RFC6022 (YANG Module for NETCONF Monitoring)

[SCadilhac]

RFC6022 (YANG Module for NETCONF Monitoring)

[olofhagsand]

No, I have not received a request on this before. Was it any specific sub-part you were interested in, I am interested in the use-case?
Best
--Olof

[SCadilhac]

I'm using a network orchestration tool which (optionally) relies on this RFC to automatically retrieve the supported models from the device, for a smooth integration.

1. The supported Yang models must be advertised as capabilities by the device in the hello message.
2. The orchestrator issues a get > filter > netconf-state > schemas to check the list of available Yang (or Yin, or XSD...) models directly served by the device.
3. Then the orchestrator can issue a get-schema for each schema it needs to download.

[olofhagsand]

Ok, I need to look at that a little more closely to understand how much work this would be. Would you be interestwd in contributing?

[SCadilhac]

Im looking at the Clixon's code in order to evaluate the needed work too...
By the way, it seems that the announced capabilities follow the older form, e.g. urn:ietf:params:xml:ns:netconf:capability:candidate:1:0 instead of urn:ietf:params:netconf:capability:candidate:1.0. Any reason for that?

[olofhagsand]

No reason, apart from following the original RFC.

[olofhagsand]

@SCadilhac clixon will be updated with new announced netconf capability: urn:ietf:params:netconf:capability:yang-library:1.0 in 3.8 (October). Following RFC 7895 and RFC 7950.
Aim for RFC 6022 in 3.9.

[SCadilhac]

Hi @olofhagsand, great work thank you!
I had started coding a part of it too, but had to temporary switch back to another project, sorry I haven't been of more help so far.

[SCadilhac]

Hi @olofhagsand, I had a quick look at the develop branch, and I'm not sure to understand why you removed the basic capabilities, such as urn:ietf:params:netconf:base:1.0. My understanding is that yang-library should be an additional capability, not the single one.

Example:

<hello xmlns="urn:ietf:params:xml:ns:netconf:base:1.0">
  <capabilities>
    <capability>urn:ietf:params:netconf:base:1.1</capability>
    <!-- Add here other capabilities like validate, candidate, etc. -->
    <capability>urn:ietf:params:xml:ns:yang:ietf-netconf-monitoring?module=ietf-netconf-monitoring&revision=2010-10-04</capability><!-- As per RFC 6022 -->
    <capability>urn:ietf:params:netconf:capability:yang-library:1.0?revision=2016-06-21&amp;module-set-id=[id]</capability><!-- As per RFC 7895 and 7950 -->
  </capabilities>
</hello>

Notes:

• It seems that special character escaping is required in Netconf, e.g. & becomes &.
• The namespace from the Yang module must match the xmlns attribute of the root branch modeled by that module in the configuration (RPC response to get-config).

[olofhagsand]

@SCadilhac thanks. Added back the netconf announcements of capabilities in hello. When reading RFC7950 Sec 5.6.4 I interpreted it first as yang module library as replacing the capability announcement in Netconf RFC6241 Sec 8. But it doesnt say that- RFC7895 seems to be an additional way to announce capabilities - or actually modules and their features which seem to be similar to the netconf capability announcement, but not the same. It is a little confusing.
I am glad for your assistance.

[SCadilhac]

Thank you for the very quick fix! I will do some interoperability testing based on the develop branch and will let you know the result.

[SCadilhac]

Hi @olofhagsand,

I've made some testing, and have seen the following issues:

• clixon only understands the legacy ]]>]]>-based framing mechanism (see RFC6242, section 4), so it should only advertises urn:ietf:params:netconf:base:1.0 capability (and not 1.1).
• Single quotes should be accepted around XML parameters such as <blabla xmlns='namespace'>.
• Character escaping like & shouldn't be followed by a space (character escaping but also parsing).
• The revision shouldn't be between quotes in urn:ietf:params:netconf:capability:yang-library:1.0?revision=%s&module-set-id=%s.
• The urn:ietf:params:netconf:capability:writable-running:1.0 capability should be announced.
• The namespace (xmlns) attribute of modules-state is not passed by the Netconf agent.

I've changed the code for a few of these items already, do you want a PR?

[olofhagsand]

Yes please!

[SCadilhac]

PR submitted. For the last issue (namespace attribute of modules-state), I have to dig into the code to understand why xmlns is lost on the way...

[SCadilhac]

xml_merge doesn't seem to preserve xmlns. I guess anyway when printing out the XML back via Netconf following a Yang model, the namespace of each level should be printed (via xmlns attribute) whenever it's different than the parent.

[SCadilhac]

Hi @olofhagsand,

I have patched xml_merge as follows: d858f4c
But the other Netconf operations such as get-config also need to add the right namespace (xmlns) at the roots of the tree. Need a global change here?

Also noticed that conformance-type, although mandatory, is not announced by the current code. See d858f4c.

[olofhagsand]

cherry-picked two of the commits but have some comments on the other, see #45
I know I have been sloppy with namespace handling. Had plans to do that after 3.8 release, along with some basic yang sanity checking.
But preserving xmlns in xml_merge needs to be fixed.

[olofhagsand]

@SCadilhac maybe you can submit individual issues on the remaining errors (not in #45)? Such as:

• legacy ]]>]]>-based framing
• Single quotes should be accepted around XML parameters
  etc, since it is easier to keep track of them that way. I can do it too otherwise.

[SCadilhac]

Hi @olofhagsand,
Thanks for the feedback. I've logged new issues so that each error can be followed in a clear way.

[olofhagsand]

Closing this since it diverged and split into sub-items, eg #50