Hi Wietse,

  I see your point. But current MX support will also cause this problem.
For example, if we configure the transport: "example.com  smtp:
mx.example.com:25" which is supported by  all Postfix version. smtp
daemon(smtp/smtp_connect.c) will query all MX records of "mx.example.com"
and try these mx records one by one according to preference.
  My suggestion is that can we put such MX records into transport directly
to avoid DNS querying?
  Suppose "mx.example.com" will return two MX records with IP addresses:
192.168.0.1, 192.168.0.2, can we set them in transport directly like:
example.com smtp:192.168.0.1:25#10, 192.168.0.1:10025#20. (#10, #20 means
the preference)

  Currently we are hosting mail service, so each domain need to have
different setting like above example, the workaround is to setup DNS server
but it's more convenient if we can configure them in transport directly.
  I have also searched the internet and there are also several similar
requirement for such multiple destination support, and all answers are DNS
format. So I think this support will be useful in some scenarios.

  BTW, smtp_fallback_relay also support multiple destination? Will this
feature enabled also impact the queue's manager? Because we also use this
feature as backup solution for mail service.

Regards,
King Cao


2013/7/16 Wietse Venema <wie...@porcupine.org>

> King Cao:
> > adding several codes in "smtp/smtp_connect.c", I think it's more
> convenient
> > for user to set different destinations. That's why I want to add such
> patch
> > to current Postfix source code.
>
> Multiple destinations per nexthop would break queue manager's
> concurrency scheduling policies or it would would break the queue
> manager's independence of nexthop syntax.
>
> I am suspicious about applications that require manual configuration
> of different fallback routes for different destinations. If you
> really need such things then you can use different transports, each
> having their own smtp_fallback_relay, smtp_bind_address and so on.
>
> Got it?
>
>         Wietse
>

Reply via email to