Robert Klemme wrote:
>
> >
> > Add all new messages since the highest saved message-id.
> > Yes, it keeps track.
> > NG -> ML direction seems to be flawless. [1]
>
> This doesn't seem to fit Martin's analysis: there seem to be messages
> missing on both sides, don't they?
>

There were problems last year which were fixed by using the more
reliable feed.  I think there was a brief problem in the NG -> ML
direction very recently.  If folks are trying to fix things, this
can be expected.  Martin's data are too broad.  The problem we're
addressing started around Feb. this year.


> Can we really be sure that message ID's follow this pattern?

The message-id used is numeric; local to the news service.  It's nothing
to do with those strings in the mail headers.

>
> > I'm confident that if the problem was further downstream than nntp.rb,
> > there would be losses from sources other than ours.
>
> Do we know that there aren't?
>

The hardcore Usenet crew are really vocal when things go wrong with their
communication network.  I don't know that there aren't problems but I
can't contemplate that a serious snag with propagation would go unreported
for 4 months.

> >
> > NNTP seems to be a sub-protocol of SMTP.
>
> As I understand NNTP [...] rather different in nature.
>
> Maybe copying is not sufficient, maybe some header contents have to be
> massaged in order to fit with NNTP.
>

<from RFC 1036>
"The USENET News standard is more restrictive than the Internet standard,
placing additional requirements on each message and forbidding use of
certain Internet features. However, it should always be possible to use
a tool expecting an Internet message to process a news message."
</>

> > One thing that bothers me is fixes to problems (e.g. floods) that
> > we've had in the past.  There has to be a possibility that the
> > ruby-lang mailer has imposed restrictions on messages sent to the
> > "gateway ML member" and inadvertently introduced a corner-case.
>
> But wouldn't that mean loss of postings in the mailing list as well?  I
> mean, if the mailer did selective forwarding of messages then everybody (NG
> readers as well as ML readers) would miss them.
>

No?
Send to all ML members except the g/way, because the g/way is broken
and it might mirror them back to the ML resulting in the ML getting the
messages twice.
Or this ... http://www.ruby-talk.org/92529 ?
(I have no idea what happened here.)


> > This is a problem with the gateway although (as David pointed out
> > in a post that I just caught up on in the -talk archive because it
> > didn't get through) that doesn't mean the gateway software;  rather,
> > it means the path from the ruby-lang mailer to the first Usenet
> > news server.
>
> Do you mean the selective transmission works only for mail -> news and not
> mail -> mail?  Now I'm confused here...
>

Not by me, surely ;-)
All ML members post to a single recipient ruby-talk / ruby-lang.org.  It's
up to the bulk mailer where it goes next.  The g/way is just one member.
One scenario is that the missing messages are not getting to the gateway.
If they did, they would be forwarded to c.l.ruby like the rest.
That's my favourite suspicion and could easily be disproved by a single
log entry for a message that failed to make the NG.

With this single piece of evidence (which shouldn't be so difficult to
obtain ...) we have instant progress rather than looking at every other
non-possibility (which I, and others, have already done).
If the log says a phantom message has been forwarded, it cuts the
upstream right out of the equation and IMHO leaves nntp.rb only.
Debugging is a lot easier when you know you're focusing on the right area
and it's a pointless waste of time when you strongly suspect that the
problem lies elsewhere.


>
> Kind regards
>
>     robert
>

To you too,

daz