--- loncom/loncnew 2003/06/11 02:04:35 1.8 +++ loncom/loncnew 2003/10/27 10:09:21 1.30 @@ -2,13 +2,12 @@ # The LearningOnline Network with CAPA # lonc maintains the connections to remote computers # -# $Id: loncnew,v 1.8 2003/06/11 02:04:35 foxr Exp $ +# $Id: loncnew,v 1.30 2003/10/27 10:09:21 foxr 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 requestors spread out bver m connections to londs. +# new lonc handles n request out bver m connections to londs. # This module is based on the Event class. # Development iterations: # - Setup basic event loop. (done) @@ -41,10 +40,101 @@ # - Add Configuration file I/O (done). # - Add management/status request interface. # - Add deferred request capability. (done) +# - Detect transmission timeouts. # # Change log: # $Log: loncnew,v $ +# Revision 1.30 2003/10/27 10:09:21 foxr +# Tighten up a few compares to eq and flip a few debug levels around... nothing +# critical +# +# Revision 1.29 2003/10/21 14:24:42 foxr +# Fix little typo that may explain growth of connections +# +# Revision 1.28 2003/10/14 15:36:21 albertel +# - making it easier to run loncnew, +# /etc/init.d/loncontrol startnew +# /etc/init.d/loncontrol restartnew +# will now start loncnew in place of lonc +# +# Revision 1.27 2003/10/07 11:23:03 foxr +# Installed and tested code to process reinit in parent server. +# +# Revision 1.26 2003/09/30 11:11:17 foxr +# Add book-keeping hashes to support the re-init procedure. +# +# Revision 1.25 2003/09/23 11:22:14 foxr +# Tested ability to receive sigusr2 This is now logged and must be +# properly implemented as a re-read of hosts and re-init of appropriate +# children. +# +# Revision 1.24 2003/09/16 09:46:42 foxr +# Added skeletal infrastructure to support SIGUSR2 update hosts request. +# +# Revision 1.23 2003/09/15 09:24:49 foxr +# Add use strict and fix all the fallout from that. +# +# Revision 1.22 2003/09/02 10:34:47 foxr +# - Fix errors in host dead detection logic (too many cases where the +# retries left were not getting incremented or just not checked). +# - Added some additional status to the ps axuww display: +# o Remaining retries on a host. +# o >>> DEAD <<< indicator if I've given up on a host. +# - Tested the SIGHUP will reset the retries remaining count (thanks to +# the above status stuff, and get allow the loncnew to re-try again +# on the host (thanks to the log). +# +# Revision 1.21 2003/08/26 09:19:51 foxr +# How embarrassing... put in the SocketTimeout function in loncnew and forgot +# to actually hook it into the LondTransaction. Added this to MakeLondConnection +# where it belongs... hopefully transactions (not just connection attempts) will +# timeout more speedily than the socket errors will catch it. +# +# Revision 1.20 2003/08/25 18:48:11 albertel +# - fixing a forgotten ; +# +# Revision 1.19 2003/08/19 09:31:46 foxr +# Get socket directory from configuration rather than the old hard coded test +# way that I forgot to un-hard code. +# +# Revision 1.18 2003/08/06 09:52:29 foxr +# Also needed to remember to fail in-flight transactions if their sends fail. +# +# Revision 1.17 2003/08/03 00:44:31 foxr +# 1. Correct handling of connection failure: Assume it means the host is +# unreachable and fail all of the queued transactions. Note that the +# inflight transactions should fail on their own time due either to timeout +# or send/receive failures. +# 2. Correct handling of logs for forced death signals. Pull the signal +# from the event watcher. +# +# 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 the 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 @@ -54,14 +144,16 @@ # complete coding to support deferred transactions. # # - +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; @@ -72,17 +164,10 @@ use LONCAPA::LondTransaction; use LONCAPA::Configuration; use LONCAPA::HashIterator; -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 @@ -95,35 +180,113 @@ my %perlvar = %{$perlvarref}; # parent and shared variables. my %ChildHash; # by pid -> host. +my %HostToPid; # By host -> pid. +my %HostHash; # by loncapaname -> IP. -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 = 2; +my $DebugLevel = 0; +my $NextDebugLevel= 2; # So Sigint can toggle this. my $IdleTimeout= 3600; # Wait an hour before pruning connections. # # 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 $UnixSocketDir= $perlvar{'lonSockDir'}; my $IdleConnections = Stack->new(); # Set of idle connections my %ActiveConnections; # Connections to the remote lond. 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=2; # Number of connection retries allowed. +my $ConnectionRetriesLeft=2; # Number of connection retries remaining. # -# This disconnected socket makes posible a bit more regular -# code when processing delayed requests: -# -my $NullSocket = IO::Socket->new(); +# The hash below gives the HTML format for log messages +# given a severity. +# +my %LogFormats; + +$LogFormats{"CRITICAL"} = "CRITICAL: %s"; +$LogFormats{"SUCCESS"} = "SUCCESS: %s"; +$LogFormats{"INFO"} = "INFO: %s"; +$LogFormats{"WARNING"} = "WARNING: %s"; +$LogFormats{"DEFAULT"} = " %s "; + + + +=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"); + print $fh "$now:$message:$local\n"; +} + +=pod + +=head2 Log + +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 = shift; + my $message = shift; + + if(!$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] "; + my $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 @@ -141,7 +304,7 @@ sub GetPeername { 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; @@ -162,7 +325,7 @@ sub Debug { my $level = shift; my $message = shift; if ($level <= $DebugLevel) { - print $message." host = ".$RemoteHost."\n"; + Log("INFO", "-Debug- $message host = $RemoteHost"); } } @@ -179,11 +342,32 @@ sub SocketDump { =head2 ShowStatus Place some text as our pid status. + and as what we return in a SIGUSR1 =cut sub ShowStatus { - my $status = shift; - $0 = "lonc: ".$status; + my $state = shift; + my $now = time; + my $local = localtime($now); + $Status = $local.": ".$state; + $0='lonc: '.$state.' '.$local; +} + +=pod + +=head 2 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); # A transaction timeout also counts as + # a connection failure: + $ConnectionRetriesLeft--; } =pod @@ -197,11 +381,11 @@ Invoked each timer tick. sub Tick { my $client; - ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount); - Debug(6, "Tick"); - Debug(6, " Current connection count: ".$ConnectionCount); - foreach $client (keys %ActiveClients) { - Debug(7, " Have client: with id: ".$ActiveClients{$client}); + if($ConnectionRetriesLeft > 0) { + ShowStatus(GetServerHost()." Connection count: ".$ConnectionCount + ." Retries remaining: ".$ConnectionRetriesLeft); + } else { + ShowStatus(GetServerHost()." >> DEAD <<"); } # Is it time to prune connection count: @@ -210,24 +394,41 @@ 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); } } else { $IdleSeconds = 0; # Reset idle count if not idle. } - + # + # For each inflight transaction, tick down its timeout counter. + # + foreach my $item (keys %ActiveTransactions) { + my $Socket = $ActiveTransactions{$item}->getServer(); + $Socket->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. # my $Requests = $WorkQueue->Count(); - if (($ConnectionCount == 0) && ($Requests > 0)) { - my $Connections = ($Requests <= $MaxConnectionCount) ? - $Requests : $MaxConnectionCount; - Debug(1,"Work but no connections, starting ".$Connections." of them"); - for ($i =0; $i < $Connections; $i++) { - MakeLondConnection(); + if (($ConnectionCount == 0) && ($Requests > 0)) { + if ($ConnectionRetriesLeft > 0) { + my $Connections = ($Requests <= $MaxConnectionCount) ? + $Requests : $MaxConnectionCount; + Debug(5,"Work but no connections, start ".$Connections." of them"); + my $successCount = 0; + for (my $i =0; $i < $Connections; $i++) { + $successCount += MakeLondConnection(); + } + if($successCount == 0) { # All connections failed: + Debug(5,"Work in queue failed to make any connectiouns\n"); + EmptyQueue(); # Fail pending transactions with con_lost. + } + } else { + ShowStatus(GetServerHost()." >>> DEAD!!! <<<"); + Debug(5,"Work in queue, but gave up on connections..flushing\n"); + EmptyQueue(); # Connections can't be established. } } @@ -273,19 +474,19 @@ sub ServerToIdle { my $Socket = shift; # Get the socket. 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); } } @@ -331,7 +532,7 @@ 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"); @@ -349,6 +550,12 @@ 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!"); + $Watcher->cancel; + $Socket->shutdown(2); + $Socket->close(); + } } } else { # Error of some sort... @@ -396,7 +603,7 @@ The transaction that is being completed. =cut sub CompleteTransaction { - &Debug(6,"Complete transaction"); + &Debug(5,"Complete transaction"); my $Socket = shift; my $Transaction = shift; @@ -404,9 +611,8 @@ sub CompleteTransaction { my $data = $Socket->GetReply(); # Data to send. StartClientReply($Transaction, $data); } else { # Delete deferred transaction file. - &Debug(4, "Deferred transaction complete: ".$Transaction->getFile(). - " request: ".$Transaction->getRequest(). - " answer: ".$Socket->GetReply()); + Log("SUCCESS", "A delayed transaction was completed"); + LogPerm("S:$Transaction->getClient() :".$Transaction->getRequest()); unlink $Transaction->getFile(); } } @@ -430,12 +636,12 @@ sub StartClientReply { my $Transaction = shift; my $data = shift; + 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", @@ -466,13 +672,16 @@ Parameters: sub FailTransaction { my $transaction = shift; - my $Lond = $transaction->getServer(); - if (!$client->isDeferred()) { # If the transaction is deferred we'll get to it. - my $client = $transcation->getClient(); - StartClientReply($client, "con_lost"); + Log("WARNING", "Failing transaction ".$transaction->getRequest()); + Debug(1, "Failing transaction: ".$transaction->getRequest()); + 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"); + } + if($ConnectionRetriesLeft <= 0) { + Log("CRITICAL", "Host marked dead: ".GetServerHost()); } -# not needed, done elsewhere if active. -# delete $ActiveTransactions{$Lond}; } @@ -484,14 +693,29 @@ sub FailTransaction { =cut sub EmptyQueue { + $ConnectionRetriesLeft--; # Counts as connection failure too. while($WorkQueue->Count()) { - my $request = $Workqueue->dequeue(); # This is a transaction + 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) { + KillSocket($Socket); + } +} +=cut + +=pod + =head2 KillSocket Destroys a socket. This function can be called either when a socket @@ -515,6 +739,9 @@ nonzero if we are allowed to create a ne sub KillSocket { my $Socket = 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!!! @@ -533,7 +760,7 @@ sub KillSocket { # work queue, the work all gets failed with con_lost. # if($ConnectionCount == 0) { - EmptyQueue; + EmptyQueue(); } } @@ -602,23 +829,28 @@ 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(); - if($Socket->Readable() != 0) { + &Debug(2, "Socket->Readable returned: $status"); + + if($status != 0) { # 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}); } $Watcher->cancel(); KillSocket($Socket); + $ConnectionRetriesLeft--; # Counts as connection failure return; } SocketDump(6,$Socket); @@ -652,11 +884,18 @@ sub LondReadable { } elsif ($State eq "Idle") { # 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 " + .$RemoteHost." now ready for action"); } ServerToIdle($Socket); # Next work unit or idle. @@ -759,7 +998,9 @@ sub LondWritable { 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"); + FailTransaction($ActiveTransactions{$Socket}); $Watcher->cancel(); KillSocket($Socket); return; @@ -859,8 +1100,7 @@ sub QueueDelayed { Debug(4, "Delayed path: ".$path); opendir(DIRHANDLE, $path); - @alldelayed = grep /\.$RemoteHost$/, readdir DIRHANDLE; - Debug(4, "Got ".$alldelayed." delayed files"); + my @alldelayed = grep /\.$RemoteHost$/, readdir DIRHANDLE; closedir(DIRHANDLE); my $dfname; my $reqfile; @@ -870,7 +1110,7 @@ sub QueueDelayed { my $Handle = IO::File->new($reqfile); my $cmd = <$Handle>; chomp $cmd; # There may or may not be a newline... - $cmd = $cmd."\ny"; # now for sure there's exactly one newline. + $cmd = $cmd."\n"; # now for sure there's exactly one newline. my $Transaction = LondTransaction->new($cmd); $Transaction->SetDeferred($reqfile); QueueTransaction($Transaction); @@ -897,23 +1137,27 @@ sub MakeLondConnection { my $Connection = LondConnection->new(&GetServerHost(), &GetServerPort()); - if($Connection == undef) { # Needs to be more robust later. - Debug(0,"Failed to make a connection with lond."); + if($Connection eq undef) { # Needs to be more robust later. + Log("CRITICAL","Failed to make a connection with lond."); + $ConnectionRetriesLeft--; + return 0; # Failure. } else { + # 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) { + if($Socket eq undef) { die "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, @@ -925,6 +1169,9 @@ sub MakeLondConnection { if($ConnectionCount == 1) { # First Connection: QueueDelayed; } + Log("SUCESS", "Created connection ".$ConnectionCount + ." to host ".GetServerHost()); + return 1; # Return success. } } @@ -966,7 +1213,7 @@ 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, @@ -1005,11 +1252,18 @@ 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(); + if($ConnectionRetriesLeft > 0) { + Debug(5,"Starting additional lond connection"); + if(MakeLondConnection() == 0) { + EmptyQueue(); # Fail transactions, can't make connection. + } + } else { + ShowStatus(GetServerHost()." >>> DEAD !!!! <<<"); + EmptyQueue(); # It's worse than that ... he's dead Jim. + } } } else { # Can start the request: Debug(8,"Can start..."); @@ -1022,7 +1276,6 @@ sub QueueTransaction { =pod =head2 ClientRequest - Callback that is called when data can be read from the UNIX domain socket connecting us with an apache server process. @@ -1041,17 +1294,24 @@ 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 eq "close_connection_exit\n") { + Log("CRITICAL", + "Request Close Connection ... exiting"); + CloseAllLondConnections(); + exit; + } Debug(8, "Complete transaction received: ".$data); my $Transaction = LondTransaction->new($data); $Transaction->SetClient($socket); @@ -1161,8 +1421,67 @@ sub SetupLoncListener { fd => $socket); } +=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'}; + my $fh = IO::File->new(">>$docdir/lon-status/loncstatus.txt"); + print $fh $$."\t".$RemoteHost."\t".$Status."\t". + $RecentLogEntry."\n"; + $ConnectionRetriesLeft = $ConnectionRetries; +} + =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\""); + exit 0; + +} + +=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"); + +} + =head2 ChildProcess This sub implements a child process for a single lonc daemon. @@ -1171,16 +1490,23 @@ This sub implements a child process for sub ChildProcess { - print "Loncnew\n"; - # For now turn off signals. - - $SIG{QUIT} = IGNORE; - $SIG{HUP} = IGNORE; - $SIG{USR1} = IGNORE; - $SIG{INT} = IGNORE; - $SIG{CHLD} = IGNORE; - $SIG{__DIE__} = IGNORE; + # + # Signals must be handled by the Event framework... +# + + Event->signal(signal => "QUIT", + cb => \&SignalledToDeath, + data => "QUIT"); + Event->signal(signal => "HUP", + cb => \&ChildStatus, + data => "HUP"); + Event->signal(signal => "USR1", + cb => \&ChildStatus, + data => "USR1"); + Event->signal(signal => "INT", + cb => \&ToggleDebug, + data => "INT"); SetupTimer(); @@ -1192,12 +1518,9 @@ sub ChildProcess { # Setup the initial server connection: - &MakeLondConnection(); + # &MakeLondConnection(); // let first work requirest do it. + - if($ConnectionCount == 0) { - Debug(1,"Could not make initial connection..\n"); - Debug(1,"Will retry when there's work to do\n"); - } Debug(9,"Entering event loop"); my $ret = Event::loop(); # Start the main event loop. @@ -1208,16 +1531,23 @@ sub ChildProcess { # Create a new child for host passed in: sub CreateChild { + my $sigset = POSIX::SigSet->new(SIGINT); + sigprocmask(SIG_BLOCK, $sigset); my $host = shift; $RemoteHost = $host; - Debug(3, "Forking off child for ".$RemoteHost); - sleep(5); - $pid = fork; + Log("CRITICAL", "Forking server for ".$host); + my $pid = fork; if($pid) { # Parent - $ChildHash{$pid} = $RemoteHost; + $RemoteHost = "Parent"; + $ChildHash{$pid} = $host; + $HostToPid{$host}= $pid; + sigprocmask(SIG_UNBLOCK, $sigset); + } else { # child. ShowStatus("Connected to ".$RemoteHost); - ChildProcess; + $SIG{INT} = 'DEFAULT'; + sigprocmask(SIG_UNBLOCK, $sigset); + ChildProcess; # Does not return. } } @@ -1252,11 +1582,13 @@ if ($childpid != 0) { # ShowStatus("Parent writing pid file:"); -$execdir = $perlvar{'lonDaemons'}; +my $execdir = $perlvar{'lonDaemons'}; open (PIDSAVE, ">$execdir/logs/lonc.pid"); print PIDSAVE "$$\n"; close(PIDSAVE); + + if (POSIX::setsid() < 0) { print "Could not create new session\n"; exit -1; @@ -1264,29 +1596,240 @@ if (POSIX::setsid() < 0) { ShowStatus("Forking node servers"); +Log("CRITICAL", "--------------- Starting children ---------------"); + my $HostIterator = LondConnection::GetHostIterator; while (! $HostIterator->end()) { - $hostentryref = $HostIterator->get(); + my $hostentryref = $HostIterator->get(); CreateChild($hostentryref->[0]); + $HostHash{$hostentryref->[0]} = $hostentryref->[4]; $HostIterator->next(); } +$RemoteHost = "Parent Server"; # Maintain the population: ShowStatus("Parent keeping the flock"); +# +# Set up parent signals: +# + +$SIG{INT} = \&Terminate; +$SIG{TERM} = \&Terminate; +$SIG{HUP} = \&Restart; +$SIG{USR1} = \&CheckKids; +$SIG{USR2} = \&UpdateKids; # LonManage update request. + while(1) { - $deadchild = wait(); + my $deadchild = wait(); if(exists $ChildHash{$deadchild}) { # need to restart. - $deadhost = $ChildHash{$deadchild}; + my $deadhost = $ChildHash{$deadchild}; + delete($HostToPid{$deadhost}); delete($ChildHash{$deadchild}); - Debug(4,"Lost child pid= ".$deadchild. + Log("WARNING","Lost child pid= ".$deadchild. "Connected to host ".$deadhost); + Log("INFO", "Restarting child procesing ".$deadhost); CreateChild($deadhost); } } + + +=pod + +=head1 CheckKids + + Since kids do not die as easily in this implementation +as the previous one, there is no need to restart the +dead ones (all dead kids get restarted when they die!!) +The only thing this function does is to pass USR1 to the +kids so that they report their status. + +=cut + +sub CheckKids { + Debug(2, "Checking status of children"); + my $docdir = $perlvar{'lonDocRoot'}; + my $fh = IO::File->new(">$docdir/lon-status/loncstatus.txt"); + my $now=time; + my $local=localtime($now); + print $fh "LONC status $local - parent $$ \n\n"; + foreach my $pid (keys %ChildHash) { + 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"); + + # Just in case we need to kill our own lonc, we wait a few seconds to + # give it a chance to receive and relay lond's response to the + # re-init command. + # + + sleep(2); # Wait a couple of seconds. + + my %hosts; # Indexed by loncapa hostname, value=ip. + + # Need to re-read the host table: + + + LondConnection::ReadConfig(); + my $I = LondConnection::GetHostIterator; + while (! $I->end()) { + my $item = $I->get(); + $hosts{$item->[0]} = $item->[4]; + $I->next(); + } + + # The logic below is written for clarity not for efficiency. + # Since I anticipate that this function is only rarely called, that's + # appropriate. There are certainly ways to combine the loops below, + # and anyone wishing to obscure the logic is welcome to go for it. + # Note that we don't re-direct sigchild. Instead we do what's needed + # to the data structures that keep track of children to ensure that + # when sigchild is honored, no new child is born. + # + + # For each existing child; if it's host doesn't exist, kill the child. + + foreach my $child (keys %ChildHash) { + my $oldhost = $ChildHash{$child}; + if (!(exists $hosts{$oldhost})) { + Log("CRITICAL", "Killing child for $oldhost host no longer exists"); + delete $ChildHash{$child}; + delete $HostToPid{$oldhost}; + kill 'QUIT' => $child; + } + } + # For each remaining existing child; if it's host's ip has changed, + # Restart the child on the new IP. + + foreach my $child (keys %ChildHash) { + my $oldhost = $ChildHash{$child}; + my $oldip = $HostHash{$oldhost}; + if ($hosts{$oldhost} ne $oldip) { + + # kill the old child. + + Log("CRITICAL", "Killing child for $oldhost host ip has changed..."); + delete $ChildHash{$child}; + delete $HostToPid{$oldhost}; + kill 'QUIT' => $child; + + # Do the book-keeping needed to start a new child on the + # new ip. + + $HostHash{$oldhost} = $hosts{$oldhost}; + CreateChild($oldhost); + } + } + # Finally, for each new host, not in the host hash, create a + # enter the host and create a new child. + # Force a status display of any existing process. + + foreach my $host (keys %hosts) { + if(!(exists $HostHash{$host})) { + Log("INFO", "New host $host discovered in hosts.tab..."); + $HostHash{$host} = $hosts{$host}; + CreateChild($host); + } else { + kill 'HUP' => $HostToPid{$host}; # status display. + } + } +} + + +=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. + Log("CRITICAL", "Restarting"); + my $execdir = $perlvar{'lonDaemons'}; + unlink("$execdir/logs/lonc.pid"); + exec("$execdir/loncnew"); +} + +=pod + +=head1 KillThemAll + +Signal handler that kills all children by sending them a +SIGHUP. Responds to sigint and sigterm. + +=cut + +sub KillThemAll { + Debug(2, "Kill them all!!"); + local($SIG{CHLD}) = 'IGNORE'; # Our children >will< die. + foreach my $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 'QUIT' => $pid; + delete($ChildHash{$pid}); + } + my $execdir = $perlvar{'lonDaemons'}; + unlink("$execdir/logs/lonc.pid"); + +} + +=pod + +=head1 Terminate + +Terminate the system. + +=cut + +sub Terminate { + KillThemAll; + Log("CRITICAL","Master process exiting"); + exit 0; + +} +=pod + =head1 Theory The event class is used to build this as a single process with an