too-llong expires

Ron Jarrell jarrell at solaris.cc.vt.edu
Mon Jul 2 21:33:53 UTC 2001


Ok, isn't shlock supposed to prevent more than one news.daily
from running at a time?  Or does killing and restarting innd clear
that lock?

My innd wedged up from the db->put error I reported earlier at
about 11pm june 29.  news.daily started at midnight.  Operations 
finally figured out that the blinking read light that said "news is down"
meant that news was down, and that they really ought to do something,
like call me, or my backups, about 12 hours later at 11am on June 30.
(Don't get me started...)  I didn't answer my phone, so ratehr than paging
me, they tried my back up first.  He, knowing that when this happens,
we have to kill innd and restart it did so, not noticing that news.daily
was still running.  (although it looks like it was stuck in innstat.  He'd
actually checked for expire running, which it wasn't. yet.)

So expireover started, and never ended.  Jul 1st, another news.daily
started, and had made it down to *it's* expire.  Expire, as of 5pm
today, Jul 2nd, is still running.  News.daily, this morning, properly
figured out that another news.daily was running, this time.

Any suggestions for making the expire go faster (or figuring out why
it's stuck) greatfully accepted.  This isn't an overloaded machine, and
the spool and db files are split across multiple io channels and spindles;
but if I get more than about 80G of articles in a day, expire takes 30
hours.

I'm considering reserving next sunday (I've got an engineering window
I can use from 3am to 3pm) to regenerate the history and the overview
db.



More information about the inn-workers mailing list