[omniORB] Fallback from SSL to TCP on CA failure?

[Duncan Grisby]

On Wed, 2009-10-07 at 17:58 +0200, Peter Klotz wrote:

[...]
> These lines look strange:
> 
> Switching to TCP:
> 
>    omniORB: (0) 2009-10-07 14:56:55.889763: Switch rope to use address 
> giop:tcp:10.18.2.48:11260
> 
> Immediately afterwards failing using SSL:
> 
>    omniORB: (0) 2009-10-07 14:56:55.889861: Unable to open new 
> connection: giop:ssl:10.18.2.48:11261

That's normal. Due to the way the exceptions are handled, the log
message to say that the "rope" (which represents communication with a
particular set of endpoints) is switched to a new address happens before
the log message to say that opening a new connection failed.

Now I look at your trace, I realise that I wasn't correct in saying that
the fallback to a second address is automatic. When opening a connection
fails, the rope is switched to the next address in the available
addresses, but the connection attempt is not automatically retried. If
you try the call again, it will try the second address. That's why you
see that when you try two calls to a particular server, the second one
succeeds after the first has failed.

It's possible to register a TRANSIENT exception handler that
automatically retries in these situations. That's why I told you the
wrong thing before -- the system I spend most of my time using does have
such a handler, and I forgot that the standard one doesn't retry.

Cheers,

Duncan.

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