Re: [asio-users] asio + URDL: understanding tcp connect errors

[Christopher K. <ch...@ko...>]

On Wed, 22 Jul 2009 10:40 +0200, "Jose" <jm...@gm...> wrote:
> On Wed, Jul 22, 2009 at 9:58 AM, Christopher Kohlhoff<ch...@ko...>
> wrote:
> 
> > TCP should already retry a connection after 10s (from memory). However,
> > I'd consider adding something like a min_connect_tries property that
> > would put a floor on the number of iterations in async_connect() for
> > sites with a small number of addresses.
> 
> I think I didn't communicate properly. I meant having a way to retry
> temporary errors beyond what TCP does.
> This is a feature request for URDL rather than asio.

Yes, I know. I was referring to async_connect in
include/urdl/detail/connect.hpp.

What I meant was adding an additional test to the for-loop based on a
min_connect_tries property. It will keep trying if there have been too
few attempts.

> Wouldn't it be better to have:
> 
> - a connection timeout  (which can indicate a connection to be retried)

In time I will consider adding a connection timeout.

> - an open timeout (after the connection is open - this could be
> retried also, it would occur in overloaded servers)

I think the current "open" timeout is better than a timeout "after the
connection is open". For the client program's control flow, it is
simpler to timeout after the overall request is initiated, i.e. a
deadline.

> > Perhaps, but I think a status callback indicating what stage it is up to
> > would be more useful for that. Then the client software would be in a
> > better position to know whether what had happened so far may have had
> > any side effects, and whether it can safely re-issue the request.
> 
> YES. Can you elaborate ? Will this be available ?

All in due course :)

Cheers,
Chris