[omniORB] Source of COMM_FAILURE_MarshalArguments on narrow

[Duncan Grisby]

On Wed, 2013-07-03 at 14:53 -0400, Jeff Frontz wrote:

> Every so often, a narrow on an object will throw a
> COMM_FAILURE_MarshalArguments (1096024067) exception.  After reviewing
> the exception trace (which I've unfortunately deleted and am trying to
> reproduce), I poked through the omniORB source (4.1.2) and the initial
> obvious source is a timeout -- except all of our timeouts are set to
> "0" (forever).  Looking further, it seems the next likely culprit is
> send(2) experiencing some sort of a (transient?) error.

COMM_FAILURE_MarshalArguments means that the communication failed while
omniORB was sending a request message. As you say, the most likely
reason for that is the send() system call returning an error. Quite
aside from the fact that you haven't set any call timeouts, if you did
get a timeout, you would get a TRANSIENT exception, not a COMM_FAILURE.

>   Since these processes are on the same machine, I can't imagine there
> being any sort of intramachine congestion in the TCP stack.  There
> doesn't seem to be any obvious processor/resource overload (per sar)
> -- that other (different application) clients simultaneously running
> on the same machine continue to execute perfectly would seem to refute
> any obvious resource issue.

I have seen occasional errors from send() even for in-machine calls on
some platforms. What platform are you using?

Cheers,

Duncan.

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