Re: [netconf] YANG attributes in a datastore

On Tue, 2019-12-10 at 08:25 -0800, Andy Bierman wrote:
> 
> 
> On Tue, Dec 10, 2019 at 7:50 AM Ladislav Lhotka <lhotka@nic.cz> wrote:
> > On Tue, 2019-12-10 at 07:42 -0800, Andy Bierman wrote:
> > > Hi,
> > > 
> > > YANG already has the "annotation" statement to define metadata.
> > > YANG has data-def-stmt to define data.  It would be very unwise
> > > to use metadata inside a leaf as if it were a leaf inside a container.
> > 
> > A typical example is a "comment" annotation - you want to permit it
> > basically
> > everywhere but don't want to clutter the schema with "comment" leaves.
> > 
> 
> I understand the difference between annotation-stmt and leaf-stmt.
> If the semantics can apply to all data nodes then annotation is appropriate.
> We should not change the protocols and YANG to reinvent a leaf-stmt.

Then I don't understand the point of your previous message. I think we've been
talking about annotations as per RFC 7952 from the beginning of this thread
(Michal just called it somewhat confusingly "attribute").

Lada

> 
> When we created YANG 1.0, we explicitly rejected the then-current style of XML
> 
>    <address street="123 Elm" city="Anytown" zipcode="100036" />
> 
> We rejected the use of attributes as leafs. We should not change that decision
> now.
> 
> 
> > Lada 
> > 
> 
> Andy
>  
> > > 
> > > 
> > > Andy
> > > 
> > > 
> > > On Tue, Dec 10, 2019 at 7:31 AM Ladislav Lhotka <lhotka@nic.cz> wrote:
> > > > On Tue, 2019-12-10 at 14:58 +0000, Balázs Lengyel wrote:
> > > > > Hello,
> > > > > If we try to standardize editing metadata, we will face a number of
> > issues
> > > > > that need to be clarified:
> > > > > 
> > > > > - Will you model whether the attribute (metadata) is writable?  Config
> > is
> > > > > not an allowed substatement in 
> > > > https://tools.ietf.org/html/rfc7952#section-3
> > > > 
> > > > In my view annotations should never be configuration because they should
> > not
> > > > influence the device state. The read-only/read-write propertly might be
> > > > useful.
> > > > 
> > > > > - Can a metadata have other additional properties? (mandatory,
> > > > cardinality,
> > > > > uniqueness of some kind?)
> > > > 
> > > > By design, annotations do not have a specific place in the schema, so
> > they
> > > > cannot be mandatory.
> > > > 
> > > > Cardinality: if you mean annotations with a list of values, this was
> > > > discussed
> > > > but rejected due to difficulties with XML encoding.
> > > > 
> > > > > - If it is modified, will it be subject to validation, error-messages?
> > > > Will
> > > > > a bad metadata result in a failed error-config?
> > > > 
> > > > I would say yes because otherwise it makes no sense to define them and
> > their
> > > > properties.
> > > > 
> > > > Lada
> > > > 
> > > > > 
> > > > > Regards Balazs
> > > > > 
> > > > > -----Original Message-----
> > > > > From: netconf <netconf-bounces@ietf.org> On Behalf Of Martin Bjorklund
> > > > > Sent: 2019. december 10., kedd 15:04
> > > > > To: mvasko@cesnet.cz
> > > > > Cc: netconf@ietf.org
> > > > > Subject: Re: [netconf] YANG attributes in a datastore
> > > > > 
> > > > > Hi,
> > > > > 
> > > > > Michal Vaško <mvasko@cesnet.cz> wrote:
> > > > > > Hello,
> > > > > > we are thinking about adding support for arbitrary attributes [1]
> > in 
> > > > > > our YANG/NETCONF datastore. To my knowledge, only the
> > representation 
> > > > > > in XML or JSON is defined and standardized but there are no
> > guidelines 
> > > > > > for editing (CRUD) these attributes.
> > > > > > 
> > > > > > So, my question is a rather general one, regarding implementations 
> > > > > > that support editing these attributes, would you mind sharing some 
> > > > > > basic ideas or principles used? Or is this outside any consensus
> > and 
> > > > > > each implementation is free to handle this completely in its own 
> > > > > > fashion? Thanks for any input.
> > > > > 
> > > > > I think this depends on the attribute defintion.  Some attributes are
> > > > > read-only from the northbound's protocol point of view.  Internally in
> > the
> > > > > server you may need some API to set/change them of course.  Some other
> > > > > attributes are read-write.
> > > > > 
> > > > > In our implementation we support a few read-write attributes (and some
> > > > > read-only).  Here's a simple example (in an edit-config):
> > > > > 
> > > > >   <foo annotation="hello"/>
> > > > > 
> > > > > This would set/change the "annotation" attribute's value to "hello",
> > > > > regardless of its previous value.
> > > > > 
> > > > > To delete an annotation, you would do:
> > > > > 
> > > > >   <foo annotation=""/>
> > > > > 
> > > > > Each of our attributes has a special value that means "delete" (in
> > most
> > > > > cases it is the empty string.
> > > > > 
> > > > > 
> > > > > /martin
> > > > > 
> > > > > 
> > > > > 
> > > > > 
> > > > > > Regards,
> > > > > > Michal
> > > > > > 
> > > > > > [1] https://tools.ietf.org/html/rfc7952
> > > > > > 
> > > > > > _______________________________________________
> > > > > > netconf mailing list
> > > > > > netconf@ietf.org
> > > > > > https://www.ietf.org/mailman/listinfo/netconf
> > > > > > 
> > > > > _______________________________________________
> > > > > netconf mailing list
> > > > > netconf@ietf.org
> > > > > https://www.ietf.org/mailman/listinfo/netconf
> > > > > _______________________________________________
> > > > > netconf mailing list
> > > > > netconf@ietf.org
> > > > > https://www.ietf.org/mailman/listinfo/netconf
-- 
Ladislav Lhotka
Head, CZ.NIC Labs
PGP Key ID: 0xB8F92B08A9F76C67