[ietf-nntp] OVER message-id

Francois Petillon fantec at proxad.net
Mon Dec 22 04:34:56 PST 2003


Charles Lindsey wrote:
>>No, you don't. You only have it if the article is numbered before
>>being stored on a spooler (with diablo, you may number articles on
>>a header feed which is much more light than having to fullfeed a
>>single server) with a single numbering system (you may have an
>>agreement with another company to use [each other|their] spoolers).
> Ah! So you are saying that Company A stores the full articles, and
> Companies B, C and D just store a database of articles in each group,
> including the overview, and using their own numbering system. So when a
> client of Company B, say, retrieves an article it gets the body from
> Company A, and constructs an Xref header consistent with its own
> numberings on the fly.

Not exactly, dreaderd keeps articles headers (thus, it will return local
header followed by spooler body).

> But in that case, how did it manage to construct its overview database, or
> did it have to retrieve the headers of all articles (but not their
> bodies) from A?

dreaderd needs a header feed to build its overview.

> I can see that in such a system B, C and D do not need to keep history
> files (though how do they then know when articles have expired from A, or
> been cancelled)?

There is no direct synchronization between spooler and frontal on 
article expiration. Expiration is done by batch (on spooler, il will
remove old spool files, on dreaderd, it will, group per group, mark
old entries as deleted). Direct synchronisation (ie mark an entry as
deleted when an article is removed) would hit disk I/O.

   François




More information about the ietf-nntp mailing list