NETMOD Minutes IETF98 (Chicago)
Version: April 17, 2017
TUESDAY, March 28, 2017
09:00-11:30 Tuesday Morning Session I (2:30 hours)
Audio stream: http://ietf98streaming.dnsalias.net/ietf/ietf988.m3u (in session only)
Num Start Duration Information
0 9:00 20 Title: Intro, Charter & WG Status
- Ladislav Lhotka: Yes - go faster on the revised datastores.
1 9:20 15 Title: YANG Next
Discussion on options:
Eric Voit: What is Yang? There is a lot of tension on the expectations meeting exposure as a device? What is the domain that YANG is modeling?
Kent Watsen: My personal thought is a controller is a kind of "device" (e.g., router), but also many people talk about using yang to model artifacts or standalone data structures that necessary tie to configuration
1)Please go down the rules: YANG should be a modeling language, not tie to any application.
2)Please make some progress for yang next now.
Lou Berger: What about extensions versus whole revision update?
Rob Shakir: Fastest way to get the additional features.
Lou Berger: There is a trade-off between new features and revision.
Ladislav Lhotka: While it may be faster, it is just as easy to bump the version. Otherwise all vendors will define their extensions.
Lou Berger: Did you argue for deprecating extension?
Ladislav Lhotka: Only application that can support the extension are workable. Changes may destroy the validity?
Rob Shakir: The different version is something we need to deal with as code-writers and implementers. I understand the ideas that you are suggesting, but if we have to wait for 2 years, we never find out which these ideas look good to some solution. We should allow testing now, and solve the combine problem.
All of the issues are about when we want something. It would work better to edit. We are in revisions 3 or 4 to our models.
Lou Berger: There are different models that people work on.
Rob Shakir: We should have massive updates. And we had a update 2 years ago that indicate this point.
John Messenger: If you want to do the 2 features, then just do these as features. Do not accept distraction. When you agree upon the features, then the feature documents are rolled into the main specification.
Lou Berger: It is just like code ready for release. Then it is just rolled forward.
Jeff Tantsura: We are going past the level of no-return.
2 9:35 40 Title: Revised Datastores
Presenter: Martin Bjorklund
Andy Bierman: What is the benefit of the operational state datastore?
Martin Bjorklund: The proposal focuses on the characteristics of the data.
Hank Birkhotz: Draft differentiates between state VS dynamic configuration, state and dynamic configuration look the same, but they are not. Pull/align the distinction into the origin definition.
Sue Hares: Did you decide to have different identities (e.g. for I2RS)
Martin Bjorklund: I2RS origins should derive from dynamic. This should be clarified in the text.
Alex Clemm: Query about no changes are required, what about references between datatores?
Martin Bjorklund: It doesn't make sense to have references between datatstores, it is better if the validation is within a datastore. Between which datastore would you need such reference?
Chris Hopps: Interested in the between state. Recommending that we design models in anticipation of this (i.e. combined config/state), this is possible in most models (except for state values that are shadowed). Get models to be the way that we want them to be.
Martin Bjorklund: I agree.
Guangying Zheng: Keep the models the same. Just uses netconf filter as condition.
Martin Bjorklund: Lots of models would augment the current models, so we need to revise the base data models.
Benoit Claise: Commenting about the transition: Can we agree that all models being published have a combined tree.
Martin Bjorklund: We should update the guidelines, which have a section to introduce how to update the split data tree.
Benoit Claise: I agree.
Jeff Tantsura: Propose discussing this in the Routing WG.
Rob Shakir: Cross datastore references are useful in some cases.
E.g, when components are dynamically plugged in, and when statement in the model needs to refer to the operational datastore rather than the configuration.
Martin Bjorklund: We already have this issue in currently published models.
Chris Hopps: It's not a datastore issue. It is just being able to refer to the state.
Rob Shakir: We should like to have a discussion in routing regarding how to lay out models.
Lou Berger: Where do we capture the guidelines?
Rob Shakir: Already gone through making different models consistency.
Martin Bjorklund: 2 issues: (1) how do you structure your data models, (2) changing the recommendation isn't a noop because it changes the current recommendation.
Sue Hares: I2RS recommendations implies cross datastores references.
Rick Taylor: can we document the decision (re: near-term recommendation)
Lou Berger: 3 options (put in 6087bis, do a 6087bisbis, do an update to 6087bis)
Benoit Claise: let's wait until we have a decision first
Kent Watsen: When ready for last call
Martin Bjorklund: Depends on comments from the WG, some edits planned based on comments and per plan. We should prepare a new version soon.
Chris Hopps: Need to address current comments, including cross datastore references
Mehmet Ersue: Do we need restate to a MUST/SHOULD (RFC2119) language.
Kent Watsen: hope to have document in 2-3 months
Kent Watsen: The WG needs to consider the guidelines that go along with this document.
Chris Hopps: 99% of works to have intended/applied relate to the same information. Only specific types that are a problem are the intended and applied are not aligned. We already have this anyone. There is no reason to block the model.
Lou Berger: It requires the publication of the support in the protocol as well as Yang models.
Chris Hopps: Our recommendations here will be viewed by other people. It will not block the data models.
Rob Shakir: 99% of the use case will not use these functions. The open-config do not split at the root, do they apply.
Martin Bjorklund: If you look at the text details of 6087 bis, there is one section talk about why we doing this works.
Rob Shakir: I want to a specific design on the models. Otherwise, the users cannot understand the work. There are not clear that the actual.
How many think they have enough information to make a recommendation now? – Enough number.
How many think we should recommend IETF WGs adopt the YANG model strategy proposed by the revised datastore draft now (if they can)? -- A reasonable number.
How many think we should not? -- Very few.
3 10:15 25 Title: Schema Mount
Presenter: Ladislav Lhotka
Chris Hopps: Not all the models need the mount parent.
Martin Bjorklund: this was support to avoid the need for mounted entire parent module?
Ladislav Lhotka: This only changes the reference to some schema. The data will be available in only one place.
Chris Hopps: I do not use a duplication of data, so much as a namespace issue, like having two pointers in C
Ladislav Lhotka: You can have a “client(netconf/restconf)” that tries to edit this exposed data in one place, and another client try to edit it in different location.
Chris Hopps: Let's leave out the special use casing. [IOW no restrictions to absolute reference only]
Lou Berger: This is the mechanism to miss the special casing.
Chris Hopps: (The slides said) parent schema is only with absolute path.
Lou Berger: What you cannot do, is from a client reference the parent data.
Ladislav Lhotka: I have an example on the next slide, maybe discuss more then.
Alex Clemm: two questions:
1) Peer mount can work for some instances.
2) “uses grouping” may another solutions.
Ladislav Lhotka: We explains this question in the document.
Phil Shafer: 3 questions:
1)Is there only schema data,not instance data?
Ladislav Lhotka: There is no instance data.
Martin Bjorklund: This document only covers the schema data, and does not specify anything about what happens to the instance data.
Lou Berger: There is a document that does what you are talking about.
2) If the reference inside the schema, how it will be satisfied? It is satisfied in the subtree you mounted or in the root?
Ladislav Lhotka: inside mount schema (take it to the list)
Phil Shafer: For example, A reference B, B reference C, where the B, C, be evaluated?
Ladislav Lhotka: These are absolute modules.
Lou Berger: The mount modules are top level modules.
3) Jail, no way to escape the jail
Ladislav Lhotka: Except parent references
Rick: I like this model that gets me out of the jails.
Chris Hopps: Parent has specific interface lists. I heard the schema does not have to refer to each other.
Phil Shafer: Reuse the schema under the mount.
Lou Berger: I think it is exist under this solution.
1) Use extension.
Lou Berger: We do not need the module structure for inline. Without a mount point, it is difficult to reader/implementer to know where to put it.
Ladislav Lhotka: In my mind, it is.
Rick Taylor: Actually this is a complex work - so making inline vs. reference mount clear.
Ladislav Lhotka: I agree.
Lou Berger: In case where you have inline, it is ok. In the case when you do not have inline, this is useful.
Rick Taylor: This is a main point.
Ladislav Lhotka: I do agree to have a place where you can have authentication.
Lou Berger: NI- use case can be inline or this specific reference.
2) Restrictions on parent modules
Chris Hopps: Make clearly, we can pull in reference from top level module.
Lou Berger: We will be holding an interim on the mount. Let's move the topic.
4 10:40 10 Title: draft-ietf-netmod-syslog-model
Presenter: Clyde Wildes
Kent Watsen: The draft is past WG LC, but the -13 changes means the WG will need to put it through another WG LC.
5 11:00 15 Title: draft-openconfig-netmod-model-catalog-02
Presenter: Anees Shaikh
Benoit Claise: In Hackthon, we develop the catalog tool, populating all modules we known.
<AD hat on.> I want to know what modules are supported by vendors.
IETF yang library should be populated in catalog model.
John Messanger: The issue you mentioned that don’t implement whole modules but a little bit. This is a big problem. We need to formalize it, but how?
Rob Shakir: Don’t need to use “if feature” everywhere.
How many interested in adopting this document? --- A good number.
How many think there are issues adopting? --- None.
THURSDAY, March 30, 2017
17:40-18:40 Thursday Afternoon Session III (1 hour)
Audio stream: http://ietf98streaming.dnsalias.net/ietf/ietf986.m3u (In session only)
Audio recording: https://www.ietf.org/audio/ietf98/ietf98-zurichd-20170330-1740.mp3
Num Start Duration Information
0 17:40 5 Title: Intro & WG Status
1 17:45 8 Title: Interface Models update
Presenter: Robert Wilton
<Originally scheduled for Tuesday's session>
Acee Lindem: Re open issue 1 -- good to keep it in.
Acee Lindem: Re Open issue 2 -- wrong to cover in protocols, keep on interface.
Chris Hopps: Seems like a TE thing, should it be in a TE model that would be augmenting it into the interface model.
Pavan Beerem: Already covered in TE interface model via augmentation
Mahesh Jethanandani: Prefers list (Re: vlan model).
2 17:53 8 Title: YANG Tree Diagrams
Presenter: Lou Berger presenting for Martin
<originally scheduled for Tuesday's session>
Jeff Haas: Should have some boilerplate (be careful on how might render in SVG).
Lou Berger: Now that 6087bis is back in the WG, we can add language directing this there.
Rob Shakir: Be very careful, don't make changes unnecessarily, lots of existing documents refer to current definition.
Lou Berger: Speaking as individual contributor, not in favor of making changes inconsistent with past RFCs.
Chris Hopps: There are changes being discussed on list.
Lou Berger: Once a WG doc, the WG can discuss.
Rob Wilton: Like the idea of documenting in one place.
Chris Hopps: Bonus: once it's an RFC you have versioning, references to old definitions refer to the old RFC, no confusion.
Benoit Claise: I pulled back 6087bis in part for this
Kent Watsen: <vote>
If you would like to see this draft adopted, please raise your hand --- A fair (good) number.
If you would not like to see this draft adopted, please raise your hand --- None.
To be confirmed on list.
3 18:01 5 Title: draft-lhotka-netmod-yang-markup-00
18:00 Presenter: Ladislav Lhotka
<originally scheduled for Tuesday's session>
<RE: 1st open issue>
Jeff Haas: Something about restricting something.
Anees sheik: Overall this would be very useful to help generate documents from text.
Rob Wilson: Same comment as Anees, but just pick one markup language
William Lupton: Supports.
Lou Berger: <vote>
How many think this type of feature is good to have? --- Good number.
How many do not? -- 1, sort of.
How many have read? [Less then raised hand on question #1]
4 18:06 8 Title: A YANG Data Model for Configuration Scheduling
Presenter: Xufeng Liu
Kent Watsen (as contributor): There was a previous draft that shared the same problem, but it was less predictable than what you are proposing.
Lou Berger: <vote>
In the last meeting, there was general interest in this draft. How many in the room interested in this draft today? [A few/not many]
Lou Berger: Let’s move discussion to the list.
5 18:14 8 Title: YANG Module Tags
16:16 Presenter: Chris Hopps
Eric Voit: Is there a reason why you’re tagging the whole module as opposed to a more granular level?
Ladislav Lhotka: Seems similar to metadata annotations [it's a form of metadata].
Ladislav Lhotka: Would it be expected to be carried in XML attributes?
Chris Hopps: This is more for clarification, this is give me all your routing modules, and you would keep that offline.
Benoit Claise: Metadata from modules is more and more important, need to combine with catalog effort.
Eric Voit: What is the difference between this and annotations.
Xufeng: This is on module (schema), Lada's (annotation) is on node (data).
Kent Watsen: Let's take it to the list, it's a new document and clearly it needs to be discussed further.
6 18:22 8 Title: draft-bertz-netmod-commonaugment-00.
18:26 Presenter: Lyle Bertz
Rob Shakir: Don't understand the problem being solved
Lyle: We did not realize controllers would treat augmentations differently.
Rob Shakir: This is a general issue.
Lyle: Both a developer and manager, with these optimizations 50% time saving
How many have read the draft? [3-5]
How many understands the problem space? [3-4]
7 18:30 5 Title: (IEEE 802.3cf) YANG Data Model Definitions Task Force
Presenter: Rob Wilton
8 18:35 5 Title: Mounting YANG-Defined Information from Remote Datastores
Presenter: Alex Clemm
Kent Watsen: <vote>
How many people read this document? [Good number]
How many people interested in this work? [Less than first]
9 18:40 0 Title: Yang for I2RS Protocol
Presenter: Susan Hares
<Sue withdrew her presentation>
Adjourn 18:40 0
Note takers (feel free to add your name here): Mahesh Jethanandani, Kent Watsen, Christian Hopps, Zitao Wang