buffindexed question

Katsuhiro Kondou kondou at nec.co.jp
Sat Sep 23 07:43:37 UTC 2000

In article <39CBE5A5.88BE8403 at orlandi.com>,
	Daniele Orlandi <daniele at orlandi.com> wrote;

} Following your suggestion, I created a cycbuff just for control.cancel,
} but I still see 1.500.000 articles from overview. If I try to list the
} articles in the newsgroup nnrpd grows and I need to kill it.

That's almost what I saw.  I had over 1.5M arts in
control.cancel.  This means buffindexed allocates memory
1.5M*300bytes(this is assumption of average data size) =
450MB when retrieving data.  This leads to that disaster.

} This cnfsstat looks weird:

For your case, I suspect old articles in control.cancel
still remain in other buffers, whilst newly incoming
articles are stored in the new cycbuff.  You still need
to nuke those old arts to reduce whole overview data in

} My history is 880MB, probably full of cancels, 30+ days old that never
} go away.

Here is mine fyi.

# ls -l history
-rw-rw-r--   1 news     news     402083435 Sep 23 16:36 history
# wc -l history
 4788401 history
# inndf -o
4 % overview space used
# inndf -n
4374762 overview data stored
# telnet localhost nntp
Connected to localhost.
Escape character is '^]'.
200 newsfeed.mesh.ad.jp InterNetNews NNRP server INN 2.4.0 (20000917 CVS prerelease) ready (posting ok).
group control.cancel
211 81937 59789356 59871294 control.cancel
205 .
Connection closed by foreign host.
# cnfsstat -a -c CANCELAREA
Class CANCELAREA for groups matching "control.cancel"
 Buffer CNCL, size:  40.0 MBytes, position:  27.5 MBytes  254.00 cycles
  Newest: 2000-09-23 16:41:12,    0 days,  0:00:28 ago
  Oldest: 2000-09-23  7:09:03,    0 days,  9:32:37 ago

Katsuhiro Kondou

More information about the inn-workers mailing list