Re: Xref and relaying agents

[Russ Allbery <rra@xxxxxxxxxxxx>]

Charles Lindsey <chl@xxxxxxxxxxxxxxxx> writes:

> There are essentially two ways in which things may be implemented:

I think that you're considering anything that has article storage to be a
serving agent, which is causing you to talk at cross-purposes to what the
draft actually says.  That's not the definition of a serving agent.  A
serving agent is an agent that makes articles available to reading
agents.  Full stop.

Relaying agents also have article storage.  That does *not* mean that
they're combined with a serving agent.

Stanford's news system doesn't look anything like either of your models.
It looks like this:


 <other news peers> <-+-----> relaying agent <--+-> serving/injecting agent
                       \                       /
                        ----> relaying agent <-

Those two relaying agents both have article storage so that they can hold
articles for a while if a peer is slow, or if the internal serving agent
is slow.  They both generate Xref headers because INN just always
generates Xref headers no matter what -- not necessarily ideal, but in
practice completely harmless.  Neither of those relaying agents accept any
connections from reading agents and could not serve articles to reading
agents because they both have overview and nnrpd disabled.

> But some no doubt follow the RIGHT scheme in which there are unlikely to
> be any Xref headers for the relaying agent to remove. The Model in
> USEPRO essentially sets out the RIGHT scheme, which then necessitates
> some mention of relayers removing Xrefs which, according to the model,
> should never have been there.

You're focusing way too much on the removal part.  I know of no agent that
removes Xref headers except when adding its own Xref header.  The draft
allows removal without adding a new one, just because there's no reason
not to, but what happens in practice is that relaying agents remove the
existing Xref header and add a new one.

This is a purposeless modification to the article, and back in the mists
of time in the working group, Brad Templeton used to argue for disallowing
it.  It turns out, however, to be much easier to write a multi-function
news server if one just always generates Xref headers, and Netnews has
functioned in that environment for over ten years.  Everything is now used
to that possibility, and I don't see any reason to pick a fight and try to
get software to change over something that's quite minor and unimportant.

> This is evident from the words "relaying agents that are also serving
> agents" which, according to our model, is actually a contradiction in
> terms. There are relaying agents and there are serving agents, and some
> *news-servers* may include both (but that does not cause the relaying
> agent to also "be" a serving agent).

I'm happy to consider a minor change wording proposal that rewords that in
some fashion to avoid the apparent contradiction.  I think the existing
wording is clear, but I don't mind tweaking it if other people don't feel
the same.

> Hence my suggeted wording:

>    Any Xref header, whether present on input or added by an associated
>    local serving agent, MAY be deleted before relaying.

This, however, is too broken to be used.

-- 
Russ Allbery (rra@xxxxxxxxxxxx)             <http://www.eyrie.org/~eagle/>