server disconnect leads to "article not found"

[Matthias Andree <ma@dt.e-technik.uni-dortmund.de>]

Some strange occurrences with Oort 0.7 from CVS:

I use leafnode with a rather short timeout. When I read a longer mail,
maybe reply, leafnode (NNTP server part, internal development version)
may disconnect. This leafnode version silently disconnects, i. e. the
client just sees EOF (read) or EPIPE (write). With Oort Gnus 0.6,
everything was fine.

With the current version (maybe half a day old), one (random) article is
claimed to be "not found" (bogus message), when it's in fact still there
(I can retrieve the "missing" article by leaving and re-entering the
group and for instance reconstruct the whole thread). I've observed this
for some days now.

The whole thing is maybe related to a POST that takes place, maybe the
disconnect prior to the post messes Gnus' internal state.

What happens:

- timeout, leafnode disconnects (5 min timeout)
- Gnus is supposed to post, connects again, sends MODE READER, POST, the
  article (which is posted properly) and GROUP (to the group it was in).
- At that time, leafnode crashes when reading the updated files (xinetd
  logs SIG 11). Clearly a leafnode bug.
- In the log, I see that Gnus immediately reconnects to leafnode, issues
  MODE READER, GROUP, ARTICLE n

I believe Gnus is confused at this point, not handling the disconnect
after "GROUP" properly, the logs don't yield more information ATM. Is
there any way to track and log Gnus keypresses or actions along with a
timestamp? I've now set up leafnode to run from tcpserver with recordio,
logging into multilog, in the hope that this will reveal more
information should that be necessary.

-- 
Matthias Andree