Re: [netmod] schema mount and YANG library

[Martin Bjorklund <[e-mail]>]

Lou Berger <[e-mail]> wrote:
> Lada,
> 
> 
> On January 16, 2018 7:07:15 AM Ladislav Lhotka <[e-mail]> wrote:
> 
> > On Tue, 2018-01-16 at 06:30 -0500, Lou Berger wrote:
> >> Lada,
> >>
> >> It sounds like you are proposing in (1) a fairly significant change in
> >> the
> >> direction of the draft and in (2) a basic approach that has been
> >
> > It is no change in direction, just a simplification of the
> > schema-describing
> > state data. Given the recent developments in 7895bis it makes no sense
> > to me to
> > have two "schema" lists if we can have just one.
> >
> 
> Managing transition is hard. It's also highlights why Yang Library
> this needs to be at least equally discussed in this group.
> 
> I will talk with my co-chairs and perhaps the ADs to get their opinion
> on making such a change this point in the process.
> 
> 
> >>
> >> rejectected by the WG multiple times.  FWIW there are drafts already
> >> with
> >
> > No at all. The first and last time I proposed this was on 15 December
> > 2017:
> >
> > https://www.ietf.org/mail-archive/web/netmod/current/msg19753.html
> >
> 
> Oh, I certainly would call you proposing that the schema for inline be
> part of the rest of the schema Mount module well before that. I'm sure
> I can dig up mail / slides it really necessary...

I don't think this has been proposed before.  All previous proposals
were basically variants on what is now "use-schema", which works fine
when all instances have the same schema.  This new proposal solves the
issue with different schemas in different instances.

> > The only reply was from you. To me, it is the cleanest solution of the
> > inline
> > case. Of course, I am open to technical objections.
> >
> 
> I'm sure I can find material on this as well....

Ok.


/martin


> 
> Lou
> 
> > If it's not clear what I mean, I can make up some examples.
> >
> > Lada
> >
> >
> >> the iesg that will need to be returned to their WGs if either change
> >> is made.
> >>
> >> Martin,
> >>
> >> Do share Lada's view?
> >>
> >> Lou
> >>
> >>
> >> On January 16, 2018 2:14:42 AM Ladislav Lhotka <[e-mail]> wrote:
> >>
> >> > Hi Lou,
> >> >
> >> > in my view, we should do the following two (significant) changes:
> >> >
> >> > 1. Instead of borrowing a grouping from ietf-yang-library and having a
> >> > parallel
> >> > list of mounted schemas, we should keep *all* mounted schemas directly
> >> > in
> >> > the
> >> > YANG library and refer to them from schema-mounts structures. Juergen
> >> > suggested
> >> > this change and it is IMO the right thing to do.
> >> >
> >> > 2. Define a metadata annotation (e.g. @schema-ref) that would be
> >> > required
> >> > for
> >> > inline mount point instances and specify the inline-mounted schema
> >> > also by
> >> > referring to a schema specified in YANG library.
> >> >
> >> > The advantage of #2 is that an annotation can be attached equally well
> >> > to
> >> > both
> >> > state an configuration data. So, instead of papering over the issue
> >> > that
> >> > YANG
> >> > library (state data) cannot appear in configuration datastores, we can
> >> > use
> >> > this
> >> > general and straightforward approach. This also allows for defining
> >> > different
> >> > mounted schemas for instances of the same mount point in different
> >> > datastores.
> >> >
> >> > I strongly believe that these changes (along with the new YANG library
> >> > schema
> >> > and NMDA) make for a simple and elegant datastore architecture in
> >> > which
> >> > schema
> >> > mount would be an optional feature.
> >> >
> >> > Lada
> >> >
> >> >
> >> >
> >> > On Mon, 2018-01-15 at 16:20 -0500, Lou Berger wrote:
> >> > > Lada/Martin,
> >> > >
> >> > > I don't believe we reached closure on this discussion.  The open
> >> > > issues
> >> > > relate to proposed new text (slightly modified):
> >> > >
> >> > > at the end of the section [3.2] adding a new paragraph along the
> >> > > lines of:
> >> > >
> >> > >    The use of mount points does not impact the nature of the
> >> > >    mounted data or in which data store information is made
> >> > >    available. For example, mounted YANG Library modules define
> >> > >    only operational state data and, as such, the information in
> >> > >    these modules is available from operational data stores using
> >> > >    the appropriate protocol operations.  It is also worth
> >> > >    noting that the Schema Mount module itself parallels the
> >> > >    YANG Library module and only defines operational state data.
> >> > >
> >> > > Is this change acceptable?
> >> > >
> >> > > What other issues related to SM are outstanding?
> >> > >
> >> > > Thank you,
> >> > >
> >> > > Lou
> >> > >
> >> > > On 12/19/2017 8:26 AM, Ladislav Lhotka wrote:
> >> > > > On Tue, 2017-12-19 at 07:49 -0500, Lou Berger wrote:
> >> > > > > On 12/19/2017 7:36 AM, Ladislav Lhotka wrote:
> >> > > > > > On Tue, 2017-12-19 at 06:43 -0500, Lou Berger wrote:
> >> > > > > > > Hi Lada,
> >> > > > > > >
> >> > > > > > > On 12/19/2017 6:23 AM, Ladislav Lhotka wrote:
> >> > > > > > > > On Tue, 2017-12-19 at 06:20 -0500, Lou Berger wrote:
> >> > > > > > > > > Lada,
> >> > > > > > > > >
> >> > > > > > > > >
> >> > > > > > > > > On December 19, 2017 1:12:35 AM Ladislav Lhotka
> >> > > > > > > > > <[e-mail]
> >> > > > > > > > > >
> >> > > > > > > > > wrote:
> >> > > > > > > > >
> >> > > > > > > > > > On Mon, 2017-12-18 at 15:30 -0500, Lou Berger wrote:
> >> > > > > > > > > > > lada,
> >> > > > > > > > > > >
> >> > > > > > > > > > >      See below.
> >> > > > > > > > > > >
> >> > > > > > > > > > >
> >> > > > > > > > > > > On 12/15/2017 8:59 AM, Ladislav Lhotka wrote:
> >> > > > > > > > > > > > Hi,
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > unfortunately, using an action for querying embedded
> >> > > > > > > > > > > > YANG
> >> > > > > > > > > > > > library
> >> > > > > > > > > > > > data
> >> > > > > > > > > > > > (needed for the "inline" case of schema mount)
> >> > > > > > > > > > > > doesn't
> >> > > > > > > > > > > > work
> >> > > > > > > > > > > > either
> >> > > > > > > > > > > > because now under NMDA actions can be used only on
> >> > > > > > > > > > > > instances
> >> > > > > > > > > > > > in
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > <operational> datastore.
> >> > > > > > > > > > >
> >> > > > > > > > > > > but the inline/embedded library would (only) be present
> >> > > > > > > > > > > in
> >> > > > > > > > > > > the
> >> > > > > > > > > > > in
> >> > > > > > > > > > > the
> >> > > > > > > > > > > operational datastore, so what's the issue?
> >> > > > > > > > > >
> >> > > > > > > > > > Well, the issue is described in my initial mail of this
> >> > > > > > > > > > thread:
> >> > > > > > > > > > the
> >> > > > > > > > > > current
> >> > > > > > > > > > text
> >> > > > > > > > > > requires that every instance of an inline mount point
> >> > > > > > > > > > contains
> >> > > > > > > > > > the
> >> > > > > > > > > > embedded
> >> > > > > > > > > > YANG library. Tha latter is state data, so the above
> >> > > > > > > > > > requirement
> >> > > > > > > > > > cannot
> >> > > > > > > > > > be
> >> > > > > > > > > > satisfied if the mount point instance is in a
> >> > > > > > > > > > configuration
> >> > > > > > > > > > datastore.
> >> > > > > > > > > >
> >> > > > > > > > >
> >> > > > > > > > > That's not how I read the intent of the current text.  I
> >> > > > > > > > > don't
> >> > > > > > > > > see
> >> > > > > > > > > SM
> >> > > > > > > > > impacting which data stores information is presented.  Just
> >> > > > > > > > > like
> >> > > > > > > > > use
> >> > > > > > > > > of
> >> > > > > > > > > scheme mount doesn't transform RO configuration information
> >> > > > > > > > > into
> >> > > > > > > > > operational information.  I sent you a couple of sentences
> >> > > > > > > > > clarifying
> >> > > > > > > > > this
> >> > > > > > > > > at one point, I'll dig up the proposed text and resend.
> >> > > > > > > >
> >> > > > > > > > Please do, this has to be discussed in the WG mailing list.
> >> > > > > > >
> >> > > > > > > Agreed - that's why I asked to start this thread!
> >> > > > > > >
> >> > > > > > > Here's the original proposal:
> >> > > > > > >
> >> > > > > > >    How about at the end of the section [3.2] adding a new
> >> > > > > > >    paragraph along the lines of:
> >> > > > > > >
> >> > > > > > >    It is important to note that both YANG Library and Schema
> >> > > > > > >    Mount Modules contain only operational state data. As such,
> >> > > > > >
> >> > > > > > s/contain/define/
> >> > > > > >
> >> > > > > > >    the information in these modules should be retrieved by
> >> > > > > > >    clients from operational data stores using the appropriate
> >> > > > > >
> >> > > > > > This is based on two assumptions:
> >> > > > > >
> >> > > > > > 1. For every configuration datastore there is a corresponding
> >> > > > > > operational
> >> > > > > > datastore.
> >> > > > >
> >> > > > > well the text is revised below.  In any case, "these modules"
> >> > > > > refers
> >> > > > > to
> >> > > > > yang library, and yes, I'm assuming YL is always and only in
> >> > > > > operational.  If the revised text below isn't clear s/these/YANG
> >> > > > > Library/
> >> > > > > -
> >> > > >
> >> > > > The thing is that we have the top-level YANG library in
> >> > > > <operational>,
> >> > > > and
> >> > > > then
> >> > > > embedded YANG libraries scattered inside inline mount point
> >> > > > instances.
> >> > > >
> >> > > > > > 2. For every mount point instance in any configuration datastore
> >> > > > > > there
> >> > > > > > is a
> >> > > > > > corresponding mount point instance (with the same path) in an
> >> > > > > > operational
> >> > > > > > datastore.
> >> > > > > >
> >> > > > > > I think that neither of these has to be true in general.
> >> > > > >
> >> > > > > agreed in general, but for inline, where YL is required, it must be
> >> > > > > true.
> >> > > >
> >> > > > How do you know? I provided an example in Singapore where a mount
> >> > > > point
> >> > > > instance
> >> > > > in <intended> is a part of pre-provisioned data (for non-existent
> >> > > > hardware).
> >> > > > Then, according to the NMDA rules there is no corresponding instance
> >> > > > in
> >> > > > <operational>, hence no place where the embedded YANG library can be
> >> > > > placed.
> >> > > > (I can easily provide a concrete example if needed).
> >> > > >
> >> > > >
> >> > > > Dean replied that this cannot happen, so it seems there are some
> >> > > > assumptions
> >> > > > how
> >> > > > the inline method of schema mount may be applied. If so, these
> >> > > > assumptions
> >> > > > have
> >> > > > to be explicitly stated.
> >> > > >
> >> > > > > > >    protocol operations.
> >> > > > > >
> >> > > > > > In contrast, the substance of my proposal with metadata
> >> > > > > > annotations
> >> > > > > > is
> >> > > > > > to be
> >> > > > > > able to retrieve all schemas from a well-known location in *the*
> >> > > > > > <operational>
> >> > > > > > datastore, namely from the top-level YANG library.
> >> > > > >
> >> > > > > What about a schema that is based on dll that contains modules that
> >> > > > > isn't loaded until a mount point is instantiated -- this is
> >> > > > > certainly
> >> > > > > a
> >> > > > > valid approach for supporting LNEs, but would be precluded in this
> >> > > > > approach.  I really don't think a top level approach works for all
> >> > > > > inline (managed) types of mounts.
> >> > > >
> >> > > > It isn't precluded: when the mount point is instantiated (no matter
> >> > > > which
> >> > > > datastore it is in), the server adds the schema as a new entry to the
> >> > > > "schema"
> >> > > > list in the top level YANG library (with a unique key), and annotates
> >> > > > the
> >> > > > mount
> >> > > > point instance with a leafref pointing to that key. So different
> >> > > > instances
> >> > > > of
> >> > > > the same mount point can have different schemas.
> >> > > >
> >> > > > > > > Given this discussion, we can generalize it further to:
> >> > > > > > >
> >> > > > > > >    The use of mount points does not impact the nature of the
> >> > > > > > >    mounted data or in which data store information is made
> >> > > > > > >    available. For example, mounted YANG Library modules contain
> >> > > > > > >    only operational state data and, as such, the information in
> >> > > > > > >    these modules is available from operational data stores
> >> > > > > > >    using
> >> > > > > > >    the appropriate protocol operations.
> >> > > > > >
> >> > > > > > The whole question here is whether and how we can locate the
> >> > > > > > schema
> >> > > > > > for
> >> > > > > > an
> >> > > > > > inline mount point in any configuration datastore.
> >> > > > >
> >> > > > > Why is a mounted YL different than a top level YL?  What works for
> >> > > > > and
> >> > > >
> >> > > > It is not different, but it can be only in an operational datastores,
> >> > > > and so
> >> > > > for
> >> > > > mount point instances inside configuration datastores we need a way
> >> > > > how
> >> > > > to
> >> > > > locate the schema for that mount point, because it cannot be found
> >> > > > directly
> >> > > > under the mount point instance (as the current text assumes).
> >> > > >
> >> > > > > is sufficient for the normal case of YL shouldn't be impacted or
> >> > > > > modified by SM -- at least that's how I thought we've been talking
> >> > > > > about
> >> > > > > since SM was started.  Again, we never made any special provisions
> >> > > > > for
> >> > > > > any other rw/ro/state data, assuming top level YL is not handled as
> >> > > > > metadata, why start now?
> >> > > > >
> >> > > > > I'm getting the impression that your argument may be more about if
> >> > > > > YL
> >> > > > > should be treated as something other than operational data, is this
> >> > > > > wrong?
> >> > > >
> >> > > > This is wrong. My argument is that there should be only one top-level
> >> > > > YANG
> >> > > > library (state data) and each inline mount point instance just points
> >> > > > to
> >> > > > a
> >> > > > schema inside it by means of a metadata annotation attached to the
> >> > > > mount
> >> > > > point
> >> > > > (in any datastore).
> >> > > >
> >> > > > Lada
> >> > > >
> >> > > > > Thanks,
> >> > > > > Lou
> >> > > > >
> >> > > > > > Lada
> >> > > > > >
> >> > > > > > > Lou
> >> > > > > > >
> >> > > > > > > > Lada
> >> > > > > > > >
> >> > > > > > > > > Lou
> >> > > > > > > > > > > > However, a good alternative seems to be a metadata
> >> > > > > > > > > > > > annotation
> >> > > > > > > > > > > > along
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > lines of RFC 7952, for example with the alternative B
> >> > > > > > > > > > > > of
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > newly
> >> > > > > > > > > > > > proposed YANG library schema:
> >> > > > > > > > > > > >
> >> > > > > > > > > > > >       md:annotation schema-ref {
> >> > > > > > > > > > > >         type leafref {
> >> > > > > > > > > > > >           path "/yanglib:yang-
> >> > > > > > > > > > > > library/yanglib:schema/yanglib:name";
> >> > > > > > > > > > > >         }
> >> > > > > > > > > > > >       }
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > In other words, all inline mounted schemas would be
> >> > > > > > > > > > > > included
> >> > > > > > > > > > > > in
> >> > > > > > > > > > > > the
> >> > > > > > > > > > > > top-level YANG library, and mount point instances in
> >> > > > > > > > > > > > all
> >> > > > > > > > > > > > datastores
> >> > > > > > > > > > > > would be annotated with leafref pointing to the
> >> > > > > > > > > > > > actual
> >> > > > > > > > > > > > schema.
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > Unlike regular state data, it is IMO no problem to
> >> > > > > > > > > > > > permit
> >> > > > > > > > > > > > such
> >> > > > > > > > > > > > annotations in configuration datastores.
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > Opinions?
> >> > > > > > > > > > >
> >> > > > > > > > > > > I'm not sure this will work for all architectures of
> >> > > > > > > > > > > LNEs
> >> > > > > > > > > > > as
> >> > > > > > > > > > > well
> >> > > > > > > > > > > as
> >> > > > > > > > > > > other possible future use cases.  In short, this seems
> >> > > > > > > > > > > *very*
> >> > > > > > > > > > > restrictive.
> >> > > > > > > > > >
> >> > > > > > > > > > I don't understand, IMO it is not restrictive at
> >> > > > > > > > > > all. What
> >> > > > > > > > > > kind
> >> > > > > > > > > > of
> >> > > > > > > > > > restrictions
> >> > > > > > > > > > do you see?
> >> > > > > > > > > >
> >> > > > > > > > > > Lada
> >> > > > > > > > > >
> >> > > > > > > > > > > Lou
> >> > > > > > > > > > >
> >> > > > > > > > > > > > Thanks, Lada
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > Ladislav Lhotka <[e-mail]> writes:
> >> > > > > > > > > > > >
> >> > > > > > > > > > > > > Hi,
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > the following text in sec. 3.2 of schema-mount-08
> >> > > > > > > > > > > > > is
> >> > > > > > > > > > > > > wrong
> >> > > > > > > > > > > > > for
> >> > > > > > > > > > > > > traditional
> >> > > > > > > > > > > > > datastores, and even more so for NDMA:
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > >     In case 1 ["inline"], the mounted schema is
> >> > > > > > > > > > > > > determined
> >> > > > > > > > > > > > > at
> >> > > > > > > > > > > > > run
> >> > > > > > > > > > > > > time:
> >> > > > > > > > > > > > > every
> >> > > > > > > > > > > > >     instance of the mount point that exists in the
> >> > > > > > > > > > > > > parent
> >> > > > > > > > > > > > > tree
> >> > > > > > > > > > > > > MUST
> >> > > > > > > > > > > > >     contain a copy of YANG library data [RFC7895]
> >> > > > > > > > > > > > >     that
> >> > > > > > > > > > > > > defines
> >> > > > > > > > > > > > > the
> >> > > > > > > > > > > > >     mounted schema exactly as for a top-level data
> >> > > > > > > > > > > > > model.  A
> >> > > > > > > > > > > > > client
> >> > > > > > > > > > > > > is
> >> > > > > > > > > > > > >     expected to retrieve this data from the
> >> > > > > > > > > > > > >     instance
> >> > > > > > > > > > > > > tree,
> >> > > > > > > > > > > > > possibly
> >> > > > > > > > > > > > > after
> >> > > > > > > > > > > > >     creating the mount point.  Instances of the
> >> > > > > > > > > > > > >     same
> >> > > > > > > > > > > > > mount
> >> > > > > > > > > > > > > point
> >> > > > > > > > > > > > > MAY
> >> > > > > > > > > > > > > use
> >> > > > > > > > > > > > >     different mounted schemas.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > An instance of the mount point in any
> >> > > > > > > > > > > > > *configuration*
> >> > > > > > > > > > > > > datastores
> >> > > > > > > > > > > > > cannot
> >> > > > > > > > > > > > > contain
> >> > > > > > > > > > > > > YANG library (being state data), and so the MUST
> >> > > > > > > > > > > > > cannot
> >> > > > > > > > > > > > > hold.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > It is not clear to me how to repair this without
> >> > > > > > > > > > > > > considerable
> >> > > > > > > > > > > > > complications
> >> > > > > > > > > > > > > and/or a lot of handwaving. There is actually one
> >> > > > > > > > > > > > > good
> >> > > > > > > > > > > > > solution
> >> > > > > > > > > > > > > but it
> >> > > > > > > > > > > > > has
> >> > > > > > > > > > > > > impact on YANG library: the server could provide it
> >> > > > > > > > > > > > > in
> >> > > > > > > > > > > > > a
> >> > > > > > > > > > > > > reply
> >> > > > > > > > > > > > > to
> >> > > > > > > > > > > > > an
> >> > > > > > > > > > > > > operation,
> >> > > > > > > > > > > > > say "get-yang-library" rather than as state
> >> > > > > > > > > > > > > data. Then
> >> > > > > > > > > > > > > everything
> >> > > > > > > > > > > > > would be
> >> > > > > > > > > > > > > fine
> >> > > > > > > > > > > > > - this operation would turn into an action for the
> >> > > > > > > > > > > > > mount
> >> > > > > > > > > > > > > point,
> >> > > > > > > > > > > > > and it
> >> > > > > > > > > > > > > can
> >> > > > > > > > > > > > > be
> >> > > > > > > > > > > > > used equally well for config true and false mount
> >> > > > > > > > > > > > > points.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > So my proposal is to move from YANG library as
> >> > > > > > > > > > > > > state
> >> > > > > > > > > > > > > data
> >> > > > > > > > > > > > > to
> >> > > > > > > > > > > > > an
> >> > > > > > > > > > > > > operation.
> >> > > > > > > > > > > > > It
> >> > > > > > > > > > > > > could be done along with changing the YANG library
> >> > > > > > > > > > > > > structure,
> >> > > > > > > > > > > > > so
> >> > > > > > > > > > > > > there
> >> > > > > > > > > > > > > will be
> >> > > > > > > > > > > > > little extra impact on implementations.
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > Lada
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > --
> >> > > > > > > > > > > > > Ladislav Lhotka
> >> > > > > > > > > > > > > Head, CZ.NIC Labs
> >> > > > > > > > > > > > > PGP Key ID: 0xB8F92B08A9F76C67
> >> > > > > > > > > > > > >
> >> > > > > > > > > > > > > _______________________________________________
> >> > > > > > > > > > > > > netmod mailing list
> >> > > > > > > > > > > > > [e-mail]
> >> > > > > > > > > > > > > https://www.ietf.org/mailman/listinfo/netmod
> >> > > > > > > > > > >
> >> > > > > > > > > > > _______________________________________________
> >> > > > > > > > > > > netmod mailing list
> >> > > > > > > > > > > [e-mail]
> >> > > > > > > > > > > https://www.ietf.org/mailman/listinfo/netmod
> >> > > > > > > > > >
> >> > > > > > > > > > --
> >> > > > > > > > > > Ladislav Lhotka
> >> > > > > > > > > > Head, CZ.NIC Labs
> >> > > > > > > > > > PGP Key ID: 0xB8F92B08A9F76C67
> >> > > > > > > > > >
> >> > >
> >> > >
> >> >
> >> > --
> >> > Ladislav Lhotka
> >> > Head, CZ.NIC Labs
> >> > PGP Key ID: 0xB8F92B08A9F76C67
> >> >
> >>
> >>
> > --
> > Ladislav Lhotka
> > Head, CZ.NIC Labs
> > PGP Key ID: 0xB8F92B08A9F76C67
> >
> 
>