemon.
+ Optional parameter:
+    $socket  - if provided, this is a socket already open for listening
+               on the client socket. Otherwise, a new listener is set up.
  =cut
  sub ChildProcess {
+     #  We've inherited all the
+     #  events of our parent and those have to be cancelled or else
+     #  all holy bloody chaos will result.. trust me, I already made
+     #  >that< mistake.
+     my $host = GetServerHost();
+     foreach my $listener (keys %parent_dispatchers) {
+ 	my $watcher = $parent_dispatchers{$listener};
+ 	my $s       = $watcher->fd;
+ 	if ($listener ne $host) { # Close everyone but me.
+ 	    Debug(5, "Closing listen socket for $listener");
+ 	    $s->close();
+ 	}
+ 	Debug(5, "Killing watcher for $listener");
-     print "Loncnew\n";
+ 	$watcher->cancel();
+ 	delete($parent_dispatchers{$listener});
-     # For now turn off signals.
+     }
-     $SIG{QUIT}  = IGNORE;
-     $SIG{HUP}   = IGNORE;
-     $SIG{USR1}  = IGNORE;
-     $SIG{INT}   = IGNORE;
-     $SIG{CHLD}  = IGNORE;
-     $SIG{__DIE__}  = IGNORE;
-     SetupTimer();
+     #  kill off the parent's signal handlers too!
+     #
-     SetupLoncListener();
+     for my $handler (keys %parent_handlers) {
+ 	my $watcher = $parent_handlers{$handler};
+ 	$watcher->cancel();
+ 	delete($parent_handlers{$handler});
+     }
+     $I_am_child    = 1;		# Seems like in spite of it all I may still getting
+                                 # parent event dispatches.. flag I'm a child.
+     #
+     #  Signals must be handled by the Event framework...
+     #
+     Event->signal(signal   => "QUIT",
+ 		  cb       => \&SignalledToDeath,
+ 		  data     => "QUIT");
+     Event->signal(signal   => "HUP",
+ 		  cb       => \&ChildStatus,
+ 		  data     => "HUP");
+     Event->signal(signal   => "USR1",
+ 		  cb       => \&ChildStatus,
+ 		  data     => "USR1");
+     Event->signal(signal   => "USR2",
+ 		  cb       => \&ToggleTransactionLogging);
+     Event->signal(signal   => "INT",
+ 		  cb       => \&ToggleDebug,
+ 		  data     => "INT");
+     # Block the pipe signal we'll get when the socket disconnects.  We detect
+     # socket disconnection via send/receive failures. On disconnect, the
+     # socket becomes readable .. which will force the disconnect detection.
+     my $set = POSIX::SigSet->new(SIGPIPE);
+     sigprocmask(SIG_BLOCK, $set);
+     #  Figure out if we got passed a socket or need to open one to listen for
+     #  client requests.
+     my ($socket) = @_;
+     if (!$socket) {
+ 	$socket =  SetupLoncListener();
+     }
+     #  Establish an event to listen for client connection requests.
+     Event->io(cb   => \&NewClient,
+ 	      poll => 'r',
+ 	      desc => 'Lonc Listener Unix Socket',
+ 	      fd   => $socket);
-     $Event::Debuglevel = $DebugLevel;
+     $Event::DebugLevel = $DebugLevel;
      Debug(9, "Making initial lond connection for ".$RemoteHost);
  # Setup the initial server connection:
-     &MakeLondConnection();
+      # &MakeLondConnection(); // let first work request do it.
+     #  need to accept the connection since the event may  not fire.
+     &accept_client($socket);
      Debug(9,"Entering event loop");
      my $ret = Event::loop();		#  Start the main event loop.
-     die "Main event loop exited!!!";
+     &child_exit (-1,"Main event loop exited!!!");
  }
  #  Create a new child for host passed in:
  sub CreateChild {
-     my $host = shift;
+     my ($host, $hostid, $defhostid, $loncaparev) = @_;
+     my $sigset = POSIX::SigSet->new(SIGINT);
+     sigprocmask(SIG_BLOCK, $sigset);
      $RemoteHost = $host;
-     Debug(3, "Forking off child for ".$RemoteHost);
+     ShowStatus('Parent keeping the flock'); # Update time in status message.
-     sleep(5);
+     Log("CRITICAL", "Forking server for ".$host);
-     $pid          = fork;
+     my $pid          = fork;
      if($pid) {			# Parent
- 	$ChildHash{$pid} = $RemoteHost;
+ 	$RemoteHost = "Parent";
+ 	$ChildPid{$pid} = $host;
+ 	sigprocmask(SIG_UNBLOCK, $sigset);
+ 	undef(@all_host_ids);
      } else {			# child.
- 	ChildProcess;
+ 	$RemoteHostId = $hostid;
+ 	$RemoteDefHostId = $defhostid;
+         $RemoteLoncapaRev = $loncaparev;
+ 	ShowStatus("Connected to ".$RemoteHost);
+ 	$SIG{INT} = 'DEFAULT';
+ 	sigprocmask(SIG_UNBLOCK, $sigset);
+ 	&ChildProcess();		# Does not return.
      }
+ }
+ # parent_client_connection:
+ #    Event handler that processes client connections for the parent process.
+ #    This sub is called when the parent is listening on a socket and
+ #    a connection request arrives.  We must:
+ #     Start a child process to accept the connection request.
+ #     Kill our listen on the socket.
+ # Parameter:
+ #    event       - The event object that was created to monitor this socket.
+ #                  event->w->fd is the socket.
+ # Returns:
+ #    NONE
+ #
+ sub parent_client_connection {
+     if ($I_am_child) {
+ 	#  Should not get here, but seem to anyway:
+ 	&Debug(5," Child caught parent client connection event!!");
+ 	my ($event) = @_;
+ 	my $watcher = $event->w;
+ 	$watcher->cancel();	# Try to kill it off again!!
+     } else {
+ 	&Debug(9, "parent_client_connection");
+ 	my ($event)   = @_;
+ 	my $watcher   = $event->w;
+ 	my $socket    = $watcher->fd;
+ 	my $connection = $socket->accept();	# Accept the client connection.
+ 	Event->io(cb      => \&get_remote_hostname,
+ 		  poll    => 'r',
+ 		  data    => "",
+ 		  fd      => $connection);
+     }
  }
+ sub get_remote_hostname {
+     my ($event)   = @_;
+     my $watcher   = $event->w;
+     my $socket    = $watcher->fd;
+     my $thisread;
+     my $rv = $socket->recv($thisread, POSIX::BUFSIZ, 0);
+     Debug(8, "rcv:  data length = ".length($thisread)." read =".$thisread);
+     if (!defined($rv) || length($thisread) == 0) {
+ 	# Likely eof on socket.
+ 	Debug(5,"Client Socket closed on lonc for p_c_c");
+ 	close($socket);
+ 	$watcher->cancel();
+ 	return;
+     }
+     my $data    = $watcher->data().$thisread;
+     $watcher->data($data);
+     if($data =~ /\n$/) {	# Request entirely read.
+ 	chomp($data);
+     } else {
+ 	return;
+     }
+     &Debug(5,"Creating child for $data (parent_client_connection)");
+     (my $hostname,my $lonid,@all_host_ids) = split(':',$data);
+     my $remotelcrev;
+     if ((scalar(@all_host_ids) > 1) && ($all_host_ids[0] =~ /^\d+\.\d+\.[\w.]+$/)) {
+         $remotelcrev = shift(@all_host_ids);
+     }
+     $ChildHost{$hostname}++;
+     if ($ChildHost{$hostname} == 1) {
+ 	&CreateChild($hostname,$lonid,$all_host_ids[-1],$remotelcrev);
+     } else {
+ 	&Log('WARNING',"Request for a second child on $hostname");
+     }
+     # Clean up the listen since now the child takes over until it exits.
+     $watcher->cancel();		# Nolonger listening to this event
+     $socket->send("done\n");
+     $socket->close();
+ }
+ # parent_listen:
+ #    Opens a socket and starts a listen for the parent process on a client UNIX
+ #    domain socket.
+ #
+ #    This involves:
+ #       Creating a socket for listen.
+ #       Removing any socket lock file
+ #       Adding an event handler for this socket becoming readable
+ #         To the parent's event dispatcher.
+ # Parameters:
+ #    loncapa_host    - LonCAPA cluster name of the host represented by the client
+ #                      socket.
+ # Returns:
+ #    NONE
+ #
+ sub parent_listen {
+     my ($loncapa_host) = @_;
+     Debug(5, "parent_listen: $loncapa_host");
+     my ($socket,$file);
+     if (!$loncapa_host) {
+ 	$loncapa_host = 'common_parent';
+ 	$file         = $perlvar{'lonSockCreate'};
+     } else {
+ 	$file         = &GetLoncSocketPath($loncapa_host);
+     }
+     $socket = &SetupLoncListener($loncapa_host,$file);
+     $listening_to{$socket} = $loncapa_host;
+     if (!$socket) {
+ 	die "Unable to create a listen socket for $loncapa_host";
+     }
+     my $lock_file = $file.".lock";
+     unlink($lock_file);		# No problem if it doesn't exist yet [startup e.g.]
+     my $watcher =
+ 	Event->io(cb    => \&parent_client_connection,
+ 		  poll  => 'r',
+ 		  desc  => "Parent listener unix socket ($loncapa_host)",
+ 		  data => "",
+ 		  fd    => $socket);
+     $parent_dispatchers{$loncapa_host} = $watcher;
+ }
+ sub parent_clean_up {
+     my ($loncapa_host) = @_;
+     Debug(1, "parent_clean_up: $loncapa_host");
+     my $socket_file = &GetLoncSocketPath($loncapa_host);
+     unlink($socket_file);	# No problem if it doesn't exist yet [startup e.g.]
+     my $lock_file   = $socket_file.".lock";
+     unlink($lock_file);		# No problem if it doesn't exist yet [startup e.g.]
+ }
+ #    This sub initiates a listen on the common unix domain lonc client socket.
+ #    loncnew starts up with no children, and only spawns off children when a
+ #    connection request occurs on the common client unix socket.  The spawned
+ #    child continues to run until it has been idle a while at which point it
+ #    eventually exits and once more the parent picks up the listen.
+ #
+ #  Parameters:
+ #      NONE
+ #  Implicit Inputs:
+ #    The configuration file that has been read in by LondConnection.
+ #  Returns:
+ #     NONE
+ #
+ sub listen_on_common_socket {
+     Debug(5, "listen_on_common_socket");
+     &parent_listen();
+ }
+ #   server_died is called whenever a child process exits.
+ #   Since this is dispatched via a signal, we must process all
+ #   dead children until there are no more left.  The action
+ #   is to:
+ #      - Remove the child from the bookeeping hashes
+ #      - Re-establish a listen on the unix domain socket associated
+ #        with that host.
+ # Parameters:
+ #    The event, but we don't actually care about it.
+ sub server_died {
+     &Debug(9, "server_died called...");
+     while(1) {			# Loop until waitpid nowait fails.
+ 	my $pid = waitpid(-1, WNOHANG);
+ 	if($pid <= 0) {
+ 	    return;		# Nothing left to wait for.
+ 	}
+ 	# need the host to restart:
+ 	my $host = $ChildPid{$pid};
+ 	if($host) {		# It's for real...
+ 	    &Debug(9, "Caught sigchild for $host");
+             &clear_childpid($pid);
+ 	    delete($ChildPid{$pid});
+ 	    delete($ChildHost{$host});
+ 	    &parent_clean_up($host);
+ 	} else {
+ 	    &Debug(5, "Caught sigchild for pid not in hosts hash: $pid");
+ 	}
+     }
+ }
  #
  #  Parent process logic pass 1:
  #   For each entry in the hosts table, we will
- Line 853  sub CreateChild {
+ Line 2068  sub CreateChild {
  #  Each exit gets logged and the child gets restarted.
  #
- my $HostIterator = LondConnection::GetHostIterator;
+ #
- while (! $HostIterator->end()) {
+ #   Fork and start in new session so hang-up isn't going to
+ #   happen without intent.
+ #
-     $hostentryref = $HostIterator->get();
-     CreateChild($hostentryref->[0]);
+ ShowStatus("Forming new session");
-     $HostIterator->next();
+ my $childpid = fork;
+ if ($childpid != 0) {
+     sleep 4;			# Give child a chacne to break to
+     exit 0;			# a new sesion.
+ }
+ #
+ #   Write my pid into the pid file so I can be located
+ #
+ ShowStatus("Parent writing pid file:");
+ my $execdir = $perlvar{'lonDaemons'};
+ open (PIDSAVE, ">$execdir/logs/lonc.pid");
+ print PIDSAVE "$$\n";
+ close(PIDSAVE);
+ if (POSIX::setsid() < 0) {
+     print "Could not create new session\n";
+     exit -1;
  }
+ ShowStatus("Forking node servers");
+ Log("CRITICAL", "--------------- Starting children ---------------");
+ LondConnection::ReadConfig;               # Read standard config files.
+ $RemoteHost = "[parent]";
+ &listen_on_common_socket();
+ $RemoteHost = "Parent Server";
  # Maintain the population:
- while(1) {
+ ShowStatus("Parent keeping the flock");
-     $deadchild = wait();
-     if(exists $ChildHash{$deadchild}) {	# need to restart.
- 	$deadhost = $ChildHash{$deadchild};
+ # We need to setup a SIGChild event to handle the exit (natural or otherwise)
- 	delete($ChildHash{$deadchild});
+ # of the children.
- 	Debug(4,"Lost child pid= ".$deadchild.
- 	      "Connected to host ".$deadhost);
+ Event->signal(cb       => \&server_died,
- 	CreateChild($deadhost);
+ 	      desc     => "Child exit handler",
+ 	      signal   => "CHLD");
+ # Set up all the other signals we set up.
+ $parent_handlers{INT} = Event->signal(cb       => \&Terminate,
+ 				      desc     => "Parent INT handler",
+ 				      signal   => "INT");
+ $parent_handlers{TERM} = Event->signal(cb       => \&Terminate,
+ 				       desc     => "Parent TERM handler",
+ 				       signal   => "TERM");
+ $parent_handlers{HUP}  = Event->signal(cb       => \&KillThemAll,
+ 				       desc     => "Parent HUP handler.",
+ 				       signal   => "HUP");
+ $parent_handlers{USR1} = Event->signal(cb       => \&CheckKids,
+ 				       desc     => "Parent USR1 handler",
+ 				       signal   => "USR1");
+ $parent_handlers{USR2} = Event->signal(cb       => \&UpdateKids,
+ 				       desc     => "Parent USR2 handler.",
+ 				       signal   => "USR2");
+ #  Start procdesing events.
+ $Event::DebugLevel = $DebugLevel;
+ Debug(9, "Parent entering event loop");
+ my $ret = Event::loop();
+ die "Main Event loop exited: $ret";
+ =pod
+ =head1 CheckKids
+   Since kids do not die as easily in this implementation
+ as the previous one, there is no need to restart the
+ dead ones (all dead kids get restarted when they die!!)
+ The only thing this function does is to pass USR1 to the
+ kids so that they report their status.
+ =cut
+ sub CheckKids {
+     Debug(2, "Checking status of children");
+     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";
+     foreach my $host (keys %parent_dispatchers) {
+ 	print $fh "LONC Parent process listening for $host\n";
+     }
+     foreach my $pid (keys %ChildPid) {
+ 	Debug(2, "Sending USR1 -> $pid");
+ 	kill 'USR1' => $pid;	# Tell Child to report status.
+     }
+ }
+ =pod
+ =head1  UpdateKids
+ parent's SIGUSR2 handler.  This handler:
+ =item
+ Rereads the hosts file.
+ =item
+ Kills off (via sigint) children for hosts that have disappeared.
+ =item
+ QUITs  children for hosts that already exist (this just forces a status display
+ and resets the connection retry count for that host.
+ =item
+ Starts new children for hosts that have been added to the hosts.tab file since
+ the start of the master program and maintains them.
+ =cut
+ sub UpdateKids {
+     Log("INFO", "Updating connections via SIGUSR2");
+     #  I'm not sure what I was thinking in the first implementation.
+     # someone will have to work hard to convince me the effect is any
+     # different than Restart, especially now that we don't start up
+     # per host servers automatically, may as well just restart.
+     # The down side is transactions that are in flight will get timed out
+     # (lost unless they are critical).
+     &KillThemAll();
+     LondConnection->ResetReadConfig();
+     ShowStatus('Parent keeping the flock');
+ }
+ =pod
+ =head1 Restart
+ Signal handler for HUP... all children are killed and
+ we self restart.  This is an el-cheapo way to re-read
+ the config file.
+ =cut
+ sub Restart {
+     &KillThemAll;		# First kill all the children.
+     LondConnection->ResetReadConfig();
+     Log("CRITICAL", "Restarting");
+     my $execdir = $perlvar{'lonDaemons'};
+     unlink("$execdir/logs/lonc.pid");
+     exec("$executable");
+ }
+ =pod
+ =head1 KillThemAll
+ Signal handler that kills all children by sending them a
+ SIGHUP.  Responds to sigint and sigterm.
+ =cut
+ sub KillThemAll {
+     Debug(2, "Kill them all!!");
+     #local($SIG{CHLD}) = 'IGNORE';
+     # Our children >will< die.
+     # but we need to catch their death and cleanup after them in case this is
+     # a restart set of kills
+     my @allpids = keys(%ChildPid);
+     foreach my $pid (@allpids) {
+ 	my $serving = $ChildPid{$pid};
+ 	ShowStatus("Nicely Killing lonc for $serving pid = $pid");
+ 	Log("CRITICAL", "Nicely Killing lonc for $serving pid = $pid");
+ 	kill 'QUIT' => $pid;
+         &clear_childpid($pid);
+     }
+     ShowStatus("Finished killing child processes off.");
+ }
+ #
+ #  Kill all children via KILL.  Just in case the
+ #  first shot didn't get them.
+ sub really_kill_them_all_dammit
+ {
+     Debug(2, "Kill them all Dammit");
+     local($SIG{CHLD} = 'IGNORE'); # In case some purist reenabled them.
+     foreach my $pid (keys %ChildPid) {
+ 	my $serving = $ChildPid{$pid};
+ 	&ShowStatus("Nastily killing lonc for $serving pid = $pid");
+ 	Log("CRITICAL", "Nastily killing lonc for $serving pid = $pid");
+ 	kill 'KILL' => $pid;
+ 	delete($ChildPid{$pid});
+         delete($ChildKeyMode{$pid});
+ 	my $execdir = $perlvar{'lonDaemons'};
+ 	unlink("$execdir/logs/lonc.pid");
+     }
+ }
+ =pod
+ =head1 Terminate
+ Terminate the system.
+ =cut
+ sub Terminate {
+     &Log("CRITICAL", "Asked to kill children.. first be nice...");
+     &KillThemAll;
+     #
+     #  By now they really should all be dead.. but just in case
+     #  send them all SIGKILL's after a bit of waiting:
+     sleep(4);
+     &Log("CRITICAL", "Now kill children nasty");
+     &really_kill_them_all_dammit;
+     Log("CRITICAL","Master process exiting");
+     exit 0;
+ }
+ =pod
+ =cut
+ sub my_hostname {
+     use Sys::Hostname::FQDN();
+     my $name = Sys::Hostname::FQDN::fqdn();
+     &Debug(9,"Name is $name");
+     return $name;
+ }
+ sub record_childpid {
+     my ($Socket) = @_;
+     my $docdir = $perlvar{'lonDocRoot'};
+     my $authmode = $Socket->GetKeyMode();
+     my $peer = $Socket->PeerLoncapaHim();
+     if (($authmode eq 'ssl') || ($authmode eq 'insecure')) {
+         my $childpid = $$;
+         if ($childpid) {
+             unless (exists($ChildKeyMode{$childpid})) {
+                 $ChildKeyMode{$childpid} = $authmode;
+             }
+             if (-d "$docdir/lon-status/loncchld") {
+                 unless (-e "$docdir/lon-status/loncchld/$childpid") {
+                     if (open (my $pidfh,'>',"$docdir/lon-status/loncchld/$childpid")) {
+                         print $pidfh "$peer:$authmode\n";
+                         close($pidfh);
+                     }
+                 }
+             }
+         }
+     }
+     return;
+ }
+ sub clear_childpid {
+     my ($childpid) = @_;
+     my $docdir = $perlvar{'lonDocRoot'};
+     if (-d "$docdir/lon-status/loncchld") {
+         if ($childpid =~ /^\d+$/) {
+             if (($ChildKeyMode{$childpid} eq 'insecure') ||
+                 ($ChildKeyMode{$childpid} eq 'ssl')) {
+                 if (-e "$docdir/lon-status/loncchld/$childpid") {
+                     unlink("$docdir/lon-status/loncchld/$childpid");
+                 }
+             }
+         }
+     }
+     if (exists($ChildKeyMode{$childpid})) {
+         delete($ChildKeyMode{$childpid});
      }
+     return;
  }
+ =pod
  =head1 Theory
-    The event class is used to build this as a single process with
-    an event driven model.  The following events are handled:
+ The event class is used to build this as a single process with an
+ event driven model.  The following events are handled:
  =item UNIX Socket connection Received
- Line 890  All sockets are run in non-blocking mode
+ Line 2385  All sockets are run in non-blocking mode
  handler prevents hung connections.
  Key data structures:
- =item RequestQueue - A queue of requests received from UNIX sockets that are
-     waiting for a chance to be forwarded on a lond connection socket.
- =item ActiveConnections - A hash of lond connections that have transactions
+ =item RequestQueue
-     in process that are available to be timed out.
+ A queue of requests received from UNIX sockets that are
+ waiting for a chance to be forwarded on a lond connection socket.
+ =item ActiveConnections
+ A hash of lond connections that have transactions in process that are
+ available to be timed out.
+ =item ActiveTransactions
+ A hash indexed by lond connections that contain the client reply
+ socket for each connection that has an active transaction on it.
+ =item IdleConnections
+ A hash of lond connections that have no work to do.  These connections
+ can be closed if they are idle for a long enough time.
+ =cut
+ =pod
+ =head1 Log messages
+ The following is a list of log messages that can appear in the
+ lonc.log file.  Each log file has a severity and a message.
+ =over 2
+ =item Warning  A socket timeout was detected
+ If there are pending transactions in the socket's queue,
+ they are failed (saved if critical).  If the connection
+ retry count gets exceeded by this, the
+ remote host is marked as dead.
+ Called when timeouts occurred during the connection and
+ connection dialog with a remote host.
+ =item Critical Host makred DEAD <hostname>
+ The numer of retry counts for contacting a host was
+ exceeded. The host is marked dead an no
+ further attempts will be made by that child.
+ =item Info lonc pipe client hung up on us
+ Write to the client pipe indicated no data transferred
+ Socket to remote host is shut down.  Reply to the client
+ is discarded.  Note: This is commented out in &ClientWriteable
+ =item Success  Reply from lond: <data>
+ Can be enabled for debugging by setting LogTransactions to nonzero.
+ Indicates a successful transaction with lond, <data> is the data received
+ from the remote lond.
+ =item Success A delayed transaction was completed
+ A transaction that must be reliable was executed and completed
+ as lonc restarted.  This is followed by a mesage of the form
+   S: client-name : request
+ =item WARNING  Failing transaction <cmd>:<subcmd>
+ Transaction failed on a socket, but the failure retry count for the remote
+ node has not yet been exhausted (the node is not yet marked dead).
+ cmd is the command, subcmd is the subcommand.  This results from a con_lost
+ when communicating with lond.
+ =item WARNING Shutting down a socket
+ Called when a socket is being closed to lond.  This is emitted both when
+ idle pruning is being done and when the socket has been disconnected by the remote.
+ =item WARNING Lond connection lost.
+ Called when a read from lond's socket failed indicating lond has closed the
+ connection or died.  This should be followed by one or more
+  "WARNING Failing transaction..." msgs for each in-flight or queued transaction.
+ =item WARNING No SSL channel (verification failed), will try with insecure channel.
+ Called when promotion of a socket to SSL failed because SSL certificate verification failed.
+ Domain configuration must also permit insecure channel use for key exchange. Connection
+ negotiation will start again from the beginning, but with Authentication Mode not set to ssl.
+ =item INFO Connected to lond version:  <version>
+ When connection negotiation is complete, the lond version is requested and logged here.
+ =item SUCCESS Connection n to host now ready for action
+ Emitted when connection has been completed with lond. n is then number of
+ concurrent connections and host, the host to which the connection has just
+ been established.
+ =item WARNING Connection to host has been disconnected
+ Write to a lond resulted in failure status.  Connection to lond is dropped.
+ =item SUCCESS Created connection n to host host
+ Initial connection request to host..(before negotiation).
+ =item CRITICAL Request Close Connection ... exiting
+ Client has sent "close_connection_exit"   The loncnew server is exiting.
+ =item INFO Resetting Connection Retries
+ Client has sent "reset_retries" The lond connection retries are reset to zero for the
+ corresponding lond.
+ =item SUCCESS Transaction <data>
+ Only emitted if the global variable $LogTransactions was set to true.
+ A client has requested a lond transaction <data> is the contents of the request.
+ =item SUCCESS Toggled transaction logging <LogTransactions>
+ The state of the $LogTransactions global has been toggled, and its current value
+ (after being toggled) is displayed.  When non zero additional logging of transactions
+ is enabled for debugging purposes.  Transaction logging is toggled on receipt of a USR2
+ signal.
+ =item CRITICAL Abnormal exit. Child <pid> for <host> died thorugh signal.
+ QUIT signal received.  lonc child process is exiting.
+ =item SUCCESS New debugging level for <RemoteHost> now <DebugLevel>
+ Debugging toggled for the host loncnew is talking with.
+ Currently debugging is a level based scheme with higher number
+ conveying more information.  The daemon starts out at
+ DebugLevel 0 and can toggle back and forth between that and
+ DebugLevel 2  These are controlled by
+ the global variables $DebugLevel and $NextDebugLevel
+ The debug level can go up to 9.
+ SIGINT toggles the debug level.  The higher the debug level the
+ more debugging information is spewed.  See the Debug
+ sub in loncnew.
+ =item CRITICAL Forking server for host
+ A child is being created to service requests for the specified host.
+ =item WARNING Request for a second child on hostname
+ Somehow loncnew was asked to start a second child on a host that already had a child
+ servicing it.  This request is not honored, but themessage is emitted.  This could happen
+ due to a race condition.  When a client attempts to contact loncnew for a new host, a child
+ is forked off to handle the requests for that server.  The parent then backs off the Unix
+ domain socket leaving it for the child to service all requests.  If in the time between
+ creating the child, and backing off, a new connection request comes in to the unix domain
+ socket, this could trigger (unlikely but remotely possible),.
+ =item CRITICAL ------ Starting Children ----
+ This message should probably be changed to "Entering event loop"  as the loncnew only starts
+ children as needed.  This message is emitted as new events are established and
+ the event processing loop is entered.
+ =item INFO Updating connections via SIGUSR2
+ SIGUSR2 received. The original code would kill all clients, re-read the host file,
+ then restart children for each host.  Now that children are started on demand, this
+ just kills all child processes and lets requests start them as needed again.
+ =item CRITICAL Restarting
+ SigHUP received.  all the children are killed and the script exec's itself to start again.
+ =item CRITICAL Nicely killing lonc for host pid = <pid>
+ Attempting to kill the child that is serving the specified host (pid given) cleanly via
+ SIGQUIT.  The child should handle that, clean up nicely and exit.
+ =item CRITICAL Nastily killing lonc for host pid = <pid>
+ The child specified did not die when requested via SIGQUIT.  Therefore it is killed
+ via SIGKILL.
+ =item CRITICAL Asked to kill children.. first be nice..
+ In the parent's INT handler.  INT kills the child processes.  This inidicate loncnew
+ is about to attempt to kill all known children via SIGQUIT.  This message should be followed
+ by one "Nicely killing" message for each extant child.
+ =item CRITICAL Now kill children nasty
+ In the parent's INT handler. remaining children are about to be killed via
+ SIGKILL. Should be followed by a Nastily killing... for each lonc child that
+ refused to die.
+ =item CRITICAL Master process exiting
+ In the parent's INT handler. just prior to the exit 0 call.
+ =back
- =item ActiveTransactions - A hash indexed by lond connections that
-     contain the client reply socket for each connection that has an active
-     transaction on it.
- =item IdleConnections - A hash of lond connections that have no work
-     to do.  These connections can be closed if they are idle for a long
-     enough time.
  =cut

FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>

Removed from v.1.2
changed lines
	Added in v.1.110