--- loncom/loncnew 2003/07/02 01:12:35 1.12 +++ loncom/loncnew 2003/08/26 09:19:51 1.21 @@ -2,13 +2,12 @@ # The LearningOnline Network with CAPA # lonc maintains the connections to remote computers # -# $Id: loncnew,v 1.12 2003/07/02 01:12:35 foxr Exp $ +# $Id: loncnew,v 1.21 2003/08/26 09:19:51 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) @@ -46,10 +45,44 @@ # Change log: # $Log: loncnew,v $ -# Revision 1.12 2003/07/02 01:12:35 foxr -# - Add some debugging to killthemall -# - Add better error handling to LondReadable -# - Remove tick logging in the timer handler. +# 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. @@ -95,12 +128,12 @@ 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; +#$SIG{QUIT} = IGNORE; +#$SIG{HUP} = IGNORE; +#$SIG{USR1} = IGNORE; +#$SIG{INT} = IGNORE; +#$SIG{CHLD} = IGNORE; +#$SIG{__DIE__} = IGNORE; # Read the httpd configuration file to get perl variables @@ -119,13 +152,14 @@ my $MaxConnectionCount = 10; # Will get 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. # # 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. @@ -134,6 +168,7 @@ my $WorkQueue = Queue->new(); # Qu my $ConnectionCount = 0; my $IdleSeconds = 0; # Number of seconds idle. my $Status = ""; # Current status string. +my $RecentLogEntry = ""; my $ConnectionRetries=5; # Number of connection retries allowed. my $ConnectionRetriesLeft=5; # Number of connection retries remaining. @@ -210,6 +245,7 @@ sub Log { my $execdir = $perlvar{'lonDaemons'}; my $fh = IO::File->new(">>$execdir/logs/lonc.log"); my $msg = sprintf($finalformat, $message); + $RecentLogEntry = $msg; print $fh $msg; @@ -253,7 +289,7 @@ sub Debug { my $level = shift; my $message = shift; if ($level <= $DebugLevel) { - print $message." host = ".$RemoteHost."\n"; + Log("INFO", "-Debug- $message host = $RemotHost"); } } @@ -283,6 +319,21 @@ sub ShowStatus { =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); +} + +=pod + =head2 Tick Invoked each timer tick. @@ -307,7 +358,13 @@ sub Tick { } else { $IdleSeconds = 0; # Reset idle count if not idle. } - + # + # For each inflight transaction, tick down its timeout counter. + # + foreach $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. # @@ -445,6 +502,12 @@ 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... @@ -562,6 +625,7 @@ Parameters: sub FailTransaction { my $transaction = shift; + Log("WARNING", "Failing transaction ".$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(); @@ -624,6 +688,7 @@ 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, @@ -720,14 +785,16 @@ sub LondReadable { 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 # 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}); } $Watcher->cancel(); @@ -877,6 +944,7 @@ sub LondWritable { # We'll treat this as if the socket got disconnected: Log("WARNING", "Connection to ".$RemoteHost. " has been disconnected"); + FailTransaction($ActiveTransactions{$Socket}); $Watcher->cancel(); KillSocket($Socket); return; @@ -1032,7 +1100,8 @@ sub MakeLondConnection { &Debug(9,"MakeLondConnection got socket: ".$Socket); } - + $Connection->SetTimeoutCallback(\&SocketTimeout); + $event = Event->io(fd => $Socket, poll => 'w', cb => \&LondWritable, @@ -1132,7 +1201,9 @@ sub QueueTransaction { $WorkQueue->enqueue($requestData); if($ConnectionCount < $MaxConnectionCount) { Debug(4,"Starting additional lond connection"); - MakeLondConnection(); + if(MakeLondConnection() == 0) { + EmptyQueue(); # Fail transactions, can't make connection. + } } } else { # Can start the request: Debug(8,"Can start..."); @@ -1290,6 +1361,25 @@ sub SetupLoncListener { fd => $socket); } +=pod + +=head2 ChildStatus + +Child USR1 signal handler to report the most recent status +into the status file. + +=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 $$."\t".$RemoteHost."\t".$Status."\t". + $RecentLogEntry."\n"; +} + =pod =head2 SignalledToDeath @@ -1300,17 +1390,35 @@ Called in response to a signal that caus sub SignalledToDeath { - Debug(2,"Signalled to death!"); - my ($signal) = @_; + my $event = shift; + my $watcher= $event->w; + + Debug(2,"Signalled to death! via ".$watcher->data); + my ($signal) = $watcher->data; chomp($signal); Log("CRITICAL", "Abnormal exit. Child $$ for $RemoteHost " ."died through "."\"$signal\""); LogPerm("F:lonc: $$ on $RemoteHost signalled to death: " ."\"$signal\""); - die("Signal abnormal end"); exit 0; } + +=head2 ToggleDebug + +This sub toggles trace debugging on and off. + +=cut + +sub ToggleDebug { + my $Current = $DebugLevel; + $DebugLevel = $NextDebugLevel; + $NextDebugLevel = $Current; + + Log("SUCCESS", "New debugging level for $RemoteHost now $DebugLevel"); + +} + =head2 ChildProcess This sub implements a child process for a single lonc daemon. @@ -1320,14 +1428,22 @@ This sub implements a child process for sub ChildProcess { - # For now turn off signals. - - $SIG{QUIT} = \&SignalledToDeath; - $SIG{HUP} = IGNORE; - $SIG{USR1} = IGNORE; - $SIG{INT} = DEFAULT; - $SIG{CHLD} = IGNORE; - $SIG{__DIE__} = \&SignalledToDeath; + # + # Signals must be handled by the Event framework... +# + + Event->signal(signal => "QUIT", + cb => \&SignalledToDeath, + data => "QUIT"); + Event->signal(signal => "HUP", + cb => \&ChildStatus, + data => "HUP"); + Event->signal(signal => "USR1", + cb => \&ChildStatus, + data => "USR1"); + Event->signal(signal => "INT", + cb => \&ToggleDebug, + data => "INT"); SetupTimer(); @@ -1339,7 +1455,7 @@ sub ChildProcess { # Setup the initial server connection: - # &MakeLondConnection(); // let first work requirest do it. + # &MakeLondConnection(); // let first work requirest do it. Debug(9,"Entering event loop"); @@ -1359,6 +1475,7 @@ sub CreateChild { Log("CRITICAL", "Forking server for ".$host); $pid = fork; if($pid) { # Parent + $RemoteHost = "Parent"; $ChildHash{$pid} = $RemoteHost; sigprocmask(SIG_UNBLOCK, $sigset); @@ -1406,6 +1523,8 @@ open (PIDSAVE, ">$execdir/logs/lonc.pid" print PIDSAVE "$$\n"; close(PIDSAVE); + + if (POSIX::setsid() < 0) { print "Could not create new session\n"; exit -1; @@ -1432,9 +1551,10 @@ ShowStatus("Parent keeping the flock"); # Set up parent signals: # -$SIG{INT} = \&KillThemAll; -$SIG{TERM} = \&KillThemAll; - +$SIG{INT} = \&Terminate; +$SIG{TERM} = \&Terminate; +$SIG{HUP} = \&Restart; +$SIG{USR1} = \&CheckKids; while(1) { $deadchild = wait(); @@ -1448,12 +1568,58 @@ while(1) { } } + + +=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 $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 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/lonc"); +} + =pod =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 @@ -1465,14 +1631,30 @@ sub KillThemAll { 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); + kill 'QUIT' => $pid; + delete($ChildHash{$pid}); } - Log("CRITICAL", "Killing the master process."); - exit + 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