CURRENT branch

[Russ Allbery]

"Jeffrey M. Vinocur" <jeff at litech.org> writes:
> On Mon, 18 May 2009, Russ Allbery wrote:

>> I think the current tradeoff (where the article doesn't appear) is
>> probably better than breaking OVSTATICSEARCH,

> *nod* Incidentally, do you have any idea why I was the only one to
> notice corrupt XOVER responses because of this?  It seems like
> violating the OVSTATICSEARCH contract would cause trouble more
> frequently.

Well, in order to violate the contract, we have to get into a situation
where the overview data for an article can't be satisfied by the current
data map.  So my guess is that you have an Xref slaving setup using
innfeed or a similar mechanism that doesn't preserve article order.
However, that's just a wild guess.

If you're not using Xref slaving at all, then the only thing I can think
of is that you were running into cases where an article was delivered
between the start of the search and the completion of it.  Which is a
pretty tight race, so maybe you were just getting unlucky more (or had a
slower system for some reason).

> I'm trying to imagine a scenario where this would be a problem...I
> suppose if an xrefslave server were being fed a split stream
> (e.g. separate feeds for small and large articles) could do it, at
> least.

innfeed can do it even with a single stream if it ever runs a backlog,
since it prefers current articles to backlogged articles.

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>

    Please send questions to the list rather than mailing me directly.
     <http://www.eyrie.org/~eagle/faqs/questions.html> explains why.