--- loncom/loncnew 2003/07/29 02:33:05 1.16 +++ loncom/loncnew 2018/06/24 18:45:18 1.104 @@ -2,13 +2,12 @@ # The LearningOnline Network with CAPA # lonc maintains the connections to remote computers # -# $Id: loncnew,v 1.16 2003/07/29 02:33:05 foxr Exp $ +# $Id: loncnew,v 1.104 2018/06/24 18:45:18 raeburn Exp $ # # 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. @@ -27,7 +26,7 @@ # http://www.lon-capa.org/ # # -# new lonc handles n request 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) @@ -36,54 +35,16 @@ # - Add ability to create/negotiate lond connections (done). # - Add general logic for dispatching requests and timeouts. (done). # - Add support for the lonc/lond requests. (done). -# - Add logging/status monitoring. -# - Add Signal handling - HUP restarts. USR1 status report. +# - Add logging/status monitoring. (done) +# - Add Signal handling - HUP restarts. USR1 status report. (done) # - Add Configuration file I/O (done). -# - Add management/status request interface. +# - Add management/status request interface. (done) # - Add deferred request capability. (done) -# - Detect transmission timeouts. -# - -# Change log: -# $Log: loncnew,v $ -# Revision 1.16 2003/07/29 02:33:05 foxr -# Add SIGINT processing to child processes to toggle annoying trace mode -# on/off.. will try to use this to isolate the compute boud process issue. -# -# Revision 1.15 2003/07/15 02:07:05 foxr -# Added code for lonc/lond transaction timeouts. Who knows if it works right. -# The intent is for a timeout to fail any transaction in progress and kill -# off the sockt that timed out. -# -# Revision 1.14 2003/07/03 02:10:18 foxr -# Get all of the signals to work correctly. -# -# Revision 1.13 2003/07/02 01:31:55 foxr -# Added kill -HUP logic (restart). -# -# Revision 1.11 2003/06/25 01:54:44 foxr -# Fix more problems with transaction failure. -# -# Revision 1.10 2003/06/24 02:46:04 foxr -# Put a limit on the number of times we'll retry a connection. -# Start getting the signal stuff put in as well...note that need to get signals -# going or else 6the client will permanently give up on dead servers. -# -# Revision 1.9 2003/06/13 02:38:43 foxr -# Add logging in 'expected format' -# -# Revision 1.8 2003/06/11 02:04:35 foxr -# Support delayed transactions... this is done uniformly by encapsulating -# transactions in an object ... a LondTransaction that is implemented by -# LondTransaction.pm -# -# Revision 1.7 2003/06/03 01:59:39 foxr -# complete coding to support deferred transactions. -# +# - 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; @@ -99,18 +60,7 @@ use LONCAPA::Stack; use LONCAPA::LondConnection; use LONCAPA::LondTransaction; use LONCAPA::Configuration; -use LONCAPA::HashIterator; - - -# -# 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; +use Fcntl qw(:flock); # Read the httpd configuration file to get perl variables @@ -122,21 +72,31 @@ my %perlvar = %{$perlvarref}; # # parent and shared variables. -my %ChildHash; # by pid -> host. +my %ChildPid; # by pid -> host. +my %ChildHost; # by host. +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 = 10; # Will get from config later. my $ClientConnection = 0; # Uniquifier for client events. my $DebugLevel = 0; -my $NextDebugLevel= 10; # So Sigint can toggle this. -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 $UnixSocketDir= "/home/httpd/sockets"; +my $RemoteHostId; # default lonid of host child is talking to. +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; # LondTransactions in flight. @@ -148,6 +108,13 @@ my $Status = ""; # Current stat 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 @@ -155,13 +122,27 @@ my $ConnectionRetriesLeft=5; # Number of # my %LogFormats; -$LogFormats{"CRITICAL"} = "CRITICAL: %s"; -$LogFormats{"SUCCESS"} = "SUCCESS: %s"; -$LogFormats{"INFO"} = "INFO: %s"; -$LogFormats{"WARNING"} = "WARNING: %s"; +$LogFormats{"CRITICAL"} = "CRITICAL: %s"; +$LogFormats{"SUCCESS"} = "SUCCESS: %s"; +$LogFormats{"INFO"} = "INFO: %s"; +$LogFormats{"WARNING"} = "WARNING: %s"; $LogFormats{"DEFAULT"} = " %s "; +# UpdateStatus; +# Update the idle status display to show how many connections +# are left, retries and other stuff. +# +sub UpdateStatus { + if ($ConnectionRetriesLeft > 0) { + ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount + ." Retries remaining: ".$ConnectionRetriesLeft + ." ($KeyMode)"); + } else { + ShowStatus(GetServerHost()." >> DEAD <<"); + } +} + =pod @@ -170,12 +151,14 @@ $LogFormats{"DEFAULT"} = " %s "; 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"; } @@ -201,9 +184,9 @@ host and the time will be formatted into =cut sub Log { - my $severity = shift; - my $message = shift; - + + my ($severity, $message) = @_; + if(!$LogFormats{$severity}) { $severity = "DEFAULT"; } @@ -215,7 +198,7 @@ sub Log { my $now = time; my $local = localtime($now); my $finalformat = "$local ($$) [$RemoteHost] [$Status] "; - my $finalformat = $finalformat.$format."\n"; + $finalformat = $finalformat.$format."\n"; # open the file and put the result. @@ -238,14 +221,16 @@ Returns the name of the host that a sock =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; @@ -253,7 +238,6 @@ sub GetPeername { return $peerfile; } } -#----------------------------- Timer management ------------------------ =pod =head2 Debug @@ -263,18 +247,20 @@ Invoked to issue a debug message. =cut sub Debug { - my $level = shift; - my $message = shift; + + my ($level, $message) = @_; + if ($level <= $DebugLevel) { - Log("INFO", "-Debug- $message host = $RemotHost"); + 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. } } @@ -286,6 +272,7 @@ sub SocketDump { and as what we return in a SIGUSR1 =cut + sub ShowStatus { my $state = shift; my $now = time; @@ -296,31 +283,83 @@ sub ShowStatus { =pod -=head 2 SocketTimeout +=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; - - KillSocket($Socket); + 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; - ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount); + UpdateStatus(); # Is it time to prune connection count: @@ -329,8 +368,14 @@ sub Tick { ($WorkQueue->Count() == 0)) { # Idle connections and nothing to do? $IdleSeconds++; if($IdleSeconds > $IdleTimeout) { # Prune a connection... - $Socket = $IdleConnections->pop(); + 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. @@ -338,29 +383,44 @@ sub Tick { # # For each inflight transaction, tick down its timeout counter. # - foreach $item (keys %ActiveTransactions) { - my $Socket = $ActiveTransactions{$item}->getServer(); - $Socket->Tick(); + + 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)) { + if (($ConnectionCount == 0) && ($Requests > 0) && (!$LondConnecting)) { if ($ConnectionRetriesLeft > 0) { - my $Connections = ($Requests <= $MaxConnectionCount) ? - $Requests : $MaxConnectionCount; - Debug(1,"Work but no connections, start ".$Connections." of them"); - for ($i =0; $i < $Connections; $i++) { - MakeLondConnection(); + 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 { - Debug(1,"Work in queue, but gave up on connections..flushing\n"); + $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 @@ -381,7 +441,8 @@ Trigger disconnections of idle sockets. sub SetupTimer { Debug(6, "SetupTimer"); - Event->timer(interval => 1, debug => 1, cb => \&Tick ); + Event->timer(interval => 1, cb => \&Tick, + hard => 1); } =pod @@ -401,21 +462,22 @@ long enough, it will be shut down and re 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(); # This is a LondTransaction - unless($reqdata eq undef) { - Debug(9, "Queue gave request data: ".$reqdata->getRequest()); + my $reqdata = $WorkQueue->dequeue(); # This is a LondTransaction + if ($reqdata ne undef) { + Debug(5, "Queue gave request data: ".$reqdata->getRequest()); &StartRequest($Socket, $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"); $IdleConnections->push($Socket); } } @@ -426,8 +488,8 @@ sub ServerToIdle { Event callback for when a client socket is writable. -This callback is established when a transaction reponse is -avaiable from lond. The response is forwarded to the unix socket +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: @@ -442,6 +504,9 @@ the data and Event->w->fd is the socket 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; @@ -461,7 +526,7 @@ sub ClientWritable { # request. &Debug(9,"Send result is ".$result." Defined: ".defined($result)); - if(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"); @@ -480,7 +545,7 @@ sub ClientWritable { } 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!"); + # Log("INFO", "lonc pipe client hung up on us!"); $Watcher->cancel; $Socket->shutdown(2); $Socket->close(); @@ -494,7 +559,8 @@ sub ClientWritable { if($errno == POSIX::EWOULDBLOCK || $errno == POSIX::EAGAIN || $errno == POSIX::EINTR) { - # No action taken? + # 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. @@ -505,6 +571,7 @@ sub ClientWritable { } } else { $Watcher->cancel(); # A delayed request...just cancel. + return; } } @@ -522,8 +589,8 @@ Parameters: =item Socket -Socket on which the lond transaction occured. This is a -LondConnection. The data received is in the TransactionReply member. +Socket on which the lond transaction occurred. This is a +LondConnection. The data received are in the TransactionReply member. =item Transaction @@ -532,20 +599,25 @@ The transaction that is being completed. =cut sub CompleteTransaction { - &Debug(6,"Complete transaction"); - my $Socket = shift; - my $Transaction = shift; + &Debug(5,"Complete transaction"); + + 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:$Client:".$Transaction->getRequest()); - unlink $Transaction->getFile(); + LogPerm("S:".$Socket->PeerLoncapaHim().":".$Transaction->getRequest()); + unlink($Transaction->getFile()); } } + =pod + =head1 StartClientReply Initiates a reply to a client where the reply data is a parameter. @@ -558,20 +630,19 @@ sub CompleteTransaction { =item data - The data to send to apached client. + The data to send to apache client. =cut + sub StartClientReply { - my $Transaction = shift; - my $data = shift; + my ($Transaction, $data) = @_; my $Client = $Transaction->getClient(); &Debug(8," Reply was: ".$data); my $Serial = $ActiveClients{$Client}; my $desc = sprintf("Connection to lonc client %d", - $Serial); Event->io(fd => $Client, poll => "w", @@ -579,7 +650,9 @@ sub StartClientReply { cb => \&ClientWritable, data => $data); } + =pod + =head2 FailTransaction Finishes a transaction with failure because the associated lond socket @@ -589,8 +662,7 @@ sub StartClientReply { - 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. +Deleting the transaction means killing it from the %ActiveTransactions hash. Parameters: @@ -598,11 +670,19 @@ Parameters: The LondTransaction we are failing. + =cut sub FailTransaction { my $transaction = shift; - Debug(1, "Failing transaction: ".$transaction->getRequest()); + + # 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()); @@ -612,13 +692,16 @@ sub FailTransaction { } =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); @@ -632,12 +715,15 @@ sub EmptyQueue { Close all connections open on lond prior to exit e.g. =cut + sub CloseAllLondConnections { - foreach $Socket (keys %ActiveConnections) { - KillSocket($Socket); + foreach my $Socket (keys %ActiveConnections) { + if(exists($ActiveTransactions{$Socket})) { + FailTransaction($ActiveTransactions{$Socket}); + } + KillSocket($Socket); } } -=cut =pod @@ -657,13 +743,14 @@ Parameters: =item Restart -nonzero if we are allowed to create a new connection. - +non-zero if we are allowed to create a new connection. =cut + sub KillSocket { my $Socket = shift; + Log("WARNING", "Shutting down a socket"); $Socket->Shutdown(); # If the socket came from the active connection set, @@ -676,16 +763,20 @@ sub KillSocket { delete ($ActiveTransactions{$Socket}); } if(exists($ActiveConnections{$Socket})) { + $ActiveConnections{$Socket}->cancel; delete($ActiveConnections{$Socket}); + $ConnectionCount--; + if ($ConnectionCount < 0) { $ConnectionCount = 0; } } - $ConnectionCount--; - # 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) { - EmptyQueue; + $LondConnecting = 0; # No connections so also not connecting. + EmptyQueue(); + CloseAllLondConnections; # Should all already be closed but... } + UpdateStatus(); } =pod @@ -697,7 +788,7 @@ is readable. The action is state depend =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 @@ -707,9 +798,20 @@ The connection must echo the challenge b =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 @@ -722,9 +824,9 @@ The the key has been requested, now we a =head3 State=Idle The encryption key has been negotiated or we have finished -reading data from the a transaction. If the callback data has -a client as well as the socket iformation, then we are -doing a transaction and the data received is relayed to the client +reading data from the a transaction. If the callback data have +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 @@ -741,7 +843,7 @@ 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 @@ -753,25 +855,33 @@ sub LondReadable { my $Socket = $Watcher->data; my $client = undef; - &Debug(6,"LondReadable called state = ".$State); + &Debug(6,"LondReadable called state = ".$Socket->GetState()); my $State = $Socket->GetState(); # All action depends on the state. SocketDump(6, $Socket); my $status = $Socket->Readable(); + &Debug(2, "Socket->Readable returned: $status"); if($status != 0) { - # bad return from socket read. Currently this means that + # bad return from socket read. Currently this means that # The socket has become disconnected. We fail the transaction. + Log("WARNING", + "Lond connection lost."); if(exists($ActiveTransactions{$Socket})) { - Debug(3,"Lond connection lost failing transaction"); FailTransaction($ActiveTransactions{$Socket}); + } else { + # Socket is connecting and failed... need to mark + # no longer connecting. + + $LondConnecting = 0; } $Watcher->cancel(); KillSocket($Socket); + $ConnectionRetriesLeft--; # Counts as connection failure return; } SocketDump(6,$Socket); @@ -779,19 +889,34 @@ sub LondReadable { $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 "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->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 @@ -803,24 +928,41 @@ sub LondReadable { } elsif ($State eq "ReceivingKey") { } elsif ($State eq "Idle") { + + # 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 { - Log("SUCCESS", "Connection ".$ConnectionCount." to " + my $count = $Socket->GetClientData(); + Log("SUCCESS", "Connection ".$count." to " .$RemoteHost." now ready for action"); } 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); @@ -828,7 +970,7 @@ sub LondReadable { } else { - # Invalid state. + # Invalid state. Debug(4, "Invalid state in LondReadable"); } } @@ -844,9 +986,9 @@ event. The action taken is very state d =head3 State = Connected The connection is in the process of sending the 'init' hailing to the -lond on the remote end. The connection object''s Writable member is -called. On error, ConnectionError is called to destroy the connection -and remove it from the ActiveConnections hash +lond on the remote end. The Writable member of the connection object +is called. On error, call ConnectionError to destroy the connection +and remove it from the ActiveConnections hash. =head3 Initialized @@ -910,20 +1052,40 @@ sub LondWritable { 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") { - 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"); - $Watcher->cancel(); - KillSocket($Socket); - return; - } # "init" is being sent... - - + } elsif ($State eq "Initialized") { # Now that init was sent, we switch @@ -931,40 +1093,43 @@ sub LondWritable { $Watcher->cb(\&LondReadable); $Watcher->poll("r"); - + } 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) { - - $Watcher->cancel(); - KillSocket($Socket); - return; - } - } 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. - - $Watcher->cancel(); - KillSocket($Socket); - return; - - } } elsif ($State eq "ReceivingKey") { # Now we need to wait for the key # to come back from the peer: @@ -973,20 +1138,10 @@ sub LondWritable { $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) { - - if(exists($ActiveTransactions{$Socket})) { - Debug(3, "Lond connection lost, failing transactions"); - FailTransaction($ActiveTransactions{$Socket}); - } - $Watcher->cancel(); - KillSocket($Socket); - return; - - } } elsif ($State eq "ReceivingReply") { # The send has completed. Wait for the @@ -1005,9 +1160,12 @@ sub LondWritable { } } + =pod =cut + + sub QueueDelayed { Debug(3,"QueueDelayed called"); @@ -1015,20 +1173,19 @@ sub QueueDelayed { Debug(4, "Delayed path: ".$path); opendir(DIRHANDLE, $path); - - @alldelayed = grep /\.$RemoteHost$/, readdir DIRHANDLE; - Debug(4, "Got ".$alldelayed." delayed files"); + + my $host_id_re = '(?:'.join('|',map {quotemeta($_)} (@all_host_ids)).')'; + my @alldelayed = grep(/\.$host_id_re$/, readdir(DIRHANDLE)); closedir(DIRHANDLE); - my $dfname; - my $reqfile; - foreach $dfname (sort @alldelayed) { - $reqfile = "$path/$dfname"; - Debug(4, "queueing ".$reqfile); + 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($cmd); + my $Transaction = LondTransaction->new("sethost:$host_id:$cmd"); $Transaction->SetDeferred($reqfile); QueueTransaction($Transaction); } @@ -1040,7 +1197,7 @@ sub QueueDelayed { =head2 MakeLondConnection Create a new lond connection object, and start it towards its initial -idleness. Once idle, it becomes elligible to receive transactions +idleness. Once idle, it becomes eligible to receive transactions from the work queue. If the work queue is not empty when the connection is completed and becomes idle, it will dequeue an entry and start off on it. @@ -1052,40 +1209,46 @@ sub MakeLondConnection { .GetServerPort()); my $Connection = LondConnection->new(&GetServerHost(), - &GetServerPort()); + &GetServerPort(), + &GetHostId()); - if($Connection == undef) { # Needs to be more robust later. + if($Connection eq undef) { Log("CRITICAL","Failed to make a connection with lond."); $ConnectionRetriesLeft--; return 0; # Failure. } else { - $ConnectionRetriesLeft = $ConnectionRetries; # success resets the count + + $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 == undef) { - die "did not get a socket from the connection"; + if($Socket eq undef) { + &child_exit(-1, "did not get a socket from the connection"); } else { &Debug(9,"MakeLondConnection got socket: ".$Socket); } - - $event = Event->io(fd => $Socket, + $Connection->SetTimeoutCallback(\&SocketTimeout); + + 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... + } $ConnectionCount++; + $Connection->SetClientData($ConnectionCount); Debug(4, "Connection count = ".$ConnectionCount); if($ConnectionCount == 1) { # First Connection: QueueDelayed; } - Log("SUCESS", "Created connection ".$ConnectionCount + Log("SUCCESS", "Created connection ".$ConnectionCount ." to host ".GetServerHost()); return 1; # Return success. } @@ -1106,7 +1269,7 @@ reply. =item $Client -Connection to the client that is making this request We got the +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. @@ -1118,8 +1281,8 @@ The text of the request to send. =cut sub StartRequest { - my $Lond = shift; - my $Request = shift; # This is a LondTransaction. + + my ($Lond, $Request) = @_; Debug(6, "StartRequest: ".$Request->getRequest()); @@ -1129,7 +1292,7 @@ sub StartRequest { $ActiveTransactions{$Lond} = $Request; $Lond->InitiateTransaction($Request->getRequest()); - $event = Event->io(fd => $Socket, + my $event = Event->io(fd => $Socket, poll => "w", cb => \&LondWritable, data => $Lond, @@ -1168,11 +1331,23 @@ sub QueueTransaction { my $LondSocket = $IdleConnections->pop(); if(!defined $LondSocket) { # Need to queue request. - Debug(8,"Must queue..."); + Debug(5,"Must queue..."); $WorkQueue->enqueue($requestData); - if($ConnectionCount < $MaxConnectionCount) { - Debug(4,"Starting additional lond connection"); - MakeLondConnection(); + Debug(5, "Queue Transaction startnew $ConnectionCount $LondConnecting"); + if(($ConnectionCount < $MaxConnectionCount) && (! $LondConnecting)) { + + 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..."); @@ -1180,11 +1355,11 @@ sub QueueTransaction { } } -#-------------------------- 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. @@ -1203,7 +1378,7 @@ 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); @@ -1214,14 +1389,26 @@ sub ClientRequest { Debug(8,"Data: ".$data." this read: ".$thisread); $data = $data.$thisread; # Append new data. $watcher->data($data); - if($data =~ /(.*\n)/) { # Request entirely read. - if($data eq "close_connection_exit\n") { + 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); + if ($LogTransactions) { + Log("SUCCESS", "Transaction: '$data'"); # Transaction has \n. + } my $Transaction = LondTransaction->new($data); $Transaction->SetClient($socket); QueueTransaction($Transaction); @@ -1230,6 +1417,40 @@ sub ClientRequest { } +# +# 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) = @_; + + Debug(8, "Entering accept for lonc UNIX socket\n"); + my $connection = $socket->accept(); # Accept the client connection. + Debug(8,"Connection request accepted from " + .GetPeername($connection, AF_UNIX)); + + + my $description = sprintf("Connection to lonc client %d", + $ClientConnection); + Debug(9, "Creating event named: ".$description); + Event->io(cb => \&ClientRequest, + poll => 'r', + desc => $description, + data => "", + fd => $connection); + $ActiveClients{$connection} = $ClientConnection; + $ClientConnection++; +} =pod @@ -1238,7 +1459,7 @@ sub ClientRequest { 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 +an fd member that is the Listener socket. We accept the connection and register a new event on the readability of that socket: =cut @@ -1248,21 +1469,8 @@ sub 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)); - - my $description = sprintf("Connection to lonc client %d", - $ClientConnection); - Debug(9, "Creating event named: ".$description); - Event->io(cb => \&ClientRequest, - poll => 'r', - desc => $description, - data => "", - fd => $connection); - $ActiveClients{$connection} = $ClientConnection; - $ClientConnection++; + &accept_client($socket); } =pod @@ -1272,10 +1480,20 @@ sub NewClient { 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 @@ -1292,6 +1510,18 @@ sub GetServerHost { =pod +=head2 GetServerId + +Returns the hostid whose lond we talk with. + +=cut + +sub GetHostId { + return $RemoteHostId; # Setup by the fork. +} + +=pod + =head2 GetServerPort Returns the lond port number. @@ -1309,25 +1539,52 @@ sub GetServerPort { 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 communcations channel), that int turn will establish +(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); + + my $socket; unless ($socket =IO::Socket::UNIX->new(Local => $SocketName, - Listen => 10, + 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, - poll => 'r', - desc => 'Lonc listener Unix Socket', - fd => $socket); + return $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 @@ -1337,16 +1594,42 @@ sub SetupLoncListener { 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'}; - my $fh = IO::File->new(">>$docdir/lon-status/loncstatus.txt"); - print $fh $$."\t".$RemoteHost."\t".$Status."\t". + + open(LOG,">>$docdir/lon-status/loncstatus.txt"); + flock(LOG,LOCK_EX); + print LOG $$."\t".$RemoteHost."\t".$Status."\t". $RecentLogEntry."\n"; + # + # Write out information about each of the connections: + # + if ($DebugLevel > 2) { + print LOG "Active connection statuses: \n"; + my $i = 1; + print STDERR "================================= Socket Status Dump:\n"; + foreach my $item (keys %ActiveConnections) { + my $Socket = $ActiveConnections{$item}->data; + my $state = $Socket->GetState(); + print LOG "Connection $i State: $state\n"; + print STDERR "---------------------- Connection $i \n"; + $Socket->Dump(-1); # Ensure it gets dumped.. + $i++; + } + } + flock(LOG,LOCK_UN); + close(LOG); + $ConnectionRetriesLeft = $ConnectionRetries; + UpdateStatus(); } =pod @@ -1363,17 +1646,18 @@ sub SignalledToDeath { my $watcher= $event->w; Debug(2,"Signalled to death! via ".$watcher->data); - my ($signal) = @_; + my ($signal) = $watcher->data; chomp($signal); Log("CRITICAL", "Abnormal exit. Child $$ for $RemoteHost " ."died through "."\"$signal\""); - LogPerm("F:lonc: $$ on $RemoteHost signalled to death: " - ."\"$signal\""); - die("Signal abnormal end"); + #LogPerm("F:lonc: $$ on $RemoteHost signalled to death: " +# ."\"$signal\""); exit 0; } +=pod + =head2 ToggleDebug This sub toggles trace debugging on and off. @@ -1389,18 +1673,54 @@ sub ToggleDebug { } +=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"); + + $watcher->cancel(); + delete($parent_dispatchers{$listener}); + + } + + # kill off the parent's signal handlers too! + # + + 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, @@ -1411,51 +1731,258 @@ sub ChildProcess { Event->signal(signal => "USR1", cb => \&ChildStatus, data => "USR1"); + Event->signal(signal => "USR2", + cb => \&ToggleTransactionLogging); Event->signal(signal => "INT", cb => \&ToggleDebug, data => "INT"); - SetupTimer(); - - SetupLoncListener(); + # 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(); // let first work requirest do it. + # &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, $hostid) = @_; + my $sigset = POSIX::SigSet->new(SIGINT); sigprocmask(SIG_BLOCK, $sigset); - my $host = shift; $RemoteHost = $host; + ShowStatus('Parent keeping the flock'); # Update time in status message. Log("CRITICAL", "Forking server for ".$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. + $RemoteHostId = $hostid; ShowStatus("Connected to ".$RemoteHost); - $SIG{INT} = DEFAULT; + $SIG{INT} = 'DEFAULT'; sigprocmask(SIG_UNBLOCK, $sigset); - ChildProcess; # Does not return. + &ChildProcess(); # Does not return. + } +} + +# parent_client_connection: +# Event handler that processes client connections for the parent process. +# This sub is called when the parent is listening on a socket and +# a connection request arrives. We must: +# Start a child process to accept the connection request. +# Kill our listen on the socket. +# Parameter: +# event - The event object that was created to monitor this socket. +# event->w->fd is the socket. +# Returns: +# NONE +# +sub parent_client_connection { + if ($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); + $ChildHost{$hostname}++; + if ($ChildHost{$hostname} == 1) { + &CreateChild($hostname,$lonid); + } 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 $host"); + delete($ChildPid{$pid}); + delete($ChildHost{$host}); + &parent_clean_up($host); + + } else { + &Debug(5, "Caught sigchild for pid not in hosts hash: $pid"); + } } } + # # Parent process logic pass 1: # For each entry in the hosts table, we will @@ -1487,11 +2014,13 @@ if ($childpid != 0) { # ShowStatus("Parent writing pid file:"); -$execdir = $perlvar{'lonDaemons'}; +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; @@ -1501,48 +2030,57 @@ ShowStatus("Forking node servers"); Log("CRITICAL", "--------------- Starting children ---------------"); -my $HostIterator = LondConnection::GetHostIterator; -while (! $HostIterator->end()) { +LondConnection::ReadConfig; # Read standard config files. + +$RemoteHost = "[parent]"; +&listen_on_common_socket(); - $hostentryref = $HostIterator->get(); - CreateChild($hostentryref->[0]); - $HostIterator->next(); -} $RemoteHost = "Parent Server"; # Maintain the population: ShowStatus("Parent keeping the flock"); -# -# Set up parent signals: -# - -$SIG{INT} = \&Terminate; -$SIG{TERM} = \&Terminate; -$SIG{HUP} = \&Restart; -$SIG{USR1} = \&CheckKids; - -while(1) { - $deadchild = wait(); - if(exists $ChildHash{$deadchild}) { # need to restart. - $deadhost = $ChildHash{$deadchild}; - delete($ChildHash{$deadchild}); - Log("WARNING","Lost child pid= ".$deadchild. - "Connected to host ".$deadhost); - Log("INFO", "Restarting child procesing ".$deadhost); - CreateChild($deadhost); - } -} +# We need to setup a SIGChild event to handle the exit (natural or otherwise) +# of the children. +Event->signal(cb => \&server_died, + 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 +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. @@ -1556,29 +2094,75 @@ sub CheckKids { my $now=time; my $local=localtime($now); print $fh "LONC status $local - parent $$ \n\n"; - foreach $pid (keys %ChildHash) { + 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. - sleep 1; # Wait so file doesn't intermix. } + +} + +=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(); } + =pod =head1 Restart Signal handler for HUP... all children are killed and -we self restart. This is an el-cheapo way to re read +we self restart. This is an el-cheapo way to re-read the config file. =cut sub Restart { - KillThemAll; # First kill all the children. + &KillThemAll; # First kill all the children. + LondConnection->ResetReadConfig(); Log("CRITICAL", "Restarting"); my $execdir = $perlvar{'lonDaemons'}; unlink("$execdir/logs/lonc.pid"); - exec("$execdir/lonc"); + exec("$executable"); } =pod @@ -1586,25 +2170,45 @@ sub Restart { =head1 KillThemAll Signal handler that kills all children by sending them a -SIGINT. Responds to sigint and sigterm. +SIGHUP. Responds to sigint and sigterm. =cut sub KillThemAll { Debug(2, "Kill them all!!"); - local($SIG{CHLD}) = 'IGNORE'; # Our children >will< die. - foreach $pid (keys %ChildHash) { - my $serving = $ChildHash{$pid}; - Debug(2, "Killing lonc for $serving pid = $pid"); - ShowStatus("Killing lonc for $serving pid = $pid"); - Log("CRITICAL", "Killing lonc for $serving pid = $pid"); - kill('INT', $pid); - delete($ChildeHash{$pid}); + + #local($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' => $pid; + } + ShowStatus("Finished killing child processes off."); +} + + +# +# Kill all children via KILL. Just in case the +# first shot didn't get them. + +sub really_kill_them_all_dammit +{ + Debug(2, "Kill them all Dammit"); + local($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}); + my $execdir = $perlvar{'lonDaemons'}; + unlink("$execdir/logs/lonc.pid"); } - my $execdir = $perlvar{'lonDaemons'}; - unlink("$execdir/logs/lonc.pid"); - ShowStatus("Killing the master process"); - Log("CRITICAL", "Killing the master process."); } =pod @@ -1616,10 +2220,27 @@ Terminate the system. =cut sub Terminate { - KillThemAll; - exit; + &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; + +} +sub my_hostname { + use Sys::Hostname::FQDN(); + my $name = Sys::Hostname::FQDN::fqdn(); + &Debug(9,"Name is $name"); + return $name; } + =pod =head1 Theory @@ -1661,3 +2282,183 @@ A hash of lond connections that have no 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 + +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: + +Can be enabled for debugging by setting LogTransactions to nonzero. +Indicates a successful transaction with lond, 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 : + +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 INFO Connected to lond 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 + +Only emitted if the global variable $LogTransactions was set to true. +A client has requested a lond transaction is the contents of the request. + +=item SUCCESS Toggled transaction logging + +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 for died thorugh signal. + +QUIT signal received. lonc child process is exiting. + +=item SUCCESS New debugging level for now + +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 = + +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 = + +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 + +=cut