post...@ptld.com:
> >> It does not complicate the code. I am more concerned about
> >> discoverability (how would a user even find out that the behavior
> >> has become configurable).
> >
> > The best we can do is cross-reference the new parameter under
> > smtp_bind_address (and IPv6 equivalent), and then sufficiently
> > motivated
> > users can find out how to make the setting a mandate rather than a
> > suggestion.
> >
> > To me this looks like a reasonable feature request, despite the fact
> > that nobody else had asked for it in the first 24 years of Postfix.
> > So by no means urgent, but sensible.
>
>
> Just to confirm I'm following along correctly. The current postfix
> behavior is to try to bind to smtp_bind_address and smtp_bind_address6.
> If it can't then postfix will attempt to bind to any address it can
> before failing?
Postfix tries to bind the socket source IP address to the specified
address. Currently, if the bind request fails, Postfix behaves as
if there was no smtp_bind_address (or smtp_bind_address6) setting.
The TCP/IP stack will choose a source IP address when a connection
is made without a successful bind() call.
With the proposed defer-after-bind-failure, the queue manager will
decide when to retry delivery, and when to return the message as
undeliverable.
Wietse
