nnrpd issue

[Alex Kiernan]

"Pavel V. Knyazev" <pasha at surnet.ru> writes:

> ----- Original Message -----
> From: "Russ Allbery" <rra at stanford.edu>
> To: <inn-workers at isc.org>
> 
> 
> > > I've got problems even after 5-6 hours after an expiration process has
> > > gone.  We refuse all cancel articles. As i understand checkart doesn't
> > > play any role over here - 'next' must know if an article exists in the
> > > spool. I didn't say there's a problem retreiving articles without 'next'
> > > - nope it's all ok! Only 'next' slows down my system without giving any
> > > answer in a proper time.
> >
> > If you have nnrpdcheckart turned on, it's still doing work, even if it
> > doesn't need to.  I'd try turning that off and seeing if that helps any.
> 
> Believe me, i wouldn't ask such stupid questions that may sound like
> "Who's guilty and what to do?". nnrpdcheckart has been disabled
> a long time ago. nnrpdcheckart doesn't influence on how many time
> 'next' does its work. And this time depends on how many time has passed
> since expireover has gone:
> 4.00am - exp finished,
> 3.20pm - 30 secs,
> 7.15pm - 55 secs,
> 9.10pm - 62 secs.
> 

Instant test against our production platform showed the same
thing:

real	0m19.038s
user	0m0.300s
sys	0m0.000s

CMDnextlast just takes the current article number and runs through
each article number attempting to open it until it gets lucky. If
you're running CNFS of course you have many polls which miss if you're
in a high traffic can.

One approach would be to get it to use the overview database and
return the standard porkies you get with that approach with
nnrpdcheckart turned off.

-- 
Alex Kiernan, Principal Engineer, Development, THUS plc