ACE Tutorial 007
Creating a thread-pool server

client_handler.cpp shows some of the changes due to the thread-pool.   Just a few though.


// page06.html,v 1.11 1999/09/22 03:13:41 jcej Exp

/* Since this is the third time we've seen most of this, I'm going to
   strip out almost all of the comments that you've already seen.
   That way, you can concentrate on the new items.  */

#include "client_acceptor.h"
#include "client_handler.h"

/* We're going to be registering and unregistering a couple of times.
   To make sure that we use the same flags every time, I've created
   these handy macros.  */
#define	REGISTER_MASK ACE_Event_Handler::READ_MASK
#define REMOVE_MASK (ACE_Event_Handler::READ_MASK | ACE_Event_Handler::DONT_CALL)

/* Our constructor still doesn't really do anything.  We simply
   initialize the acceptor pointer to "null" and get our current
   thread id.  The static self() method of ACE_Thread will return you
   a thread id native to your platform.  */
Client_Handler::Client_Handler (void)
  : client_acceptor_(0),
    creator_ (ACE_Thread::self ())
{
}

Client_Handler::~Client_Handler (void)
{
  this->peer().close();
}

/* Query our acceptor for the concurrency strategy.  Notice that we
   don't bother to check that our acceptor pointer is valid.  That is
   proably a bad idea...  */
int
Client_Handler::concurrency(void)
{
  return this->client_acceptor ()->concurrency ();
}

/* And here we ask the acceptor about the thread pool.  */
Thread_Pool *
Client_Handler::thread_pool (void)
{
  return this->client_acceptor ()->thread_pool ();
}

/* Back to our open() method.  This is straight out of Tutorial 6.
   There's nothing additional here for the thread-pool implementation.  */
int
Client_Handler::open (void *acceptor)
{
  client_acceptor ((Client_Acceptor *) acceptor);

  if (concurrency () == Client_Acceptor::thread_per_connection_)
    return this->activate (THR_DETACHED);

  this->reactor (client_acceptor()->reactor ());

  ACE_INET_Addr addr;

  if (this->peer ().get_remote_addr (addr) == -1)
    return -1;

  if (this->reactor ()->register_handler (this,
                                          REGISTER_MASK) == -1)
    ACE_ERROR_RETURN ((LM_ERROR,
                       "(%P|%t) can't register with reactor\n"),
                      -1);

  ACE_DEBUG ((LM_DEBUG,
              "(%P|%t) connected with %s\n",
              addr.get_host_name ()));
  return 0;
}

/* The destroy() method will remove us from the reactor (with the
   DONT_CALL flag set!) and then free our memory.  This allows us to
   be closed from outside of the reactor context without any danger.  */
void
Client_Handler::destroy (void)
{
  this->reactor ()->remove_handler (this, REMOVE_MASK);
  delete this;
}

/* As mentioned in the header, the typical way to close an object in a
  threaded context is to invoke it's close() method.  */
int
Client_Handler::close (u_long flags)
{
  /*
    We use the destroy() method to clean up after ourselves.
    That will take care of removing us from the reactor and then
    freeing our memory.
    */
  this->destroy ();

  /* Don't forward the close() to the baseclass!  handle_close() above
    has already taken care of delete'ing.  Forwarding close() would
    cause that to happen again and things would get really ugly at
    that point!  */
  return 0;
}

/* We will be called when handle_input() returns -1.  That's our queue
  to delete ourselves to prevent memory leaks.  */
int
Client_Handler::handle_close (ACE_HANDLE handle,
                              ACE_Reactor_Mask mask)
{
  ACE_UNUSED_ARG (handle);
  ACE_UNUSED_ARG (mask);

  delete this;

  return 0;
}

/* In the open() method, we registered with the reactor and requested
   to be notified when there is data to be read.  When the reactor
   sees that activity it will invoke this handle_input() method on us.
   As I mentioned, the _handle parameter isn't useful to us but it
   narrows the list of methods the reactor has to worry about and the
   list of possible virtual functions we would have to override.

   You've read that much before...  Now we have to do some extra stuff
   in case we're using the thread-pool implementation.  If we're
   called by our creator thread then we must be in the reactor.  In
   that case, we arrange to be put into the thread pool.  If we're not
   in the creator thread then we must be in the thread pool and we can
   do some work.  */
int
Client_Handler::handle_input (ACE_HANDLE handle)
{
  ACE_UNUSED_ARG (handle);

  /* Check our strategy.  If we're using the thread pool and we're in
    the creation thread then we know we were called by the reactor.  */
  if (concurrency () == Client_Acceptor::thread_pool_)
    {
      if (ACE_OS::thr_equal (ACE_Thread::self(),
                             creator_))
	{
          /* Remove ourselves from the reactor and ask to be put into
            the thread pool's queue of work.  (You should be able to
            use suspend_handler() but I've had problems with that.)

            By removing ourselves from the reactor, we're guaranteed
            that we won't be called back until the thread pool picks
            us up out of the queue.  If we didn't remove ourselves,
            then the reactor would continue to invoke handle_input()
            and we don't want that to happen.  */
          this->reactor ()->remove_handler (this, REMOVE_MASK);
          return this->thread_pool ()->enqueue (this);
	}
    }

  /* Any strategy other than thread-per-connection will eventually get
    here.  If we're in the single-threaded implementation or the
    thread-pool, we still have to pass this way.  */

  char buf[BUFSIZ];

  /* Invoke the process() method to do the work but save it's return
    value instead of returning it immediately.  */

  int rval = this->process (buf, sizeof (buf));

  /* Now, we look again to see if we're in the thread-pool
    implementation.  If so then we need to re-register ourselves with
    the reactor so that we can get more work when it is available.
    (If suspend_handler() worked then we would use resume_handler()
    here.)  */
  if (concurrency () == Client_Acceptor::thread_pool_)
    {
      if (rval != -1)
        /* If we don't remember to re-register ourselves, then we
            won't be able to respond to any future client requests.  */
        this->reactor ()->register_handler (this,
                                            REGISTER_MASK);
    }

  /* Return the result of process() */
  return rval;
}

/* Remember that when we leave our svc() method, the framework will
   take care of calling our close() method so that we can cleanup
   after ourselves.  */
int
Client_Handler::svc (void)
{
  char buf[BUFSIZ];

  while (1)
    if (this->process (buf, sizeof (buf)) == -1)
      return -1;

  return 0;
}

/* Once again, we see that the application-level logic has not been at
   all affected by our choice of threading models.  Of course, I'm not
   sharing data between threads or anything.  We'll leave locking
   issues for a later tutorial.  */
int
Client_Handler::process (char *rdbuf,
                         int rdbuf_len)
{
  switch (this->peer ().recv (rdbuf, rdbuf_len))
    {
    case -1:
      ACE_ERROR_RETURN ((LM_ERROR,
                         "(%P|%t) %p bad read\n",
                         "client"),
                        -1);
    case 0:
      ACE_ERROR_RETURN ((LM_ERROR,
                         "(%P|%t) closing daemon (fd = %d)\n",
                         this->get_handle ()),
                        -1);
    default:
      ACE_DEBUG ((LM_DEBUG,
                  "(%P|%t) from client: %s",
                  rdbuf));
    }

  return 0;
}

Ok, now we've gone and changed handle_input() so that it knows when to do work and when to enqueue itself.  Beyond that, we're still about the same.


[Tutorial Index] [Continue This Tutorial]