Re: re-connecting problems ...

[Jules Colding <colding 42tools com>]

Hi Michael,


On Mon, 2008-01-28 at 11:58 +0000, Michael Meeks wrote:
> Hi guys,
> 
> 	I just got:
> 
> 		http://bugzilla.gnome.org/show_bug.cgi?id=512520
> 
> 	and I wonder if this is one of the issues that the timeout stuff was
> added to get around ;-)

The timeout stuff was specifically added to get around the problem of
being blocked forever if a remote server is powered down or physically
disconnected. It only applies for AF_INET and AF_INET6 (but not for
oneways).


> 	It makes me wonder - why do we bother trying to re-connect so
> vigorously ? [ I add a few queries at the end of the bug ].
> 
> 	I think it should be the case that we distinguish between a connection
> that we received CLOSECONNECTION on before the connection died, and one
> where the cnx just died unexpectedly.
> 
> 	The latter we should not try to re-connect (or at least not so
> hard ;-), the former we should.
> 
> 	What do people think ?

It might complicate matters that a CLOSECONNECTION is sent when we get a
timeout - giop-recv-buffer.c:giop_timeout() - i.e. no clear distinction
between a timeout and other types of connection failures except if we
look at the link state.

But yes, some kind of distinction between the various ways a connection
could die would properly be a good idea, when decide whether we should
try to re-connect, although I've no clear idea of what precisely to do
implementation-wise.

Best regards,
  jules