[omniORB] Help needed to interpret COMM_FAILURE

[Renzo Tomaselli]

Duncan,
    we are using 4.06, and the overall issue is due to the infamous 
WSAENOBUF error I already met in 2002 while sending large buffers.
Since we control the transport, it should be fairly trivial to segment 
down plain TCP segments to reasonable chunks, since OmniORB already 
loops over them.
Thanks,

Renzo

Duncan Grisby wrote:

>On Monday 28 November, Renzo Tomaselli wrote:
>
>  
>
>>    is there any behavior which might affect a sending in a way that
>>connection would be closed on either side while sending a very long
>>message ? I mean > 1 Gigabyte message, which takes quite a number of
>>seconds.
>>Client/server timeouts are left to their defaults,  and the same for
>>in/outConScanPeriod (120/180).
>>Is it possible that on either side a connection is considered idle,
>>and thus closed even while sending the message ? This would explain
>>our troubles, I should just increase in/outConScanPeriod.
>>    
>>
>
>No, strands (i.e. connections) are only considered for closure when they
>are really idle. They definitely won't be closed when a message is still
>being sent, or when a reply is being waited for.
>
>What omniORB version are you using?  There was a bug in versions before
>4.0.4 that meant large messages could be considered too big for the max
>message size even when they weren't, due to the address pointers
>wrapping around. That could cause mysterious errors if you are unlucky
>enough with the way your OS allocates heap addresses.
>
>Cheers,
>
>Duncan.
>
>  
>