[Greylist-users] Does Greylisting *always* work?

[Eric S]

On Mon, 23 Jun 2003, martin dempsey wrote:

> If you have a busy "normal" email server that sends a new message from
> potentially a different user to to a different user every 20 minutes, each
> new message will get "temp failed" since its new and the one hour clock
> starts. However, from the point of view of the originating server every
> twenty minutes they contact the greylisting server and a message is "temp
> failed". If the server includes logic that says "don't bother running the
> queue for a server that reported a failure less than 30 minutes ago", then
> since the grelisting server always has failed less than twenty minutes ago,
> the originating server may decide not to bother processing the queue of
> previously failed messages (that would now be ok).  Or it might be past the
> four hour window by the time it retries a previously tried message.
>
> In this case, although the originating server is following all RFCs, all
> messages to the greylisting server may fail. Can this happen? Has anyone seen
> it? I may have an example but I'm not sure if this is the problem yet.

I'd argue that the originating server isn't following the RFCs as intended
(though possibly as written) because greylisting doesn't temp-fail the
connection (which is an option under SMTP, and is what sendmail does if
the loadaverage gets too high), but each individual recipient.  Therefore
the originating server shouldn't make assumptions about the state of the
server, only those mailboxes.  I can certainly see where a polite MTA
would delay all email bound for a certain server if that server was
temp-failing the connections rather than the mailboxes, however.

On the other hand, I can see a possible gotcha similar to this, but it
should be so rare as to not be an issue.  This would be where it isn't
practical to share the database amongst your MX MTAs, and you have enough
MX machines that someone could connect to one, get a defer, wait, hit a
different MX, get a defer, etc., never coming back to an MX that was hit
in the last 4 hours.  Mind you, I don't know many places that would need
that many MX servers that couldn't share the database, and whoever was
sending would have to have a high delay set (I think sendmail defaults to
retrying once an hour).