--- loncom/loncnew 2003/07/02 01:31:55 1.13 +++ loncom/loncnew 2003/07/29 02:33:05 1.16 @@ -2,7 +2,7 @@ # The LearningOnline Network with CAPA # lonc maintains the connections to remote computers # -# $Id: loncnew,v 1.13 2003/07/02 01:31:55 foxr Exp $ +# $Id: loncnew,v 1.16 2003/07/29 02:33:05 foxr Exp $ # # Copyright Michigan State University Board of Trustees # @@ -27,7 +27,7 @@ # 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) @@ -46,6 +46,18 @@ # 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). # @@ -93,12 +105,12 @@ 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 @@ -117,6 +129,7 @@ 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. # @@ -132,6 +145,7 @@ 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. @@ -208,6 +222,7 @@ 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; @@ -251,7 +266,7 @@ sub Debug { my $level = shift; my $message = shift; if ($level <= $DebugLevel) { - print $message." host = ".$RemoteHost."\n"; + Log("INFO", "-Debug- $message host = $RemotHost"); } } @@ -281,6 +296,21 @@ 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. @@ -305,7 +335,13 @@ 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. # @@ -443,6 +479,12 @@ 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... @@ -1288,6 +1330,25 @@ 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 $$."\t".$RemoteHost."\t".$Status."\t". + $RecentLogEntry."\n"; +} + =pod =head2 SignalledToDeath @@ -1298,7 +1359,10 @@ Called in response to a signal that caus sub SignalledToDeath { - Debug(2,"Signalled to death!"); + my $event = shift; + my $watcher= $event->w; + + Debug(2,"Signalled to death! via ".$watcher->data); my ($signal) = @_; chomp($signal); Log("CRITICAL", "Abnormal exit. Child $$ for $RemoteHost " @@ -1309,6 +1373,22 @@ sub SignalledToDeath { 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. @@ -1318,14 +1398,22 @@ 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; - $SIG{__DIE__} = \&SignalledToDeath; + # + # Signals must be handled by the Event framework... +# + + Event->signal(signal => "QUIT", + cb => \&SignalledToDeath, + data => "QUIT"); + Event->signal(signal => "HUP", + cb => \&ChildStatus, + data => "HUP"); + Event->signal(signal => "USR1", + cb => \&ChildStatus, + data => "USR1"); + Event->signal(signal => "INT", + cb => \&ToggleDebug, + data => "INT"); SetupTimer(); @@ -1337,7 +1425,7 @@ sub ChildProcess { # Setup the initial server connection: - # &MakeLondConnection(); // let first work requirest do it. + # &MakeLondConnection(); // let first work requirest do it. Debug(9,"Entering event loop"); @@ -1430,9 +1518,10 @@ ShowStatus("Parent keeping the flock"); # Set up parent signals: # -$SIG{INT} = \&KillThemAll; -$SIG{TERM} = \&KillThemAll; +$SIG{INT} = \&Terminate; +$SIG{TERM} = \&Terminate; $SIG{HUP} = \&Restart; +$SIG{USR1} = \&CheckKids; while(1) { $deadchild = wait(); @@ -1446,6 +1535,34 @@ 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 @@ -1482,13 +1599,29 @@ sub KillThemAll { ShowStatus("Killing lonc for $serving pid = $pid"); Log("CRITICAL", "Killing lonc for $serving pid = $pid"); kill('INT', $pid); + delete($ChildeHash{$pid}); } + my $execdir = $perlvar{'lonDaemons'}; + unlink("$execdir/logs/lonc.pid"); + ShowStatus("Killing the master process"); Log("CRITICAL", "Killing the master process."); - exit } =pod +=head1 Terminate + +Terminate the system. + +=cut + +sub Terminate { + KillThemAll; + exit; + +} +=pod + =head1 Theory The event class is used to build this as a single process with an