Re: [Netconf] nmda-restconf operations (was: netconf-binary-encoding comments)

On Wed, 2018-07-11 at 08:40 -0700, Andy Bierman wrote:
> 
> 
> On Wed, Jul 11, 2018 at 4:22 AM, Ladislav Lhotka <lhotka@nic.cz> wrote:
> > Kent Watsen <kwatsen@juniper.net> writes:
> > 
> > >>> This isn't what I meant.  To be more specific, I'm wondering if the
> > nmda-restconf
> > >>> draft would benefit from having a sentence like:
> > >>> 
> > >>>    A RESTCONF server supporting NMDA datastores MAY implement the
> > >>>    "ietf-netconf" [RFC6241] and "ietf-netconf-nmda" [I-D. ietf-netconf-
> > nmda-
> > >>>    netconf] modules to enable the NETCONF operations defined in those
> > >>>    drafts to appear {+restconf}/operations resource.
> > >>> 
> > >>> Note: I put "MAY" as RESTCONF may someday have a more native way to do
> > this.
> > >>>
> > >>
> > >> Well, "ietf-netconf" does not really support NMDA well and this is why
> > >> we have "ietf-netconf-nmda". Does not make much sense to point to
> > >> "ietf-netconf" in an NMDA document.
> > >
> > > But we'd still need lock, unlock, commit, commit-confirmed, etc.,
> > > right?
> > 
> > These would make RESTCONF server stateful, thus violating REST principles.
> > My
> > draft
> > 
> > https://datatracker.ietf.org/doc/draft-lhotka-netconf-restconf-transactions/
> > 
> > tries to achieve similar effects in a RESTful way.
> > 
> 
> 
> I like the idea of "private candidate" datastores you call "staging".
> I would keep the term candidate.

The problem with this is that <candidate> in NETCONF has a looser semantics (it
can also be shared).

> 
> This idea was discussed during RESTCONF draft development, but
> as the client creating a "transaction" resource. I think your draft is on
> the right track, and datastores are protocol-independent so NETCONF
> can use the same datastores with RPC operations.

Yes, it can certainly be improved. I already have some pending updates, but I
wanted to keep it simple for start.

> 
> You ignore the problem of concurrent edits, which is why this idea was not
> pursued
> at the time. What happens when the altered data overlaps across private
> candidates?

I don't ignore it, the text says that the user's staging datastore has to be
atomically merged into <intended>, I just didn't want to specify the procedure.
In any case, the result has to be a valid <intended>. But I am open to a
discussion.

> What are the procedures equivalent to git pull, push, merge?
> How are collisions handled? What exactly is a collision?
> Not trivial standards issues to solve.

I think it can be left open, I even suspect there is no universal solution
suitable for all use cases. Our implementation (JetConf) does the following:

- each user's edit operation is applied to the staging datastore and also
recorded in a journal

- at commit, if <intended> hasn't been changed in the mean time, then the
staging datastore simply becomes <intended>

- otherwise, the edit operations from the journal are applied sequentially on
the actual contents of <intended>.

(Actually, we are not NMDA-compatible yet and don't have <intended> but it works
this way).

Lada 

>  
> > Lada
> > 
> 
> 
> Andy
>  
> > >
> > > Maybe it's a moot point, since ietf-netconf-nmda requires that ietf-
> > netconf
> > > is implemented too, but I thought being explicit would be helpful here. 
> > >
> > > So, adding something like this to nmda-restconf would be good?
> > >
> > >
> > > Kent // contributor
> > >
> > > _______________________________________________
> > > Netconf mailing list
> > > Netconf@ietf.org
> > > https://www.ietf.org/mailman/listinfo/netconf
> > 
-- 
Ladislav Lhotka
Head, CZ.NIC Labs
PGP Key ID: 0xB8F92B08A9F76C67