Skip to content
Snippets Groups Projects
  1. Apr 19, 2009
    • Ruud Overeem's avatar
      Bug 1000: · c332c8eb
      Ruud Overeem authored
      Some major changes are made in GCF/TM:
       1) GCFScheduler replaces all global GCFTask functions and variables.
       2) An event-queue is introduced
       3) Return value NEXT_STATE is added
       4) The function autoOpen is added to TCP ports
       5) 'fast' mode available for low level drivers.
      
      Ad. 1
      The GCFTask class had a lot of global functions and variables that were 
      not bound to one task but were multi-task items. All these things are 
      moved to a new class GCFScheduler. This class is a singleton class.
      Consequence of this is that 'init', 'run' and 'stop' are not longer 
      connected to the GCFTask class but are now part of the Scheduler.
      
      old code:
      	GCFTask::init(argc, argv, argv[0]);
      
      	myTask	a("a");
      	myTasl	b("b");
      	a.start();
      	b.start();
      
      	GCFTask::run();
      
      new code:
      	GCFScheduler::instance()->init(argc, argv, argv[0]);
      
      	myTask	a("a");
      	myTasl	b("b");
      	a.start();
      	b.start();
      
      	GCFScheduler::instance()->run();
      
      In the same the stop function should be called with GCFScheduler::instance()->stop();
      
      
      Ad. 2
      The events are now stored in a queue before they are passed to the tasks. 
      This has two benefits:
       - The code does not have to be re-entrant anymore because TRAN drops the 
         F_EXIT and F_ENTRY in the queue in stead of executing them during the TRAN-line.
       - The values for GCFEvent::TResult you return DO MATTER NOW. See Ad 3.
      
      The disadvantage of the queue is that the scheduler has to make a copy of 
      the event before placing it in the queue. To avoid needless copies all events
      that are NO framework events (F_FSM_PROTOCOL and F_PORT_PROTOCOL) are passed 
      immediately to the task. When the task returns HANDLED we save ourselves the 
      effort of a copy. When anything else is returned the event is treated accordingly. See Ad 3.
      
      
      Ad. 3
      Since there is a queue now, the value returned by the task matters:
      HANDLED: event is removed from the queue.
      NEXT_STATE: event is parked on the queue until the task changes state with TRAN(...). 
      After the F_ENTRY event all events that were parked with NEXT_STATE are passed 
      to the task again.
      NOT_HANDLED: this value will become OBSOLETE since it doesn't tell the scheduler 
      what to do. FOR NOW the event will be destroyed.
      
      Remark: In the future the return values may be extended with values like 
      HOLD_x_SEC or HOLD_x_EVENTS.
      
      
      Ad. 4
      To simplify opening or reopening a port the function autoOpen is added to TCPPorts.
      syntax: autoOpen(nrRetries, maxTimeout, reconnectInterval)
      
      This function does several retries to setup the connection. The user can choose 
      the use the nrRetries and reconnectInterval arguments or use the maxTimeout 
      (and reconnectinterval) arguments:
        nrRetries        : -1 = infinite ; How often to retry the open when it fails.
        reconnectInterval: After how many seconds after a fail a reconnect attempt will be made.
        timeout          : -1 = infinite ; How long the auto-open attempts may last.
      
      When both nrRetries and timeout are specified the condition that will be met first will 
      stop the auto open sequence.
      
      When the connection is made an F_CONNECTED events is send like open() does. When the 
      connection can not be made within the given limitations a F_DISCONNECTED is returned.
      
      Note: autoOpen(0,0) acts the same as open()
      
      
      Ad. 5
      Lowlevel drivers like the RSPDriver which heavily lean on FSMs (several 100's)  may 
      not be happy with the eventqueue because statetransitions are handled via the queue. 
      These drivers can call 'disableQueue()' after calling GCFScheduler::instance()->init(..). 
      The queue mechanisme is than completely bypassed and everything works like the 
      old-situation again.
      
      
      Internal notes:
      The boolean that the open() function returned was not really straitforward defined 
      and therefor not used anywhere (as far as I know). The value that is returned now is 
      true when a CONNECT or DISCONNECT message in placed in the queue and false when the 
      ServiceBrokerTask needs so more time to give a result.
       
      c332c8eb
  2. Jul 18, 2008
  3. Mar 04, 2008
  4. Sep 28, 2006
    • blaakmeer's avatar
      BugID: 803 · 2db6fdb9
      blaakmeer authored
      State of the MACScheduler is written to PVSS during destruction. A signal handler is installed
      (late enough to override the GCF signal handlers!) that triggers the transition to the
      finishing state. In that state, the property is written and a timer is installed because
      PVSS database writes are asynchronous. When the timer goes off (1 second) the MACScheduler
      is closed.
      2db6fdb9
  5. May 29, 2006
    • Ruud Overeem's avatar
      BugID: 679 · eff7a615
      Ruud Overeem authored
      Merge MAC-CS1-branch with HEAD. See MAC/Journal for changes.
      eff7a615
  6. Apr 28, 2006
  7. Jul 12, 2005
    • blaakmeer's avatar
      %[ER: 291]% · a1201dfb
      blaakmeer authored
      Added lofar_config.h include and LofarLogger.h include to all .cc files
      a1201dfb
  8. Apr 15, 2005
    • blaakmeer's avatar
      %[ER: 240]% · 99ac9d1e
      blaakmeer authored
      _handleTimers now sends claim, resume and suspend messages to its own state machine in stead of sending them to the server port
      99ac9d1e
  9. Apr 08, 2005
  10. Mar 29, 2005
  11. Mar 10, 2005
  12. Jul 29, 2004
  13. May 19, 2004
  14. Feb 16, 2004
  15. Nov 18, 2003
  16. Nov 14, 2003
  17. Oct 28, 2003
Loading