Hello, Le 22 juin 2013 à 16:13, Eric Faurot <[email protected]> a écrit :
> On Sat, Jun 22, 2013 at 03:25:36AM +0600, Denis Fateyev wrote: > >> At least, I've noticed, that after successful relay through one session >> "smtp-out" doesn't close others for the same message. >> You can look at a piece of log below: here we have a successful relay >> through one session (000000038ae32e21), but another one >> (00000002cf237b1e) for the same message still open until connection >> timeout. I think all the sessions (regardless of their validity or >> state) should be closed forcefully after succeeded message relay >> through one of them? >> >> (…) >> >> You can argue that in this example the second session is invalid anyway >> since it tries to use IPv6, but more important that it hasn't been >> closed after the first one succeeded. Returning to the beginning of the >> question, what would happen if they're both valid? > > It's not 2 connections for the same message, it's 2 connections for > the same relay. The message is eventually handled by one of the > connection. > > What happens here is that there is a mail to be sent to a certain > domain, the MTA resolves the MX list, which turns out to contain an > IPv6 and IPv4 address. It starts a new connection on the first, and, > after a while, on the second. The idea is that if the second is > reachable immediatly, we don't have to wait for the first one to > timeout if we can send the message on the second. So after having > sent the message the second connection closes since there is nothing > left to do. And later, the first connection attempts times out. > By the way it's not far from what is proposed by the RFC 6555: http://www.rfc-editor.org/rfc/rfc6555.txt. There exists a very good explanation of this RFC and the reasons to do that by Stéphane Bortzmeyer (for french speakers): http://www.bortzmeyer.org/6555.html Emmanuel -- You received this email because you are subscribed to mailing list: [email protected] To unsubscribe, send mail with subject: [[email protected]] unregister
