On Wed, Mar 05, 2014 at 01:30:28PM -0800, Ned Freed wrote:
On Wed, Mar 05, 2014 at 08:22:19PM +0000, Paul Smith wrote:
On 05/03/2014 18:20, Murray S. Kucherawy wrote:
How come this never got adoption? It comes up from time to time
in "We really should do this" sorts of discussions, but it doesn't
seem like anyone ever took the plunge and it just expired. Is it
just that nobody does it because nobody else does it?
Hmm, this one's a tricky one. With our mail server we've been quite
careful to do things which would either accept the "whole message",
or reject it all. I can see places where we could change
functionality to use PRDR, but the problem is that the user would
expect it to always work, not be dependent on the sending MTA. So,
we would have to 'fake' PRDR, by falling back to accepting the
message and generating bounce messages, which isn't nice, and will
lead to backscatter.
Correct me if I'm wrong, I may have missed something in the draft:
PRDR implies that deliveries be attempted during the SMTP transaction so
this cannot really be implemented by MTA that use a two-step approach to
first commit to queue and acknowledge responsibility, then later deliver
the message to the mailbox ?
No, that's incorrect. From the draft:
Furthermore, positive responses are not a
guarantee that any subsequent transfer or delivery operations
will also succeed, but instead only indicate that the message
appears to be acceptable for the recipient according to the
rules and policies that are known to the current server.
Well, my understanding of this paragraph is that if I do get a positive
response for a recipient, it is for this transaction only. Trying again
to transfer/deliver to the same recipient will not necessarily mean the
response will be positive again.
I'm quite confidence your reading is incorrect - the ending text clearly
indicates that this is talking about a specific message in the context
of a single server.
If PRDR allows you to answer positively to mails that you will possibly
fail later down the chain doesn't it mean that any PRDR positive answer
can basically be ignored because it carries no guarantees (as in people
will fake positive replies because it is easier in some cases, and they
can bounce later anyway) ?
Of course PRDR allows you to respond positively to mail that will fail later.
Nothing says a server offering PRDR is performing final delivery, and given
that mail is a store-and-forward protocol that can involve many hops before
final delivery, how could a server possibly provide such a guarante.
Besides, we have a name for PRDR used for final delivery: We call that LMTP.
And even then message can fail to make to the actual recipient; that's what
failure MDNs are for.
ietf-smtp mailing list