Is buffindexed overview corruption problem fixed in -STABLE?

Tim Ramsey tar at ksu.edu
Wed Sep 27 20:20:23 UTC 2000


Hiho,

I'm getting more and more complaints of newsgroups that are unavailable
because of buffindexed overview corruption:

  200 newsfeed.ksu.edu InterNetNews NNRP server INN 2.3.0 ready (posting ok).
  group rec.hunting
  211 2572 122158 124731 rec.hunting
  article 124731
  423 Bad article number

  Sep 27 14:56:50 cnn nnrpd[26236]: buffindexed: ovgroupmmap ovbuff is null(ovindex is 13622, ovblock is 892547126
  Sep 27 14:57:03 cnn nnrpd[26236]: fubar.cns.ksu.edu group rec.hunting 0

I upgraded from 2.2.3 to 2.3.0 on Sept. 8 and at that time decided to go
with buffindexed overview.  My overview is on three block devices, each a
1.5G partition ("inndf -o" currently reports 70% usage).  These were
initialized and the buffindexed overview rebuilt with "makehistory -O -x
-F".  That took ~10 hours to run with innd not running.

Within a day or two I started to see these "ovgroupmmap ovbuff is null"
messages.  Not too long after I started to see complaints from people who
are no longer able to read some of their newsgroups.  I've read messages
to inn-workers that say this message is harmless, but it sure doesn't
seem like that's correct.

Am I correct that this is a bug?  If so, is this bug fixed in the current
-STABLE snapshot?  Is there a reason why I should NOT fetch the latest
-STABLE and install that?

Also, what are the issues with rebuilding overview while innd/nnrpd is
running?  Will articles that arrive while makehistory is running get added
to the overview index?  What happens if a reader selects a newsgroup that
hasn't been completely rebuilt yet?  Do they see only some of the
articles, with others appearing to be missing or expired and so marked as
read by their newsreader?

Thanks for any wisdom.




More information about the inn-workers mailing list