On 18/08/2015 18:22, Andy Bierman wrote:
On Tue, Aug 18, 2015 at 9:59 AM, Robert Wilton <[email protected]
<mailto:[email protected]>> wrote:
On 11/08/2015 08:38, Ladislav Lhotka wrote:
On 10 Aug 2015, at 22:15, Andy Bierman <[email protected]
<mailto:[email protected]>> wrote:
On Mon, Aug 10, 2015 at 12:34 PM, Acee Lindem (acee)
<[email protected] <mailto:[email protected]>> wrote:
I think there is agreement that there is a problem. The
YANG Routing Design Team is addressing this with
https://www.ietf.org/id/draft-rtgyangdt-rtgwg-device-model-00.txt
(which has evolved from
https://www.ietf.org/id/draft-openconfig-netmod-model-structure-00.txt).
In essence, a place for everything and everything in its
place. However, there are those who feel that can’t
mandate a single model structure for network devices and
we need mechanisms to manage multiple models but allow for
different device structure (e.g.,
https://www.ietf.org/id/draft-bierman-netmod-yang-package-00.txt).
I hope we can agree on an approach in the coming interim
meetings.
Do you expect "everything in its place" to apply to all
other SDOs and
all vendor modules? Or do you expect just the IETF routing
modules
to follow this subtree hierarchy? If the former, does this
mean the
YANG Routing Design Team is the "YANG data placement
authority"
or all YANG modules that ever get written? How long should
it take for all other SDOs and vendors to redo all their
existing
modules to re-root them in their assigned place in the
data tree?
IMO is is not a good idea to rely on rigid data node
placement,
and a single data placement authority. It is better and
use meta-data
built from the YANG modules (or YANG packages) instead.
I think that having fixed paths may end up being too restrictive.
This is how languages like SMIv2 and YANG work.
A conceptual object is given a permanent "home" within the tree of
object identifiers.
Moving data is very expensive, since any clients working with the old data
will break as soon as the data is moved.
I am not convinced the IETF can or should come up with a set of
containers
that covers every possible topic that can be modeled in YANG.
I mostly agree, but having some more structure/advice as to where to
place YANG modules may be helpful. I'm thinking more along the lines of
broad categories rather than precise locations.
E.g. I had suggested that it might be good if the Diffserv QoS YANG
model was located in some slightly more abstract QoS containers/choices
(e.g. for queuing/shaping/policying/etc).
Mainly my reasoning for this was the observation that vendor specific
extensions would probably fit better augmenting generic QoS containers
rather than a DiffServ specific QoS model (e.g. if they were adding L2
specific QoS features which wouldn't sit cleanly on top of a specific L3
QoS model). However, I think that Norman Finn also mentioned a
requirement from Deterministic Networking where it may be necessary to
interact with some abstract QoS parameters without having to know the
precise details of the underlying QoS model.
One other observation is that if we do end up with some more hierarchy
then the "can't augment with mandatory node" issue may become a much
more widespread issue. Today, top level models can introduce mandatory
nodes, but if they were all augmentations of an existing container they
would not be able to (without a P-container).
Even if such a lake could be brought to a boil, I have no confidence
that the hierarchy will be 100% stable. It can certainly
never be 100% complete. If the IETF thinks it's a good idea
to obsolete all YANG RFCs and start over now, who is to say
the IETF won't do that again every few years?
My perception is that YANG models and how to fit them together are still
somewhat in flux. There are lots of drafts defining YANG models were
folks (myself included) are trying to figure out how these models should
all fit together, but relatively few published RFCs. So, personally I
think that there is still a potential opportunity to add some more
structure at this time if the desire is strong enough. But I'm new to
IETF, and hence perhaps this is just my naivety of the IETF processes.
Three years down the line and I doubt that will be the case - there will
be too much inertia, and unless it is very broken it won't get changed.
The reason Linux uses packages to install/update functionality
is because managing 8000 packages is hard enough.
Managing 247,000 individual files would be insane.
The actual location of files is quite configurable and
different
across distros. We could learn something from the last decade
of Linux package management, and try to apply it to YANG.
That’s an interesting idea, could a YANG package also specify,
e.g. that the root node for the package is X/Y/Z? This could
solve some use cases for relocatability, and it would also be
relatively easy to modify all absolute XPath expressions
inside the package.
If someone wants to builds a YANG controller node that is managing
the configuration for a network of devices then wouldn't they want
a particular device's interface configuration to be located
somewhere like /network/device/<device-name>/interfaces/interface?
Ideally, they would be able to use the same YANG definitions that
are defined for /interfaces/ but root them relative to
/network/device/<device-name>/.
Yes -- some of us (like Martin) have pointed this out many times.
The "device" container on an NE does not help at all wrt/
aggregation on a controller. "/device" or "/" work the same for this
purpose.
Having YANG packages being able to control the relative paths of
the imported YANG modules would possibly allow for more
flexibility in how YANG modules fit together, and also give a more
pragmatic way for how these could be changed/upgraded in future if
necessary.
YANG packages provide a way to describe a set of modules.
They do not change the top-level data nodes of any objects.
This would be very complicated and not really worth it.
IMO the "tree of NP containers" should be a resource directory,
meaning it contains links to the real resources. Kind of like an
index in a library. They don't keep the contents of every book in the
index.
They keep the location of every book in the index. The index is stable.
The resource location does not have to be stable.
What would the real resource location for
"/network/device/<device-name>/interfaces/interface" be? If the
concrete location for all interfaces (regardless of device) was the flat
list /interfaces/ then you would get clashes between the names of the
interfaces on different devices. Also, what if someone wanted two
separate lists of interfaces in two separate parts of the resource
hierarchy. Would that be feasible?
Thanks,
Rob
Cheers,
Rob
Andy
Lada
Thanks,
Acee
Andy
From: netmod <[email protected]
<mailto:[email protected]>> on behalf of "Einar
Nilsen-Nygaard (einarnn)" <[email protected]
<mailto:[email protected]>>
Date: Monday, August 10, 2015 at 5:29 AM
To: Jonathan Hansford <[email protected]
<mailto:[email protected]>>
Cc: NETMOD Working Group <[email protected]
<mailto:[email protected]>>
Subject: Re: [netmod] Y34 - root node
As someone sharing responsibilities for guiding a number
of development teams both defining new models and
implementing to some already defined models in this area,
I can only agree with this addition to what I said earlier.
Cheers,
Einar
On Aug 10, 2015, at 9:46 AM, Jonathan Hansford
<[email protected]
<mailto:[email protected]>> wrote:
And it is not just end users who need help to better
understand YANG models and how to use them. For those
still on the edge, looking to finally take the plunge
and use NETCONF/YANG to configure their devices, help
is also required to determine how best to structure
their YANG models, make use of the existing ones, etc.
For those who have "grown up" with the developments in
NETCONF and YANG, much of this is probably second
nature. But coming to it cold (in the sense of
compiling/writing a first set of YANG models for a
device; I've been following the netconf/netmod WGs for
3+ years), it still feels very much like a dark art!
It is not just the individual modules, it is how to
put them together to best manage a device (let alone a
system).
Jonathan
----- Original Message -----
From:
"Einar Nilsen-Nygaard (einarnn)" <[email protected]
<mailto:[email protected]>>
To:
"Andy Bierman" <[email protected]
<mailto:[email protected]>>
Cc:
"NETMOD Working Group" <[email protected]
<mailto:[email protected]>>
Sent:
Sat, 8 Aug 2015 11:10:15 +0000
Subject:
Re: [netmod] Y34 - root node
Andy,
I agree that there is a need for organization of
models, but I don’t have a firm position on
draft-openconfig-netmod-model-structure/draft-rtgyangdt-rtgwg-device-model
or draft-bierman-netmod-yang-package. But we
absolutely need *something* to help end-users of the
models comprehend the overall structure of models,
their relationship and how to use them.
Cheers,
Einar
On Aug 4, 2015, at 5:16 PM, Andy Bierman
<[email protected] <mailto:[email protected]>> wrote:
On Tue, Aug 4, 2015 at 2:34 AM, t.petch
<[email protected] <mailto:[email protected]>> wrote:
----- Original Message -----
From: "Andy Bierman" <[email protected]
<mailto:[email protected]>>
To: "t.petch" <[email protected]
<mailto:[email protected]>>
Sent: Monday, August 03, 2015 5:17 PM
----- Original Message -----
From: "Andy Bierman" <andy@yumaworkscom>
To: "t.petch" <[email protected]
<mailto:[email protected]>>
Sent: Monday, August 03, 2015 4:10 PM
----- Original Message -----
From: "Andy Bierman" [email protected]
<mailto:[email protected]>
Sent: Saturday, August 01, 2015 7:05 PM
On Sat, Aug 1, 2015 at 9:51 AM, Acee Lindem
(acee) <[email protected] <mailto:[email protected]>>
On 8/1/15, 2:51 AM,
[email protected]
<mailto:[email protected]>>
wrote:
Section 1.1 in
https://www.ietf.org/id/draft-openconfig-netmod-model-structure-00.txt
lists the goals of a generic model
structure that will accommodate
most
modern network devices. I guess you don’t
agree that these are
desirable?
The only objection I have to this draft is the
insertion of a
top-level
root
called "device". (Might as well be called
"self").
There are no sibling nodes planned or intended for
this node, so it serves as an extra document root.
<tp>
One aspect of YANG I have never grasped is
what a root means, if
anything.
I understand that it is needed for NETCONF
(filters) and for YANG
(augments, constraints) and so must be
somewhere and must be
relatively
stable, but has it any other significance in
the data model?
As you may recall, I was involved with SMI
first, where the root is
somewhere up in the sky and life only becomes
interesting some six
levels down the hierarchy and that may colour
my thinking.
YANG does a poor job of defining the root for YANG
data nodes.
It is sometimes called a datastore (in the abstract).
Technically, YANG borrows the definition from XPath.
YANG just defines top-level data nodes and the
parent of
these nodes is the document root
There is no protocol or encoding neutral definition,
only an XML-specific definition.
<tp>
Thanks for that.
It seems to me that much of the extensive
discussion on Y34 (all of
which I have read) is as much political as
technical, that is SMI is
hierarchical, top down, perhaps befitting its
origins in ISO, whereas
YANG is bottom up. IETF-like. YANG could have had
a single tree, but
doesn't. So when I read
https://www.ietf.org/id/draft-openconfig-netmod-model-structure-00.txt
I see a plea for a more hierarchical organisation,
and when I read
draft-bjorklund-netmod-openconfig-reply-00
I see a response that says we are not like that.
If so, I doubt that there ever will be a technical
solution.
And I am mindful that when I configure routing in
a (Cisco) router, I
have to do some of it under the interface
definitions and some of it
under the definition of the routing protocol.
Which is life, never
wholy interface-centric and never wholy routing
protocol-centric!
Are you saying the job would be easier if the
path was /device/interfaces/interface instead
of just /interfaces/interface? Are you saying
that /protocols/routing could not also be defined?
Clearly edit-config and copy-config allow both
subtrees to be accessed in the same operation,
so I don't understand your concern.
I have been trying to get the root node to be better
defined
in the protocols that use YANG (i.e., ncx:root, Y34-04).
IMO this is a better approach than defining a YANG module
with a special container that all other modules are
expected
to augment. YANG is designed such that each vendor or SDO
is not dependent on other vendors or SDOs in order to
define data nodes.
<tp>
Andy
I am agreeing with you that adding 'device' brings no
technical benefit,
rather that the motivation is the opposite of
technical (which I
referred to as political). I am also agreeing with the
current declared
consensus on Y34.
And yes, YANG is going to give us a large number of
modules, some
tightly coupled (augments) some loosely so (how many
do you need to
configure OSPF?) and work in this area will be of
benefit now and
probably essential in a few years time. That said, I
am unsure what
such work would be like; I am looking (in despair) at
50 routing area
YANG models and wondering how a user will ever get a
coherent picture of
how to do what they want to do.
The "YANG Land Grab" gives a false sense of progress.
Reaching WG consensus on every single leaf is hard work.
I don't think a collection of 100s of YANG modules is ever
going to to be easy to understand. Operators will not
examine
a NETCONF <hello> message, look at 150 module URIs,
and say "I know exactly what this device supports".
I guess it is up to client tools to do that
I wrote a draft that defines YANG Packages, which can
provide
a higher level of organization for YANG modules.
http://datatracker.ietf.org/doc/draft-bierman-netmod-yang-package/
You and I are apparently in the minority, since the
official status
is that there is no problem with the current approach,
and no need
to organize individual YANG modules into any larger
abstractions.
Tom Petch
Andy
Andy
Andy
The well-specified XPath and YANG root (/) can be
accessed by all protocol operations, exactly
the same
as a node called 'device'. The actual node
name will
depend on the RPC function (e.g. 'data' or
'config').
This is more than redundant however. It
introduces a "super-module"
into YANG that every other module is expected
to augment.
YANG is intended to be more loosely coupled
than that.
This introduces an extra node and namespace
declaration
in all protocol messages, increasing message
sizes.
It also assumes all existing YANG models will
get rewritten to
account for "/device" in all path and XPath
expressions.
This is highly disruptive to existing deployments.
Also, multiple data models to edit the same
thing causes lots
of extra complexity in the server (supporting
both old
location and new location).
IMO a resource directory approach is much more
realistic.
The /device tree can contain all the organized
NP containers.
Instead of all the actual data nodes, this
tree just has pointers
to the real location of the resource. (like
301 Moved Permanently)
Acee
Andy
** Solution Y34-02
Keep 'anyxml'. Introduce 'anydata' as above but
without the
'format' substatement.
'anyxml' would still be used to represent
unrestricted XML, as is
done in NETCONF.
'anydata' would be an unknown chunk of data that
can be modelled
with YANG. Can be encoded as xml or json.
For example:
#+BEGIN_EXAMPLE
anydata data;
#+END_EXAMPLE
** Solution Y34-05
Same as Y34-02 plus two guidelines adopted from
Y34-04:
- Keep 'anyxml' unchanged as it is defined in YANG
1.0. This
ensures backward compatibility.
- Introduce a new 'anydata' statement that
represents an unknown
chunk of data that can be modeled with YANG
- Document that implementations MAY have
restrictions for anyxml and
that anyxml is not necessarily interoperable;
data model writers
should use anydata whenever possible.
- The guidelines document should say that YANG
Doctors will review
each use of anyxml in IETF modules when YANG 1.1
is adopted to
avoid its use whenever possible.
_______________________________________________
netmod mailing list
[email protected] <mailto:[email protected]>
https://www.ietf.org/mailman/listinfo/netmod
_______________________________________________
netmod mailing list
[email protected] <mailto:[email protected]>
https://www.ietf.org/mailman/listinfo/netmod
_______________________________________________
netmod mailing list
[email protected] <mailto:[email protected]>
https://www.ietf.org/mailman/listinfo/netmod
--
Ladislav Lhotka, CZ.NIC Labs
PGP Key ID: E74E8C0C
_______________________________________________
netmod mailing list
[email protected] <mailto:[email protected]>
https://www.ietf.org/mailman/listinfo/netmod
_______________________________________________
netmod mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/netmod
