[omniORB] omniORB, posix rwlock_t and deadlock

[Attila Pletyak]

Hello,

   We have encountered a problem what I do not really understand. Maybe 
it lies in omniORB, maybe not. The situation is the following:

  When I use rwlock_t types in my distributed corba system, at some 
point the system locks, dunno why. I have narrowed the problem to the 
following:

   one servant gets a request and makes a w lock on itself, and also a w 
lock on it's so-called base class, which is shared amongst several other 
servers (using the C++ & operator, they get a reference to the base 
servant). While having the w lock, an other request is coming in 
directly to the base class. It waits for a w lock possibility. The first 
w lock unlocks, the second starts its work, but a new request is coming 
in which works like the first. Let's call it the third one. Now, the 
second W lock finishes, the function returns from the base class, the 
lock unlocks. But the third lock request is waiting and waiting and 
waiting... never gets to work again.

   There is a wrapper class around the pthread_rwlock_t types, so they 
get unlocked in every case.

   As I know that omniORB is doing thread operations (creating and 
deleting threads all the time) I started to be curious if omniORB is at 
fault or my code.

   Does anyone has experience with such a thing? Is there an omniORB 
specific thread synchronization tool which works like the rwlock? 
Previously I used the omni_mutex, but it was slow as in the base class 
there is only reading in the majority of the time.

   All this is going with omniORB 3.03, Debian Linux 2.2 and C++

  Thanks in advance,

   Attila Pletyak