system crash

[Russ Allbery]

Pavel V Knyazev <pasha at surnet.ru> writes:

> We all know well, that fullfeed usenet routers are working under heavy
> load, particularly their hard drives. Consider the situation when a
> system crashes and is uncorrectly rebooted. Even if a history files are
> damaged, they can be simply recovered by removing these files. I'm not
> worried about my logs also.  CNFS buffers can be also rewritten by
> themselves later (by new articles) IMHO.

> But what about active and overview buffers?
> Are they predisposed to be self-recovered after an expiration process?

For overview, I know that tradindexed deals very poorly with system
crashes at present.  I'm not sure about buffindexed or ovdb; I've not used
either of those overview methods.  My new tradindexed implementation
should deal with system crashes much better; for one thing, it msyncs all
of its data aggressively (I'm going to try that for starters and then only
back off from that if there's a real performance problem with that), and
for another I'm going to write some self-correction code into the
implementation that can be run on crash recovery.

The active file is periodically synced to disk, but there's a small
possibility that a corrupted line could be written out if the crash
catches it at a bad time.  A backup file is always saved, though, which
makes me think that it would probably be relatively easy to recover.

Incidentally, I don't find Usenet routers to be particularly unstable when
running INN:

windlord:~> rsh -l news newsfeed 'ps -ef | grep innd'
    news 18309     1 45   Apr 22 ?       188316:23 /var/news/bin/innd -p5

:)

-- 
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.