From: Russ Allbery (rra@stanford.edu)
Date: Sun Feb 01 2004 - 04:37:58 CST
Charles Lindsey <chl@clerew.man.ac.uk> writes:
> Russ Allbery <rra@stanford.edu> writes:
>> Charles Lindsey <chl@clerew.man.ac.uk> writes:
>>> In practice, though, history file entriess are typically kept for at
>>> least 14 days so as to avoid that problem.
>> What evidence do you have to support this statement? It does not jive
>> with my operational experience and isn't even consistent with INN's
>> defaults.
> It's the default in CNews.
Gee, that's incredibly relevant to... something.
> And in the days when transatlantic propagation delays were regularly 7
> days, it was very necessary. Agreed it could be safely less now, but 2
> days seems dangerously short, as Henry seems to agree.
So, in other words, you don't actually know how long history file entries
are typically kept in practice.
> Anyway, the purpose of this discussion is to decide what number to write
> in for how stale an article could be allowed to be at the injector. John
> Stanley said 24 hours was too short. I suggested 72 hours as a possible
> answer. There has to be _some_ number there. I am waiting for further
> suggestions or agreements.
How about we guarantee that 24 hours is safe (thus requiring sites to keep
at least 24 hours of history, which I think is reasonable and proof
against at least the forseeable future), and then explain why setting it
to longer than that may cause articles not to propagate but leave it up to
sites to make their own decisions? We can't anticipate what this setting
is going to be in another ten years.
Right now, my guess is that three days is almost certainly safe, five days
is mostly okay but may cause problems, and ten days is going to almost
definitely hit significant issues.
There is an interoperability issue here, so I agree that we have to say
something, but there isn't actually a hard upper limit that one can
justify. Really, all the numbers we might come up with are magic and
completely arbitrary.
Usenet wants the Date header to be the injection time, not the time at
which the message was written, for protocol reasons. E-mail wants the
Date header to be the time at which the message was written. It's an
unfortunate conflict in how the header is used in the two protocols, and
it causes problems with moderated newsgroups and other gateways all the
time. The safe thing for a posting agent to do is to (mildly) violate the
shared article format and just set the Date header to the injection time,
perhaps preserving the original Date header in some other header.
-- Russ Allbery (rra@stanford.edu) <http://www.eyrie.org/~eagle/>