Re: threading woes

[Bowie Owens <bowie owens csiro au>]

Bowie Owens wrote:

> Hi,
>
> I received a request for some help on the orbitcpp-list. It relates to 
> using threading as part of the implementation of a servant (but not 
> using POA threading policies). The IDL looks kind of like this:
>
> module test {
>        interface Worker {
>                void ping ();
>        };
>
>        interface Master {
>                void create_thread ();
>                void submit_work (in Worker server);
>        };
> };
>
> The Master servant in creates a new thread using glib in the 
> create_thread() method. It the receives a reference to a Worker 
> servant (in a different process) from the submit_work() method. The 
> master then calls the ping() method on the Worker object. If the 
> ping() is called in the same thread as the main loop, everythings 
> works fine. However, if the ping() is called from the thread created 
> in create_thread() the ORB drops out of its main loop and things fall 
> apart. I've attached a log of ORBIT2_DEBUG=all. For the adventurous 
> this is checked in to CVS under orbitcpp/test/cpp/threading. I was 
> wondering if anyone had any insight into why the ORB is dropping out 
> of its run loop or how I can find out why.

I've just worked it out myself. The program is using the linc_loop and 
the g_main_loop_quit (link_loop) is happening within 
link_exec_set_io_thread() which is getting called indirectly from 
giop_thread_new_check() (see backtrace below). I needed to ensure that 
link_set_io_thread(TRUE) is called before the run loop is entered. Nick, 
this can be achieved by using the passing orb identifier 
"orbit-io-thread" to CORBA::ORB_init(). However, I think this is bug 
since it seems like the intention is that we can switch transparently 
from non-threaded io to threaded io. I've attached a patch which causes 
giop_main_run() to enter the giop_main_loop after we quit the link_loop 
if giop_thread_io() has become true.

#0  link_exec_set_io_thread (data=0xbf7ff8c4, immediate=1) at linc.c:379
#1  0x4007b2a7 in link_dispatch_command (data=0xbf7ff8c4, immediate=1)
   at linc.c:447
#2  0x4007aac6 in link_exec_command (cmd=0xbf7ff8c4) at linc.c:115
#3  0x4007b22f in link_set_io_thread (io_in_thread=1) at linc.c:432
#4  0x4005493f in giop_thread_new_check (opt_self=0x80646c8) at giop.c:826
#5  0x4005c018 in ORBit_small_invoke_stub (obj=0x80671c0, m_data=0x8056800,
   ret=0x0, args=0x0, ctx=0x0, ev=0xbf7ffa3c) at orbit-small.c:629
#6  0x4005be7c in ORBit_small_invoke_stub_n (object=0x80671c0,
   methods=0x8056864, index=0, ret=0x0, args=0x0, ctx=0x0, ev=0xbf7ffa3c)
   at orbit-small.c:575
#7  0x400745e9 in ORBit_c_stub_invoke (obj=0x80671c0, methods=0x8056864,
   method_index=0, ret=0x0, args=0x0, ctx=0x0, ev=0xbf7ffa3c, class_id=1,
   method_offset=4, skel_impl=0x80531a4 
<_ORBIT_skel_small_test_Worker_ping>)
   at poa.c:2640
#8  0x08054642 in test_Worker_ping ()
#9  0x080532f6 in test::Worker::ping() ()
#10 0x08050c7f in server_thread::consumer() ()
#11 0x08050afd in server_thread::thread_consumer(void*) ()
#12 0x401931f4 in g_thread_create_proxy ()
  from /local_scratch/owe043/opt/gnome2/lib/libglib-2.0.so.0
#13 0x400ffc0e in pthread_start_thread_event () from /lib/libpthread.so.0

--
Bowie Owens

CSIRO Mathematical & Information Sciences
phone  : +61 3 9545 8055
fax    : +61 3 9545 8080
mobile : 0425 729 875
email  : Bowie Owens csiro au


Index: ./src/orb/GIOP/giop.c
===================================================================
RCS file: /cvs/gnome/ORBit2/src/orb/GIOP/giop.c,v
retrieving revision 1.50
diff -u -p -r1.50 giop.c
--- ./src/orb/GIOP/giop.c	4 Mar 2004 17:49:37 -0000	1.50
+++ ./src/orb/GIOP/giop.c	21 Dec 2004 01:06:54 -0000
@@ -597,6 +597,9 @@ static GMainLoop *giop_main_loop = NULL;
 void
 giop_main_run (void)
 {
+	if (!giop_thread_io ())
+		link_main_loop_run ();
+
 	if (giop_thread_io ()) {
 		g_assert (giop_thread_self () == giop_main_thread);
 		g_assert (giop_main_loop == NULL);
@@ -604,8 +607,7 @@ giop_main_run (void)
 		g_main_loop_run (giop_main_loop);
 		g_main_loop_unref (giop_main_loop);
 		giop_main_loop = NULL;
-	} else
-		link_main_loop_run ();
+	} 
 }
 
 void