Hi Edward,

On Thu, Jul 19, 2018 at 09:27:23PM +0200, Edward Shryane wrote:
> we plan to return objects from both the RIPE and RIPE-NONAUTH sources
> for NRTM, as it will be less disruptive when NWI-5 goes into
> production.

I am not convinced anyone can authoritatively make that statement. As
far as I know the RIPE NCC IRR would be the first to publish multiple
sources in a single stream. There is an industry convention related to
IRR sources, I see no immediate reason to deviate from this standard.

> - If clients do nothing, they still receive all updates.

And perhaps consider them corrupted, because they requested source X and
get back source Y.

> - If we do not return all updates by default, then out of region
> object updates will not appear (even though they will occur).

The 20 NRTM clients (of which we probably know most personally) just
have to configure a second source and all will be good.

> - Receiving all updates will require separate queries in separate
> connections, which need to be merged on the client side.

The whole point is to not merge them into a single source, so that we
can programmatically differentiate between the two sources. If there is
merging, it happens at a different point in the pipeline, not at
'ingress' (from the point of view of a mirror).

> - The default behaviour for Whois queries (on port 43, REST API and Web) will be to return objects from both sources.

I have no problem with that.

> On the other hand, to separate NRTM streams by source:
>
> - The NRTM query does specify the source, this implies updates from only that source.

exactly. This is a foundational assumption in all NRTM software I am
aware of.

> - It is straightforward to support separate sources on the server side.
> - There are relatively few active NRTM clients that would need to be updated.

I am happy to hear that separate sources are straight forward to
implement. On the IRRd v2, v3 and v4 side of things it is not straight
forward to implement.

> We also do not plan to separate the database dumps by source. This can
> be done, but any consumers need to make adjustments or will no longer
> find out of region objects.

yes, that is the point. We make a change, and people need to update if
they rely on this. By splitting the NRTM and database dumps, most
implementations only need to add an additional source, rather than
reprogram their client software to understand that what used to be a
single file containing a single source, now contains multiple sources.

> I welcome discussion on these points, we will implement as the DB-WG
> prefers.

Because of NWI-5 the RIPE IRR is splitting into two, and we should plan
accordingly. To me that means separate database dumps and separate NRTM
streams.

Kind regards,

Job