[omniORB] COMM_FAILURE_UnMarshalArguments vs TRANSIENT_ConnectFailed

[Duncan Grisby]

On Tuesday 3 June, Eddy Caron wrote:

> 	On a cluster, I try a CORBA connection between two Intel(R) Pentium(R) 
> III CPU family 1400MHz. I launch the omniNames on one computer (ls1 
> IP:193.253.175.223) and I try to run the CORBA application from another 
> computer (ls2 IP:193.253.175.224). The connection between ls1 and ls2 
> seems start but stop with COMM_FAILURE_UnMarshalArguments (for 
> omniNames) and TRANSIENT_ConnectFailed (for application).

The problem is that both machines think that their primary interface
is 193.253.175.185. That means that both are putting that address in
their IORs. In particular, all naming contexts in omniNames claim to
be on host 193.253.175.185. So, you contact the root context OK,
because the corbaloc URI gives the right address, but when the client
attempts to contact the sub-context, it tries to connect to itself,
and fails, hence the TRANSIENT exception.

The COMM_FAILURE is just the omniORB server thread cleaning itself up.

You can avoid this problem with suitable -ORBendPoint configuration on
the two machines, so they publish the correct addresses in their IORs.

Cheers,

Duncan.

-- 
 -- Duncan Grisby         --
  -- duncan at grisby.org     --
   -- http://www.grisby.org --