On 07/02/2015 11:56 PM, King Cao wrote:
Hi Wietse,
Actually it's our relay mta and can not know if it's deliverable or not
until bounced by downsteam...
King:
I ran into this problem when I used Postfix to front a large number of
Plesk (qmail) and CPanel (exim) systems at a Web hosting company. My
inbound MTA cluster was the focal point for inbound mail -- I did this
to centralize all the anti-spam measures. A series of small Unix shell
scripts would forward the valid mail addresses to the master MX boxes,
so that they would know what inbound mail addresses were valid and which
were not.
Incoming mail was then distributed by the inbound MX cluster to the many
Web boxes, spam already removed unless the customer didn't want spam
removal on one or more mailboxes, so those boxes couldn't be DoSed by
spam mail. Our business was Web hosting, after all.
The inbound MX clusters would reject unwanted messages during the SMTP
transaction, so that NDR bounce messages would be generated by the
upstream MTA if appropriate. No bounce traffic was created by the
inbound cluster itself.
If the inbound cluster did create new mail (don't know why, but I'm a
belt-and-suspenders man) any such mail was smart-hosted to my outbound
MX servers. The outbound MX servers had per-endpoint rate-limiting to
reduce the chance of being blocked due to mail volume. (Bulk mailers
each had their own rate-limiting outgoing MX.)
At my network edge, I blocked inbound port-25 mail to all but my inbound
MX cluster. Customers could submit mail from "outside" to their Web
server mail agent via port 587, as I recall, or by the sendmail(1)
utility from within their Web sites.
With this system, the only time the inside MTAs would generate NDR
messages was when mail quotas were exceeded. Those NDRs were
smart-hosted to my outbound MX servers. The volume of NDR traffic from
each of the Web host MTAs was low enough that I didn't bother trying to
deal with it.