Re: [netconf] get-data origin filters

[Martin Bjorklund <[e-mail]>]

Andy Bierman <[e-mail]> wrote:
> On Mon, Oct 7, 2019 at 10:39 AM Martin Bjorklund <[e-mail]> wrote:
> 
> > Andy Bierman <[e-mail]> wrote:
> > > On Mon, Oct 7, 2019 at 7:36 AM Rob Wilton (rwilton) <[e-mail]>
> > > wrote:
> > >
> > > > Hi Andy,
> > > >
> > > >
> > > >
> > > > Don’t all the filters effectively work this way?
> > > >
> > >
> > >
> > > I do not see the text that explains origin-filter and
> > negated-origin-filter
> > > working the way Martin
> > > describes it.  These filters do not say anywhere to select a node because
> > > it has descendants
> > > that match the origin filters.  It says very clearly that the filter test
> > > is on the specified node.
> >
> > Let's simplify the example somewhat.  Suppose the client sends:
> >
> >   <get-data>
> >     <datastore>ds:operational</datastore>
> >     <origin-filter>or:learned</origin-filter>
> >     <with-origin/>
> >   </get-data>
> >
> > Are you saying that b/c the top-level node doesn't have or:learned,
> > this filter will return nothing?
> >
> >
> 
> Yes -- the origin for addresses is "intended".
> The text clearly says the origin-filter has to match the origin annotation
> for the node.
> The annotation for "address" can have only 1 value.
> 
> I don't think this is implementable so I will just use
> deviation(not-supported) and ignore it

Fine; up to you!

But the question is if the spec is unclear or not.  From what I can
tell so far, it seems it is not incorrect, but can probably be
clarified, perhaps along the lines of Rob's explanation.


/martin




> 
> 
> Andy
> 
> It seems you are assuming that the server first matches the top-level
> > nodes, and if they match, then match descendants and so one.  The text
> > doesn't say that, however.  It says:
> >
> >                 A configuration node matches the filter if its
> >                 'origin' annotation is derived from or equal to any of
> >                 the given filter values.
> >
> > and:
> >
> >           Any ancestor nodes (including list keys) of nodes selected by
> >           the filters are included in the response.
> >
> > So I think it is pretty clear that the origin filter matches the
> > zipcode nodes in the example, and then the ancestors are included in
> > the reply.
> >
> >
> > > It also says the origin is derived from the origin annotation for that
> > node.
> > > Since only 1 instance of the origin annotation is allowed per node, there
> > > is no way to tag
> > > a node with multiple origins.
> >
> > Agreed.
> >
> >
> > /martin
> >
> >
> > > If implementation is too complex then people will just leave it out (w/ a
> > > deviation).
> > > It is unlikely that the instrumentation knows at any given instant all
> > the
> > > origin values
> > > of all the descendant dynamic data at the instant the <get-data> request
> > is
> > > processed.
> > >
> > >
> > >
> > >
> > > >
> > > > They select a subset of the nodes to include in the response, and must
> > > > also include all ancestor nodes and required list keys to the selected
> > > > nodes, regardless of whether those ancestor/key nodes were also
> > selected by
> > > > the query.
> > > >
> > > >
> > > >
> > >
> > > Yes. Understood.
> > > Still does not explain how a filter for the list node selects descendant
> > > nodes that match the origin filters.
> > >
> > >
> > >
> > > > E.g. a “config false” filter will still return “config true” nodes if
> > they
> > > > are ancestors or list keys to a descendant config false node.  The same
> > > > logic applies for xpath and origin filters as well.
> > > >
> > > >
> > > >
> > >
> > > No they won't.
> > > Where is that text?
> > >
> > >     get-data config=filter=false
> > >
> > > This starts from top-level YANG nodes.
> > > If the top-level YANG node is not config=false then the server will not
> > > keep looking for descendants that match.
> > >
> > >
> > >
> > > > Thanks,
> > > > Rob
> > > >
> > > >
> > >
> > > Andy
> > >
> > >
> > >
> > > >
> > > >
> > > >
> > > >
> > > > *From:* netconf <[e-mail]> *On Behalf Of *Andy Bierman
> > > > *Sent:* 07 October 2019 15:12
> > > > *To:* Martin Bjorklund <[e-mail]>
> > > > *Cc:* Netconf <[e-mail]>
> > > > *Subject:* Re: [netconf] get-data origin filters
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > On Mon, Oct 7, 2019 at 12:43 AM Martin Bjorklund <[e-mail]>
> > wrote:
> > > >
> > > > Andy Bierman <[e-mail]> wrote:
> > > > > On Sun, Oct 6, 2019 at 8:32 AM Martin Bjorklund <[e-mail]>
> > wrote:
> > > > >
> > > > > > Hi,
> > > > > >
> > > > > > Andy Bierman <[e-mail]> wrote:
> > > > > > > Hi,
> > > > > > >
> > > > > > > I am trying to figure out how to use the origin-filter and
> > > > > > > negated-origin-filter
> > > > > > > in the <get-data> operation in RFC 8526.
> > > > > > >
> > > > > > >
> > > > > > >           leaf-list origin-filter {
> > > > > > >              type or:origin-ref;
> > > > > > >              description
> > > > > > >                "Filter based on the 'origin' annotation.  A
> > > > > > >                 configuration node matches the filter if its
> > 'origin'
> > > > > > >                 annotation is derived from or equal to any of the
> > > > given
> > > > > > >                 filter values.";
> > > > > > >            }
> > > > > > >
> > > > > > >
> > > > > > > These filters seem kind of worthless if implemented according to
> > the
> > > > > > text.
> > > > > > > Consider a simple example where there is 1 learned leaf within a
> > > > list:
> > > > > > >
> > > > > > > module: address
> > > > > > >   +--rw addresses
> > > > > > >      +--rw address* [last-name first-name]
> > > > > > >         +--rw last-name     string
> > > > > > >         +--rw first-name    string
> > > > > > >         +--rw street?       string
> > > > > > >         +--rw city?         string
> > > > > > >         +--rw zipcode?      string
> > > > > > >         +--rw phone* [phone-type]
> > > > > > >            +--rw phone-type      enumeration
> > > > > > >            +--rw phone-number    string
> > > > > > >
> > > > > > > Let's say the "zipcode" field is learned in <operational>
> > > > > > > (e.g. ZIP code lookup expands missing or 5 digit zipcode to full
> > 9
> > > > digit
> > > > > > > zipcode).
> > > > > > > So /addresses and /addresses/address have origin "intended".
> > > > > > > Only the /addresses/address/zipcode leaf has origin "learned".
> > > > > > >
> > > > > > > So how does origin-filter=learned find all the learned leafs?
> > > > > >
> > > > > > Perhaps I don't understand your question; IMO you give the answer
> > to
> > > > > > this question below:
> > > > > >
> > > > > > > What filters are required to return only the learned entries +
> > > > ancestors
> > > > > > +
> > > > > > > ancestor-or-self keys?  Seems like this filter mechanism has to
> > be
> > > > used
> > > > > > > to retrieve the exact leaf that might be learned, and the client
> > > > > > > needs to know in advance all the possible nodes that might be
> > > > learned.
> > > > > > >
> > > > > > > Want to be able to retrieve an ancestor that is intended and
> > still
> > > > find
> > > > > > the
> > > > > > > learned entries
> > > > > > >
> > > > > > >    get-data xpath-filter=/addresses/address
> > origin-filtter=learned
> > > > > >
> > > > > > ... here.  So this request will return:
> > > > > >
> > > > > >    <addresses or:origin="or:intended">
> > > > > >      <address>
> > > > > >        <last-name>...</last-name>
> > > > > >        <first-name>...</first-name>
> > > > > >        <zipcode or:origin="or:learned">...</zipcode>
> > > > > >      </address>
> > > > > >      ...
> > > > > >    </addresses>
> > > > > >
> > > > > >
> > > > > I do not interpret the text the same way as you.
> > > >
> > > > Does this mean that you think that the reply is different from what I
> > > > show above?  If so, what would it be, and why?
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > Explain how the list address node has origin "learned".
> > > >
> > > >
> > > >
> > > > The filter is for /addresses/address and only origin=learned.
> > > >
> > > > How does the list node have origin=learned?
> > > >
> > > > It can only have 1 value.
> > > >
> > > > It has child nodes with both intended and learned as origin.
> > > >
> > > > I do no understand how the origin=learned matched this node.
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > >
> > > > >                      The content returned
> > > > >
> > > > >           by get-data must satisfy all filters, i.e., the filter
> > > > >           criteria are logically ANDed.
> > > > >
> > > > >
> > > > >           leaf-list origin-filter {
> > > > >              type or:origin-ref;
> > > > >              description
> > > > >                "Filter based on the 'origin' annotation.  A
> > > > >                 configuration node matches the filter if its 'origin'
> > > > >                 annotation is derived from or equal to any of the
> > given
> > > > >                 filter values.";
> > > > >            }
> > > > >
> > > > >
> > > > >               Configuration nodes that do not have an
> > > > >               'origin' annotation are treated as if they have the
> > > > >               'origin' annotation 'or:unknown'.
> > > > >
> > > > >
> > > > >
> > > > > > The draft shows an example where both "intended" and "system" are
> > given
> > > > > > > as filters.  This will work but will include all the "intended"
> > > > leafs as
> > > > > > > well.
> > > > > > > What if a "learned" node is within a "system" node within an
> > > > "intended"
> > > > > > > node?
> > > > > >
> > > > > > This works as well.  Note that the get-data description says:
> > > > > >
> > > > > >           Any ancestor nodes (including list keys) of nodes
> > selected by
> > > > > >           the filters are included in the response.
> > > > > >
> > > > > >
> > > > > >
> > > > >
> > > > > The issue is how the /iaddresses and /addresses/address nodes match
> > the
> > > > > origin "learned".
> > > >
> > > > They don't, but they are included b/c of the quoted text above (i.e.:
> > > >       Any ancestor nodes (including list keys) of nodes selected by
> > > >       the filters are included in the response.)
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > No.
> > > >
> > > >
> > > >
> > > > If the filter was for /addresses/address/zipcode then maybe that text
> > > > applies.
> > > >
> > > > It is still unclear that the XPath is fully processed and then the
> > > > origin-filter is processed.
> > > >
> > > > The RFC just says they are ANDed together.
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > > The leafs in list "address" are a mixture of "intended" and "learned"
> > > > > origin.
> > > > > The text clearly says that a node has a single origin property,
> > coupled
> > > > to
> > > > > the annotation.
> > > > >
> > > > > Issue 1: mixed origin descendant nodes
> > > > > So how does a search on /addresses/address match
> > origin-filter=learned?
> > > > > I cannot find any text that says what the origin of a list or
> > P-container
> > > > > is if it
> > > > > contains nodes of mixed origin.
> > > >
> > > > See above.
> > > >
> > > >
> > > >
> > > > No text above explains how the list origin is tagged if it has multiple
> > > > types of child nodes.
> > > >
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > > Issue 2: NP-containers
> > > > >
> > > > > Also from RFC 8342:
> > > > >
> > > > >    The origin applies to all configuration nodes except non-presence
> > > > >    containers.
> > > > >
> > > > >
> > > > > What if the top-level node is an NP-container in this case.
> > > > > I thought the top-level node MUST have an origin attribute.
> > > > >
> > > > > The text is not clear how NP-containers are handled.
> > > > > Do they have an origin attribute? If not then RFC 8526 says they have
> > > > > origin "unknown".
> > > > > Is the intent that NP-containers always pass the origin-filter tests
> > > > (test
> > > > > skipped)?
> > > >
> > > > No, since they don't have an origin value they will not be selected by
> > > > the filter.  But an NP-container will be included in the reply if it
> > > > is the ancestor of a node that is selected by the filter.
> > > >
> > > >
> > > >
> > > > The RFC text does not really say that.
> > > >
> > > > Since it is very difficult to know if a data node 5 layers deep is
> > going
> > > > to match,
> > > >
> > > > implementing these filters according to this vague interpretation is
> > > > unlikely.
> > > >
> > > >
> > > >
> > > >
> > > > /martin
> > > >
> > > >
> > > >
> > > > Andy
> > > >
> > > >
> > > >
> > > >
> > > >
> > > > >
> > > > >
> > > > >
> > > > > /martin
> > > > > >
> > > > > >
> > > > > Andy
> > > > >
> > > > >
> > > > > >
> > > > > > > Seems like the client needs to know a lot about the server
> > > > implementation
> > > > > > > details
> > > > > > > in order to use the origin filters.
> > > > > > >
> > > > > > >
> > > > > > > Andy
> > > > > >
> > > >
> > > >
> >