[omniORB] connection problems

Empting Arnd Arnd.Empting@kst.siemens.de
Thu, 19 Jul 2001 10:22:31 +0200


Hi,

we run in connectionproblems using the windows version of omniorb3.

we connect to a visibroker server who closes idle tcp-connections after some
time.
(configurable)

in our scenario we make 4 invocations toward the server in sequence.
If we make the calls in short intervalls all works fine.
If we wait the serverside idletime before we try to send data to the server
we get an 

omniORB: throw omniConnectionBroken (minor 10053) from
tcpSocketMTfactory.cc:1140

exception for the first call to the server. 
the next 3 calls work perfect.

Unfortunalty we remove the destination interface from the list 
if the destination couldnt be reached.
Therefore we make only 3 calls in the next try.

In the next try we get the same pattern - the first call gets an 
exception the next 2 calls went through 
and so we have only 2 destinations left. 
if we call them before the serverside idletime is over the calls work
alright. 

first we thought the problem would be solved with omniorb 3.0.3
because of this bugfix but the behaviour is still there and we get the
exception in the send and not in the recv case.

	Summary: Incorrect connection closure on Windows (bug number 16)  
	Date:  Mon Feb 5 12:12:14 GMT 2001 
	Fixed by:  dpg1 
	Reported by: Victor Joukov 
	Description: In extremely rare cases, a recv() could be interrupted
on Windows, 
	causing a connection closure, rather than the proper retry.  

Any ideas, why there is no connection retry on the first error and 
why there is no Transient-Exception thrown as the manual suggests? (p.70)

Regards
	Arnd