diff options
Diffstat (limited to 'docs/tutorials/006/page05.html')
-rw-r--r-- | docs/tutorials/006/page05.html | 348 |
1 files changed, 0 insertions, 348 deletions
diff --git a/docs/tutorials/006/page05.html b/docs/tutorials/006/page05.html deleted file mode 100644 index af803d9a54a..00000000000 --- a/docs/tutorials/006/page05.html +++ /dev/null @@ -1,348 +0,0 @@ -<!-- $Id$ --> -<HTML> -<HEAD> - <META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=iso-8859-1"> - <META NAME="GENERATOR" CONTENT="Mozilla/4.04 [en] (X11; I; Linux 2.0.32 i486) [Netscape]"> - <META NAME="Author" CONTENT="James CE Johnson"> - <META NAME="Description" CONTENT="A first step towards using ACE productively"> - <TITLE>ACE Tutorial 006</TITLE> -</HEAD> -<BODY TEXT="#000000" BGCOLOR="#FFFFFF" LINK="#000FFF" VLINK="#FF0F0F"> - -<CENTER><B><FONT SIZE=+2>ACE Tutorial 006</FONT></B></CENTER> - -<CENTER><B><FONT SIZE=+2>Creating a thread-per-connection server</FONT></B></CENTER> - - -<P> -<HR WIDTH="100%"> -<P><A HREF="client_handler.cpp">client_handler.cpp</A> exposes all the -things I've been hinting at. Pay special attention to the decision -made in open() as well as the bit of cleverness in svc(). - -<P> -<HR WIDTH="100%"> -<PRE> -<font color=red>// $Id$</font> - -<font color=red>/* In client_handler.h I alluded to the fact that we'll mess around - with a Client_Acceptor pointer. To do so, we need the - Client_Acceptor object declaration. - - We know that including client_handler.h is redundant because - client_acceptor.h includes it. Still, the sentry prevents - double-inclusion from causing problems and it's sometimes good to - be explicit about what we're using. - - On the other hand, we don't directly include any ACE header files - here. */</font> -<font color=blue>#include</font> "<font color=green>client_acceptor.h</font>" -<font color=blue>#include</font> "<font color=green>client_handler.h</font>" - -<font color=red>/* Our constructor doesn't do anything. That's generally a good idea. - Unless you want to start throwing exceptions, there isn't a really - good way to indicate that a constructor has failed. If I had my - way, I'd have a boolean return code from it that would cause new to - return 0 if I failed. Oh well... */</font> -<font color=#008888>Client_Handler::Client_Handler</font> (void) -{ -} - -<font color=red>/* Our destructor doesn't do anything either. That is also by design. - Remember, we really want folks to use destroy() to get rid of us. - If that's so, then there's nothing left to do when the destructor - gets invoked. */</font> -<font color=#008888>Client_Handler::~Client_Handler</font> (void) -{ -} - -<font color=red>/* The much talked about destroy() method! The reason I keep going on - about this is because it's just a Bad Idea (TM) to do real work - inside of a destructor. Although this method is void, it really - should return int so that it can tell the caller there was a - problem. Even as void you could at least throw an exception which - you would never want to do in a destructor. */</font> -void -<font color=#008888>Client_Handler::destroy</font> (void) -{ - <font color=red>/* Tell the reactor to forget all about us. Notice that we use the - same args here that we use in the open() method to register - ourselves. In addition, we use the DONT_CALL flag to prevent - handle_close() being called. Since we likely got here due to - handle_close(), that could cause a bit of nasty recursion! */</font> - this->reactor ()->remove_handler (this, - <font color=#008888>ACE_Event_Handler::READ_MASK</font> - | <font color=#008888>ACE_Event_Handler::DONT_CALL</font>); - - <font color=red>/* This is how we're able to tell folks not to use delete. By - deleting our own instance, we take care of memory leaks after - ensuring that the object is shut down correctly. */</font> - delete this; -} - -<font color=red>/* As mentioned before, the open() method is called by the - Client_Acceptor when a new client connection has been accepted. - The Client_Acceptor instance pointer is cast to a void* and given - to us here. We'll use that to avoid some global data... */</font> -int -<font color=#008888>Client_Handler::open</font> (void *void_acceptor) -{ - <font color=red>/* We need this to store the address of the client that we are now - connected to. We'll use it later to display a debug message. */</font> - ACE_INET_Addr addr; - - <font color=red>/* Our ACE_Svc_Handler baseclass gives us the peer() method as a way - to access our underlying ACE_SOCK_Stream. On that object, we can - invoke the get_remote_addr() method to get get an ACE_INET_Addr - having our client's address information. As with most ACE methods, - we'll get back (and return) a -1 if there was any kind of error. - Once we have the ACE_INET_Addr, we can query it to find out the - client's host name, TCP/IP address, TCP/IP port value and so - forth. One word of warning: the get_host_name() method of - ACE_INET_Addr may return you an empty string if your name server - can't resolve it. On the other hand, get_host_addr() will always - give you the dotted-decimal string representing the TCP/IP - address. */</font> - if (this->peer ().get_remote_addr (addr) == -1) - return -1; - - <font color=red>/* Convert the void* to a Client_Acceptor*. You should probably use - those fancy ACE_*_cast macros but I can never remember how/when to - do so. Since you can cast just about anything around a void* - without compiler warnings be very sure of what you're doing when - you do this kind of thing. That's where the new-style cast - operators can save you. */</font> - Client_Acceptor *acceptor = (Client_Acceptor *) void_acceptor; - - <font color=red>/* Our Client_Acceptor is constructed with a concurrency strategy. - Here, we go back to it to find out what that strategy was. If - thread-per-connection was selected then we simply activate a - thread for ourselves and exit. Our svc() method will then begin - executing in that thread. - - If we are told to use the single-threaded strategy, there is no - difference between this and the Tutorial 5 implementation. - - Note that if we're in thread-per-connection mode, open() is exited - at this point. Furthermore, thread-per-connection mode does not - use the reactor which means that handle_input() and it's fellows - are not invoked. */</font> - if (acceptor->thread_per_connection ()) - return this->activate (THR_DETACHED); - - <font color=red>// ************************************************************************</font> - <font color=red>// From here on, we're doing the traditional reactor thing. If</font> - <font color=red>// you're operating in thread-per-connection mode, this code does</font> - <font color=red>// not apply.</font> - <font color=red>// ************************************************************************</font> - - <font color=red>/* Our reactor reference will be set when we register ourselves but - I decided to go ahead and set it here. No good reason really... */</font> - this->reactor (acceptor->reactor ()); - - <font color=red>/* If we managed to get the client's address then we're connected to - a real and valid client. I suppose that in some cases, the client - may connect and disconnect so quickly that it is invalid by the - time we get here. In any case, the test above should always be - done to ensure that the connection is worth keeping. - - Now, regiser ourselves with a reactor and tell that reactor that - we want to be notified when there is something to read. Remember, - we took our reactor value from the acceptor which created us in - the first place. Since we're exploring a single-threaded - implementation, this is the correct thing to do. */</font> - if (this->reactor ()->register_handler (this, - <font color=#008888>ACE_Event_Handler::READ_MASK</font>) == -1) - ACE_ERROR_RETURN ((LM_ERROR, - "<font color=green>(%P|%t) can't register with reactor\n</font>"), - -1); - - <font color=red>/* Here, we use the ACE_INET_Addr object to print a message with the - name of the client we're connected to. Again, it is possible that - you'll get an empty string for the host name if your DNS isn't - configured correctly or if there is some other reason that a - TCP/IP addreess cannot be converted into a host name. */</font> - ACE_DEBUG ((LM_DEBUG, - "<font color=green>(%P|%t) connected with %s\n</font>", addr.get_host_name ())); - - <font color=red>/* Always return zero on success. */</font> - return 0; -} - -<font color=red>/* As mentioned in the header, the typical way to close an object in a - threaded context is to invoke it's close() method. Since we - already have a handle_close() method built to cleanup after us, - we'll just forward the request on to that object. */</font> -int -<font color=#008888>Client_Handler::close</font>(u_long flags) -{ - ACE_UNUSED_ARG (flags); - - <font color=red>/* 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. */</font> - this->destroy (); - - <font color=red>/* 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! */</font> - return 0; -} - -<font color=red>/* 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. - - Again, this is not used if we're in thread-per-connection mode. */</font> -int -<font color=#008888>Client_Handler::handle_input</font> (ACE_HANDLE handle) -{ - <font color=red>/* Some compilers don't like it when you fail to use a parameter. - This macro will keep 'em quiet for you. */</font> - ACE_UNUSED_ARG (handle); - - <font color=red>/* Now, we create and initialize a buffer for receiving the data. - Since this is just a simple test app, we'll use a small buffer - size. */</font> - char buf[BUFSIZ]; - - <font color=red>/* Invoke the process() method with a pointer to our data area. - We'll let that method worry about interfacing with the data. You - might choose to go ahead and read the data and then pass the - result to process(). However, application logic may require that - you read a few bytes to determine what else to read... It's best - if we push that all into the application-logic level. */</font> - return this->process (buf, sizeof (buf)); -} - -<font color=red>/* If we return -1 out of handle_input() or if the reactor sees other - problems with us then handle_close() will be called. The reactor - framework will take care of removing us (due to the -1), so we - don't need to use the destroy() method. Instead, we just delete - ourselves directly. */</font> -int -<font color=#008888>Client_Handler::handle_close</font> (ACE_HANDLE handle, - ACE_Reactor_Mask mask) -{ - ACE_UNUSED_ARG (handle); - ACE_UNUSED_ARG (mask); - - this->destroy (); - return 0; -} - -<font color=red>/* The ACE_Svc_Handler<> is ultimately derived from ACE_Task<>. If - you want to create a multi-threaded application, these are your - tools! Simply override the svc() method in your derivative and - arrange for your activate() method to be called. The svc() method - then executes in the new thread. - - Of course, this is only valid if we're in thread-per-connection - mode. If we're using the reactor model, then svc() never comes - into play. */</font> -int -<font color=#008888>Client_Handler::svc</font>(void) -{ - <font color=red>/* Like handle_input(), we create a buffer for loading the data. - Doing so in handle_input() doesn't help any but there is a small - performance increase by doing this here: the buffer is created - once when the thread is created instead of for each invocation of - process(). */</font> - char buf[BUFSIZ]; - - <font color=red>// Forever...</font> - while( 1 ) - { - <font color=red>/* Invoke the process() method to read and process the data. - This is exactly the way it is used by handle_input(). That's - the reason I created process() in the first place: so that it - can be used in either concurrency strategy. Since process() - has all of our application-level logic, it's nice that it - doesn't have to change when we decide to go multi-threaded. - - Notice that since the recv() method call in process() blocks until - there is data ready, this thread doesn't consume any CPU time until - there is actually data sent from the client. */</font> - if (this->process(buf, sizeof (buf)) == -1) - return -1; - } - - return 0; -} - -<font color=red>/* And, at last, we get to the application-logic level. Out of - everything we've done so far, this is the only thing that really - has anything to do with what your application will do. In this - method we will read and process the client's data. In a real - appliation, you will probably have a bit more in main() to deal - with command line options but after that point, all of the action - takes place here. */</font> -int -<font color=#008888>Client_Handler::process</font> (char *rdbuf, - int rdbuf_len) -{ - <font color=red>/* Using the buffer provided for us, we read the data from the - client. If there is a read error (eg -- recv() returns -1) then - it's a pretty good bet that the connection is gone. Likewise, if - we read zero bytes then something wrong has happened. The reactor - wouldn't have called us if there wasn't some kind of read activity - but there wouldn't be activity if there were no bytes to read... - - On the other hand, if we got some data then we can display it in a - debug message for everyone to see. */</font> - switch (this->peer ().recv (rdbuf, rdbuf_len)) - { - case -1: - ACE_ERROR_RETURN ((LM_ERROR, - "<font color=green>(%P|%t) %p bad read\n</font>", - "<font color=green>client</font>"), - -1); - case 0: - ACE_ERROR_RETURN ((LM_ERROR, - "<font color=green>(%P|%t) closing daemon (fd = %d)\n</font>", - this->get_handle ()), - -1); - default: - ACE_DEBUG ((LM_DEBUG, - "<font color=green>(%P|%t) from client: %s</font>", - rdbuf)); - } - - return 0; -} -</PRE> -<HR WIDTH="100%"> -<P> - Did you notice the <i>THR_DETACHED</i> flag on the call to - <i>activate()</i>? Threads, like any system resource, are a - limited resource. Unless we intend to <i>join()</i> or - <i>wait()</i> for the new thread later, we want use THR_DETACHED - so that we don't cause a leak. In fact, in most cases, you'll - want to specify THR_DETACHED because it's just easier. -<p> - Another handy flag for use with <i>activate()</i> is - <i>THR_NEW_LWP</i>. That's short for <i>Light Weight - Process</i>. If you've got a multiprocessor, this flag will - allocate a new schedulable process and decrease the odds of your - threads all fighting for the same process. Of course, if you - have a uni-processor, it will neither help nor hurt. Since I - developed these on a uni-processor, I've been a bit inconsistent - in the use of <i>THR_NEW_LWP</i>. - -<P>Well, that's it! After all the talk & the hype, you would -have expected it to be more difficult to create a multi-threaded server. -Surprise! It really is that easy. You still have to handle -contention issues which we haven't addressed here and that is a rather -nasty topic. Still, for the simple case, this is all you have to -do. - -<P>The next page is the last for this tutorial. Head on over there -& we'll round up the file list one last time. - -<P> -<P><HR WIDTH="100%"> -<CENTER>[<A HREF="../online-tutorials.html">Tutorial Index</A>] [<A HREF="page06.html">Continue This Tutorial</A>]</CENTER> |