Re: [leafnode-list] Upstream server group counts not getting updated.

["Michael O'Quinn" <michael@xxxxxxxxxxx>]

I know it's bad form to reply to my own post before others have had a
chance, but I have more to add...

Another problem I've encountered -- which I suspect is caused by the
article counts not being update properly -- is that on occasion I've had
all of the articles expire from a group, then any new articles are stored
into the spool starting from 1 again.  This of course does not work very
well on the newsreader end of things.

Also, I sometimes have to abort a long run in the middle for one reason or
another.  When that happens, the upstream server group count FOR THAT
GROUP needs to be updated correctly to reflect how far it got before
fetchnews updates.

The more I think about this, the more I'm convinced that a lot of the
frustrations and "ghosts & goblins" I've experienced are caused by this.

Hhhhmmmm...

That count is so critical that maybe it SHOULD be updated after each
article, or perhaps after each 10 articles.  Or make it configurable: let
the administrator tune it to get the best balance between performance and
reliability.

One thing that could mitigate some of the performance hit would be
updating the numbers in place like INN does.  Instead of re-writing the
entire file each time, store each number with a lot of leading zeros, and
then just overwrite the new number on top of the old.

Of course that last one would add a lot of complexity and opportunity for 
new bugs.

Michael O'Quinn


On Fri, 8 Mar 2002, Michael O'Quinn wrote:

> As I mentioned in my last email, I've yet to see fetchnews complete an
> entire run without crashing.  Setting aside for a moment the cause of the
> crash, another issue has floated to the surface because that makes the
> whole package less robust.
> 
> The file containing the upstream server's group's counts, in my case
> /var/spool/news/leaf.node/clark.oquinn.info, has never been updated.  
> Each and every group is still sitting with a count of "1".  (This, BTW, is
> what I suspect is behind much of my earlier initialfetch -vs- maxfetch
> confusion.)
> 
> I assume this happens because fetchnews is waiting until just before
> (normally) exiting before it update those counts.  I would like to request
> that each newsgroup is updated as soon as that newsgroup completes.  Even 
> better from a robustness standpoint would be to update it after each 
> article, although that might be too severe a performance hit.
> 
> In a perfect world software would never crash, and administrators would
> never need to stop a run that is only partially completed, so this would
> not be an issue.
> 
> But, as we all well know...
> 
> Thanks, 
> Michael O'Quinn
> 
> 
> 



-- 
leafnode-list@xxxxxxxxxxxxxxxxxxxxxxxxxxxx -- mailing list for leafnode
To unsubscribe, send mail with "unsubscribe" in the subject to the list