[omniORB] Help needed to interpret COMM_FAILURE

[Renzo Tomaselli]

Hi all,
    this sound interesting. We too had an apparently random problem, 
appearing as a COMM_FAILURE_MarshalArguments instead.
This occurs while the involved client manages to send very large 
messages (e.g. several hundred megabytes) to the server, co-located on 
the same Win host.
We initialize maxMsgSize to be 1 gigabyte, but we have never seen 
MARSHAL_MessageSizeExceedLimitOnClient as a minor, as one would expect 
in case of overflow failures.
But from Duncan's answer, it seems that connection might be dropped 
after overcoming max. message size, thus resulting in a shadowing comm. 
failure. After looking at sources, I saw that this comm. failure can 
only come along the chain notifyCommFailure<errorOnSend<Send, then 
transport stuff.
What is granted for sure, if we reduce message size a bit then this 
problem disappears.
Any comment is highly appreciated.
Thanks,

Renzo Tomaselli


Duncan Grisby wrote:

>On Friday 25 November, "Wernke zur Borg" wrote:
>
>  
>
>>>>giopStream::CommFailure from
>>>>        
>>>>
>>giopStream.cc:834(0,NO,COMM_FAILURE_UnMarshalArguments)
>>
>>I am trying to interpret the source code at that line but I am not 100% sure
>>if the exception is thrown due to a timeout or to a remote disconnect. If
>>this is a timeout condition, I would like to know which config parameter I
>>can use to increase it. It is known that the line between server and client
>>may sometimes become unstable, so to avoid these disonnections I would need
>>to increase the stability with larger timeout values.
>>    
>>
>
>It's not a timeout. If it was, the exception minor code would be
>TRANSIENT_CallTimedOut.
>
>The COMM_FAILURE means the TCP connection was unexpectedly closed.
>Possibly the network broke. Another alternative is that the server tried
>to send something larger than the maximum message size (which is 2MB by
>default), and therefore dropped the connection. You might learn
>something by turning traceLevel 25 on on the server.
>
>Cheers,
>
>Duncan.
>
>  
>