omniORB refcount miscount

General Installation
1

Hi,

At DESY we perform tests before switching from Tango 8 to 9.
Running some automatize tests of python servers on (Py)Tango9 I get a strange error:


omniORB: Assertion failed.  This indicates a bug in the application
using omniORB, or maybe in omniORB itself.
 file: ../../../../../src/lib/omniORB/orbcore/SocketCollection.cc
 line: 1663
 info: pd_refcount > 0
omniORB: AsyncInvoker: Warning: unexpected exception caught while executing a task.

It looks harmless, i.e. it did not crashed any of my tests, but this miscount of reference counter in the removeSocket() method


void
SocketCollection::removeSocket(SocketHolder* s)
{
  OMNIORB_ASSERT(s->pd_belong_to == this);

  int refcount = 0; // Initialise to stop over-enthusiastic compiler warnings
  {
    omni_tracedmutex_lock sync(pd_collection_lock);

    OMNIORB_ASSERT(pd_refcount > 0);
    refcount = --pd_refcount;

    *(s->pd_prev) = s->pd_next;

    if (s->pd_next)
      s->pd_next->pd_prev = s->pd_prev;

    s->pd_belong_to = 0;
  }
  if (refcount == 0) delete this;
}

looks really suspicious. This error comes randomly after all my tests.

I’m testing purely python servers on debian jessie with python-tango (9.2.0), tango (9.2.5a), libomniorb4-1 (4.1.6-2).
I’ve never seen this error even with the same tests for tango 8.

Therefore, I would like to ask if anyone has idea what is going on or what could be a problem?

Bests,
Jan

2

Hi Jan,

What I would suggest is to update your version of omniORB to version 4.2.1 and to apply the patch available on our patches web page: http://www.tango-controls.org/downloads/patches/

Kind regards
Reynald

3

Thanks Reynald.

We will try it.

4

Hi,

as an update I would like to mention that one has to be careful in creating thread local DeviceProxy instances with omniORB 4.1.

In this case we get

omniORB: Application check failed. This indicates a bug in the application
  using omniORB.  See the comment in the source code for more info.
  file: ../../../../../src/lib/omniORB/orbcore/corbaOrb.cc
  line: 989
  info: pd_destroyed
omniORB: Application check failed. This indicates a bug in the application
  using omniORB.  See the comment in the source code for more info.
  file: ../../../../../src/lib/omniORB/orbcore/corbaOrb.cc
  line: 990
  info: pd_refCount == 0

The gdb points a problem in creating of DeviceProxy.
Moving the DeviceProxy constructor to the main thread the problem disappears.

We are using debian8 with tango9. Since jessie/stretch provides only omniORB 4.1 we haven’t managed to check if the problem disappears for 4.2.1, yet.

Bests,
Jan

5

Just for info,

https://tracker.debian.org/pkg/omniorb-dfsg

the 4.2.2 version is available in the experimental repository of Debian.
So if you want you can backport it for stretch and recompile also tango, pytango etc… in order to check that it works with this version of omniorb.

Cheers

Frederic